Artigos Doenças Crônicas.pdf
Documento PDF (21.5MB)

Exibir conteúdo Baixar

                    Article

https://doi.org/10.1038/s41467-023-37353-8

Pan-cancer classification of single cells in the
tumour microenvironment
Received: 6 July 2022
Accepted: 10 March 2023

1234567890():,;

1234567890():,;

Check for updates

Ido Nofech-Mozes
Sagi Abelson 1,2,5

1,2

, David Soave1,3, Philip Awadalla

1,2,4,5

&

Single-cell RNA sequencing can reveal valuable insights into cellular heterogeneity within tumour microenvironments (TMEs), paving the way for a deep
understanding of cellular mechanisms contributing to cancer. However, high
heterogeneity among the same cancer types and low transcriptomic variation
in immune cell subsets present challenges for accurate, high-resolution confirmation of cells’ identities. Here we present scATOMIC; a modular annotation
tool for malignant and non-malignant cells. We trained scATOMIC on
>300,000 cancer, immune, and stromal cells defining a pan-cancer reference
across 19 common cancers and employ a hierarchical approach, outperforming current classification methods. We extensively confirm scATOMIC’s accuracy on 225 tumour biopsies encompassing >350,000 cancer
and a variety of TME cells. Lastly, we demonstrate scATOMIC’s practical significance to accurately subset breast cancers into clinically relevant subtypes
and predict tumours’ primary origin across metastatic cancers. Our approach
represents a broadly applicable strategy to analyse multicellular cancer TMEs.

Tumour microenvironments (TMEs) are highly complex. Various
immune and stromal cells within the TME interact with cancer cells to
regulate processes such as angiogenesis, tumour proliferation, invasion, and metastasis, as well as mediate mechanisms of therapeutic
resistance1–4. Single-cell RNA sequencing (scRNA-seq) techniques are
explicitly suitable to disentangle complex systems as they provide
transcriptome information for every cell within a sample, enabling the
study of subtle transcriptomic changes reflecting different cell types
and their functional states5.
Cell-type annotation is arguably the most critical step to derive
biological insight from scRNA-seq experiments and can be performed manually or using automatic classifiers6,7. Manual annotation
is often unfavourable as it is subjective to user definition of nonparametric clustering of cells, conducted under the assumption that
all the cells within a defined cluster are identical, and depends on preexisting knowledge of canonical genes. Although expression of
canonical markers has been used to characterise some cell types,
definitive markers are not always available8. Moreover, due to their

relatively low number and the possibility of incomplete detection
due to technical variation, the sole use of canonical gene expression
is not ideal.
Given these limitations, there has been a shift towards automatic
methods for cell classification, with over 100 classifiers described in a
recent census of available scRNA tools9. To date, most automated
annotators are focused on classification of blood or subsets of cells
from other specialised tissues, thus having limited capabilities in
deciphering complex TMEs across diverse human cancers. Indeed,
using single-cell transcriptomics to predict cancer types and differentiate between cancer and related normal tissue cells while also
classifying the large number of immune cells and stroma, is not a
straightforward task10. In the context of TMEs, cell type predictions are
challenged by high inter-patient tumour cell heterogeneity among
cancers of the same tissue11–13 and low transcriptomic variation among
related, yet different specialised immune cells14. Since cancer samples
tend to cluster by patient11–13 and transcriptional variation is often
driven by genomic instability, existing cell type classification methods

1

Ontario Institute for Cancer Research, Toronto, ON, Canada. 2Department of Molecular Genetics, University of Toronto, Toronto, ON, Canada. 3Department of
Mathematics, Wilfrid Laurier University, Waterloo, ON, Canada. 4Dalla Lana School of Public Health, University of Toronto, Toronto, ON, Canada. 5These
e-mail: philip.awadalla@oicr.on.ca; sabelson@oicr.on.ca
authors jointly supervised this work: Philip Awadalla, Sagi Abelson.

Nature Communications | (2023)14:1615

1

Article
which rely on distance correlations to a reference15–17 are expected
to fail.
The current standard for the identification of malignant cells in
scRNA-seq data relies on copy number variation (CNV) inference
methods18,19. Nevertheless, these methods are incapable of providing
definitive information concerning the cancer’s tissue of origin. Furthermore, CNV inference necessitates the presence of genetically
unstable cells, and its accuracy may suffer when lacking a sizeable,
distinctive normal cell reference within the sequenced specimen.
Solely relying on the presence of inferred CNVs to annotate malignant
cells may lead to false negatives or undefined cells in cancers with
minimal genomic structural variation or nearly diploid genomes. Thus,
a limitation in scRNA-seq analysis of tumour ecosystems is that there is
no universal method for effective, detailed classification of heterogenous non-malignant TME cell types and subtypes, and cancer cells.
Clearly, a fully automated, pan cancer classification scheme that
can easily be updated to capture additional subsets of normal cells and
clinically relevant molecular subtypes of cancer, holds promise
towards a better understanding of cancer ontogenies and the molecular interaction of diverse tumour tissues with their
microenvironments.
In this work, we present single cell annotation of tumour microenvironments in pan-cancer settings (scATOMIC), a comprehensive,
pan cancer, TME cell type classifier. We devise a structured scheme
that uses hierarchically organized models and elimination processes,
reducing the transcriptomic complexity of the TME multi-cellular
system to improve cell classification.

Results
An overview of scATOMIC
We postulated that the sheer number of publicly available single-cell
transcriptomic datasets will enable the development of a highly
accurate and thorough classifier for cancer, blood, and stromal cells.
To define a pan cancer reference, we interrogated cancer patient data,
augmented by two additional comprehensive data sources containing
transcriptomic-independent confirmation of cell identities. These
include scRNA-seq of cancer cell lines representing 19 common cancer
types20 and a CITE-seq dataset (proteomics and transcriptomics) of
diverse peripheral blood cells16. scRNA-seq of stromal cells was gathered from several tumour and normal tissue sources21–28 (Supplementary Data 1, 2). Overall, 301,662 cells were included in the training
reference dataset of scATOMIC.
Obtaining an accurate set of differentiating features is critical to
successful classification. Nonetheless, significant differentially
expressed genes (DEG) concerning non-malignant cell types are often
expressed in other related cells that are functionally distinct14,29,30
(Supplementary Fig. 1). On the other hand, inter-patient heterogeneity
among malignant cells has been repeatedly observed with different
patients forming unique clusters11–13 (Supplementary Fig. 2). To
improve cell identity predictions, we developed a method, termed
reversed hierarchical classification and repetitive elimination of parental nodes (RHC-REP) which reduces the breadth of cell types in an
ensemble of classification tasks. As compared with top-down local
hierarchical methods, here, predictions of terminal classes are
repeatedly being evaluated to infer the cells’ broader parental nodes.
During this process terminal cell classes are investigated iteratively
using multiple sets of refined differentiating features until confident
terminal annotations are achieved.
To develop this approach, we structured a pan cancer TME cellular hierarchy where each parent node represents a group of related
cells, and each terminal node represents a single-cell class of interest.
Overall, we trained 24 random forest models corresponding to the
total number of parent nodes (Fig. 1a). For every model, we selected
DEGs that distinguish each cell type from all other terminal classes
nested within the same parent. RHC-REP will then prioritise the

Nature Communications | (2023)14:1615

https://doi.org/10.1038/s41467-023-37353-8

features with the highest specificity to the interrogated cell
types (Fig. 1b).
During each classification task, every cell receives a vector of
prediction scores (PS) corresponding to the percentage of trees voting
for each terminal class in the parent node (Fig. 1c). This cell by PS
matrix is then used to calculate intermediate group scores (IGS), to
subsequently link cells to their next parental node in the hierarchy
(Fig. 1d, Supplementary Fig. 3). At each classification task, the distribution of IGSs obtained from all the cells interrogated in the model
is used to automatically define prediction cut-offs (Supplementary
Fig. 4). Each cell is then interrogated by its next associated model,
defined by a more discriminative set of features and fewer potential
terminal classes (Fig. 1e). Cells that do not pass the IGS thresholds are
given their previous parent classification and withheld from further
subclassification (Supplementary Note 1).
Given that non-malignant cells that share the cancer’s tissue of
origin can be found in cancer biospecimens (for example, normal
alveolar cells in a lung biopsy) we embedded a cancer signature
scoring and cell differentiating module in scATOMIC. Using an established transcriptional program scoring method11,16, cancer-typespecific up and down-regulated programs31 are evaluated in cells
receiving an original annotation of a cancer type by scATOMIC (Fig. 1f).

Performance evaluation and validation across internal and
external datasets
To evaluate scATOMIC’s performance, we first conducted fivefold
cross validation using the training reference dataset (Supplementary
Data 1) while keeping equal proportions of cell types in each of the five
folds. scATOMIC achieved median F1 scores ranging from 0.90 to 0.99
across all the tested cell types (Fig. 2a), implying great accuracy in
classifying the breadth of cells in the settings of pan cancer TMEs. To
ensure scATOMIC was not heavily impacted by batch effects, we also
trained 4 scATOMIC iterations, where all cells from intact technical
batches were held out from training. We found no significant difference in the performance of the models on the held out batches
(Supplementary Fig. 5a, Kruskal–Wallis rank sum test: H statistic(degrees of freedom = 3) = 6.12, P value = 0.106). We further tested scATOMIC performance across different scRNA platforms using external
melanoma datasets29,32–34 and again, found no significant difference in
F1 scores (Supplementary Fig. 5b, Kruskal–Wallis rank sum test: H
statistic(degrees of freedom = 3) = 2.18, P value = 0.537).
We next aimed to conduct a comprehensive external, trainingindependent validation of scATOMIC performance. To build a validation dataset with high-confidence cell annotations, we mined publicly
available scRNA-seq data from primary tumour biopsies and blood
samples. Overall, the curated set used for validation contained
228,460 cancer, 82,976 stroma and 46,090 blood cells from 225 primary biopsies spanning 13 cancer types (Supplementary Data 3).
Importantly, these ground truth sets include cancer cells supported by
abnormal CNV profiles, and immune cells with transcriptomicindependent identity supported by cell surface protein markers via
CITE-seq. Similar to the results obtained from internal validation, in
this independent validation process, scATOMIC achieved a median
F1 score of 0.99 (Fig. 2b).
Overall, these results demonstrate the broad abilities of scATOMICʼs core algorithm to detect cancer cells and their type, as well as
predicting non-malignant cell types and subtype.

Comparison with other cell-type classification methods
We compared scATOMIC’s performance to six commonly used scRNAseq classifiers (SingleR15, Seurat16,35, SingleCellNet36, scmap-cell17,
CHETAH37, and scType38). These annotators encompass a variety of
methods including reference correlation, label transfer using integration to a reference, random forest algorithm, flat and hierarchical
models, and marker-based classification methods making their
2

Article

https://doi.org/10.1038/s41467-023-37353-8

a

Blood Cell

Stromal Cells
Cancer Cells
Blood Cells
Parent Nodes

Mast
Cell

T/NK
Cell

Any Cell
Non-Blood
Cell

Oligodendrocyte

Enteric Stromal
Glial Cell
Cell

Endothelial
Fibroblast
Cell

Group 3
Cancers

Soft Tissue/
Brain/
Neuroblastoma
Brain/
Neuroblastoma

B
Cell

Group 1
Cancers

ASDC pDC cDC Monocyte Macrophage

Breast/ Endometrial/ Hepatobiliary/
Lung/
Ovarian/
Pancreatic/
Prostate
Kidney
Bladder

Melanoma

Cells within
parent node

CD4+/CD8+ CD8+ T Cell/
NK Cell
T Cell
CD4+ T
Cell

CD8+ T
Cell

NK
Cell

Naïve Tem Tfh
Naïve Tem Ex
MAIT
CD4 CD4 CD4
CD8 CD8 CD8

Colorectal Gastric Esophageal

Cells within
parent node
Filter DEGs
using
DES cutoffs

DEGs

Differential
Gene
Expression

Treg

HSPC

Lung Breast Prostate Ovarian Endometrial Kidney Pancreas Bile Duct Gallbladder Liver Bladder

Cells within
b parent
node

Gene
Expression
low

Dendritic Monocyte/
Cell
Macrophage

Plasma Cell

GI Tract

Bone Sarcoma Brain Neuroblastoma

Genes

Group 2
Cancers

Myeloid Cell

Stable DEGs

Soft
Tissue

B Cell/
Plasma Cell

Cancer Cell

Train
Random
Forest

Unstable
DEG

high

e

c

eg. cell 10

IGS >
cutoff
IGS <
cutoff

Cells

Prediction scores

d

Example IGS derivation:
Any Cell

Non Blood Cell
Endothelial Melanoma Sarcoma …
Cell 1
0.09
0.10
0.07
…
Cell 2
0.08
0.08
0.08
…
Cell 3
0.08
0.07
0.10
…
Cell 4
0.03
0.04
0.05
…
Cell 5
0.02
0.06
0.07
…
Cell 6
0.02
0.26
0.15
…
Cell 7
0.14
0.09
0.21
…
Cell 8
0.22
0.03
0.04
…
Cell 9
0.17
0.08
0.05
…
Cell 10
0.06
0.02
0.02
…
.
.
.
IGS Non Bloodcell 10
.
Cell n
n1
n2
n3
…

IGS Non Bloodcell n =

ΣPS

Prediction Score

Gastric
0.09
0.09
0.12
0.08
0.11
0.10
0.00
0.09
0.02
0.02

B Cell
0.20
0.08
0.02
0.04
0.06
0.06
0.05
0.01
0.16
0.25

n23

n24

pDC
0.01
0.10
0.04
0.17
0.03
0.03
0.04
0.14
0.15
0.15

Blood Cell
NK Cell
0.24
0.04
0.03
0.00
0.04
0.04
0.19
0.23
0.00
0.12

…
…
…
…
…
…
…
…
…
…
…

HSPC
0.09
0.08
0.08
0.03
0.02
0.02
0.14
0.22
0.17
0.15

IGS Bloodcell 10

cell n within
non blood cell
subgroup

n25

n26

IGS Bloodcell n =

…

ΣPS

n35

low

high

Prediction
Scores for
cell 10
Prediction
Scores for
cell n

cell n within
blood cell
subgroup

cell 10

cell 10

IGS Non Blood
across all query cells

Terminal
classification

IGS Blood
across all query cells

f

Broad
classification

Cancer
Normal Tissue

2

2
1

1
Potential Cancer
Blood or Stromal

UpRegSig

scATOMIC annotated
cancer cells

Score cancer specific
transcriptional modules

comparison with scATOMIC’s underlying RHC-REP approach highly
suitable. Each tool was provided with the same reference and external
validation dataset as scATOMIC for training and testing (Supplementary Data 1–3). All classifiers were highly accurate in classifying nonmalignant cells (median F1 > 0.85, Fig. 2c). However, for cancer cells in
particular, F1-scores were significantly lower as compared with scATOMIC (two-sided Wilcoxon rank sum tests, all P values < 2 × 10−16). The

Nature Communications | (2023)14:1615

2
1

DownRegSig

Modify scATOMIC
annotation

next best performing classifier following scATOMIC was SingleR15 with
median F1 scores of 0.92, 0.97 and 0.76, for blood, stromal, and cancer
cells, respectively (Fig. 2c). Using scATOMIC, we obtained median
F1 scores of 0.95 0.99, and 0.99 in each of these respective categories.
As existing cell type classification tools were not designed to annotate
malignant cells, this comparison highlights scATOMIC’s ability to
overcome the complexity presented in pan cancer settings to
3

Article

https://doi.org/10.1038/s41467-023-37353-8

Fig. 1 | Overview of scATOMIC training and classification. a Hierarchical structure of the pan-cancer tumour microenvironment. The cellular hierarchies in the
pan-cancer tumour microenvironment are organized into a flow chart with
increasing cell type resolution. Parent nodes represent broad classification branches, and terminal nodes represent specialised cell classes of interest. b Training of
classification branches for each parent node (n = 24). The reference datasets are
filtered based on transcriptomic-independent information to only include terminal
cell types that are found within a particular parental node. Genes that significantly
differentiate one cell type from all the others are gathered. Differentially expressed
genes (DEGs) with greater specificity to each terminal class, determined by differential expression score (DES), are kept (Methods). A random forest classifier is
trained on filtered, library size normalised count matrices to derive a model that
provides prediction scores corresponding to the proportion of trees voting for
each terminal class within the parental node. Colours on the top of the heatmap

illustrate different cell types. c–f Classification of query datasets. c Gene expression
count matrices from query tumour biopsies are inputted into the first scATOMIC
classification branch model, outputting a cell-by-prediction scores matrix.
d Prediction scores (PS) from all blood and non-blood cell subtypes are respectively
summed to derive intermediate group score (IGS) distributions associating single
cells with their appropriate parental class. e Cells are iteratively interrogated at
their next parent nodes’ corresponding models until terminal classification are
obtained. Broad classifications occur if the IGS for a cell is lower than the confidence cut-off. In this example, cell 10 is subclassified until a terminal B cell designation is derived. f Differentiating between cancer and tissue-specific nonmalignant cells through scoring of bulk RNA-seq derived differentiating gene
expression programs (Methods). scATOMIC automatically annotates population 2
as cancer cells, and population 1 as non-malignant. Heatmaps and cell illustrations
were created with BioRender.com.

accurately identify cancer cells while also having comparable or significantly better performance in annotating stroma and blood. Lastly,
with the exception of CHETAH37, mostly comparable time and memory
usages were measured for all the methods (Supplementary Fig. 6).

of “epithelial cells” (Supplementary Fig. 9a), suggesting tissue-specific
cell classes that are not represented in the current scATOMIC training
reference. Using scATOMIC’s automatic approach to set confidence
IGS cut-offs (Supplementary Fig. 4) these cells were abstained from
being falsely annotated and correctly assigned with the lower-level
annotation of non-blood cells. scATOMIC resolved the remaining
epithelial cells into lung cancer and normal tissue cells by evaluating
lung cancer associated transcriptional signatures (Fig. 1f).
Increased cellular resolution across the cell types of the TME was
also observed in other recent datasets of different cancer types
including bladder4, breast40, liver41, ovarian40, prostate42, and skin
cancer33 (Fig. 4b–g). In addition, scATOMIC identified hematopoietic
stem/progenitor cells (HSPCs) in glioblastoma43 (Supplementary
Fig. 10); a population which was shown to promote tumour cell
proliferation43.
Collectively, this analysis demonstrates the ability of scATOMIC’s
core hierarchical algorithm to resolve cell identities at high resolution,
label fine grained T cell states, identify rare cell types, abstain from
falsely classifying unknown cells, and determine cancer types.

Distinguishing between non-malignant, tissue-specific cells and
cancer
Aneuploid CNV profiles are highly associated with the development
and progression of numerous cancers by impacting gene expression
levels39. We assessed scATOMIC’s ability to distinguish malignant cells
from other normal cells of the TME, sharing the same cell-of-origin, by
comparing scATOMIC’s final cancer predictions (Fig. 1f) to their
inferred CNV-based ploidy status19. We observed strong agreement in
cells predicted as malignant and aneuploid-inferred CNV profiles
across biopsies, as well as between non-malignant detected cells and
diploid inferred profiles, with a median agreement rate of 85.9%
(Fig. 3). In addition, in silico serial dilution analysis of cancer cells in our
external dataset collection revealed that decreasing number of
malignant cells can be appropriately annotated, with scATOMIC also
providing cancer type notations (Supplementary Fig. 7). Discordant
cases, defined as cases with an agreement rate below the 1st quartile
(Q1 ≤ 63.2%), were typically attributed to tumours with low CNV levels,
intra-tumoural malignant subclones, low number of cancer cells harbouring the CNV, or low number of reference cells (Supplementary
Fig. 8). In only 7% of discordant cases more cells were inferred as
aneuploid than cells annotated as malignant by scATOMIC (Fig. 3b).
These results suggest that cancer and related normal tissue cells are
efficiently classified by their transcriptomic profiles using scRNA-seq
data, independent of their ploidy status.

scATOMIC annotations increase cellular resolution in tumour
biopsies
To further demonstrate the benefits of scATOMIC in annotating multicellular TMEs, we analysed several datasets, including scRNA-seq of
lung cancer29. Original annotations for this dataset were determined by
the authors using SingleR15 with its default references in combination
with cell type signatures and the use of canonical marker genes. Similar
to our observations (Supplementary Fig. 1) Slyper et al.29, noted overlapping expression programs between T cells and NK cells which
makes high resolution single-cell discrimination among these cell
types challenging. scATOMIC resolved NK cells and T cells, and further
subclassified the latter into fine grained subtypes including T regulatory cells, naive CD4 + T cells, CD4 + T follicular helper cells, effector/memory CD4+, effector/memory CD8 + T cells, and exhausted
CD8 + T cells (Fig. 4a). In addition, in this lung dataset, scATOMIC
identified other distinct cell types including plasma cells, and plasmacytoid dendritic cells (pDCs) which scATOMIC separated from B
cells. Unsupervised clustering and expression of cell specific markers34
supported the existence of these separated cell identities (Supplementary Fig. 9). Of note, this biopsy included two more small clusters

Nature Communications | (2023)14:1615

Extending the core scATOMIC hierarchy for novel applications
By leveraging RHC-REP, one can easily deploy new scRNA-seq data to
train extensions at any terminal branch of the hierarchy. We thought
that extending the breast cancer classification node would provide a
practical example of utilising modularity (Fig. 5a). Two sizable scRNAseq breast cancer atlases were used to train, and independently test
(Supplementary Data 4, 5) a classification model that resolves breast
cancer cells into the major ER + , HER2 + and triple negative breast
cancer (TN) histological subtypes. We applied scATOMIC to the
training-independent validation dataset containing 38 tumours spanning ER + , HER2 + , and TN breast cancer, and 2 HER2 + /ER + double
positive tumours, a class not represented in the current reference of
scATOMIC’s breast mode due to a lack of data. scATOMIC correctly
subtyped 37 of the 38 (97.4%) training-independent breast cancer
biopsies, as determined by immune-staining21,44 (Fig. 5b, Supplementary Data 5). In the two HER2 + /ER + double positive samples, scATOMIC assigned mixed annotations of HER2 + and ER + cells (Fig. 5b).
We observed different degrees of tumour cellularity, with 6
biopsies (15%) having more predicted normal breast cells than cancer
cells. In another tumour reported as ERlow (that is, <10% ER + cancer
cells by immunostaining), scATOMIC identified 8% ER + breast cancer
cells (Fig. 5c, Supplementary Data 5). Of note, scATOMIC identified
these ER + cells as malignant, in line with the histology report, however
CNV inference predicted a diploid profile (Fig. 5d). This example
highlights a distinct subpopulation of cancer cells that could have been
misinterpreted as normal tissue by strictly relying on CNV inference,
thus suggesting integrative approach for best results.
Overall, these data demonstrate scATOMIC’s practical and modular framework to further subset primary tumour classes into their
clinically relevant subtypes.
4

Article

https://doi.org/10.1038/s41467-023-37353-8

a

Blood

Cancer

Stromal

F1 Score

1.00

0.95

0.90

CD B C
4
e
CD + T C ll
8+
e
T C ll
ell
cD
Ma HS C
cro PC
p
Ma hage
st
Na
C
tur Mon ell
o
al
kill cyte
er
Ce
ll
Pl a
sm pDC
aC
ell
Bi l
eD
u
Bla ct
dd
e
Bo r
ne
Bra
B in
Co reas
En lorec t
do
ta
Es met l
op ria
Ga hage l
llbl
a
ad l
d
Ga er
st
Ki d r i c
ne
y
Liv
Ne
e
uro
Lu r
bla ng
sto
O v ma
Pa aria
nc
re n
Pr atic
os
Sa tate
rco
ma
En
Sk
do
in
the
l
Fib ial Ce
rob ll
l
Ol
igo Gli asts
de al C
nd
roc ell
yte
s

0.00

0.00

Cell Type

b

Blood

Cancer

Stromal

1.00

F1 Score

0.75
0.50
0.25

8+

CD

CD

4+

BC

ell

TC
el
TC l
ell
Ma
c
cro DC
ph
Na
Ma age
tur
al st Ce
kill
er ll
Ce
ll
Pla
pD
C
sm
aC
ell
Bla
dd
er
Bra
i
Br n
Co east
lor
ec
ta
Kid l
ne
y
Liv
e
r
Ne
L
uro
bla ung
sto
m
Ov a
Pa arian
nc
rea
Pr tic
os
Sa tate
rco
ma
En
Sk
do
in
the
lia
lC
F
Ol
igo ibrob ell
de
l
nd asts
roc
yte
s

0.00

Cell Type

c

Blood

Cancer

ns *

*

*

ns *

***
***
***
***
***
***

1.00

Stromal
ns

***
**
ns***
**

F1 Score

0.75
0.50
0.25

IC
gle
R
Sin Seur
a
gle
Ce t
llN
e
sc
t
ma
pC
e
ll
CH
ET
AH
sc
Typ
e
sc
AT
OM
IC
Sin
gle
R
Sin Seur
a
gle
Ce t
llN
et
sc
ma
pC
CH ell
ET
AH
sc
Typ
e
Sin

e

AT
OM

Typ

sc

ET
AH

CH

sc

ell

t
sc

ma

pC

t

llN
e

ura

Ce

Se

R
gle

Sin

gle
Sin

sc

AT
OM

IC

0.00

Annotation Tool

scATOMIC identifies the tumour of origin across metastatic
cancers
Given that existing single-cell annotation tools are not designed to
provide information regarding the originating tissue of a cancer cell,
we applied scATOMIC to predict the tumour origin in settings where it
may be unknown. We curated a dataset of 62 metastatic biopsies from
breast, kidney, lung, ovarian, and skin cancers from diverse anatomical

Nature Communications | (2023)14:1615

sites (Supplementary Data 6). In 52 of the 62 samples (83.9%), the
primary tissue of origin was correctly predicted by scATOMIC (Fig. 6),
demonstrating its robustness at distant sites, in cells that may have
undergone transcriptional changes associated with metastasis. In 1
kidney and 2 lung samples (additional 4.9%) scATOMIC abstained from
giving a terminal classification yet focused the prediction on the correct intermediate class. In 2 lower throughput melanoma scRNA-seq,
5

Article

https://doi.org/10.1038/s41467-023-37353-8

Fig. 2 | scATOMIC performs accurately in internal and external validation
experiments. a k-fold cross validation. The reference dataset was randomly split
into 5 sub-samples containing equal numbers of each cell type. F1 scores are shown
for each cell type in each of the 5 replicates (jitter points). Each fold contained
overall ~61,100 cells. Boxplot colours represent the major cell type classes.
b External validation in datasets not used for training. scATOMIC was validated on
CITE-seq datasets of tumour derived blood cells, datasets of aneuploid cancer cells
and stromal cells from primary tumour biopsies. F1 scores are shown for each cell
type within individual samples (jitter points). The plot represents n = 357,526 cells
from 225 samples. Red dots indicate low-confidence cell type classifications
(Methods). Boxplot colours represent the major cell type classes. c scATOMIC

outperforms other existing automatic cell type annotators, particularly when
applied to identify cancer cells and determine their type. Six existing classifiers
were provided the same training/reference and training-independent validation
datasets as scATOMIC. Combined F1 scores for each of the three major cell class,
blood, cancer, and stroma are shown (jitter points). The plot represents n = 337,790
cells from 221 samples that were given a classification output by all tools. (two-sided
Wilcoxon rank sum test comparing scATOMIC to each tool *P < 0.05, **P < 1.1 × 10−6,
***P < 2 × 10−16, are shown). Boxplot colours represent the different tools. For all
plots, boxes and whiskers represent the lower fence, first quartile (Q1), median
(Q2), third quartile (Q3), and upper fence. Source data are provided as a Source
Data file.

only 5 and 6 cancer cells were reported11 yet scATOMIC found none.
We considered these as false predictions. In 4 of the 5 remaining
samples that received incorrect terminal classifications, the predicted
cancer type and the reported primary were related cancers falling
under the same immediate parent node. For example, a mixed serous/
clear-cell ovarian carcinoma was predicted to be endometrial cancer,
with relatively low classification scores (Supplementary Fig. 11). Overall, these results show that accurate detection of metastatic cancers’
tissue of origin using single-cell transcriptomics is feasible and that
scATOMIC can aid in identifying cancers’ primary sites across a variety
of solid human tumours.

As molecular classification of cancers by tissue-of-origin is fundamental to diagnostic pathology we demonstrated scATOMIC’s
ability and high accuracy in predicting the primary origin of metastatic
tumours. Additional work is required to evaluate the limits of singlecell transcriptomics to predict cancer origin, specifically in cancers of
unknown primary and other contexts where distinguishing primary
from metastatic tumours is not trivial, such as in the case of mucinous
ovarian carcinoma47.
In summary, we have described, benchmarked, and validated a
highly accurate single-cell annotation tool across TMEs of common,
deadly cancer types. The RHC-REP classification approach underlying
scATOMIC used here to tackle the complexity associated with multicellular TMEs might be of interest in areas outside the cancer field
entailing multi-class structured systems. We highlight the benefits that
scATOMIC holds in cancer settings compared to other tools, providing
a method to standardise single-cell cancer transcriptomic studies. We
expect that scATOMIC’s abilities to accurately identify TME resident
cells with high resolution, separate between cancer and normal tissue
cells, and determine tumours’ origin will enrich and expedite broad
cancer studies seeking to refine prognostication or cell–cell communication from single-cell transcriptomes.

Discussion
The rate of scRNA-seq publications reporting major scientific insights
concerning the function of various immune and stromal cells in cancer
has increased steadily over the years40,45,46. However, the lack of
automated methods that can also standardise the identification of
single malignant cells is becoming a major obstacle to accurately study
tumour-microenvironment interactions across various cancer types.
We developed scATOMIC to effectively annotate the TME in pancancer settings. scATOMIC overcomes several classification challenges, including high inter-patient heterogeneity and highly overlapping expression profiling among specificized immune cells. By
using stably expressed transcripts as features, structured classification, and models trained using reliable and large datasets, scATOMIC
has proven to accurately identify cancer cells and their origin. Moreover, scATOMIC is comparable to or outperforms other existing
automatic cell type annotators when classifying blood and stromal
cells using our training reference. In samples with genome instability
and an appropriate reference of normal cells, we found high agreement between scATOMIC and CNV inference to pinpoint malignant
cells in scRNA-seq data. However, in samples with cancer sub-clones
defined by variable CNV burdens, cancer cells with near diploid
genomic profiles, or few normal cells to serve as controls, CNV inference may fall short. Since information concerning CNV may be useful
for cancer prognostication, and a degree of discordancy still exists,
using scATOMIC in conjunction with CNV inference to annotate cancer
cells and their type is recommended.
We designed the core, ploidy-neutral scATOMIC algorithm to
accurately identify cancer and normal tissue cells across 19 common
cancer types, including key rare populations such as plasmacytoid
dendritic cell and hematopoietic stem and progenitor cells that were
reported in cancer tissues and are associated with immunosuppressive
phenotypes29,43. To ensure that scATOMIC remains powerful, we
designed it in a way that new data can be easily interrogated to extend
the core hierarchy by adding new terminal cell classes. We demonstrated this modularity by further classifying breast cancers into their
clinically relevant molecular subtypes achieving high agreement
between transcriptomics and immunostaining. With the progressive
accumulation of high quality publicly available scRNA-seq data, future
extensions of the core hierarchy to further subclassify the other 18
cancer types and the various core, non-malignant cell types to their
more resolved classes or states will become simple.

Nature Communications | (2023)14:1615

Methods
Defining the pan-cancer tumour microenvironment cell type
hierarchy
We defined a structured hierarchy where cell types with transcriptomic similarities are grouped into nodes (Fig. 1a). We first
grouped blood cells based on existing relationships that correspond
to the hematopoietic hierarchy48, and kept stromal cells together. For
cancer cells, we derived putative groups where transcriptomic similarities are expected based on the cancers’ shared organ system,
histological subtype, hormonal tissue, or germ layer. These included
carcinomas of the digestive system (group 1: colorectal, gastric,
esophageal, liver, gallbladder, bile duct, and pancreas), carcinomas
not of the digestive system (group 2: lung, breast, prostate, endometrial, and ovarian), and non-carcinomas including soft tissue,
neuroendocrine, and nervous system cancers (group 3 cancers:
bone, sarcoma, brain, melanoma, and neuroblastoma). We then used
a subset of our data, corresponding to one patient sample from each
cancer type to evaluate random forest models. Evaluating the proportion of trees voting for each cancer type helped refine the groups
and provided guidelines concerning what the subsequent nodes
might be.
For example, for the less trivial grouping of kidney cancer, a
classification model including all cancers assigned 42.3% of tree votes
for kidney, while the majority of the remaining trees voted for various
group 2 cancers thus, suggesting linking kidney cancer with group 2. In
the other case of lung cancer, we decided to include it in both groups 2
and 3, as separate cancers from both these groups obtained a high
number of trees voting for lung. In this case, no clear distinction was
observed. In a different case concerning CD8 + T cells, we included
CD8 + T cells in both child nodes of the parent node T/NK as different
CD8 + T cell populations show more transcriptomic similarities to NK
6

Article

https://doi.org/10.1038/s41467-023-37353-8

UMAP 2

a

UMAP 1

CopyKAT prediction:
Aneuploid
Diploid
NA

5000

8000

30

29

28

27

26

25

24

23

22

21

20

19

18

16

15

14

17

8

13

6

9

12

11

10

7

1

4000

4000

6000

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46

12

9

11

10

6

8

7

5

3

2

1

8

6

7

0.0

5

0.5

0.0

4

0.5

0.0
3

0.5

0.0
1

0.5
2

1.0

4

Colorectal Cancer

1.0

3

Breast Cancer

1.0

2

Brain Cancer

Bladder Cancer
1.0

4

b

5

scATOMIC cancer signature:
Cancer
Normal Tissue Specific Cell
Blood or Stromal Cell

2000

3000

3000
2000

112
2
334
4
55
676
78
89
910
10
11
11
12
12
13
13
14
14
15
15
16
16
17
17
18
18
19
19
20
20
21
21
22
22
23
23
24
24
25
25
26
26
27
27
28
28
29
29
30
30
10

11

10

11

9

9

8

8

7

5

8

10

11

12

13

14

10

11

12

13

14

7

7

9

6

6

9

5

5

8

4

4

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17

15000

4

0.0
3

0.5

0.0

3

0.5

0.0

1

0.5

0.0

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

0.5

3

1.0

2

1.0

1

1.0

2

Prostate Cancer

1.0

2

Pancreatic Cancer

6

1

8

7

6

5

4

3

1

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18

Ovarian Cancer

Neuroblastoma

7

0
6

0

5

1000

0

4

2000

2000

4

4000

100

3

3000

3

6000

2

1

8000

400

2

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25

8

6

7

5

0.0

500

200

0

0.5

g

4000

300

1000

1 12
2
3
3
4 45
5
6 67
7
8
8
9109
10
11
11
121213
13
141415
15
161617
17
181819
19
202120
21
22
22
23
23
242524
25
26
26
272728
28
29
29
303130
31
32
32
33
33
343534
35
363637
37
38
38
394039
40
41
41
42
42
434344
44
45
45
4646

12

9

11

10

8

7

5

6

4

0.0

Melanoma
1.0

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25

2000

4

0.5

0.0
3

0.5
2

y

1.0

2

0.0

Lung Cancer

1.0

1

0.5

0

Liver Cancer

Kidney Cancer
1.0

3

2

1

8

7

6

5

0

3

0
4

0
2

1000

1

1000

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18

Agreement
Rate

1000

2000

3000
3000

Number
of Cells

2000

1

4000

2000
6000

7500
1500

10000

5000

4000

1000

0

0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

0
4

0

3

2000

2

500
1

2500

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17

5000

Patient ID

Annotation:
scATOMIC cancer
CopyKAT aneuploid

Fig. 3 | scATOMIC effectively distinguishes between malignant cells and normal
tissue specific cells. a scATOMIC predictions and inferred ploidy in breast cancer
patient CID406621. Cells are coloured by scATOMIC predictions and copy number
variation (CNV)-based inferred ploidy. scATOMIC-predicted malignant cells are
inferred as aneuploid cells while normal tissue cells are inferred as diploid.
b Comparison of scATOMIC cancer predictions and inferred ploidy statues across
the training-independent, external validation datasets. Blue bars represent the

Nature Communications | (2023)14:1615

scATOMIC normal
CopyKAT diploid

number of cells predicted as malignant (solid blue) and non-malignant (transparent
blue) by scATOMIC. Red bars represent the number of cells inferred as aneuploid
(solid red) and diploid (transparent red). Green bars represent agreement rate in
each biopsy. Rates do not include cells without a confident ploidy status (that is
received an “NA” annotation by CopyKAT). Source data are provided as a Source
Data file.

7

Article

https://doi.org/10.1038/s41467-023-37353-8

a

Endothelial Cells
Cancer Associated Fibroblasts
Lung Cancer Cells
Non Blood Cells
Normal Tissue Cells
Blood Cells

Endothelial Cells

Natural killer Cells

Fibroblast

T or NK Cells
CD8 T or NK Cells
Exhausted CD8+ T Cells

Epithelial Cells

Effector/Memory CD8+ T Cells
CD8+ T Cells
CD4 or CD8 T Cells
Tfh/Th1 helper CD4+ T Cells

T Cells
Mast Cells

T regulatory Cells
Effector/Memory CD4+ T Cells

Macrophage

CD4+ T Cells
Mast Cells

B Cells

Macrophage or Dendritic Cells

Original
Annotations

cDC
Macrophage
pDC
Plasma Cells
B Cells
Any Cells

scATOMIC Annotations

b

Endothelial Cells

c

Endothelial Cells
Stromal Cells
Cancer Associated Fibroblasts

Stromal Cells

Bladder Cancer Cells
Endothelial Cells
Non Blood Cells
Normal Tissue Cells

iCAFs

Endothelial Cells
Fibroblast

Exhausted CD8+ T Cells

Blood Cells
Natural killer Cells
CD8 T or NK Cells
Exhausted CD8+ T Cells
Effector/Memory CD8+ T Cells
CD8+ T Cells
CD4 or CD8 T Cells
Tfh/Th1 helper CD4+ T Cells
T regulatory Cells
Effector/Memory CD4+ T Cells
Naive CD4+ T Cells
Macrophage or Dendritic Cells

Cancer Cells

Effector/Memory CD8+ T Cells

Mast Cells
Myeloid

B Cells

Stromal Cells

T Cells

Ovarian Cancer Cells

Endothelial Cells

Fibroblast

CD8+ T Cells

Cancer Cells

CD4 or CD8 T Cells

Mast Cells
Dendritic Cells

Tfh/Th1 helper CD4+ T Cells

Myeloid

Effector/Memory CD4+ T Cells

B Cells

Cancer Associated Fibroblasts

Non Blood Cells
Normal Tissue Cells
Blood Cells
Natural killer Cells
CD8 T or NK Cells
MAIT Cells

myCAFs

T Cells

Endothelial Cells

Breast Cancer Cells

Cancer Associated Fibroblasts

Epithelial Cells

d

T Cells

T regulatory Cells

Myeloid

Naive CD4+ T Cells
CD4+ T Cells

B Cells

Mast Cells
Macrophage or Dendritic Cells
cDC

cDC
Macrophage

e

Any Cells
Stromal Cells
Cancer Associated Fibroblasts

f

CD8+ T Cells
CD4 or CD8 T Cells
T regulatory Cells
Effector/Memory CD4+ T Cells
Naive CD4+ T Cells
CD4+ T Cells
Mast Cells
Macrophage or Dendritic Cells
cDC

pDC

Macrophage

B Cells
Any Cells

Breast

Effector/Memory CD8+ T Cells

Macrophage

Plasma Cells

Plasma Cells

Bladder

Non Blood Cells
Normal Tissue Cells
Blood Cells
Natural killer Cells
T or NK Cells
CD8 T or NK Cells
MAIT Cells

Ovarian

pDC
B Cells
Any Cells

g
Unknown

Endothelial Cells
Endothelial Cells

Skin Cancer Cells

Prostate Cancer Cells

Endothelial Cells

Fibroblast

Stromal Cells

Cancer Cells
Non Blood Cells
Luminal

Cancer Associated Fibroblasts

Normal Tissue Cells
Blood Cells

Normal Tissue Cells

CAF

Natural killer Cells
Exhausted CD8+ T Cells

Blood Cells

Mast Cells
Monocytic

Liver Cancer Cells

CD8+ T Cells

Effector/Memory CD8+ T Cells
T Cells

Exhausted CD8+ T Cells
CD8+ T Cells

Liver Cancer Cells
Non Blood Cells

CD4 or CD8 T Cells

CD4+ T Cells

Effector/Memory CD8+ T Cells

T regulatory Cells

T Cells

Tfh/Th1 helper CD4+ T Cells

Macrophage

Effector/Memory CD4+ T Cells

Macrophage

Macrophage

B Cells

T regulatory Cells

Effector/Memory CD4+ T Cells
CD4+ T Cells

B Cells

Macrophage

Prostate

Any Cells

Skin

Fig. 4 | scATOMIC provides greater cellular resolution than original annotations across tumour datasets. a Sankey plot comparing original cell type annotations to higher resolution scATOMIC annotations in a recent lung cancer biopsy
dataset29. scATOMIC identifies lung cancer as the tissue of origin and distinguishes
these cells from normal lung tissue cells. scATOMIC identifies subtypes of blood

Nature Communications | (2023)14:1615

T Cells

Any Cells

Liver

Blood Cells
Natural killer Cells
Effector/Memory CD8+ T Cells
Effector/Memory CD4+ T Cells
Macrophage or Dendritic Cells
cDC
Macrophage

cells. b–g scATOMIC identifies the tumour origin of common cancers and deliver
relatively higher resolution in other cell types4,33,40–42. Colours represent the original
reported annotations associated with each dataset. The height of each block
represents the relative number of cells that received a respective annotation.
Source data are provided as a Source Data file.

8

Article

https://doi.org/10.1038/s41467-023-37353-8

a

ER+ Breast
Cancer
Non-Blood
Cell

Any Cell

Cancer
Cell

Breast/
Lung/
Prostate

Group 2
Cancers

HER2+ Breast
Cancer

Breast

Triple Negative
Breast Cancer

Core scATOMIC hierarchy

Histology: ER+ HER2-

Histology: HER2+ ER-

8 patients 11 patients 6 patients 1 patient

Histology: HER2+ ER+

scATOMIC Predictions:

Histology: ER low

ER+ Breast Cancer Cell
HER2+ Breast Cancer Cell
Triple Negative Breast Cancer Cell
Normal Breast Tissue Cell
Unclassified Breast Cancer Cell

1 patient

Relative
Expression

Chr20
Chr21
Chr22

Chr18

Chr19

Chr17

Chr15

Chr16

Chr12

Chr13
Chr14

Chr11

Chr10

Chr9

Chr7

Chr5

Chr6

Chr3

UMAP 2

Chr4

0.9 1.0 1.1

d

Chr1

c

1 patient 3 patients 1 patient 1 patient

1 patient

1 patient

1 patient

Histology: Triple Negative

Chr2

1 patient

2 patients 1 patient

Chr8

b

New branch

UMAP 1

Fig. 5 | Extending the core scATOMIC model to further classify breast cancer
subtypes. a The terminal breast cancer cell node from the core hierarchy of scATOMIC is extended to subclassify breast cancers into their major ER+, HER2+, and
triple negative histological subtypes. b Validation of scATOMIC predictions in an
external cohort44. Pie charts reflecting intra-tumoural breast subtype heterogeneity
according to scATOMIC classification are shown for each reported histological
subtype. Patient specimens with similar distributions of cell annotations are illustrated together in a single pie chart. c Breast cells from an ER-low tumour (Patient:

ER-AH0319) are visualised on UMAP and coloured by scATOMIC predictions. ER +
breast cancer cells represent a sub-clonal cancer cell population. d Inferred copy
number variation (CNV) profiles of cells from ER-low tumour. Red represents
inferred gains, while blue represents inferred losses of genomic regions. The y axis
is coloured according to scATOMIC prediction. Colours representing scATOMIC
predictions apply to all the panels in the figure. Source data are provided as a
Source Data file.

(i.e. cytotoxic T cells) while others resemble CD4 + T cells more (Supplementary Fig. 1). We found that this structure yields stronger performance as compared to only having CD8+ cells in one of the
branches or generating a single model to differentiate between CD4+,
CD8+ and NK all at once. Overall, given that a small random data subset
was utilised to infer transcriptomic similarities among classes of cells,
it is possible that other hierarchical structures might also be
appropriate.

counts) or with more than 25% of their reads being mapped to
mitochondrial genes.
To find DEGs between each terminal cell type and all other
terminal cell types present in the same parent node we used the
‘Seurat’ R package v4.0.116. Raw gene by cell count matrices were
normalised and variance stabilised using the SCTransform function to
remove technical variability. Principle components were found using
the RunPCA function, on the “SCT” assay. Louvain clustering was
performed by first applying the FindNeighbors function on the top 50
PCs, followed by the FindClusters function with a resolution of 2. The
identity of the resulting cell clusters was determined by the
transcriptomic-independent ground truth associated with the training
datasets. For each model, DEGs were found using the FindMarkers
function (two-sided Wilcoxon rank sum test) with ident.1 set to include
all clusters containing a particular terminal cell type and ident.2 being
all other cells in the parent node. The function returned a list of DEGs
per class that passed default Seurat filtering settings: a log2 foldchange of at least 0.25, and at least 10% of the cells in ident.1 or ident.2

Feature selection
For every classification branch within the core scATOMIC hierarchy we
selected features as a training input of a random forest model. Raw
gene by cell count matrices derived from scRNA-seq analysis were
gathered from multiple sources (Supplementary Data 1) and were
organized into 24 parent groups (Supplementary Fig. 3). To merge
matrices into a particular parent dataset, we removed genes that are
not represented in all of the data sources. In each parent dataset we
removed cells with <500 expressed genes (as defined by non-zero

Nature Communications | (2023)14:1615

9

Article

https://doi.org/10.1038/s41467-023-37353-8

Breast Cancer

Kidney Cancer

Lung Cancer

True Tumour scATOMIC
Origin
Prediction

True Tumour scATOMIC
Origin
Prediction

True Tumour scATOMIC
Origin
Prediction

Ovarian Cancer

Skin Cancer

Classification Classification
Accuracy
Confidence
Correct

Confident

Incorrect

Low

True Tumour of Origin
Breast
Clear Cell Renal
Papillary Renal
Lung Adenocarcinoma
High Grade Serous Ovarian
Mixed Serous/Clear Cell Ovarian
Melanoma

scATOMIC Prediction

True Tumour scATOMIC
Origin
Prediction

True Tumour scATOMIC
Origin
Prediction

Breast
Lung
Kidney
Endometrial
Pancreatic
Ovarian

Skin
No Cancer
Brain
Ovarian/Endometrial/Kidney
Breast/Lung/Prostate

Fig. 6 | scATOMIC accurately identifies the tissue of origin in metastatic tumour
biopsies. scATOMIC was applied to 62 metastatic tumours from breast, kidney,
lung, ovarian and skin. Metastatic sites included the brain, lungs, GI tract, liver,
adrenal glands, lymph nodes, abdomen, and peritoneal cavity. Each pair of dots
represents the true tumour origin and the predicted origin. Horizontal connected
lines represent correct predictions, while diagonal lines represent incorrect

predictions. True tumour origins are coloured by the reported cancer subtype.
Circular points represent confident annotations, while triangular points represent
low-confidence annotations (Methods). Multi-coloured points represent tumours
that received an intermediate scATOMIC annotation. Source data are provided as a
Source Data file.

expressing the respective gene. We defined a differential expression
score (DES) as the difference of the fraction of cells expressing a nonzero value for a respective DEG in ident.1 and ident.2 (DES =
pct_expr_ident.1 – pct_expr_ident.2) to capture DEGs more specifically
expressed in any particular cell type. For each terminal cell type, we
kept genes with a DES greater than the mean DES of all DEGs for that
cell type. We removed all ribosomal genes. We also removed DEGs that
had a pct_expr_ident.2 >40% to ensure high performance when interrogating datasets with large technical variation. For the same reason,
we set a minimum and maximum number of DEGs for each cell type at
50 and 200. Specifically, a minimum number of features was set to
mitigate potential issues in classifying cells with high levels of technical
dropout. In the case where there are fewer than 50 DEGs with DES
higher than the mean, we kept the top 50 DEGs ranked by DES. Features that were used for each cell class at each classification layer are
provided in Supplementary Data 7.

associated with imbalanced classification towards majority classes49,
we randomly sampled an equal number of cells from each terminal
class, with replacement. Library size of each single cell was normalised
by using the library.size.normalise function from the ‘Rmagic’ v2.0.3
package50. Prior to training each model, normalised counts were filtered to include selected features and cells within the corresponding
parental node. Read count values were transformed to a fraction of the
total filtered counts. A random forest classifier was trained on the
transformed matrix using the ‘randomForest’ R package v4.6–14 with
500 trees and default parameters51. Each random forest was trained to
classify the terminal nodes present within the corresponding parental
node. The specific cell type organization of the 24 classifiers is detailed
in Supplementary Fig. 3.

Random forest modeling
For each classification branch within the core scATOMIC hierarchy, we
trained a random forest model on cells within a respective parental
node and features selected as described above. To minimise bias

Nature Communications | (2023)14:1615

Applying scATOMIC to query datasets
Before using scATOMIC on query datasets, the interrogated data was
processed as follows. Raw gene by cell count matrices were filtered to
remove cells with non-zero counts for <500 genes or with more than
25% of their reads being mapped to mitochondrial genes. We imputed
missing values in DEG using the magic function from the ‘Rmagic’
package50, where all the cells within the query dataset act as a
10

Article
reference. In datasets where there were no reported values across all
the samples for specific selected features, we assigned a value of zero
before imputation. Following each classification task, every cell
received a vector of prediction scores corresponding to the percentage of trees voting for each terminal class in the random forest model.
The values in each vector within the next immediate parent node were
then summed to generate IGSs. For example, when the first model
interrogating all the terminal classes in the hierarchy ran (i.e. the parent model “Any Cell”), the output for each cell was composed of two
intermediate group scores. The first corresponded to the sum of trees
voting to all the terminal cell classes belonging to the “Blood Cell”
parent node and the second IGS for the “Non-Blood Cells” parent node
(Fig. 1d). Data from all the cells in the interrogated sample was used to
derive IGS parent distributions. Cells which received an IGS greater
than the defined parent threshold continued down the classification
hierarchy until they were terminally classified (Fig. 1e). At any stage, if a
cell received an IGS lower than the calculated threshold it was annotated based on the previous parental node (a less specific classification). An IGS threshold for a classification to be deemed confident was
automatically determined (Supplementary Fig. 4). Using the ajus
function from the ‘agrmt’ package v1.42.452, IGS distributions for each
IGS calculated among all cells within a parental node are classified as
either unimodal or bimodal. Unimodal distributions suggest a layer
includes one subtype, while bimodal distributions indicate there is
likely more than one subtype. For unimodal distributions we set the
IGS threshold to be the mean IGS (µ) – 3 standard deviations (σ). For
bimodal distributions, using the em function from the ‘Cutoff’ R
package53 v0.1.0, we fit a mixture model for the distributions and
predicted estimates of mean and standard deviations for both distributions using the expectation maximation algorithm. We selected a
conservative approach when a mixed cell type population exists in a
layer by setting the IGS threshold in bimodal distributions to be the
mean of the higher score modality (µ2) – 2 standard deviations (σ2). We
set the maximum IGS threshold to be 0.7, as in some distributions,
such as when a single pure population remained for classification,
unreasonably high thresholds may be obtained.
A schematic and description of the main functions is detailed in
Supplementary Note 2.

Flagging cells with lower confidence annotations
To provide a way for scATOMIC users to evaluate the confidence of
their cell annotation output, we devised a secondary post-classification
flag to warn about low-confidence annotations. To define lowconfidence annotations, we used the results obtained by external
validation (Fig. 2b). For every model throughout the hierarchy, we
determined the median IGSs (or PS for terminal nodes) across samples
for correctly annotated cells (X) and incorrectly annotated cells (Y).
Correct versus incorrect status was defined by the terminal annotation
of single cells. For example, in terminally annotated breast cancer cells,
median IGSnon-blood, IGSnon-stromal, IGSgroup2-cancer, IGSbreast/lung/prostate,
PSbreast for all the correctly and incorrectly annotated cells in the
validation data were recorded. We derived confidence thresholds
based on the overlap between the distributions of X and Y using their
quartiles (Q). When there was low overlap (defined as minimum X > Q3
Y), the threshold was set to the minimum X. When there was intermediate overlap (defined as minimum X < Q3 Y, yet Q1 X > Q3 Y), the
threshold was set to Q3 Y. In all remaining cases where there was high
overlap (defined as Q1 X < Q3 Y), segregation could not be made and
the classifications of query cells in such cases were deemed confident.
For example, all CD4 + T cells that are misclassified as CD8 + T cells will
still obtain comparable IGSblood with respect to correctly classified
CD8 + T cells, both being subtypes of blood cells. If at any model
throughout the hierarchy a low-confident IGS is observed, a flag is
applied. In addition, we assigned a sample-level confidence metric
derived from the proportion of cells receiving a confident annotation

Nature Communications | (2023)14:1615

https://doi.org/10.1038/s41467-023-37353-8

based on this flag. To maximise identification of potentially poorquality samples, in any new interrogated sample, if <75% of cells
receive confident annotations, a warning will be issued (Supplementary Fig. 12).

Scoring cancer signatures to refine cancer cell predictions and
identity of normal tissue-specific cells
To identify potential normal tissue specific cells that are not defined in
the core scATOMIC TME hierarchy, we employed a post classification
method for scoring cancer specific modules in each cell. Lists of genes
differentially expressed between different cancer types and their
matched normal tissues were obtained from OncoDB31. We selected
DEGs from OncoDB31 with a reported log2 fold change >1 or <−1 and an
adjusted P value <0.01. Since OncoDB31 is based on bulk RNA-seq, we
further filtered the DEG list to only include those with reported
expression values in the query scRNA-seq dataset. Upregulated gene
programs and downregulated gene programs31 were scored using the
AddModuleScore function from Seurat11,16 in each cell predicted as
cancer by random forests. Ward.D2 hierarchical clustering was then
performed on a Euclidean distance matrix of each cell’s upregulated
and downregulated cancer programs. Two groups were derived using
the cutree function. At this stage, scATOMIC evaluates the percentage
of normal cells in each group corresponding to those cells annotated
as either blood, stroma or cancer cells with lower upregulated program
scores compared to downregulated program scores. As the AddModuleScore function uses average expression of control feature sets
across all cells in the dataset, the calculated scores are affected by the
proportion of normal cells present. Thus, we filtered out all confident
normal cells in the cluster with a greater percentage of normal cells
and repeated the AddModuleScore pipeline to identify additional
normal cells that were overlooked in the first iteration. We repeated
scoring of cancer programs, hierarchical clustering, calculating the
percentage of normal cells and filtering until both clusters contained
no more than 20% normal cells. Cells that were initially classified as
cancer which were scored as normal cells were given a normal tissue
cell label.

Benchmarking scATOMIC
We validated scATOMIC’s performance using internal cross validation
and external validation datasets. Internal validation was performed by
splitting the training dataset (Supplementary Data 1) into five subsets
containing equal proportions of each cell type. For each iteration we
used 4 subsets as scATOMIC training dataset and applied scATOMIC to
the held out independent test subset. F1 scores were calculated for
each terminal cell type at each iteration (Supplementary Data 1). AXL+
dendritic cells (ASDC) represent a rare, recently discovered, transitional state between cDCs and pDCs54. Due to their small numbers in
the training data, these cells were omitted from the internal validation
procedure.
For external validation of scATOMIC we used 424,534 cancer,
blood and stromal cells gathered from various sources (Supplementary Data 3). For ground truth, we relied on the authors’ annotations of
stromal cells. To improve reliability, cells that were annotated as cancer by the authors were subjected to additional validation by CNV
inference using CopyKAT19. For blood cells, we used CITE-seq data with
protein surface markers supporting the author derived cell type
annotations. We excluded cell types from individual samples if their
number was <30. Per sample F1 scores were calculated for each
terminal cell type. During both scATOMIC’s internal and external
validation processes (Fig. 2a, b), we considered correct intermediate
classifications as true positives.

Comparison between scATOMIC and other tools
We compared scATOMIC’s performance to existing scRNA-seq classifiers. In this comparison only, we bypassed the use of IGS cut-offs in
11

Article
scATOMIC to enable comparison to other tools that cannot output
intermediate cell classes. Of note, a separate analysis evaluating scATOMIC’s performance under this forced mode against its default
(unforced mode) favoured the latter by indicating that the use of
intermediate annotations more frequently avoids derivation of incorrect terminal annotations than restricting the output of correct terminal classes. The same training and validation datasets provided to
scATOMIC were also used to provide a comparison of scATOMIC with
the performance of all the other tested tools. A SingleCellNet36 random
forest model was trained on a balanced reference of 2500 random
samples of each cell type, using default parameters. SingleCellNet
expects a class-balanced matrix as input. We provided the same classbalanced matrix used in scATOMIC’s first classification model, representing all the cell classes. CHETAH37 is using an alternative hierarchical
classification approach. We used CHETAH with its default settings with
the exception of the ‘thresh’ parameter that was set to zero to enforce
terminal annotations, similar to scATOMIC. Scmap-cell17 classification
was performed using default parameters. To enable SingleR processing of the large reference dataset, we applied its pseudobulk implementation by setting ‘aggr.ref’ to TRUE15. For the comparison with
Seurat16,35, we partitioned the pan-cancer TME training reference
according to batch and applied the reciprocal PCA workflow. We
transferred labels to query samples using the TransferData function.
For the comparison with scType38, we provided scType with the list of
features corresponding to each terminal class in scATOMIC’s reference
which were derived in the first classification node (Supplementary
Data 7). Otherwise, default parameters were used. To compare scATOMIC’s final cancer cell prediction with the CNV inference approach
of detecting malignant cells we used CopyKAT19 with its default settings (Fig. 3). Both aneuploid and diploid cells from the external validation biopsies were included in this analysis. Agreement rate was
defined as the simple percentage agreement using the agree function
from the irr55 R package. Cells which received an NA ploidy annotation
were omitted from the calculation.

Analysis of run time and memory usage
We applied each tool without parallel processing, as only some tools
(scATOMIC, Seurat, SingleR) provide that functionality. We considered
the time and memory usage for query datasets following model
training as some methods bypass this step and rely on a given cell
marker list (scType). Using the peakRAM56 R package we monitored
the time for classification and maximum RAM used. We compared the
time and RAM required for the classification of cell types prior to
cancer signature scoring by scATOMIC to SingleR, Seurat, scmap-cell,
SingleCellNet, CHETAH, and scType final classifications. As appropriate, we compared the time and memory for the cancer signature
scoring step that differentiates cancer from normal tissue cells to
CopyKAT.

In silico dilution assay
For each primary tumour sample in the external validation, we ran
scATOMIC and CopyKAT on all non-malignant cells while iteratively
reducing the number of malignant cells. Specifically, we sampled
10–100 malignant cells in increments of 10. Only cells that both scATOMIC annotated as cancer cells and CopyKAT inferred as aneuploid
were sampled.

https://doi.org/10.1038/s41467-023-37353-8

Cell Line Encyclopedia DepMap portal57 and the second, primary
tumour data from Wu et al.21 with immunohistochemistry information
(Supplementary Data 4). We used the clinical molecular subtypes of
HER2 + , ER + and triple negative as classes. There was not sufficient
data available to train and test HER2 + /ER + and HER2-/ER + as separated classes. We evaluated this extension’s performance on an
external set of 40 tumours from Pal et al.44 (Supplementary Data 5).
One tumour containing fewer than 100 cancer cells was omitted from
this analysis. The ground truth of ER and HER2 status in the testing set
was received by correspondence with the authors.

Visualising inferred CNVs
To visualise the inferred CNV profile in the ER-low tumour we used the
‘inferCNV’58 package with the cutoff variable set to 0.1 and all other
variables set to default. We defined normal reference cells as cells
annotated by scATOMIC as blood or stromal cells.

Applying scATOMIC to metastatic tumours
scATOMIC was applied to predict the tumour of origin in 62 metastatic
tumours. Individual samples used are described in detail in Supplementary Data 6. We defined the scATOMIC tumour of origin prediction
by taking the cancer type called in the majority of cancer cells in the
sample.

Reporting summary
Further information on research design is available in the Nature
Portfolio Reporting Summary linked to this article.

Data availability
All the scRNA-seq data used in this work are publicly available. Datasets
retrieved from the Gene Expression Omnibus can be downloaded using
the following accession numbers: GSE16437816, GSE11825724,
GSE11437425, GSE13589327, GSE14081929, GSE14867319, GSE17607821,
GSE13246522, GSE12544941, GSE13190723, GSE11597833, GSE13780459,
GSE14144542, GSE15482660, GSE12313934, GSE16152944. Datasets
retrieved from the Broad Institute Single Cell Portal are available
with the following accession numbers: SCP54220, SCP128861, SCP141532.
The remaining datasets were downloaded directly from links provided
in their corresponding publications: Madissoon et al.26 [https://data.
humancellatlas.org/explore/projects/c4077b3c-5c98-4d26-a614246d12c2e5d7], Chen et al.4 [https://static-content.springer.com/esm/
art%3A10.1038%2Fs41467-020-18916-5/MediaObjects/41467_2020_
18916_MOESM2_ESM.zip], Couturier et al.62 [https://datahub-262-c54.p.
genap.ca/GBM_paper_data/GBM_cellranger_matrix.tar.gz], Qian et al.40
[https://lambrechtslab.sites.vib.be/en/pan-cancer-blueprint-tumourmicroenvironment-0], Young et al.63 [https://www.science.org/doi/
suppl/10.1126/science.aat1699/suppl_file/aat1699_datas1.gz.zip], Peng
et al.64 [https://zenodo.org/record/3969339], Zheng et al.45 [https://
zenodo.org/record/5461803]. Additional descriptions of these datasets are provided in Supplementary Data 8. All re-processed data used
for training and validation have been deposited in Zenodo and are
available through the following link: https://doi.org/10.5281/zenodo.
741923665. Bulk RNA sequencing cancer specific signatures were
obtained from OncoDB31 [https://oncodb.org/data_download.html].
Subtypes of cancer cell lines were derived from the Cancer Cell Line
Encyclopedia in DepMap57 [https://depmap.org/portal/ccle/]. Source
data are provided with this paper.

Breast cancer subclassification
We extended the core scATOMIC model to subclassify breast cancer
cells into their histological subtypes. The model extending breast
cancer into its molecular subtypes was trained using a combination of
two datasets. The first included the breast cancer cell lines data within
the core training dataset (Supplementary Data 1, 2), where the assigned
molecular subtypes were based on annotations found in the Cancer

Nature Communications | (2023)14:1615

Code availability
The scATOMIC R package, associated code and user manual are
available at the abelson-lab/scATOMIC GitHub repository: https://
github.com/abelson-lab/scATOMIC66. Additional scripts to reproduce
the figures in the manuscript are deposited in Zenodo and are available
through the following link: https://doi.org/10.5281/zenodo.741923665.

12

Article

References
1.

Karaayvaz, M. et al. Unravelling subclonal heterogeneity and
aggressive disease states in TNBC through single-cell RNA-seq. Nat.
Commun. 9, 1–10 (2018).
2. Maynard, A. et al. Therapy-Induced Evolution of Human Lung
Cancer Revealed by Single-Cell RNA Sequencing. Cell 182,
1232–1251.e22 (2020).
3. Sade-Feldman, M. et al. Defining T Cell States Associated with
Response to Checkpoint Immunotherapy in Melanoma. Cell 175,
998–1013.e20 (2018).
4. Chen, Z. et al. Single-cell RNA sequencing highlights the role of
inflammatory cancer-associated fibroblasts in bladder urothelial
carcinoma. Nat. Commun. 11, 1–12 (2020).
5. Valdes-Mora, F. et al. Single-cell transcriptomics in cancer immunobiology: The future of precision oncology. Front. Immunol. 9,
2582 (2018).
6. Luecken, M. D. & Theis, F. J. Current best practices in single‐cell
RNA‐seq analysis: a tutorial. Mol. Syst. Biol. https://doi.org/10.
15252/msb.20188746 (2019).
7. Clarke, Z. A. et al. Tutorial: guidelines for annotating single-cell
transcriptomic maps using automated and manual methods. Nat.
Protoc. 16, 2749–2764 (2021).
8. Xie, B., Jiang, Q., Mora, A. & Li, X. Automatic cell type identification
methods for single-cell RNA sequencing. Comput. Struct. Biotechnol. J. 19, 5874–5887 (2021).
9. Zappia, L. & Theis, F. J. Over 1000 tools reveal trends in the singlecell RNA-seq analysis landscape. Genome Biol. 22, 301 (2021).
10. Fan, J., Slowikowski, K. & Zhang, F. Single-cell transcriptomics in
cancer: computational challenges and opportunities. Exp. Mol.
Med. 52, 1452–1465 (2020).
11. Tirosh, I. et al. Dissecting the multicellular ecosystem of metastatic
melanoma by single-cell RNA-seq. Science 352, 189–196 (2016).
12. Puram, S. V. et al. Single-Cell Transcriptomic Analysis of Primary
and Metastatic Tumor Ecosystems in Head and Neck Cancer. Cell
171, 1611–1624.e24 (2017).
13. Vázquez-García, I. et al. Ovarian cancer mutational processes drive
site-specific immune evasion. Nature 612, 1–9 (2022).
14. Domínguez Conde, C. et al. Cross-tissue immune cell analysis reveals
tissue-specific features in humans. Science 376, eabl5197 (2022).
15. Aran, D. et al. Reference-based analysis of lung single-cell
sequencing reveals a transitional profibrotic macrophage. Nat.
Immunol. 20, 163–172 (2019).
16. Hao, Y. et al. Integrated analysis of multimodal single-cell data. Cell
184, 3573–3587.e29 (2021).
17. Kiselev, V. Y., Yiu, A. & Hemberg, M. Scmap: Projection of single-cell
RNA-seq data across data sets. Nat. Methods https://doi.org/10.
1038/nmeth.4644 (2018).
18. Patel, A. P. et al. Single-cell RNA-seq highlights intratumoral heterogeneity in primary glioblastoma. Science 344, 1396–1401 (2014).
19. Gao, R. et al. Delineating copy number and clonal substructure in
human tumors from single-cell transcriptomes. Nat. Biotechnol.
1–10 https://doi.org/10.1038/s41587-020-00795-2 (2021).
20. Kinker, G. S. et al. Pan-cancer single-cell RNA-seq identifies recurring programs of cellular heterogeneity. Nat. Genet. https://doi.org/
10.1038/s41588-020-00726-6 (2020).
21. Wu, S. Z. et al. A single-cell and spatially resolved atlas of human
breast cancers. Nat. Genet. 53, 1334–1347 (2021).
22. Lee, H. O. et al. Lineage-dependent gene expression programs
influence the immune landscape of colorectal cancer. Nat. Genet.
52, 594–603 (2020).
23. Kim, N. et al. Single-cell RNA sequencing demonstrates the molecular and cellular reprogramming of metastatic lung adenocarcinoma. Nat. Commun. 11, 1–15 (2020).
24. Jäkel, S. et al. Altered human oligodendrocyte heterogeneity in
multiple sclerosis. Nature 566, 543–547 (2019).

Nature Communications | (2023)14:1615

https://doi.org/10.1038/s41467-023-37353-8
25. Kinchen, J. et al. Structural Remodeling of the Human Colonic
Mesenchyme in Inflammatory Bowel Disease. Cell 175,
372–386.e17 (2018).
26. Madissoon, E. et al. scRNA-seq assessment of the human lung,
spleen, and esophagus tissue stability after cold preservation.
Genome Biol. 21, 1 (2019).
27. Habermann, A. C. et al. Single-cell RNA sequencing reveals profibrotic roles of distinct epithelial and mesenchymal lineages in
pulmonary fibrosis. Sci. Adv. 6, eaba1972 (2020).
28. Regev, A. et al. The human cell atlas. Elife https://doi.org/10.7554/
eLife.27041 (2017).
29. Slyper, M. et al. A single-cell and single-nucleus RNA-Seq toolbox
for fresh and frozen human tumors. Nat. Med. 26, 792–802 (2020).
30. Kiselev, V. Y., Andrews, T. S. & Hemberg, M. Challenges in unsupervised clustering of single-cell RNA-seq data. Nat. Rev. Genet.
2018 205 20, 273–282 (2019).
31. Tang, G., Cho, M. & Wang, X. OncoDB: an interactive online database for analysis of gene expression and viral infection in cancer.
Nucleic Acids Res. 50, D1334–D1339 (2022).
32. Wu, S. Z. et al. Cryopreservation of human cancers conserves
tumour heterogeneity for single-cell multi-omics analysis. Genome
Med. 13, 1–17 (2021).
33. Jerby-Arnon, L. et al. A Cancer Cell Program Promotes T Cell
Exclusion and Resistance to Checkpoint Blockade. Cell 175,
984–997.e24 (2018).
34. Li, H. et al. Dysfunctional CD8 T Cells Form a Proliferative, Dynamically Regulated Compartment within Human Melanoma. Cell
176, 775–789.e18 (2019).
35. Stuart, T. et al. Comprehensive Integration of Single-Cell Data. Cell
https://doi.org/10.1016/j.cell.2019.05.031 (2019).
36. Tan, Y. & Cahan, P. SingleCellNet: A Computational Tool to Classify
Single Cell RNA-Seq Data Across Platforms and Across Species.
Cell Syst. 9, 207–213.e2 (2019).
37. de Kanter, J. K., Lijnzaad, P., Candelli, T., Margaritis, T. & Holstege, F.
C. P. CHETAH: a selective, hierarchical cell type identification
method for single-cell RNA sequencing. Nucleic Acids Res. 47,
e95 (2019).
38. Ianevski, A., Giri, A. K. & Aittokallio, T. Fully-automated and ultra-fast
cell-type identification using specific marker combinations from
single-cell transcriptomic data. Nat. Commun. 13, 1–10 (2022).
39. Shao, X. et al. Copy number variation is highly correlated with differential gene expression: a pan-cancer study. BMC Med. Genet.
20, 1–14 (2019).
40. Qian, J. et al. A pan-cancer blueprint of the heterogeneous tumor
microenvironment revealed by single-cell profiling. Cell Res. 30,
745–762 (2020).
41. Ma, L. et al. Tumor Cell Biodiversity Drives Microenvironmental
Reprogramming in Liver Cancer. Cancer Cell 36,
418–430.e6 (2019).
42. Chen, S. et al. Single-cell analysis reveals transcriptomic remodellings in distinct cell types that contribute to human prostate
cancer progression. Nat. Cell Biol. 23, 87–98 (2021).
43. Lu, I. N. et al. Tumor-associated hematopoietic stem and progenitor
cells positively linked to glioblastoma progression. Nat. Commun.
12, 1–16 (2021).
44. Pal, B. et al. A single-cell RNA expression atlas of normal, preneoplastic and tumorigenic states in the human breast. EMBO J. 40,
e107333 (2021).
45. Zheng, L. et al. Pan-cancer single-cell landscape of tumorinfiltrating T cells. Science 374, abe6474 (2021).
46. Cheng, S. et al. A pan-cancer single-cell transcriptional atlas of
tumor infiltrating myeloid cells. Cell 184, 792–809.e23 (2021).
47. Dundr, P. et al. Primary mucinous ovarian tumors vs. ovarian
metastases from gastrointestinal tract, pancreas and biliary tree: a
review of current problematics. Diagn. Pathol. 16, 1–17 (2021).
13

Article
48. Doulatov, S., Notta, F., Laurenti, E. & Dick, J. E. Hematopoiesis: A
Human Perspective. Cell Stem Cell 10, 120–136 (2012).
49. Liu, X. Y., Wu, J. & Zhou, Z. H. Exploratory undersampling for classimbalance learning. IEEE Trans. Syst. Man. Cybern. B. Cybern. 39,
539–550 (2009).
50. van Dijk, D. et al. Recovering Gene Interactions from Single-Cell
Data Using Data Diffusion. Cell 174, 716–729.e27 (2018).
51. Liaw, A. & Wiener, M. Classification and Regression by randomForest. R. N. 2, 18–22 (2002).
52. Ruedin, D. agrmt: Calculate Concentration and Dispersion in
Ordered Rating Scales. R package version 1.42.8. https://CRAN.Rproject.org/package=agrmt (2021).
53. Trang, N. V. et al. Determination of cut-off cycle threshold values in
routine RT-PCR assays to assist differential diagnosis of norovirus in
children hospitalized for acute gastroenteritis. Epidemiol. Infect.
143, 3292–3299 (2015).
54. Villani, A. C. et al. Single-cell RNA-seq reveals new types of human
blood dendritic cells, monocytes, and progenitors. Science 356,
eaah4573 (2017).
55. Gamer, M., Lemon, J. & Singh, I. F. P. irr: Various Coefficients of
Interrater Reliability and Agreement. R package version 0.84.1.
https://CRAN.R-project.org/package=irr (2019).
56. Quinn, T. peakRAM: Monitor the Total and Peak RAM Used by an
Expression or Function. R package version 1.0.3. http://github.com/
tpq/peakRAM (2022).
57. Ghandi, M. et al. Next-generation characterization of the Cancer
Cell Line Encyclopedia. Nature 569, 503–508 (2019).
58. Tickle, T. I., Georgescu, C., Brown, M. & Haas, B. inferCNV of the
Trinity CTAT Project. https://github.com/broadinstitute/
inferCNV (2019).
59. Dong, R. et al. Single-Cell Characterization of Malignant Phenotypes and Developmental Trajectories of Adrenal Neuroblastoma.
Cancer Cell 38, 716–733.e6 (2020).
60. Leader, A. M. et al. Single-cell analysis of human non-small cell lung
cancer lesions refines tumor classification and patient stratification.
Cancer Cell 39, 1594–1609.e12 (2021).
61. Bi, K. et al. Tumor and immune reprogramming during immunotherapy in advanced renal cell carcinoma. Cancer Cell 39,
649–661.e5 (2021).
62. Couturier, C. P. et al. Single-cell RNA-seq reveals that glioblastoma
recapitulates a normal neurodevelopmental hierarchy. Nat. Commun. 2020 111 11, 1–19 (2020).
63. Young, M. D. et al. Single-cell transcriptomes from human kidneys
reveal the cellular identity of renal tumors. Science 361,
594–599 (2018).
64. Peng, J. et al. Single-cell RNA-seq highlights intra-tumoral heterogeneity and malignant progression in pancreatic ductal adenocarcinoma. Cell Res 29, 725–738 (2019). 2019 299.
65. Nofech-Mozes, I., Soave, D., Awadalla, P. & Abelson, S. Data and
Codes for Pan-cancer classification of single cells in the tumour
microenvironment. https://doi.org/10.5281/zenodo.7419236 (2022).
66. Nofech-Mozes, I., Soave, D., Awadalla, P. & Abelson, S. abelson-lab/
scATOMIC: scATOMIC v1.1.0. https://doi.org/10.5281/zenodo.
7689011 (2023).

Nature Communications | (2023)14:1615

https://doi.org/10.1038/s41467-023-37353-8

Acknowledgements
This work was supported by a grant from The Banting Research Foundation (Discovery Award #2021-1418) and Investigator Awards received
from the Ontario Institute for Cancer Research with funds from the
province of Ontario to S.A. and P.A. I.N.M. obtained funds from the
Ontario Graduate Scholarship Program – University of Toronto. We thank
Salman Basrai for his assistance checking the validity and accuracy of
the scATOMIC code, as well as on his comments and suggestions concerning the package documentation.

Author contributions
All authors discussed the results, wrote, and commented on the paper.
I.N.M. developed the concept, performed data analysis, developed the
tool, and wrote the code. D.S. contributed to statistical analysis and
model development. S.A. and P.A. conceived the idea, contributed to
data analysis, led, and supervised all aspects of the study.

Competing interests
The authors declare no competing interests.

Additional information
Supplementary information The online version contains
supplementary material available at
https://doi.org/10.1038/s41467-023-37353-8.
Correspondence and requests for materials should be addressed to
Philip Awadalla or Sagi Abelson.
Peer review information Nature Communications thanks Maria Tsagiopoulou and the other anonymous reviewer(s) for their contribution to the
peer review of this work. Peer review reports are available.
Reprints and permissions information is available at
http://www.nature.com/reprints
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as
long as you give appropriate credit to the original author(s) and the
source, provide a link to the Creative Commons license, and indicate if
changes were made. The images or other third party material in this
article are included in the article’s Creative Commons license, unless
indicated otherwise in a credit line to the material. If material is not
included in the article’s Creative Commons license and your intended
use is not permitted by statutory regulation or exceeds the permitted
use, you will need to obtain permission directly from the copyright
holder. To view a copy of this license, visit http://creativecommons.org/
licenses/by/4.0/.
© The Author(s) 2023

14

nutrients
Systematic Review

The Effect of Diet and Exercise Interventions on Body
Composition in Liver Cirrhosis: A Systematic Review
Heidi E. Johnston 1,2, * , Tahnie G. Takefala 1 , Jaimon T. Kelly 3,4 , Shelley E. Keating 5 , Jeff S. Coombes 5 ,
Graeme A. Macdonald 2,6 , Ingrid J. Hickman 1,2 and Hannah L. Mayr 1,2,7,8
1
2
3
4

5

6

7
8

*

Citation: Johnston, H.E.; Takefala,
T.G.; Kelly, J.T.; Keating, S.E.;
Coombes, J.S.; Macdonald, G.A.;
Hickman, I.J.; Mayr, H.L. The Effect
of Diet and Exercise Interventions on
Body Composition in Liver Cirrhosis:
A Systematic Review. Nutrients 2022,
14, 3365. https://doi.org/
10.3390/nu14163365
Academic Editors: Aldo J.
Montano-Loza and Maryam Ebadi
Received: 21 July 2022
Accepted: 12 August 2022
Published: 17 August 2022
Publisher’s Note: MDPI stays neutral

Department of Nutrition and Dietetics, Princess Alexandra Hospital, Woolloongabba, QLD 4102, Australia
Faculty of Medicine, The University of Queensland, Brisbane, QLD 4072, Australia
Centre for Online Health, Faculty of Medicine, The University of Queensland, Brisbane, QLD 4072, Australia
Centre for Health Services Research, Faculty of Medicine, The University of Queensland, Brisbane,
QLD 4072, Australia
School of Human Movement and Nutrition Sciences, The University of Queensland, Brisbane,
QLD 4072, Australia
Department of Gastroenterology and Hepatology, Princess Alexandra Hospital, Woolloongabba,
QLD 4102, Australia
Centre for Functioning and Health Research, Metro South Health, Brisbane, QLD 4102, Australia
Bond University Nutrition and Dietetics Research Group, Faculty of Health Sciences and Medicine,
Bond University, Gold Coast, QLD 4226, Australia
Correspondence: heidi.johnston@health.qld.gov.au; Tel.: +61-7-3176-7938

Abstract: Alterations in body composition, in particular sarcopenia and sarcopenic obesity, are
complications of liver cirrhosis associated with adverse outcomes. This systematic review aimed to
evaluate the effect of diet and/or exercise interventions on body composition (muscle or fat) in adults
with cirrhosis. Five databases were searched from inception to November 2021. Controlled trials of
diet and/or exercise reporting at least one body composition measure were included. Single-arm
interventions were included if guideline-recommended measures were used (computed tomography/magnetic resonance imaging, dual-energy X-ray absorptiometry, bioelectrical impedance
analysis, or ultrasound). A total of 22 controlled trials and 5 single-arm interventions were included.
Study quality varied (moderate to high risk of bias), mainly due to lack of blinding. Generally, sample
sizes were small (n = 6–120). Only one study targeted weight loss in an overweight population.
When guideline-recommended measures of body composition were used, the largest improvements
occurred with combined diet and exercise interventions. These mostly employed high protein diets with aerobic and or resistance exercises for at least 8 weeks. Benefits were also observed with
supplementary branched-chain amino acids. While body composition in cirrhosis may improve
with diet and exercise prescription, suitably powered RCTs of combined interventions, targeting
overweight/obese populations, and using guideline-recommended body composition measures are
needed to clarify if sarcopenia/sarcopenic obesity is modifiable in patients with cirrhosis.
Keywords: liver cirrhosis; sarcopenia; sarcopenic obesity; nutrition; exercise; body composition

with regard to jurisdictional claims in
published maps and institutional affiliations.

1. Introduction
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/

Advanced liver disease is a complex major health problem, impacting more than
1.5 billion individuals worldwide [1]. Cirrhosis is the end stage of chronic liver disease and
is characterised by severe hepatic fibrosis with potential impacts on hepatic function. Once
patients develop cirrhosis, they are at risk of dying from decompensated liver disease or
hepatocellular carcinoma (HCC) [2]. Liver transplantation offers the opportunity to cure
both. During the progression to cirrhosis, many aspects of health deteriorate, increasing
the risk of malnutrition and loss of muscle mass [3,4], which in turn are associated with
adverse outcomes for patients with cirrhosis and those awaiting transplant [5,6].

4.0/).

Nutrients 2022, 14, 3365. https://doi.org/10.3390/nu14163365

https://www.mdpi.com/journal/nutrients

Nutrients 2022, 14, 3365

2 of 28

There are two key issues relating to body composition for people with liver cirrhosis.
Firstly, sarcopenia is a condition characterised by a significant depletion of skeletal muscle in
combination with low muscle strength and/or physical performance [7]. Sarcopenia is often
interrelated with malnutrition [8]. In general, sarcopenia in liver disease literature refers
to reduced muscle mass alone, which has a prevalence in cirrhosis between 40–70% [9].
Sarcopenia is associated with increased mortality in patients with cirrhosis, and in those
who receive a liver transplant [10]. The second issue is an elevated body mass index in
people with cirrhosis. Comorbid sarcopenia with obesity, where low muscle mass may be
masked due to excess adiposity, increases the risk of hepatic decompensation and death
in patients with cirrhosis [11,12]. Additionally, surgical risk is increased for obese liver
transplant recipients [13,14]. The proportion of patients being referred for liver transplant
with comorbid obesity is increasing [15]. Interventions to reduce adiposity may ameliorate
the severity of their underlying liver disease, but also needs to be considered to improve
transplant outcomes. The challenge in achieving weight loss in this patient group is to
preserve or increase muscle mass whilst losing fat mass.
The first challenge in addressing low muscle mass and/or high adiposity in patients
with cirrhosis is accurately assessing body composition, which can be complicated by fluid
retention with ascites and oedema. Triceps skinfold thickness (TSF) and mid-arm muscle
circumference (MAMC) appear less affected by fluid overload than other anthropometric
measures in this population [16]. While there is evidence that these measures have good
intra- and inter-rater reliability for the diagnosis of malnutrition [17], there remain concerns
about their reproducibility [18] and their reliability in identifying subtle changes [7]. Recent guidelines have recommended several reference methods for the assessment of body
composition in patients with cirrhosis, specifically computerised tomography (CT) and
magnetic resonance imaging (MRI) techniques [16,19]. The use of dual-energy Xray absorptiometry (DXA), bioelectrical impedance analysis (BIA), and ultrasound are also supported
when CT/MRI are unavailable. These may be more readily available in clinical settings,
although the reliability of BIA and some DXA measures may be adversely impacted by
fluid retention in decompensated cirrhosis [20,21].
It is well known that both diet and exercise have positive effects on health outcomes
across multiple chronic health conditions. Low muscle mass and high adiposity are attractive therapeutic targets in advanced liver disease because they may be modifiable through
diet and/or exercise interventions. Exercise training is known to reduce the progression, or
reverse muscle wasting [22] and has been shown to improve physical function and frailty in
cirrhosis [23]. According to current guidelines [16,19], a high protein, high energy diet has
been recommended for people with cirrhosis, due to catabolic effects of cirrhosis that can
lead to protein degradation and therefore muscle loss. Minimising fasting times, and the
inclusion of a late evening carbohydrate rich snack to prevent overnight catabolism have
also been recommended [24]. It is still unclear how to accurately estimate energy needs for
individuals with cirrhosis who are obese. There have also been several studies exploring
the effect of Branched Chain Amino Acids (BCAAs) in this population; however, there
has been heterogeneity in BCAA dosage type [25]. While advice about combined diet and
exercise in cirrhosis is beginning to appear in more detail in practice guidelines [26], there
remains a gap in current knowledge relating to improving body composition in cirrhosis,
especially in obese persons. There is currently no comprehensive synthesis of evidence to
guide interventions to slow progression or potentially reverse muscle wasting or reduce
adiposity for patients with cirrhosis.
Therefore, we aimed to systematically evaluate the evidence on the effect of diet and/or
exercise interventions on body composition in adults with cirrhosis, with a particular
interest in the impact of these interventions on patients with obesity and liver cirrhosis to
determine whether muscle mass can be preserved concurrently with fat loss.

Nutrients 2022, 14, 3365

3 of 28

2. Materials and Methods
This systematic review followed the Preferred Reporting Items for Systematic Reviews and
Meta-analyses (PRISMA) statement [27] (see Supplementary Materials Supplementary File S1),
and the protocol was registered with the international Prospective Register of Systematic
Reviews (PROSPERO ID: CRD42020176547).
2.1. Eligibility Criteria
Table 1 summarises the population, intervention, control, outcomes, and study design
(PICOS) for the study selection.
Table 1. PICOS for study selection and eligibility criteria.
Criteria

Inclusion and Exclusion Details

Population

-

Liver cirrhosis, including potential transplant candidates.

-

Diet or exercise intervention (alone or combination), of at least four
weeks duration.
Studies excluded if the intervention was a single nutrient (e.g.,
vitamin D, omega-3 fatty acid), or nutrition was exclusively
administered intravenously without oral nutrition support.

Intervention

-

Control

-

No specified control.
Studies without a control group were included if they reported
specific body composition measures (see below).

-

At least one body composition measure, via imaging (CT, MRI, or
DXA), BIA, ultrasound, or anthropometry (TSF, MAMC, MAC, thigh,
or calf circumference).
Single-arm interventions were included if they had one of the
guideline-recommended measures (CT, MRI, DXA [19]; or BIA if in
compensated cirrhosis).
Waist circumference was not included due to the confounding effect
of any ascites.

Outcomes

-

-

Study Design

-

RCTs, non-randomised controlled trials and single-arm interventions
were eligible.
Articles excluded: case report, letter to the editor, abstract only, or
non-English.

CT: computerised tomography, MRI: magnetic resonance imaging, DXA: dual-energy Xray absorptiometry,
BIA: bioelectrical impedance analysis, TSF: triceps skinfold thickness, MAMC: mid-arm muscle circumference,
MAC: mid-arm circumference, RCT: randomised controlled trials.

2.2. Search Strategy
Databases were searched from inception to 15 November 2021 (PubMed, Embase,
Web of Science, CINAHL, and CENTRAL). Reference lists of relevant review articles
were hand-searched to identify further articles. The strategy utilised a combination of
key words and controlled vocabulary combining terms related to liver cirrhosis AND
diet/exercise AND intervention/trial (see Supplementary Materials Supplementary File S2
for full search strategy). The final search was de-duplicated using reference management
software, Endnote [28]. References were screened in Rayyan [29]. Two reviewers (H.J. and
T.T.) independently screened approximately half of the title and abstracts using a screening
tool. Twenty studies were piloted with the tool to determine agreement before completing
the screening. For potentially eligible articles, full texts were retrieved and independently
screened by two of three reviewers (H.J., T.T., or H.M.). Disagreements were resolved by
consensus or referral to the third reviewer.

Nutrients 2022, 14, 3365

4 of 28

2.3. Data Extraction
Extracted data included study authors, publication year, country, population, setting,
intervention, control, and body composition outcomes. If data were not available, an
attempt to contact authors was made to retrieve information. Data were initially extracted
by either of two reviewers (H.J. or T.T.) in a standardised extraction table. Extraction was
piloted across three different study designs (RCT, non-randomised controlled trial, and
single-arm intervention studies) to ensure consistency. Where present, we extracted body
composition change data between study treatment groups. If unavailable, we recorded
the within-group change. All extraction was cross-checked by a second reviewer with
disagreements discussed to reach consensus.
2.4. Quality Assessment
For each included study, risk of bias was assessed independently by two of three
reviewers (H.J., T.T., or H.M.) using the Cochrane risk of bias tool (Rob2) [30] for RCTs,
and the ROBINS-I tool [31] for non-randomised controlled and single-arm studies. Rob2
evaluates five domains including risk of bias from: randomisation process, deviations
from intended interventions, missing outcome data, measurement of the outcome, and
selection of the reported result. The ROBINS-I tool evaluates seven domains including risk
of bias due to: confounding, participant selection, classification of interventions, deviations
from intended interventions, missing data, measurement of outcomes, and selection of the
reported result. For the domains considering deviations from intended interventions, where
intervention blinding is considered, we allocated ‘some concerns’ rather than ‘serious’ if
participants were not blinded. This is due to the nature of diet and/or exercise interventions,
where it is often not feasible for intervention allocation blinding. Conflicts were resolved
by consensus or a third reviewer. The certainty of the body of evidence based on outcomes
using the Grading of Recommendations Assessment Development and Evaluation was
not possible due to significant variability in study design, interventions used, outcome
measures employed, and statistical methodologies used across the studies.
2.5. Data Synthesis
A meta-analysis was unable to be performed due to variability in study interventions,
control groups, tools to assess body composition, and reporting of means and medians
across studies. Narrative synthesis was conducted based on type of intervention and body
composition measures. Where a study reported on multiple body composition measures
the guideline-recommended measures were prioritised in the text results (CT or MRI,
followed by DXA, BIA, or ultrasound, then anthropometry).
3. Results
3.1. Characteristics of Studies
The final search contained 10,099 articles, including three articles from hand searches
(Figure 1). A total of 152 full text articles were retrieved and 27 studies included in this
review. Thirty-two studies were excluded for not reporting on body composition measures.
The characteristics and outcomes of the included studies are summarised in Table 2. Of the
27 studies, 19 were RCTs, 3 were non-randomised controlled trials, and 5 were single-arm
intervention studies. Most studies were relatively small, with participant numbers ranging
from 6 to 120, totalling 1263 participants. Intervention duration ranged between 4 and
56 weeks and populations included patients with both compensated and decompensated
cirrhosis. Only one study specifically targeted an overweight population [32], however
the primary outcomes of interest were weight loss and portal hypertension changes. Thirteen studies in total reported populations with a mean BMI either overweight [33–40] or
obese [32,41–44]. Others did not report on BMI [45–51]. None of the studies specifically
targeted sarcopenic obesity in cirrhosis.

Nutrients 2022, 14, x FOR PEER REVIEW
Nutrients 2022, 14, 3365

5 of 28

Identification of studies via databases and registers

Identification

Records identified from:
PubMed (n = 2002)
Cochrane (n = 1141)
Embase (n = 4805)
Web of Science (1824)
CINAL (n = 324)
Hand searches (n = 3)

Records

removed

before

screening:
Duplicate records removed
(n = 3174)

Total (n = 10,099)

Records screened (title and
abstract)

Screening

(n = 6925)

Records excluded
(n = 6773)

Reports sought for retrieval

Reports not retrieved

(n =152)

(n = 0)

Full text reports assessed for
eligibility
(n = 152)

Reports excluded (n = 125):
Incorrect outcome (n = 32)
Incorrect intervention (n =19)
Full text not English (n = 11)
Incorrect population (n = 11)
Incorrect duration (n = 12)

Included

Clinical trial/registration
Studies included in review
(n = 27)

(n = 14)
Did not meet study design
criteria (n = 26)

Figure 1. Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow
diagram of the study selection process.

Figure 1. Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PR
agram of the study selection process

For most of the included studies, the change in muscle or fat mass wa
outcome, and factors such as muscle strength [29], aerobic/exercise capaci
vival [39,47], quality of life [48], portal hypertension [28], hepatic venous pr
liver function [49] were primary outcomes. Fourteen studies were combine

Nutrients 2022, 14, 3365

6 of 28

Table 2. Study characteristics and outcomes for diet and/or exercise interventions in cirrhosis.

Study
Citation,
Country

Study
Design

Aaman et al.
[40]
2019
Denmark

RCT

Chen et al. [44]
2020
USA

Pilot
RCT

Body Composition Outcomes
↑
=
Significantly Increased or Higher
Dietary
Control
Exercise Intervention
Population
↓ = Significantly Decreased or Lower
Intervention
Group
↔ = No Significant Difference (Pre/Post or
vs. Control)
Combined intervention studies (n = 9 RCTs, n = 2 non-randomised studies, n = 3 single arm intervention trials)
Intervention n = 20
Age 61.7 ± 7.8 years
Supervised resistance training 3
80% male
Intervention versus control:
days/week for 60 min at a
Oral nutrition
BMI 26 ± 3.0 kg/m2
↑ Cross sectional area of quadriceps via MRI
supplements
(125
moderate
level.
5
min
warm
up,
Child Pugh Class:
↑ Body cell mass via BIA
mL, 14.4 g protein
then
A 50%, B 50%
↔ Dry lean mass via BIA
and 2.9 g
7 whole body exercises, (3 sets
MELD 10.8 ± 2.7
No change to
↔ Lean mass via BIA
BCAA/100 g)
Control n = 19
for legs, 2 sets for arms/chest, 1
current
↔ Calf circumference
Age 63 ± 7 years
provided if
exercise or diet
set lower back, 1 for
↔ MAC
74% male
abdominals), starting at 15–12
protein intake <
↔ Thigh circumference
2
BMI 25 ± 4.2 kg/m
repetitions at the start down to 8
1.2 g/kg/day at
↔ Mid arm muscle area
Child Pugh Class:
baseline
by week 12
↔ TSF
A 53%, B 47%
Duration:12 weeks
MELD 10.7 ± 2.8
Outpatients
Intervention versus control:
↑ Psoas muscle index via CT
Intervention n = 9
↔ Total skeletal muscle index via CT
Age 55 ± 7 years
↔ Intramuscular adipose tissue via CT
56% male
↔ Total abdominal adipose tissue via CT
BMI 30 ± 6 kg/m2
↔ Total thigh muscle volume via CT
Education
on
exercise,
and
Standardised diet
Child Pugh Class:
↔ Thigh muscle index via CT
behavioural
counselling
provided
1.2–1.5
B 78%, C 22%
↔ Cross sectional area, 50% of femur length
bi-weekly for first 8 weeks.
g/kg/day of
MELD-Na 16 ± 4
Standardised
via CT
Self-directed exercise increasing
protein + late
Control n = 8
diet (same as
↔
Thigh
adipose
tissue volume via CT
500 steps/day weekly to
Age 54 ± 11 years
evening snack +
intervention
↔ Fat free mass via DXA
biweekly.
75% male
oral nutrition
group) only
↔ Fat mass via DXA
BMI 31 ± 8 kg/m2
Daily to weekly motivational
supplement (6 g
↔ Lean muscle index via DXA
Child Pugh Class:
phone calls.
essential amino
↔ Lower extremities lean muscle index
B 50%, C 50%
acids) twice a day
Duration: 12 weeks
via DXA
MELD-Na 19 ±3
↔ Fat free mass via BIA
Portal hypertension and MELD
↔ Fat mass via BIA
≥ 10
↔
Skeletal
muscle mass via BIA
Outpatients
↔ Skeletal muscle index via BIA
↔ Phase angle via BIA

Nutrients 2022, 14, 3365

7 of 28

Table 2. Cont.

Study
Citation,
Country

HernandezConde et al.
[39] 2021

Kruger et al.
[47]
2018
Canada

Study
Design

Population

Pilot,
doubleblind
RCT

Intervention n = 15
Age 69 ± 9.7 years
86.7% male
BMI 29 ± 4.6 kg/m2
MELD 10.7 ± 4.4
Child Pugh Class:
A 78.6%, B 21.4%
Control n = 17
Age 61 ± 9.4 years
88.2% male
BMI 26 ± 4.7 kg/m2
MELD 11 ± 3.4
Child Pugh Class:
A 59%, B 29%,
C 12%
Compensated outpatients

RCT

Intervention n = 20
Age 53 ± 8 years
50% male
MELD 9.05
Child Pugh Class:
A 70%, B 30%
Control n = 18
Age 56.4 ± 8.5 years
65% male
MELD 9.7
Child Pugh Class:
A 70%, B 30%
BMI not reported
Outpatients

Exercise Intervention

Dietary
Intervention

Control
Group

Body Composition Outcomes
↑ = Significantly Increased or Higher
↓ = Significantly Decreased or Lower
↔ = No Significant Difference (Pre/Post or
vs. Control)

Personalised exercise
instructions with use of
accelerometers in wristbands or
smartphones to include
5000–10,000 steps/day with
gradual increments of 2000–2500
steps/day + moderate intensity
exercise in 30-min sessions (goal
at least 150 min/week) +
verbal reinforcement at reviews.
Duration: 12 weeks

Personalised diet
recommendations +
instructed to eat 7
meals/day
including late
evening snack plus
BCAA supplement
100 g dissolved in
500 mL water
throughout the day
(15 g protein, 8.5 g
fat, 68 g of
carbohydrates, 2.61
g of leucine, 1.01 g
of isoleucine, and
1.62 g of valine) +
verbal
reinforcement at
reviews

Same exercise
and diet recommendations
as intervention
group except
took placebo
supplement
100 g
dissolved in
500 mL water
throughout
day
(maltodextrin
99.63%)
instead of
BCAA

Intervention versus control:
↑ Skeletal muscle index via CT
↓ % total body fat via BIA
↔ Phase angle via BIA

Supervised at home, moderate
to high intensity aerobic exercise
(60–80% of heart rate reserve) on
cycle ergometer 3 days/week
(30 min sessions gradually
increased to 60 min). Visited
bi-weekly for session
observation.
Duration: 8 weeks

Dietary counselling
on optimal protein
(1.2–1.5 g/kg/day,
ideal body weight
for BMI > 30) and
energy intake
(35–40 kcal/kg for
BMI 20–30,
25–35 kcal/kg for
BMI 30–40, and
20–25 kcal/kg for
BMI > 40. Advised
on exercise days to
consume an extra
250–300 kcal.

Usual care

Intervention versus control:
↔ Thigh muscle mass via ultrasound
↔ Thigh circumference

Nutrients 2022, 14, 3365

8 of 28

Table 2. Cont.

Study
Citation,
Country

Lattanzi et al.
[38] 2021

MaciasRodriguez
et al. [37] 2020

Study
Design

Population

Pilot
single
blind
RCT

Intervention n = 14
Age: 59.2 ± 8.4 years
64% male
BMI 29.8 ± 4.3 kg/m2
Child Pugh Class:
A 86%, B 14%
MELD 9 ± 2.7
Control n = 10
Age: 56 ± 4.6 years
60% male
BMI 29.6 ± 6.8 kg/m2
Child Pugh Class:
A 90%, B 10%
MELD 9.8 ± 3.2
Outpatients with portal
hypertension

RCT

Intervention n = 22
Age 53.5 ± 7.6 years
47% male
BMI 29.8 ± 4.8 kg/m2
Child Pugh Class:
A 82%, B 18%
MELD 8.5 (7–10)
Control n = 21
Age 53.7 ± 8.2 years
43% male
BMI 29.2 ± 3.7 kg/m2
Child Pugh Class:
A 95%, B 5%
MELD 8 (7.5–9.5)
Compensated cirrhosis,
outpatients

Exercise Intervention

Dietary
Intervention

Control
Group

Body Composition Outcomes
↑ = Significantly Increased or Higher
↓ = Significantly Decreased or Lower
↔ = No Significant Difference (Pre/Post or
vs. Control)

Motivational interviewing with
information on physical activity
at baseline

Motivational
interview at
baseline with
information and
counselling on diet
in line with EASL
clinical guidelines
(2019) + HMB
supplement
(3 g/day)

Same exercise
and diet as
intervention
group +
placebo
supplement
(Sorbitol
3 g/day)

Within group changes:
↑ Thigh muscle thickness via ultrasound
↔Fat free mass via BIA
↔Phase Angle via BIA

Given wrist-worn accelerometer
as activity tracker. Aim to
gradually increase physical
activity to reach >2500
steps/day above baseline. Total
5000 steps/day. Light to
moderate intensity.
Duration: 10 weeks

Harris–Benedict
equation was
utilised to calculate
energy
requirements + 10%
extra for thermic
effect of food and
20% extra for
exercise.
Diet 60%
carbohydrates,
1.3–1.5 g
protein/kg/day +
remainder from fats
+ 1.5–2 g sodium
restriction/day
restriction +
non-alcoholic beer
at lunch
(330 mL/day)

The same diet
and exercise
prescribed as
intervention
group without
non-alcoholic
beer (given a
330 mL bottle
of water
instead)

Within group changes:
↔ Phase Angle via BIA
↑ Thigh circumference
↔MAMC
↔TSF

Nutrients 2022, 14, 3365

9 of 28

Table 2. Cont.

Study
Citation,
Country

MaciasRodriguez
et al. [36]
2016
Mexico

Roman et al.
[33]
2014
Spain

Study
Design

Population

Pilot
open
RCT

Intervention n = 13
Age 53 (48–55) years
69% male
BMI 27.5 (22.4–28.9) kg/m2
Child Pugh score
6 (5–7)
MELD 9 (8–12)
Control n = 12
Age 51 (38–57) years
83% male
BMI 27.4 (25–30) kg/m2
Child Pugh score
6 (5–7)
MELD: 12 (7–14)
Compensated outpatients

Supervised exercise
3 days/week of 60–70% max
heart rate, for 40 min of aerobic
training using cycle ergometer +
kinesiotherapy/rhythmic
activities)
Duration: 14 weeks

Pilot
RCT

Intervention n = 8
Age 65.5 (46–72) years
62% male
BMI 26.7 (18.3–34.7) kg/m2
Child Pugh Class:
A 87%, B 13%
MELD 9.5 (7–12)
Control n = 9
Age 61 (43–75) years
78% male
BMI 27.6 (19.5–35.3) kg/m2
Child Pugh Class:
A 78%, B 22%,
MELD 9 (7–13)
Outpatients with a previous
episode of decompensation

Supervised exercise
3 days/week, moderate
intensity (60–70% max heart
rate) for 60 min. Cycle
ergometry and
treadmill walking
Duration: 12 weeks

Exercise Intervention

Dietary
Intervention

Control
Group

Body Composition Outcomes
↑ = Significantly Increased or Higher
↓ = Significantly Decreased or Lower
↔ = No Significant Difference (Pre/Post or
vs. Control)

Instructed to
consume 30% extra
calories (65%
carbohydrates,
1.2 g/kg/day
protein) + no added
salt diet of
1.5–2 g/day

Same recommendations as
intervention;
consume 10%
extra calories
(65%
carbohydrates,
1.2 g/kg/day
protein) + no
added salt diet
of 1.5–2 g/day.
Continue
regular
activities, no
new exercise

Intervention versus control:
↑ Phase angle via BIA

10 g oral leucine
supplementation
daily

10 g oral
leucine supplementation
daily, no
exercise recommendations

Within group changes:
↑ Lower thigh circumference (intervention
compared to baseline, ↔ control)
↔ Mid or upper thigh circumference
(intervention or control)
↔ MAMC (intervention or control)
↔ Mid-arm circumference (intervention
or control)
↔ TSF (intervention or control)

Nutrients 2022, 14, 3365

10 of 28

Table 2. Cont.

Study
Citation,
Country

Zenith et al.
[35]
2014
Canada

Morkane et al.
[43]
2020
United
Kingdom

Exercise Intervention

Dietary
Intervention

Control
Group

Body Composition Outcomes
↑ = Significantly Increased or Higher
↓ = Significantly Decreased or Lower
↔ = No Significant Difference (Pre/Post
or vs. Control)

RCT

Intervention n = 9
Age 56 ± 8 years
78% male
BMI 27.7 ± 3.8 kg/m2
Child Pugh score: 6.2 ± 1.4
MELD 9.7 ± 2.4
Control n = 10
Age 59 ± 6 years
80% male
BMI 28.9 ± 4.1 kg/m2
Child Pugh score:
6.3 ± 1.4
MELD 10.2 ± 1.9
Outpatients, Child Pugh A or B

Supervised exercise
3 days/week, 60–80% of peak
VO2 , 30 min session, increased
by 2.5 min per session each
week, 5 min warm up and cool
down using cycle ergometer
Duration: 8 weeks

Baseline dietetic
counselling to reach
1.2–1.5 g/kg of
protein (for BMI >
30 adjustments
made based on
ideal body weight),
calories BMI
specific (between 14
up to 30 kcal/kg)
and instructed to
consume an extra
250–300 calories on
exercise days

Baseline
counselling by
dietitian
(same as
intervention)
but no formal
exercise
regimen

Intervention versus control:
↑ Quadricep muscle thickness via
ultrasound
↑ Thigh circumference

Nonrandomised
controlled
trial

Intervention n = 16
Age 55.6 ± 7.8 years
87.5% male
MELD 13.7 ± 4.6
BMI 30.9 ± 5.6 kg/m2
Control n = 17
Age 55.6 ± 7.8 years
82.7% Male
MELD 13.2 ± 3.7
BMI: 27 ± 4.6 kg/m2
Outpatients, transplant
candidates

Supervised 40 min interval
training on cycle ergometer (4–6
× 3 min intervals at 80% of AT
(moderate intensity) and 4–6 ×
2 min intervals at 50% of
difference between VO2 at peak
and VO2 at AT (‘severe’
intensity) with 5 min warm up
and cool down)
Duration: 6 weeks

Standardised
nutrition
assessment and
advice by transplant
dietitian at baseline
and 6 weeks

Standard care,
no initiation of
exercise.
Standardised
nutrition
assessment
and advice by
transplant
dietitian at
baseline and
6 weeks

Within group changes:
↔ Mid-arm circumference (intervention
or control)
↔ MAMC (intervention or control)

Study
Design

Population

Nutrients 2022, 14, 3365

11 of 28

Table 2. Cont.

Study
Citation,
Country

Schmidt et al.
[42] 2021

Berzigotti et al.
[32]
2017
Spain

Hiraoka et al.
[52]
2017
Japan

Study
Design

Population

Nonrandomised
controlled
trial

Intervention n = 11
Age 56.6 ± 9.9 years
63.6% male
BMI 30.3 ± 5.4 kg/m2
Child Pugh Class:
A 91%, B 9%
Control n = 22
Age 58.7 ± 12.9 years
59.1% male
BMI 32.4 ± 5.1 kg/m2
Child Pugh Class:
A 86%, B 14%
MELD—not reported
Compensated outpatients

Total n = 50
Age 56 ± 8 years
Multi62% male
centre
BMI 33.3 ± 3.2 kg/m2
single
MELD 9 ± 3
arm
Child Pugh Class:
intervention
A 92%, B 8%
pilot
Compensated outpatients with
study
BMI ≥ 26 kg/m2

Single
arm intervention
study

Total n = 33
Age 67 (63–71) years
39% men
BMI 23.2 (20.8–25.1) kg/m2
Child Pugh Class:
A 90%, B 10%
Compensated outpatients

Exercise Intervention

Dietary
Intervention

Control
Group

Body Composition Outcomes
↑ = Significantly Increased or Higher
↓ = Significantly Decreased or Lower
↔ = No Significant Difference (Pre/Post
or vs. Control)

Supervised exercise
3 days/week, aerobic, moderate
intensity (5 min warm up,
30 min walking/running
60–70% VO2 max). Increasing
session by 2 min until reaching
50 mins by week 8.
Duration: 12 weeks

Diet advice to aim
for 25–30 kcal/day
and 1.2–1.5 g of
protein/kg/day—
using estimated dry
body weight.

The same diet
advice without
any exercise
intervention

Intervention versus control:
↔ Phase Angle via BIA
↔ Lean mass via BIA
↔ Fat mass via BIA
↔ MAMC
↓ MAC

Supervised exercise 1 day/week
for 60 min moderate intensity
(10–12 Borg Scale of Perceived
Effort) in groups of 1–5 +
increase daily step activity
Duration: 16 weeks

Reduction of
500–1000 kcal/day.
Protein intake
maintained at
20–50% of total kcal
and within 0.8 g/kg
ideal
bodyweight/day.
Carbohydrates
45–50% and fat
<35% of total kcal.
20 g/day
alimentary fibre
recommended.

No control

↓ Fat mass via BIA
↔ Lean mass via BIA

Walking (an additional 2000
steps/day on top of usual
average steps)
Duration: 12 weeks

Late evening BCAA
supplement
provided once daily
(13.5 g protein, 210
kcal/day)

No control

↑ Muscle volume via BIA (reported as
change ratio)

Nutrients 2022, 14, 3365

12 of 28

Table 2. Cont.

Study
Citation,
Country

Nishida et al.
[53]
2017
Japan

Study
Design

Population

Single
arm intervention
study

Total n = 6
Age from 51–79 years
100% female
BMI 24.3 (19.6–26.1) kg/m2
Child Pugh Class:
A 100%
Compensated
outpatients

Exercise Intervention

Dietary
Intervention

Control
Group

Body Composition Outcomes
↑ = Significantly Increased or Higher
↓ = Significantly Decreased or Lower
↔ = No Significant Difference (Pre/Post
or vs. Control)

Instructed to undertake bench
step activity at anaerobic
threshold level at home. Aim
140 min/week.
Duration: 12 months

BCAA supplement
(3 sachets/day =
12.45 g of BCAA),
no specific nutrition
advice except to
maintain usual
dietary intake

No control

↔ % fat via BIA
↔ Visceral fat area via CT
↔ Intramuscular adipose tissue content
via CT

Diet-only intervention studies (n = 9 RCTs, n = 1 non-randomised study, n = 2 single arm interventions)

Dupont et al.
[45]
2012
France

Hirsh et al.
[54]
1983
Chile

Multicentre
RCT

Intervention n = 44
Age 56.1 ± 9.6 years
68% male
Child Pugh score: 11.2 ± 1.3
Control n = 55
Age 54.6 ± 9.6 years
64% male
Child Pugh score: 10.5 ± 1.5
BMI or MELD—not reported
Inpatients with ARLD and
jaundice (without alcoholic
hepatitis)

RCT

Intervention n = 26
Age 49.9 ± 8.6 years
81% male
Control n = 25
Age 46.1 ± 8.0 years
84% male
BMI, Child Pugh, or
MELD—not reported
Decompensated outpatients

NA

Enteral nutrition
3–4 weeks
(30–55 kcal/kg/day
through nasogastric
tube). Subsequent
3 oral nutrition
supplements/day
for 2 months
Duration: 12 weeks
with outcomes
reported at
12 months

Standard
hospital oral
diet

Intervention versus control:
↔ MAMC
↔ TSF

NA

1 L oral nutrition
supplement /day
(1000 kcal, 34 g
protein) + usual diet
Duration:
12 months

Placebo tablet
daily

Intervention versus control:
↔ TSF
↔ Mid-arm circumference

Nutrients 2022, 14, 3365

13 of 28

Table 2. Cont.

Study
Citation,
Country

Le Cornu et al.
[55]
2000
England

Les et al.
[48]
2011
Spain

Study
Design

Population

RCT

Intervention n = 42
Age 52 (27–67) years
69% male
Child Pugh Class:
A 7%, B 48%, C 45%
Control n = 40
Age 50 (24–68) years
78% male
Child Pugh Class:
A 10%, B 28%, C 62%
BMI or MELD not reported
Outpatient transplant
candidates with MAMC < 25%
percentile

Multicentre
RCT

Intervention n = 58
Age 64.1 ± 10.4 years
78% male
Child Pugh 8.3 ± 2.0
MELD 16.1 ± 4.5
Control n = 58
Age 62.5 ± 10.4 years
74% male
Child Pugh 8.1 ± 1.7
MELD 16.2 ± 3.9
BMI—not reported
Outpatients with previous
episode of hepatic
encephalopathy

Body Composition Outcomes
↑ = Significantly Increased or Higher
↓ = Significantly Decreased or Lower
↔ = No Significant Difference (Pre/Post
or vs. Control)

Exercise Intervention

Dietary Intervention

Control
Group

NA

Oral nutrition
supplement of 500
mL/day (750 kcal, 20 g
protein) was given +
dietary counselling to
adapt/increase their
calories and protein
based on their medical
condition until
transplantation
Duration: until
transplantation.
Median wait 77 (1–395)
days intervention and
45 (1–424) control

Standard
dietary advice
to
adapt/increase
their calories
and protein
based on their
medical
condition until
transplantation

Intervention versus control:
↔ MAMC
↔ Mid-arm circumference
↔ TSF

NA

Diet of 35 kcal/kg +
0.7 g/kg of protein/day
adjusted to ideal
weight + late evening
BCAA supplement
2/day (120 kcal).
Enteral nutrition if
admitted for episode of
hepatic encephalopathy
and oral intake in
hospital was poor.
Duration: mean 32 ±
22 weeks intervention
and 36 ± 2 weeks
control

Same diet but
with
maltodextrin
supplement
2/day instead
of BCAA.
Enteral
nutrition
provided if
episode of
hepatic encephalopathy
and oral intake
was poor

Within group changes:
↑ MAMC (intervention compared
to baseline)
↔ MAMC (control compared to baseline)

Nutrients 2022, 14, 3365

14 of 28

Table 2. Cont.

Study
Citation,
Country

Manguso et al.
[34]
2005
Italy

Okabayashi
et al. [56]
2011
Japan

Study
Design

Randomised,
double
period
crossover
trial

RCT

Population

Group 1: n = 45
Age 60 ± 9 years
67% male
BMI 28.5 ± 3.2 kg/m2
Child Pugh Class:
A 33%, B 77%
Group 2: n = 45
Age 60 ± 7 years
49% male
BMI 27.8 ± 2.1 kg/m2
Child Pugh Class:
A 33%, B 77%
Outpatients with HCV cirrhosis

Intervention n = 40
Age 68 ± 7.6 years
28% male
BMI 23.6 ± 3.2 kg/m2
Child Pugh Class:
A 70%, B 30%
Control n = 36
Age 65.1 ± 11.3 years
31% male
BMI 22.7 ± 3.2 kg/m2
Child Pugh Class:
A 71%, B 29%
Outpatients with scheduled
HCC surgery

Exercise Intervention

Dietary Intervention

NA

Group 1: Prescribed diet
of 30–40 kcal/kg/day
based on calculated
desirable weight
(total calories split into
16% protein, 55%
carbohydrates, 28–30%
fat) +
low sodium
1000 mg/day
Followed by usual
diet after.
Group 2:
Usual diet first. Followed
by prescribed diet second.
Duration: 3 months per
diet (6 months total)

NA

Carbohydrate and BCAA
enriched supplement
morning and night.
(420 kcal, 13 g free amino
acids, 13 g of gelatine
hydrolysate, 62 g
carbohydrates, 7 g lipids)
+ dietitian education to
modify intake to reduce
420 kcal/day to account
for the supplement and
match caloric intake
to controls
Duration: supplements
for at least 6 months,
with a follow up at
12 months

Control
Group

Body Composition Outcomes
↑ = Significantly Increased or Higher
↓ = Significantly Decreased or Lower
↔ = No Significant Difference (Pre/Post
or vs. Control)

Within group changes:
↑ MAMC (Group 1 at 3 months post
prescribed diet vs baseline)
↑ MAMC (Group 2 at 6/12, post
prescribed diet vs baseline and vs 3/12)
↓ MAMC (Group 1 at 6 months post
usual diet vs 3 months post
prescribed diet)
↔ MAMC (Group 2 at 3 months post
usual diet vs baseline)
↔ TSF (Group 1 or Group 2 after both
diet interventions at 3 and 6 months)

Usual diet. No
supplements

Intervention versus control:
↑ MAMC (at 6, 8, 10, 12 months)
↔ TSF no change post-operatively in
both groups (data not reported)

Nutrients 2022, 14, 3365

15 of 28

Table 2. Cont.

Study
Citation,
Country

Poon et al. [50]
2004
China

Sorrentino
et al. [51]
2012
Italy

Study
Design

Population

RCT

Intervention n = 41
Age 59 (24–84) years
95% male
Control n = 43
Age 59 (27–80) years
90% male.
No BMI, Child Pugh or MELD
reported. Outpatients with
unresectable HCC

RCT

Group A: n = 40
Age 64 ± 6.3 years
65% male
Child Pugh Class:
B 28%, C 72%
MELD 12.1 ± 0.7
Group B: n = 40
Age 66 ± 7.5 years
67% male
Child Pugh Class:
B 30%, C 70%
MELD 11.7 ± 0.7
Group C: n = 40
Age: 65 ± 7.6 years
70% male
Child Pugh Class:
B 25%, C 75%
MELD 12.4 ± 0.9
BMI not reported
In/outpatients with
refractory ascites

Body Composition Outcomes
↑ = Significantly Increased or Higher
↓ = Significantly Decreased or Lower
↔ = No Significant Difference (Pre/Post
or vs. Control)

Exercise Intervention

Dietary Intervention

Control
Group

NA

BCAA supplement
morning and night (420
kcal, 13 g amino acids, 13
g peptides, 62 g
carbohydrates, 7 g lipids)
+ unrestricted diet unless
HE—protein was
restricted
Duration: 1 week prior to
surgery, up to 12 months

Usual diet

Intervention versus control:
↔ Mid-arm circumference
↔ TSF

NA

Group A: Instructed to
consume 1–1.3 g
protein/kg/day,
30–35 kcal/kg/day + low
sodium diet (80
mEq/day) + BCAA
evening snack (210 kcal,
13.5 g protein, 3.5 g fat) +
instructed to adjust
energy intake to account
for BCAA supplement +
post LVP parenteral
nutrition for 24 hrs post
paracentesis during
hospital admission +
Dietitian advice monthly.
Group B: same as group
A without parenteral
nutrition post
paracentesis.
Duration: 12 months,
follow up at 3, 6,
12 months

Group C: Low
sodium diet
(80 mEq /day)
+ Dietitian
advice
monthly

Between group changes:
↓ TSF (Group C versus Group A at 3, 6,
and 12 months and Group C versus
Group B at 6 months only)
↓ MAC (Group C versus Group A and
Group B at 6 and 12 months)

Nutrients 2022, 14, 3365

16 of 28

Table 2. Cont.

Study
Citation,
Country

Tangkijvanich
et al. [57]
2000
Thailand

Okabayashi
et al. [49]
2008
Japan

Kitajima et al.
[59]
2018
Japan

Study
Design

Population

RCT

Group 1: n = 14
Age: 53 ± 11 years
71% male
BMI 23.7 ± 3.4 kg/m2
Child Pugh score:
5–7: 64%, score 8–15: 36%.
Group 2: n = 15
Age: 53 ± 13 years
80% male
BMI: 25 ± 4.1 kg/m2
Child Pugh score:
5–7: 60%, score 8–15: 40%
Outpatients

Intervention n = 13
Age 66.2 ± 9.1 years
54% male
NonChild Pugh Class:
randomised
A 77%, B 23%
study
Control n = 28
with
Age 65.6 ± 8.2 yrs
historical
75% male
control
Child Pugh Class:
group
A 82%, B 18%
BMI not reported Outpatients
for HCC surgery

Single
arm intervention
study

Total n = 21
Age 71.3 ± 7.9 years
42% male
BMI 23.9 ± 4.0 kg/m2
Child Pugh Class:
A 48%, B 52%
MELD—not reported
Outpatients with
hypoalbuminaemia

Body Composition Outcomes
↑ = Significantly Increased or Higher
↓ = Significantly Decreased or Lower
↔ = No Significant Difference (Pre/Post
or vs. Control)

Exercise Intervention

Dietary Intervention

Control
Group

NA

Group 1: received
standard diet (40 g
protein/day) + 150 g
BCAA supplement/day
= total of ~2000 kcal/day.
Duration: 4 weeks

Group 2:
standard diet
(80 g
protein/day =
total of ~2000
kcal/day)

Within group changes:
↔ MAMC (Group 1 or Group 2)

NA

Carbohydrate and BCAA
enriched supplement
morning and night.
(420 kcal, 13 g free amino
acids, 13 g gelatin
hydrolysate, 62 g
carbohydrates, 7 g lipids)
Duration: 2 weeks prior
to surgery and at least
6 months post

Usual
care—no supplementation

Within group changes:
↑ MAMC (baseline to 6 months for
intervention, not reported for control)

NA

BCAA supplement 3/day
after meals. Dietitian
advised intakes of
25–35 kcal/kg/day and
protein 1–1.4 g/kg/day.
Adherence monitored
monthly.
Duration: 48 weeks

No control

↔ Skeletal muscle index via CT
↔ Intramuscular adipose tissue content
via CT
↔ Subcutaneous fat area via CT
↔ Visceral fat area via CT

Nutrients 2022, 14, 3365

17 of 28

Table 2. Cont.

Study
Citation,
Country

Putadechakum
et al. [58]
2012
Thailand

Study
Design

Population

Single
arm intervention
study

n = 22
Age 52.9 ± 12.8 years
55% male
BMI 21.4 ± 0.6 kg/m2
Child Pugh Class:
A 63%, B 23%, C 14%
Outpatients with ARLD

Body Composition Outcomes
↑ = Significantly Increased or Higher
↓ = Significantly Decreased or Lower
↔ = No Significant Difference (Pre/Post
or vs. Control)

Exercise Intervention

Dietary Intervention

Control
Group

NA

20 g protein (soy based)
oral nutrition
supplement daily
(420 kcal, 20 g protein,
65 g CHO, 10.6 g fat) +
regular diet.
Duration: 8 weeks

No control

↑ Lean mass via BIA
↔ Fat mass via BIA
↔ TSF

Sham
intervention
1h3
days/week of
cephalocaudal
muscle
relaxation, and
breathing,
visualisation,
and
concentration
exercises

Within group changes:
↑ Lean appendicular mass via DXA
(intervention compared to baseline,
↔ control)
↑ Lean leg mass via DXA (intervention
compared to baseline, ↔ control)
↑ Lean body mass via DXA (intervention
compared to baseline, ↔ control)
↓ Fat body mass via DXA (intervention
compared to baseline, ↔ control)
↑ Upper thigh circumference
(intervention compared to baseline,
↔ control)
↔ Lower thigh circumference
(intervention or control)
↓ Mid-arm circumference and mid-arm
skinfold thickness (intervention
compared to baseline, ↔ control)
↓ Mid-thigh skinfold thickness
(intervention compared to baseline,
↔ control)
↔ MAMC (intervention or control)

Exercise only intervention (n = 1 RCT)

Roman et al.
[41]
2016
Spain

RCT

Intervention n = 14 Age
62 ± 2.4 years
71% male
BMI 31.5 ± 1.6 kg/m2
Child Pugh score:
5.4 ± 0.2
MELD 8.2 ± 0.4
Control n = 9
Age 63.1 ± 2.3 years 85% male
BMI 30.3 ± 1.4 kg/m2
Child Pugh score:
5.4 ± 0.2
MELD 9.1 ± 0.4
Outpatients with a previous
episode of decompensation

Supervised exercise
3 days/week, 60 min of
cycle ergometry and
treadmill walking +
5–10 min of upper body
resistance exercise +
10–15 min balance,
coordination, stretching
and relaxation.
Moderate intensity
(60–70%) of max heart rate.
Duration: 12 weeks

NA

Outcome data presented for controlled trials are the between group differences (where reported) and the within group differences if the significance of between group data were not
reported. Data presented as mean SD or median (range/inter-quartile range). RCT: randomised controlled trial, AT: anaerobic threshold, MELD: model for end-stage liver disease, BMI:
body mass index, ARLD: alcohol related liver disease, BCAA: branched-chain amino acid, CT: computed tomography, DXA: dual-energy X-ray absorptiometry, BIA: bio-electrical
impedance analysis, MAMC: mid-arm muscle circumference, TSF: triceps skinfold thickness, MAC: mid arm circumference, HE: hepatic encephalopathy, LVP: large volume paracentesis,
EASL; European Association of the Study of the Liver, NA: not applicable, VO2 max: maximum amount of oxygen your body is able to use during exercise. Child Pugh score [60].

Nutrients 2022, 14, 3365

18 of 28

For most of the included studies, the change in muscle or fat mass was a secondary
outcome, and factors such as muscle strength [29], aerobic/exercise capacity [41,46], survival [39,47], quality of life [48], portal hypertension [28], hepatic venous pressure [30],
or liver function [49] were primary outcomes. Fourteen studies were combined diet
and exercise interventions [32,33,35–40,42–44,47,52,53], all in outpatient settings. Their
exercise components varied, with most delivering supervised sessions in a clinic setting [32,33,35,36,40,42,43], although one study was supervised by a clinician at the patient’s
home [47] and others were self-directed at home [37–39,44,52,53]. Most exercise was moderate to high intensity for 30–60 min sessions on 1 to 3 days a week and utilised either
aerobic or resistance training, or a combination of these. Otherwise, some self-directed sessions focused on increasing step counts. The dietary component of five of these combined
interventions used a high protein and energy diet [35,36,40,42,44]. Four of those studies provided the same dietary intervention to the control group, with the only difference between
treatment groups being exercise in the intervention arms [35,36,42,44], while only one study
provided “usual care” to the control participants [40]. Another combined study followed
this style, however providing ‘standard dietary advice’ to both intervention and control
arms, while the intervention arm also received supervised exercise training [43]. Three
other combined studies delivered exercise and diet interventions to both groups, with the
difference being a specific dietary product, either non-alcoholic beer [37], branched-chain
amino acids (BCAAs) [33,39], or beta-hydroxy-beta-methylbutyrate (HMB, a metabolite
from leucine) [38]. Of the three diet and exercise single-arm interventions, two provided
BCAAs, with self-directed exercise [52,53] while the third study in overweight cirrhotic
patients focused on a hypocaloric, moderate protein diet with supervised exercise [32].
Twelve diet-only studies [34,45,48–51,54–59] were included: ten in an outpatient
setting [34,48–50,54–59], one in inpatients [45], and the twelfth commenced in inpatients
with outpatient follow up [51]. Most (n = 9) interventions prescribed a high protein
and energy diet plus oral nutritional supplements either with [48–50,56,57,59] or without
BCAAs [54,55,58]. Out of the three remaining studies, one prescribed a high energy diet
without supplementation [34], one study utilised 3–4 weeks of enteral nutrition follow
by oral supplementation [45], and one study utilised short-term parenteral nutrition in
combination with a high protein and energy diet [51].
One exercise-only intervention met the eligibility criteria. This RCT involved supervised aerobic and resistance exercise sessions of moderate intensity for 60 min three times
weekly, versus a relaxation program for the control group of the same frequency and
duration [41].
Across all studies, ten different methodologies were used to measure body composition
(see Table 2). Most combined diet/exercise interventions used guideline-recommended measures: CT plus DXA and BIA [38], CT plus BIA [46], MRI plus BIA [29], ultrasound [37,41], or
BIA alone [28,30,50]. Two diet-only studies used CT [54] or BIA [53], while the exercise-only
intervention utilised DXA [51]. Anthropometric measures on their own were used predominantly in diet only studies, with the most frequent variables measured being MAMC in
12 (44%) and TSF in 11 (41%) studies. Some other anthropometric measures including calf
and thigh circumference were utilised alongside guideline-recommended measures.
3.2. Quality Assessment
Plots summarising the risk of bias are presented in Figures 2 and 3. For the 19 RCTs,
high risk of bias was most prevalent in domain 4 (bias in the measurement of the outcome),
where assessors were often not blinded to the intervention. Almost all studies were low
risk for domain 1 (randomisation and concealment processes). For domain 2, evaluating
if participants and/or interventionists were blinded to the intervention allocation, the
majority were allocated ‘some concerns’. For the eight non-randomised studies, high risk
of bias was most common in domain 3 (classification of interventions), because five of these
studies were uncontrolled with no group allocations.

Nutrients 2022, 14, x FOR PEER REVIEW
Nutrients 2022, 14, 3365

19 of 28

Figure 2. Risk of bias summaries for RCTs using Cochrane Risk of Bias 2 Tool [33–41,44,45,47,48,50,51,54–57].

3.3. Outcomes
Combined
Diet and
Intervention
Studies
Figure
2. Risk for
of bias
summaries
forExercise
RCTs using
Cochrane
Risk of Bias 2 Tool [33–
From the nine combined diet and exercise RCTs, four showed significant improve41,44,45,47,48,51,54,57].
ments in lean mass measured by CT [39], MRI [40], BIA [36], and quadricep ultrasound [35]
compared to controls. One study also observed significant reductions in fat mass [35].
Three of these four studies had similar interventions of supervised, moderate intensity
exercise (aerobic and/or resistance) on 3 days/week over 8–14 weeks plus targeted protein
intakes above 1.2 g/kg/day through either provision of oral nutrition supplements in
addition to diet, or dietetic counselling [35,37,40]. The intervention of the fourth diet and
exercise RCT [39] that demonstrated an increase in skeletal muscle mass relied on frequent
meals plus BCAA supplementation, with the exercise component being an increase in the
number of daily steps. The participants in the control arm of this study were exposed to the

14 of 23

Nutrients 2022, 14, 3365

20 of 28

Nutrients 2022, 14, x FOR PEER REVIEW

15 of 23

same diet and exercise intervention, but received a placebo instead of BCAAs, implicating
these in the improvement in muscle mass.

Figure 3. Risk of bias summaries for non-RCTs using Cochrane ROBINS-I (risk of bias tool to assess
non-randomised studies of interventions) [32,42,43,52,53,56,58,59].

Figure 3. Risk of bias summaries for non-RCTs using Cochrane ROBINS-I (risk of bias tool to assess
A combined
diet/exercise
RCT [44]
which used counselling for self-directed exercise
non-randomised
studies
of interventions)
[32,42,43,52,53,56,58,59].

and a high protein diet with BCAA supplementation demonstrated a significant improvement
compared
to a diet-only
group
in psoas muscle
3.3.
Outcomes
for Combined
Diet control
and Exercise
Intervention
Studiesindex via CT, but not in any
other measures of muscle/lean mass (CT, MRI, or DXA). While the intervention group
From the nine combined diet and exercise RCTs, four showed significant improvesignificantly increased daily number of steps compared to the control group, this small
ments in lean mass measured by CT [39], MRI [40], BIA [36], and quadricep ultrasound
study population (n = 17) may have limited the power to detect change in some measures.
[35] compared to controls. One study also observed significant reductions in fat mass [35].
This cohort of transplant candidates also had more advanced liver disease compared to the
Three of these four studies had similar interventions of supervised, moderate intensity
other combined interventions.
exercise
(aerobic
resistance)
on 3 days/week
over 8–14
weeks plus
targeted
Four
of theand/or
remaining
diet/exercise
RCTs reported
significant
increases
inprotein
muscle
intakes
above
1.2
g/kg/day
through
either
provision
of
oral
nutrition
supplements
admass, however this was only reported within study groups. These four studies usedineither
dition
to
diet,
or
dietetic
counselling
[35,37,40].
The
intervention
of
the
fourth
diet
and
supplements in combination with a diet and exercise intervention, (including HMB [38],
exercise
RCT [39]
that
demonstrated
an increase
in skeletal
muscle
mass
on frequent
non-alcoholic
beer
[37],
or the amino
acid leucine—a
BCAA
[33]);
orrelied
provided
dietetic
meals
plus
BCAA
supplementation,
with
the
exercise
component
being
an
increase
in the
counselling adjusted for BMI categories [47]. While two studies indicated good adherence
number
of and
dailyexercise
steps. The
participants
in the
armdid
of this
study adherence
were exposed
to
to the diet
interventions
[37,38],
thecontrol
other two
not report
[33,47].
the same
diet
and
exercise
intervention,
but
received
a
placebo
instead
of
BCAAs,
impliBoth of the non-randomised combined diet/exercise intervention studies found no
cating
thesechanges
in the improvement
muscle
mass. via BIA [42] or MAMC [43]. Both these
significant
in lean or fatinmass
measured
A
combined
diet/exercise
RCT
[44]
which
for participants
self-directedcomplete
exercise
studies had study population numbers of <40.used
Onecounselling
had only nine
and
high protein(39%
diet attrition
with BCAA
supplementation
a significant
improvethe aintervention
rate),
so sample sizesdemonstrated
may have been
too small to
identify
ment
compared
to a diet-only
group in psoas
muscle
index via CT,
but not
in any
significant
changes
[43]. The control
three single-arm
combined
diet/exercise
studies
assessed
other
measures
muscle/lean cirrhosis
mass (CT,
MRI, or
DXA).
While
the intervention
group
outpatients
withofcompensated
showed
mixed
results
[32,52,53].
An intervention
2
significantly
increased
daily
number
of
steps
compared
to
the
control
group,
this
small
targeting overweight participants (BMI > 26 kg/m ) with a 16-week program of supervised
study
population
(n
=
17)
may
have
limited
the
power
to
detect
change
in
some
measures.
exercised with reduced caloric intake observed a significant reduction fat mass with no
This
cohortchange
of transplant
also
had single-arm
more advanced
liver intervention
disease compared
to
significant
in lean candidates
mass [32]. A
second
combined
targeting
the
other combined
interventions.
increased
step activity
and BCAA supplementation reported a significant increase in muscle
Fourvia
ofBIA,
the remaining
diet/exercise
increases
inreported
muscle
volume
expressed only
as changeRCTs
ratioreported
[52]. The significant
final single-arm
study
mass,
however changes
this was in
only
reported
within
groups.
These four
studiessupplements
used either
no significant
skeletal
muscle
viastudy
CT, after
12 months
of BCAA
supplements in combination with a diet and exercise intervention, (including HMB [38],
non-alcoholic beer [37], or the amino acid leucine—a BCAA [33]); or provided dietetic
counselling adjusted for BMI categories [47]. While two studies indicated good adherence

Nutrients 2022, 14, 3365

21 of 28

and prescribed bench step activity [53]. Compliance was not reported, and this study was
limited by a small (n = 6) all-female cohort.
3.4. Outcomes for Diet-Only Intervention Studies
Three of nine diet-only RCTs found a significant increases in lean mass assessed by
MAMC [34,48,56], while another showed a decline in the control group without change in
the intervention cohort [51]. Okabayashi et al. [56] demonstrated an increase in MAMC in
an intervention group using a carbohydrate enriched BCAA supplement over 12 months,
combined with dietary advice to reduce energy intakes to offset the extra energy supplied
with the supplement. The aim was to match dietary energy intakes to the control participants who received no supplementation; however, no dietary compliance data were
reported. The second RCT [34] observed an increase in MAMC with a 12-week prescribed
high energy (35–40 kcal/kg/day), low sodium diet compared to usual diet. This was a
two-period cross over trial where two groups followed a prescribed diet and usual diet.
The within-group change data indicated both groups significantly increased MAMC after the prescribed diet, while MAMC either declined or remained stable with usual diet.
Compliance to the prescribed diet was reportedly high in both groups. The third study
was of hospitalised patients. This study utilised BCAAs versus a maltodextrin supplement
in the control group [48]. Short-term enteral nutrition was also provided in both groups
if hepatic encephalopathy occurred and continued until oral intake was well established.
The BCAA group had a significant within-group increase in MAMC, but not a significant
change compared to controls.
The five remaining RCTs of diet-only interventions mostly assessed lean mass using MAMC and found no significant changes with the intervention [45,50,54,55,57]. An
RCT [45] of inpatients receiving enteral nutrition for 4 weeks as a component of their
intervention found no significant changes in MAMC or TSF compared to inpatients receiving a usual hospital diet. A four week diet-only RCT [57] provided an isocaloric diet for
both intervention and controls (2000 kcal and 80 g of protein/day), with the intervention
group receiving BCAA supplementation and a reduced diet to achieve the same energy
and protein intake as the control. Only within-group changes were reported and no diet
compliance data were presented. MAMC did not significantly change in participants
given the BCAA supplement over 12 months compared to usual diet [50]. Average intakes
declined marginally in both groups even though BCAA compliance was satisfactory. A
study in transplant candidates [55] with a MAMC below the 25th percentile also saw no
improvement in this measure following supplementation and dietary counselling until
transplantation, versus dietary counselling alone. The RCT by Hirsch et al. [54]. provided
supplements over 12 months to patients with decompensated cirrhosis. While mean oral
intakes appeared significantly higher in the intervention versus control, there were no
significant changes in MAMC or TSF.
The final diet-only RCT [51] evaluated the effect of three diets in people with decompensated cirrhosis and ascites on lean and fat mass assessed by anthropometry. The first
diet (Group A) prescribed 24 h of parenteral nutrition in addition to a high energy and
protein diet with monthly dietitian advice. Group B received the same diet without parenteral nutrition while the third (control) group were prescribed a “sodium free” diet with
dietitian advice. The control group had a significant decline in TSF and MAC compared
to Groups A and B. MAMC was not reported in this study. Unfortunately, the control
group had mean dietary protein and energy (0.6 ± 3 g/kg/day and 25 ± 8 kcal/kg/day
respectively), considerably below guidelines for decompensated cirrhosis [16]. This is likely
the cause for the changes in the control group and highlights the potential negative impact
of a restrictive low sodium diet without a protein or energy prescription in patients with
decompensated cirrhosis.
Three non-randomised or single arm studies of diet-only interventions yielded mixed
results [49,58,59]. One non-randomised study [49] assessed patients who had undergone
surgery for HCC and reported a within-group increase in MAMC after 6 months of BCAA

Nutrients 2022, 14, 3365

22 of 28

supplementation twice daily, using a historical control group who received no supplementation. No MAMC data were reported for the control and MAMC was not compared
between groups. One of the two single-arm diet-only interventions [58] saw a significant
improvement in lean mass via BIA with a soy-based nutritional supplement over 8 weeks
plus usual diet. Dietary intake changes were not reported. Participants had predominantly
Child-Pugh A cirrhosis, allowing a reasonably reliable interpretation of BIA. The final
single-arm diet-only study [59] reported no significant change in skeletal muscle via CT.
The intervention of BCAA supplementation over 48-weeks was said to have 100% adherence to the supplement. Intramuscular adipose tissue was also assessed via CT with no
significant change observed.
3.5. Outcomes for Exercise-Only Interventions
The one exercise-only study [41] was an RCT involving 12 weeks of supervised
moderate intensity aerobic exercise 3 days/week, compared to a “sham intervention” of
relaxation exercises. The exercise group, which reported high attendance rates, significantly
increased lean mass via DXA. Additionally, there was a significant within-group reduction
in fat mass in the intervention group as well as an increase in upper thigh circumference
and reduction in mid-arm circumference. There were no significant changes in the “sham”
group, but comparisons between the active and sham groups were not reported.
4. Discussion
The aim of this systematic review was to assess the impact of diet and/or exercise
interventions on body composition in patients with liver cirrhosis. While published reviews exist on nutrition and/or exercise interventions in cirrhosis, there are none, to our
knowledge, that reviewed studies which specifically measured body composition across
both diet and exercise interventions. Secondly, this review also sought to determine the
effect of these interventions in patients with cirrhosis and obesity, given the increasing
prevalence of obesity and the risks associated with this [12], versus the potential deleterious
impact of calorie restriction on muscle mass in this population.
Unfortunately, the 27 studies identified for this review were too heterogeneous in
terms of design and outcome measures to allow meta-analysis. Small study size and
failure to report adherence to interventions also impacted data synthesis. Nonetheless,
on systematic review, the combined diet and exercise interventions appeared to show the
greatest potential to increase muscle mass. To demonstrate an increase in muscle mass with
exercise, these interventions needed to be of ≥8 weeks duration and comprise 30–60 min of
moderate intensity supervised exercise (aerobic and/or resistance), on at least 3 days per
week combined with protein intakes of 1.2–2 g/kg/day. In addition, there appeared to be a
benefit to muscle mass from BCAA supplementation [35,36,39,40]. Interestingly, several of
the combined RCTs [35,37–39,44] provided the control group with either a diet or exercise
intervention. Based on these studies it appears that there is a synergistic effect when both
diet and exercise interventions are delivered to increase muscle mass.
Obesity is known to impact patients with cirrhosis as an important contributor
to progression of liver disease. In patients undergoing liver transplant, severe obesity
(BMI > 35 kg/m2 ) increases the risk of peri-transplant complications and death [61]. The
prevalence of obesity is increasing in the whole population and in patients with advanced
liver disease, and so the impacts of obesity in patients with advanced liver disease are likely
to become increasingly important. A challenge addressing obesity in patients with cirrhosis
is that the catabolic metabolism found in advanced liver disease could potentially result in
significant muscle loss with calorie restriction. Of the 27 studies included in this systematic
review, 19 reported on dry weight BMI, with the mean BMI of patients in 13 of these studies
being in the overweight [33–40] or obese [32,41–44] ranges. Only two studies reported on
changes in fat mass [32,39]. In the RCT by Hernandez-Conde and colleagues [39], the mean
BMI of patients in intervention and control arm were in the overweight range. Although
weight loss was not a specific goal of their study, they showed that a combined intervention

Nutrients 2022, 14, 3365

23 of 28

of diet, exercise, and BCAA supplementation led to a reduction in fat mass while muscle
mass improved. The one study targeting weight loss in overweight and obese patients with
cirrhosis was promising in that it demonstrated a fall in body weight with maintenance
of lean mass after 16-weeks of a combined intervention of exercise with a reduced energy,
moderate protein diet [32].
In relation to the heterogeneity of studies, one issue that impacted the ability to
synthesise the findings of this review was that 10 different methods of assessing body
composition were used across the 27 studies. Current guidelines recommend CT or MRI as
optimal body composition assessment methods, in part because they are less impacted by
the fluid overload and ascites that occur in decompensated cirrhosis than some of the other
methods [16,19]. While CT/MRI are expensive and not always available, they are often
part of standard of care for patients undergoing transplant evaluation to assess hepatic
vasculature or for HCC monitoring. Although these routine measures are not performed
specifically to assess body composition, they can additionally be used to assess muscle and
fat mass; and it is possible for allied health clinicians to perform these analyses [62].
When abdominal CT or MRI are not available, guidelines recommend using DXA
and BIA to assess body composition, on the provision that fluid retention is not an issue [19]; however, this restricts their utility in the group of cirrhotic patients most at risk
of sarcopenia, those with decompensated disease. Muscle mass quantification by DXA
has been shown to correlate with CT in cirrhosis [63]. Ultrasound is promising yet requires further exploration in this population [6,16]. While the accuracy of BIA can be
affected by hydration [21], the use of Phase Angle from BIA may provide a more reliable
assessment of nutritional status in cirrhosis than other BIA modalities [19], with results
comparable to CT [64]. Several studies only utilised MAMC and TSF as outcome measures,
particularly in the diet only interventions. Anthropometry is routinely used in the clinical
assessment of nutritional status. However, the utility in clinical studies is less clear as
these measures suffer in regard to reliability [18], and cannot distinguish small changes
in body composition [7,26]. This makes them less than ideal for studies conducted over
8–12 weeks like a number of the studies reported here. Additional issues with their use
in the studies included in this review were that outcome assessors were frequently not
blinded to intervention arm, or for these very operator dependent measures, that several
assessors may have been involved in the serial measurements. This increases the impact of
interobserver variability on findings. Interestingly, while there is a strong body of evidence
indicating the deleterious effects of sarcopenia in cirrhosis, very few of the included studies
assessed if the patient’s level of baseline muscle mass was indicative of sarcopenia prior
to conducting the intervention. This highlights the need for future studies to evaluate
baseline muscle mass and therefore sarcopenia to understand the true effect of diet and/or
exercise interventions.
An additional issue was that most diet and exercise studies in cirrhosis have small
study populations with body composition measures generally underpowered and as mentioned, were often included as secondary outcomes of the studies. Some of the challenges
to increasing participant numbers in studies in this area are the complexities of conducting lifestyle interventions in a population with advanced liver disease who may be quite
unwell. Another factor which can impact drop-out rates in this population is inclusion
of participants who are potential transplant candidates. In one of the RCTs included in
this review [43], a 6-week exercise intervention was completed in just over half (56%) of
potential liver transplant recipients, and this was largely because of study participants
receiving a liver transplant rather than not adhering to the program. We faced a similar
issue in an 8-week pilot feasibility RCT of exercise in patients on a liver transplant waiting
list [65], only 50% of participants completed the study, largely because participants received
their liver transplant within the study period.
This review also highlighted the sparsity of relevant intervention studies which have
targeted patients with decompensated liver disease. Patients with decompensated disease
are a complex and high-risk population, who are more likely to experience muscle wasting

Nutrients 2022, 14, 3365

24 of 28

and adverse outcomes [6]. Chen et al. [44], is one of the few studies in this review that
included only decompensated cirrhosis patients that used a combined diet and exercise
intervention. This study was small; however, they were able to demonstrate that homebased exercise is safe in this population. This is promising, and future studies should
focus on these populations to better understand how body composition can be improved
pre-transplant to improve morbidity and mortality.
While this systematic literature review focused on changes in body composition measured using methodology validated in liver cirrhosis, there are other diet and or exercise
intervention studies that have added value to the management of patients with advanced
liver disease. Several exercise RCTs in patients with cirrhosis have measured aspects of
physical performance including strength, exercise capacity, and/or physical function and
therefore did not meet the inclusion criteria for this review. Measures such as hand grip
strength, anaerobic threshold by cardiopulmonary exercise testing and functional performance assessments such as the Short Physical Performance Battery and the Liver Frailty
Index have demonstrated associations with patient outcomes [5,66–68]. They can be useful
as screening tools to identify patients at risk of complications [69] and are recommended
as part of the evaluation of nutritional status in people with cirrhosis [16,19,26]. Consideration should be given to including these measures in future studies that address body
composition alongside functional status.
The field of diet and exercise interventions in patients with cirrhosis is obviously at an
early and evolving stage. An important goal for future studies is to determine the significance of modest improvements in body composition both in terms of clinical outcomes,
but also in patient-important outcomes and their quality of life. Given the potential range
and combination of diet and exercise interventions, defining minimal clinically important
differences for muscle and fat mass and thresholds for adverse outcomes patients should
be a goal to facilitate comparisons between interventions.
5. Conclusions
In summary, effective interventions to improve body composition in cirrhosis appear
more likely to succeed if diet and exercise components are combined. There remains a
paucity of studies in patients with cirrhosis and obesity despite the increasing prevalence of
obesity in this population. At present, the evidence supporting diet and exercise approaches
to improve body composition in cirrhosis is impacted by underpowered, short-term interventions. Future research should be directed at appropriately powered combined diet
and exercise RCTs of at least 8 weeks duration. Ideally assessments of changes in muscle
mass, particularly in patients with decompensated cirrhosis should rely on guidelinerecommended methods in this population, specifically CT or MRI. These studies should
ideally be large enough to allow for the potentially high rates of patient drop-out and include formal assessments of patient adherence to interventions to identify strategies that do
and do not work in this cohort. An important goal for future studies should be to determine
what are clinically meaningful changes in body composition in patients with cirrhosis as
this will facilitate comparison between intervention strategies. These approaches will help
clarify if sarcopenia and sarcopenic obesity are modifiable risk factors in cirrhosis.
Supplementary Materials: The following supporting information can be downloaded at: https:
//www.mdpi.com/article/10.3390/nu14163365/s1, Supplementary File S1: PRISMA 2020 Checklist
for Systematic Reviews; Supplementary File S2: Systematic review search strategy.
Author Contributions: Conceptualisation, H.E.J., T.G.T., I.J.H., and H.L.M.; methodology, H.E.J.,
T.G.T., J.T.K., I.J.H., and H.L.M.; software, H.E.J.; validation, H.E.J., G.A.M., I.J.H., and H.L.M.; formal
analysis, H.E.J., T.G.T., I.J.H., and H.L.M.; investigation, H.E.J., T.G.T., and H.L.M.; resources, H.E.J.,
I.J.H., and H.L.M.; data curation, H.E.J., T.G.T., and H.L.M.; writing—original draft preparation,
H.E.J., I.J.H., and H.L.M.; writing—review and editing, H.E.J., T.G.T., S.E.K., G.A.M., J.S.C., J.T.K.,
I.J.H., and H.L.M.; visualisation, H.E.J., I.J.H., G.A.M., and H.L.M.; supervision, I.J.H., G.A.M., and
H.L.M.; project administration, H.E.J. and H.L.M.; funding acquisition, H.E.J. All authors have read
and agreed to the published version of the manuscript.

Nutrients 2022, 14, 3365

25 of 28

Funding: Heidi Johnston was supported by an Australian Government Research Training Program
and Living Stipend Scholarship via the University of Queensland.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: The data that support the findings of this study are available from the
corresponding author upon reasonable request.
Acknowledgments: We thank the University of Queensland Librarian Marcos Riba for assistance
refining the search strategy process and the Department of Nutrition and Dietetics at the Princess
Alexandra Hospital for in-kind support.
Conflicts of Interest: The authors declare no conflict of interest.

References
1.
2.
3.
4.
5.
6.
7.

8.
9.
10.

11.

12.

13.
14.

15.

16.
17.
18.
19.

Moon, A.M.; Singal, A.G.; Tapper, E.B. Contemporary epidemiology of chronic liver disease and cirrhosis. Clin. Gastroenterol.
Hepatol. 2020, 18, 2650–2666. [CrossRef] [PubMed]
D’Amico, G.; Garcia-Tsao, G.; Pagliaro, L. Natural history and prognostic indicators of survival in cirrhosis: A systematic review
of 118 studies. J. Hepatol. 2006, 44, 217–231. [CrossRef]
Cheung, K.; Lee, S.S.; Raman, M. Prevalence and mechanisms of malnutrition in patients with advanced liver disease, and
nutrition management strategies. Clin. Gastroenterol. Hepatol. 2012, 10, 117–125. [CrossRef] [PubMed]
Bhanji, R.A.; Carey, E.J.; Yang, L.; Watt, K.D. The long winding road to transplant: How sarcopenia and debility impact morbidity
and mortality on the waitlist. Clin. Gastroenterol. Hepatol. 2017, 15, 1492–1497. [CrossRef] [PubMed]
Sinclair, M.; Poltavskiy, E.; Dodge, J.L.; Lai, J.C. Frailty is independently associated with increased hospitalisation days in patients
on the liver transplant waitlist. World J. Gastroenterol. 2017, 23, 899. [CrossRef]
Tandon, P.; Montano-Loza, A.J.; Lai, J.C.; Dasarathy, S.; Merli, M. Sarcopenia and frailty in decompensated cirrhosis. J. Hepatol.
2021, 75, S147–S162. [CrossRef]
Cruz-Jentoft, A.J.; Baeyens, J.P.; Bauer, J.M.; Boirie, Y.; Cederholm, T.; Landi, F.; Martin, F.C.; Michel, J.-P.; Rolland, Y.; Schneider,
S.M.; et al. Sarcopenia: European consensus on definition and diagnosisReport of the European Working Group on Sarcopenia in
Older People. Age Ageing 2010, 39, 412–423. [CrossRef]
Laube, R.; Wang, H.; Park, L.; Heyman, J.K.; Vidot, H.; Majumdar, A.; Strasser, S.I.; McCaughan, G.W.; Liu, K. Frailty in advanced
liver disease. Liver Int. 2018, 38, 2117–2128. [CrossRef]
Kim, G.; Kang, S.H.; Kim, M.Y.; Baik, S.K. Prognostic value of sarcopenia in patients with liver cirrhosis: A systematic review and
meta-analysis. PLoS ONE 2017, 12, e0186990. [CrossRef]
Van Vugt, J.; Levolger, S.; de Bruin, R.; van Rosmalen, J.; Metselaar, H.; IJzermans, J. Systematic review and meta-analysis
of the impact of computed tomography–assessed skeletal muscle mass on outcome in patients awaiting or undergoing liver
transplantation. Am. J. Transplant. 2016, 16, 2277–2292. [CrossRef]
Berzigotti, A.; Garcia-Tsao, G.; Bosch, J.; Grace, N.D.; Burroughs, A.K.; Morillas, R.; Escorsell, A.; Garcia-Pagan, J.C.; Patch, D.;
Matloff, D.S. Obesity is an independent risk factor for clinical decompensation in patients with cirrhosis. Hepatology 2011, 54,
555–561. [CrossRef] [PubMed]
Montano-Loza, A.J.; Angulo, P.; Meza-Junco, J.; Prado, C.M.; Sawyer, M.B.; Beaumont, C.; Esfandiari, N.; Ma, M.; Baracos, V.E.
Sarcopenic obesity and myosteatosis are associated with higher mortality in patients with cirrhosis. J. Cachexia Sarcopenia Muscle
2016, 7, 126–135. [CrossRef] [PubMed]
Spengler, E.K.; O’Leary, J.G.; Te, H.S.; Rogal, S.; Pillai, A.A.; Al-Osaimi, A.; Desai, A.; Fleming, J.N.; Ganger, D.; Seetharam, A.; et al.
Liver Transplantation in the Obese Cirrhotic Patient. Transplantation 2017, 101, 2288–2296. [CrossRef] [PubMed]
Vidot, H.; Kline, K.; Cheng, R.; Finegan, L.; Lin, A.; Kempler, E.; Strasser, S.I.; Bowen, D.G.; McCaughan, G.W.; Carey, S. The
relationship of obesity, nutritional status and muscle wasting in patients assessed for liver transplantation. Nutrients 2019, 11,
2097. [CrossRef]
Calzadilla-Bertot, L.; Jeffrey, G.P.; Jacques, B.; McCaughan, G.; Crawford, M.; Angus, P.; Jones, R.; Gane, E.; Munn, S.; Macdonald,
G. Increasing incidence of nonalcoholic steatohepatitis as an indication for liver transplantation in Australia and New Zealand.
Liver Transpl. 2019, 25, 25–34. [CrossRef]
Plauth, M.; Bernal, W.; Dasarathy, S.; Merli, M.; Plank, L.D.; Schütz, T.; Bischoff, S.C. European Society of Enteral and Parenteral
Nutrition Guideline on Clinical Nutrition in Liver Disease. Clin. Nutr. 2019, 38, 485–521. [CrossRef]
Morgan, M.Y.; Madden, A.M.; Soulsby, C.T.; Morris, R.W. Derivation and validation of a new global method for assessing
nutritional status in patients with cirrhosis. Hepatology 2006, 44, 823–835. [CrossRef]
Ulijaszek, S.J.; Kerr, D.A. Anthropometric measurement error and the assessment of nutritional status. Br. J. Nutr. 1999, 82,
165–177. [CrossRef]
European Association for the Study of the Liver. EASL Clinical Practice Guidelines on nutrition in chronic liver disease. J. Hepatol.
2019, 70, 172–193. [CrossRef]

Nutrients 2022, 14, 3365

20.
21.
22.
23.
24.
25.
26.

27.
28.
29.
30.
31.
32.

33.

34.
35.

36.

37.

38.

39.

40.

41.

42.
43.

26 of 28

Sinclair, M.; Hoermann, R.; Peterson, A.; Testro, A.; Angus, P.W.; Hey, P.; Chapman, B.; Gow, P.J. Use of dual X-ray absorptiometry
in men with advanced cirrhosis to predict sarcopenia-associated mortality risk. Liver Int. 2019, 39, 1089–1097. [CrossRef]
Morgan, M.Y.; Madden, A.M.; Jennings, G.; Elia, M.; Fuller, N.J. Two-component models are of limited value for the assessment of
body composition in patients with cirrhosis. Am. J. Clin. Nutr. 2006, 84, 1151–1162. [CrossRef] [PubMed]
Bowen, T.S.; Schuler, G.; Adams, V. Skeletal muscle wasting in cachexia and sarcopenia: Molecular pathophysiology and impact
of exercise training. J. Cachexia Sarcopenia Muscle 2015, 6, 197–207. [CrossRef] [PubMed]
Williams, F.R.; Berzigotti, A.; Lord, J.M.; Lai, J.C.; Armstrong, M.J. Impact of exercise on physical frailty in patients with chronic
liver disease. Aliment. Pharmacol. Ther. 2019, 50, 988–1000. [CrossRef] [PubMed]
Toshikuni, N.; Arisawa, T.; Tsutsumi, M. Nutrition and exercise in the management of liver cirrhosis. World J. Gastroenterol. 2014,
20, 7286–7297. [CrossRef] [PubMed]
Ooi, P.H.; Gilmour, S.M.; Yap, J.; Mager, D.R. Effects of branched chain amino acid supplementation on patient care outcomes in
adults and children with liver cirrhosis: A systematic review. Clin. Nutr. ESPEN 2018, 28, 41–51. [CrossRef]
Lai, J.C.; Tandon, P.; Bernal, W.; Tapper, E.B.; Ekong, U.; Dasarathy, S.; Carey, E.J. Malnutrition, Frailty, and Sarcopenia in
Patients With Cirrhosis: 2021 Practice Guidance by the American Association for the Study of Liver Diseases. Hepatology 2021, 74,
1611–1644. [CrossRef] [PubMed]
Page, M.J.; McKenzie, J.E.; Bossuyt, P.M.; Boutron, I.; Hoffmann, T.C.; Mulrow, C.D.; Shamseer, L.; Tetzlaff, J.M.; Akl, E.A.;
Brennan, S.E. The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. BMJ 2021, 372, n71. [CrossRef]
The EndNote Team. EndNote; Endnote X9; Clarivate: Philadelphia, PA, USA, 2013.
Ouzzani, M.; Hammady, H.; Fedorowicz, Z.; Elmagarmid, A. Rayyan—A web and mobile app for systematic reviews. Syst. Rev.
2016, 5, 210. [CrossRef]
Sterne, J.A.; Savović, J.; Page, M.J.; Elbers, R.G.; Blencowe, N.S.; Boutron, I.; Cates, C.J.; Cheng, H.-Y.; Corbett, M.S.; Eldridge, S.M.
RoB 2: A revised tool for assessing risk of bias in randomised trials. BMJ 2019, 366, l4898. [CrossRef]
Sterne, J.A.; Hernán, M.A.; McAleenan, A.; Reeves, B.C.; Higgins, J.P. ROBINS-I: A tool for assessing risk of bias in non-randomized
studies of interventions. BMJ. 2016, 355, i4919. [CrossRef]
Berzigotti, A.; Albillos, A.; Villanueva, C.; Genescá, J.; Ardevol, A.; Augustín, S.; Calleja, J.L.; Bañares, R.; García-Pagán, J.C.;
Mesonero, F. Effects of an intensive lifestyle intervention program on portal hypertension in patients with cirrhosis and obesity:
The SportDiet study. Hepatology 2017, 65, 1293–1305. [CrossRef] [PubMed]
Román, E.; Torrades, M.T.; Nadal, M.J.; Cárdenas, G.; Nieto, J.C.; Vidal, S.; Bascunana, H.; Juárez, C.; Guarner, C.; Córdoba, J.
Randomized pilot study: Effects of an exercise programme and leucine supplementation in patients with cirrhosis. Dig. Dis. Sci.
2014, 59, 1966–1975. [CrossRef] [PubMed]
Manguso, F.; D’ambra, G.; Menchise, A.; Sollazzo, R.; D’agostino, L. Effects of an appropriate oral diet on the nutritional status of
patients with HCV-related liver cirrhosis: A prospective study. Clin. Nutr. 2005, 24, 751–759. [CrossRef] [PubMed]
Zenith, L.; Meena, N.; Ramadi, A.; Yavari, M.; Harvey, A.; Carbonneau, M.; Ma, M.; Abraldes, J.G.; Paterson, I.; Haykowsky, M.J.
Eight weeks of exercise training increases aerobic capacity and muscle mass and reduces fatigue in patients with cirrhosis. Clin.
Gastroenterol. Hepatol. 2014, 12, 1920–1926.e2. [CrossRef]
Macías-Rodríguez, R.U.; Ilarraza-Lomelí, H.; Ruiz-Margáin, A.; Ponce-de-León-Rosales, S.; Vargas-Vorácková, F.; García-Flores, O.;
Torre, A.; Duarte-Rojo, A. Changes in hepatic venous pressure gradient induced by physical exercise in cirrhosis: Results of a
pilot randomized open clinical trial. Clin. Transl. Gastroenterol. 2016, 7, e180. [CrossRef]
Macías-Rodríguez, R.U.; Ruiz-Margáin, A.; Román-Calleja, B.M.; Espin-Nasser, M.E.; Flores-García, N.C.; Torre, A.; GaliciaHernández, G.; Rios-Torres, S.L.; Fernández-del-Rivero, G.; Orea-Tejeda, A. Effect of non-alcoholic beer, diet and exercise on
endothelial function, nutrition and quality of life in patients with cirrhosis. World J. Hepatol. 2020, 12, 1299. [CrossRef]
Lattanzi, B.; Bruni, A.; Di Cola, S.; Molfino, A.; De Santis, A.; Muscaritoli, M.; Merli, M. The Effects of 12-Week Beta-HydroxyBeta-Methylbutyrate Supplementation in Patients with Liver Cirrhosis: Results from a Randomized Controlled Single-Blind Pilot
Study. Nutrients 2021, 13, 2296. [CrossRef]
Hernández-Conde, M.; Llop, E.; Gómez-Pimpollo, L.; Carrillo, C.F.; Rodríguez, L.; Van Den Brule, E.; Perelló, C.;
López-Gómez, M.; Abad, J.; Martínez-Porras, J.L. Adding Branched-Chain Amino Acids to an Enhanced Standard-ofCare Treatment Improves Muscle Mass of Cirrhotic Patients With Sarcopenia: A Placebo-Controlled Trial. Off. J. Am. Coll.
Gastroenterol. 2021, 116, 2241–2249. [CrossRef]
Aamann, L.; Dam, G.; Borre, M.; Drljevic-Nielsen, A.; Overgaard, K.; Andersen, H.; Vilstrup, H.; Aagaard, N.K. Resistance
training increases muscle strength and muscle size in patients with liver cirrhosis. Clin. Gastroenterol. Hepatol. 2019, 18, 1179–1187.
[CrossRef]
Román, E.; García-Galcerán, C.; Torrades, T.; Herrera, S.; Marín, A.; Doñate, M.; Alvarado-Tapias, E.; Malouf, J.; Nácher, L.;
Serra-Grima, R. Effects of an exercise programme on functional capacity, body composition and risk of falls in patients with
cirrhosis: A randomized clinical trial. PLoS ONE 2016, 11, e0151652. [CrossRef]
Schmidt, N.P.; Fernandes, S.A.; Silveira, A.T.; Rayn, R.G.; Henz, A.C.; Rossi, D.; Galant, L.H.; Marroni, C.A. Nutritional and
functional rehabilitation in cirrhotic patients. J. Gastroenterol. Hepatol. Res. 2021, 10, 3470–3477.
Morkane, C.M.; Kearney, O.; Bruce, D.A.; Melikian, C.N.; Martin, D.S. An outpatient hospital-based exercise training program for
patients with cirrhotic liver disease awaiting transplantation: A feasibility trial. Transplantation 2020, 104, 97–103. [CrossRef]
[PubMed]

Nutrients 2022, 14, 3365

44.

45.

46.

47.

48.

49.

50.
51.

52.

53.

54.

55.
56.

57.
58.
59.

60.
61.

62.

63.

64.

27 of 28

Chen, H.W.; Ferrando, A.; White, M.G.; Dennis, R.A.; Xie, J.; Pauly, M.; Park, S.; Bartter, T.; Dunn, M.A.; Ruiz-Margain, A.
Home-Based Physical Activity and Diet Intervention to Improve Physical Function in Advanced Liver Disease: A Randomized
Pilot Trial. Dig. Dis. Sci. 2020, 65, 3350–3359. [CrossRef] [PubMed]
Dupont, B.; Dao, T.; Joubert, C.; Dupont-Lucas, C.; Gloro, R.; Nguyen-Khac, E.; Beaujard, E.; Mathurin, P.; Vastel, E.; Musikas, M.
Randomised clinical trial: Enteral nutrition does not improve the long-term outcome of alcoholic cirrhotic patients with jaundice.
Aliment. Pharmacol. Ther. 2012, 35, 1166–1174. [CrossRef]
Debette-Gratien, M.; Tabouret, T.; Antonini, M.-T.; Dalmay, F.; Carrier, P.; Legros, R.; Jacques, J.; Vincent, F.; Sautereau, D.; Samuel,
D. Personalized adapted physical activity before liver transplantation: Acceptability and results. Transplantation 2015, 99, 145–150.
[CrossRef]
Kruger, C.; McNeely, M.L.; Bailey, R.J.; Yavari, M.; Abraldes, J.G.; Carbonneau, M.; Newnham, K.; DenHeyer, V.; Ma, M.;
Thompson, R. Home exercise training improves exercise capacity in cirrhosis patients: Role of exercise adherence. Sci. Rep. 2018,
8, 99. [CrossRef]
Les, I.; Doval, E.; García-Martínez, R.; Planas, M.; Cárdenas, G.; Gómez, P.; Flavià, M.; Jacas, C.; Mínguez, B.; Vergara, M. Effects
of branched-chain amino acids supplementation in patients with cirrhosis and a previous episode of hepatic encephalopathy: A
randomized study. Off. J. Am. Coll. Gastroenterol. 2011, 106, 1081–1088. [CrossRef]
Okabayashi, T.; Nishimori, I.; Sugimoto, T.; Iwasaki, S.; Akisawa, N.; Maeda, H.; Ito, S.; Onishi, S.; Ogawa, Y.; Kobayashi, M.
The benefit of the supplementation of perioperative branched-chain amino acids in patients with surgical management for
hepatocellular carcinoma: A preliminary study. Dig. Dis. Sci. 2008, 53, 204–209. [CrossRef]
Poon, R.P.; Yu, W.C.; Fan, S.T.; Wong, J. Long-term oral branched chain amino acids in patients undergoing chemoembolization
for hepatocellular carcinoma: A randomized trial. Aliment. Pharmacol. Ther. 2004, 19, 779–788. [CrossRef]
Sorrentino, P.; Castaldo, G.; Tarantino, L.; Bracigliano, A.; Perrella, A.; Perrella, O.; Fiorentino, F.; Vecchione, R.; D’Angelo,
S. Preservation of nutritional-status in patients with refractory ascites due to hepatic cirrhosis who are undergoing repeated
paracentesis. J. Gastroenterol. Hepatol. 2012, 27, 813–822. [CrossRef]
Hiraoka, A.; Michitaka, K.; Kiguchi, D.; Izumoto, H.; Ueki, H.; Kaneto, M.; Kitahata, S.; Aibiki, T.; Okudaira, T.; Tomida, H.
Efficacy of branched-chain amino acid supplementation and walking exercise for preventing sarcopenia in patients with liver
cirrhosis. Eur. J. Gastroenterol. Hepatol. 2017, 29, 1416–1423. [CrossRef] [PubMed]
Nishida, Y.; Ide, Y.; Okada, M.; Otsuka, T.; Eguchi, Y.; Ozaki, I.; Tanaka, K.; Mizuta, T. Effects of home-based exercise and
branched-chain amino acid supplementation on aerobic capacity and glycemic control in patients with cirrhosis. Hepatol. Res.
2017, 47, E193–E200. [CrossRef] [PubMed]
Hirsch, S.; Bunout, D.; De La Maza, P.; Iturriaga, H.; Petermann, M.; Icazar, G.; Gattas, V.; Ugarte, G. Controlled trial on nutrition
supplementation in outpatients with symptomatic alcoholic cirrhosis. J. Parenter. Enter. Nutr. 1993, 17, 119–124. [CrossRef]
[PubMed]
Le Cornu, K.A.; McKiernan, F.J.; Kapadia, S.A.; Neuberger, J.M. A prospective randomized study of preoperative nutritional
supplementation in patients awaiting elective Orthotopic liver Transplantation. Transplantation 2000, 69, 1364–1369. [CrossRef]
Okabayashi, T.; Iyoki, M.; Sugimoto, T.; Kobayashi, M.; Hanazaki, K. Oral supplementation with carbohydrate-and branchedchain amino acid-enriched nutrients improves postoperative quality of life in patients undergoing hepatic resection. Amino Acids
2011, 40, 1213–1220. [CrossRef]
Tangkijvanich, P.; Mahachai, V.; Wittayalertpanya, S.; Ariyawongsopon, V.; Isarasena, S. Short-term effects of branched-chain
amino acids on liver function tests in cirrhotic patients. Southeast Asian J. Trop. Med. Public Health 2000, 31, 152–157.
Putadechakum, S.; Klangjareonchai, T.; Soponsaritsuk, A.; Roongpisuthipong, C. Nutritional status assessment in cirrhotic
patients after protein supplementation. Int. Sch. Res. Not. 2012, 2012, 690402. [CrossRef]
Kitajima, Y.; Takahashi, H.; Akiyama, T.; Murayama, K.; Iwane, S.; Kuwashiro, T.; Tanaka, K.; Kawazoe, S.; Ono, N.; Eguchi,
T. Supplementation with branched-chain amino acids ameliorates hypoalbuminemia, prevents sarcopenia, and reduces fat
accumulation in the skeletal muscles of patients with liver cirrhosis. J. Gastroenterol. 2018, 53, 427–437. [CrossRef]
Pugh, R.; Murray-Lyon, I.; Dawson, J.; Pietroni, M.; Williams, R. Transection of the oesophagus for bleeding oesophageal varices.
J. Br. Surg. 1973, 60, 646–649. [CrossRef]
Molina Raya, A.; García Navarro, A.; San Miguel Méndez, C.; Domínguez Bastante, M.; Villegas Herrera, M.T.; Granero, K.;
Becerra Massare, A.; Villar Del Moral, J.M.; Expósito, M.; Fundora Suárez, Y. Influence of Obesity on Liver Transplantation
Outcomes. Transplant. Proc. 2016, 48, 2503–2505. [CrossRef]
Johnston, H.E.; de Crom, T.; Hargrave, C.; Adhyaru, P.; Woodward, A.J.; Pang, S.; Ali, A.; Coombes, J.S.; Keating, S.E.; McLean, K.
The inter-and intrarater reliability and feasibility of dietetic assessment of sarcopenia and frailty in potential liver transplant
recipients: A mixed-methods study. Clin. Transplant. 2021, 35, e14185. [CrossRef]
Georgiou, A.; Papatheodoridis, G.V.; Alexopoulou, A.; Deutsch, M.; Vlachogiannakos, I.; Ioannidou, P.; Papageorgiou, M.-V.;
Papadopoulos, N.; Yannakoulia, M.; Kontogianni, M.D. Validation of cutoffs for skeletal muscle mass index based on computed
tomography analysis against dual energy X-ray absorptiometry in patients with cirrhosis: The KIRRHOS study. Ann. Gastroenterol.
2020, 33, 80. [CrossRef] [PubMed]
Ruiz-Margáin, A.; Xie, J.J.; Román-Calleja, B.M.; Pauly, M.; White, M.G.; Chapa-Ibargüengoitia, M.; Campos-Murguía, A.;
González-Regueiro, J.A.; Macias-Rodríguez, R.U.; Duarte-Rojo, A. Phase Angle From Bioelectrical Impedance for the Assessment
of Sarcopenia in Cirrhosis With or Without Ascites. Clin. Gastroenterol. Hepatol. 2021, 19, 1941–1949.e2. [CrossRef]

Nutrients 2022, 14, 3365

65.

66.
67.

68.

69.

28 of 28

Wallen, M.P.; Keating, S.E.; Hall, A.; Hickman, I.J.; Pavey, T.G.; Woodward, A.J.; Skinner, T.L.; Macdonald, G.A.; Coombes, J.S.
Exercise training is safe and feasible in patients awaiting liver transplantation: A Pilot Randomized Controlled Trial. Liver Transpl.
2019, 25, 1576–1580. [CrossRef] [PubMed]
Lai, J.C.; Feng, S.; Terrault, N.A.; Lizaola, B.; Hayssen, H.; Covinsky, K. Frailty predicts waitlist mortality in liver transplant
candidates. Am. J. Transplant. 2014, 14, 1870–1879. [CrossRef] [PubMed]
Sinclair, M.; Chapman, B.; Hoermann, R.; Angus, P.W.; Testro, A.; Scodellaro, T.; Gow, P.J. Handgrip strength adds more prognostic
value to the Model for End-Stage Liver Disease score than imaging-based measures of muscle mass in men with cirrhosis. Liver
Transpl. 2019, 25, 1480–1487. [CrossRef] [PubMed]
Tandon, P.; Tangri, N.; Thomas, L.; Zenith, L.; Shaikh, T.; Carbonneau, M.; Ma, M.; Bailey, R.J.; Jayakumar, S.; Burak, K.W.; et al. A
Rapid Bedside Screen to Predict Unplanned Hospitalization and Death in Outpatients With Cirrhosis: A Prospective Evaluation
of the Clinical Frailty Scale. Am. J. Gastroenterol. 2016, 111, 1759–1767. [CrossRef]
Lai, J.C.; Dodge, J.L.; Kappus, M.R.; Dunn, M.A.; Volk, M.L.; Duarte-Rojo, A.; Ganger, D.R.; Rahimi, R.S.; McCulloch, C.E.;
Haugen, C.E. Changes in frailty are associated with waitlist mortality in patients with cirrhosis. J. Hepatol. 2020, 73, 575–581.
[CrossRef]

ORIGINAL ARTICLE

The Journal of Nursing Research

▪ VOL. 32, NO. 1, FEBRUARY 2024

DOI: https://doi.org/10.1097/jnr.0000000000000595

Downloaded from http://journals.lww.com/jnr-twna by BhDMf5ePHKav1zEoum1tQfN4a+kJLhEZgbsIHo4XMi0hCywC
X1AWnYQp/IlQrHD3i3D0OdRyi7TvSFl4Cf3VC1y0abggQZXdgGj2MwlZLeI= on 10/29/2024

Effects of a Health Literacy Education Program
on Mental Health and Renal Function in
Patients With Chronic Kidney Disease: A
Randomized Controlled Trial
Hsiao-Ling HUANG1 • Ya-Hui HSU2 • Chung-Wei YANG3 • Min-Fang HSU4 • Yu-Chu CHUNG5*

ABSTRACT
Background: Chronic kidney disease (CKD) refers to permanent damage to the kidneys that occurs gradually over time.
Further progression may be preventable depending on its
stage.
Purpose: This study was developed to evaluate the effect of a
health literacy education program (HLEP) on mental health and
renal functioning in patients with CKD.
Methods: A single-blind, randomized controlled trial study was
conducted. Data were collected from March 25 to December
18, 2021. Participants were randomly assigned to either the experimental group (n = 42), which received multidisciplinary care
and HLEP, or the control group (n = 42), which received multidisciplinary care only. Data were collected at baseline (T1),
Month 3 (T2), and Month 6 (T3), and the data included patient
characteristics, estimated glomerular filtration rate, and responses to the Mandarin Multidimensional Health Literacy
Questionnaire and Beck Depression Inventory.
Results: After 6 months of the HLEP intervention, the results of
generalized estimating equations analysis showed that, compared with the control group, the experimental group had significantly higher health literacy at Month 3 (β = −3.37, 95% CI
[−5.68, −1.06]), significantly improved depression at Month 3
(β = −2.24, 95% CI [−4.11, −0.37]) and Month 6 (β = −4.36,
95% CI [−6.60, −2.12]), and a significantly higher estimated glomerular filtration rate at Month 6 (β = 5.87, 95% CI [1.35,
10.38]).
Conclusions/Implications for Practice: The findings of this
study may provide a reference for healthcare providers to educate patients with Stage 3–4 CKD using the HLEP. Positive effects on health literacy, depression, and renal function in patients with Stage 3–4 CKD were observed in the short term.
Findings from this study may facilitate the implementation of
multidisciplinary and nurse-led strategies in primary care to reinforce patients' health literacy, self-care ability, and adjustment
to CKD as well as delay disease progression.

KEY WORDS:
chronic kidney disease, depression, health literacy,
renal function, self-management health education.

Introduction
Chronic kidney disease (CKD) is a global public health issue
that affects many people around the world. Of the estimated
14.4% of adults in the United States who have CKD, up to
90% are unaware of their disease (U.S. Renal Data System,
2021). A study on 3,713 patients with Stage 3–4 CKD categorized the sample into four risk levels for renal failure
within 5 years: minimum risk (< 2%), low risk (2%–4%),
moderate risk (5%–14%), and high risk (≥ 15%). The study
concluded that up to 51% of patients in the moderate- and
high-risk populations were unaware of their CKD. Patients
have a much lower awareness of CKD than of diabetes and
hypertension (Chu et al., 2020). However, the onset and progression of a chronic disease tend to be slow, and patients
with CKD often have one or more concurrent chronic diseases (Elliott et al., 2020). The progression of chronic disease
also affects the mental health of patients. Studies have shown
that up to 75.5% of patients with CKD experience depression, which in turn compromises their quality of life (QOL;
Kunwar et al., 2020). This emphasizes the importance of aggressive screening and early intervention.
The definition of CKD includes all individuals with an estimated glomerular filtration rate (eGFR) < 60 ml/min per
1.73 m2 over a 3-month period, an albumin-to-creatinine ratio ≥ 30 mg/g, or other markers of kidney damage regardless
of kidney injury status. In those with CKD, the kidneys do
not function properly and treatment, for example, kidney
1
PhD, Associate Professor, Department of Healthcare Management,
Yuanpei University of Medical Technology • 2MSN, RN, Nephrology
Case Manager, Department of Internal Medicine, National Taiwan
University Hospital Hsin-Chu Branch • 3PhD, MD, Assistant Professor
and Attending Physician, Division of Nephrology, Department of
Internal Medicine, National Taiwan University Hospital Hsin-Chu
Branch • 4PhD, RN, Assistant Professor, Department of Nursing,
Yuanpei University of Medical Technology • 5PhD, RN, Professor,
Department of Nursing, Yuanpei University of Medical Technology.

Copyright © 2024 The Authors. Published by Wolters Kluwer Health,
Inc.
This is an open access article distributed under the Creative Commons
Attribution License 4.0 (CCBY), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.

1

The Journal of Nursing Research

Downloaded from http://journals.lww.com/jnr-twna by BhDMf5ePHKav1zEoum1tQfN4a+kJLhEZgbsIHo4XMi0hCywC
X1AWnYQp/IlQrHD3i3D0OdRyi7TvSFl4Cf3VC1y0abggQZXdgGj2MwlZLeI= on 10/29/2024

transplantation, is required to avoid kidney failure (National
Kidney Foundation, 2022). Therefore, delaying CKD progression is the key focus of medical care. In Taiwan, since
the enactment of the National Health Insurance's Healthcare
and Health Education for Pre-End Stage Renal Disease Patients Program in 2003, the high-risk population with CKD
(patients in Stages 3b–5) has been subject to case management and receives care from a professional medical team responsible for assessing each patient's self-care and self-management
abilities. Furthermore, this team follows up on patient health
status to help maintain residual renal function. As part of the
program, the CKD Clinical Practice Guidance recommends
initiation of a low-protein diet (0.6–0.8 g/kg a day) and ketogenic amino acid treatment for patients with Stage 3 CKD to
reduce kidney damage caused by nitrogenous wastes and delay dialysis or death (Xu, 2015).
Health literacy is a relatively new field of research in the area
of medicine and health. Different age groups require different
health literacy programs to empower them to implement beneficial health behaviors (Quaglio et al., 2017). A systematic review by Sørensen et al. (2012) defined health literacy as the
“knowledge, motivation, and competencies of accessing, understanding, appraising, and using health-related information
within the healthcare, disease prevention and health promotion setting in daily life to make judgment and decisions in
order to maintain or improve the overall QOL.” Therefore,
health literacy influences self-care efficacy and disease prognosis. Lack of health literacy has been reported in 17.7% of
patients with Stage 1–5 CKD (Schrauben et al., 2020), and
low or absent health literacy has been reported in 22.5% of
patients with Stage 3–4 CKD (Hanpaiboon & Pratoomsoot,
2019). Wei et al. (2017) has evaluated the validity of the Mandarin Multidimensional Health Literacy Questionnaire
(MMHLQ) on a sample of 2,394 adults in Taiwan, finding
the highest score in the domain of “understanding health information,” followed by “accessing health information,”
“communication and interaction,” “applying health information,” and “appraising health information.” Factors that affect health literacy include age, gender, educational level, marital status, spouse cohabitation status, family income, CKD
stage, duration of CKD, and number of comorbidities (Y.-C.
Chen et al., 2018; Hanpaiboon & Pratoomsoot, 2019; Wong
et al., 2018). Patients with higher health literacy show better
self-care behaviors (Schrauben et al., 2020).
Patients with CKD tend to experience depression because
they are forced to attend numerous hospital visits and face
complex treatment plans, drug side effects, dietary restrictions, and uncontrollable clinical symptoms as their disease
progresses (S. F. V. Wu et al., 2018). The prevalence of depression is related to CKD stage. A meta-analysis of 22 studies that investigated the correlation between depression and
death in patients with CKD reported an average depression
prevalence of 27.4% in predialysis patients with CKD
(Palmer et al., 2013). Patients with CKD experiencing depression exhibit poor compliance to drug treatment and
poor QOL, resulting in increased utilization of medical
2

Hsiao-Ling HUANG et al.

resources and higher rates of morbidity and mortality
(Palmer et al., 2013).
A study conducted by S. F. V. Wu et al. (2018) applied an innovative health education program promoting self-management
in a sample of patients with Stage 3b–5 CKD. The program
delivered one 100-minute session per week for 4 weeks, and
the participants were followed up for 3 months. Outcomes
included significantly improved blood urea, nitrogen, and
creatinine; reduced depression; and higher self-efficiency
and self-management. However, the intervention had no effect
on eGFR. Wang et al. (2018) conducted a cross-sectional study
to compare the effect of participation in a comprehensive
healthcare program on self-care behaviors and kidney function
in patients with CKD. The results revealed a slower rate of deterioration in kidney function and better self-management behaviors in patients participating in the healthcare program.
Machida et al. (2019) studied the effects of a 1-week inpatient
education program on kidney function in patients with Stage
3–5 CKD. The patients were followed from 6 months before
hospitalization to 24 months after discharge. Implementation
of the program delayed kidney function deterioration during
the 2-year observation period, especially in patients with low
proteinuria (urinary protein < 0.5 g/gCr). Thus, the authors recommended the program be initiated in patients with low proteinuria. A randomized clinical trial conducted by Lin et al.
(2021) investigated patients with Stage 1–3a CKD, with the
study group receiving routine care and health coaching for
6 weeks in addition to 12 months of postintervention followup. The findings indicate health coaching improves QOL,
self-management, patient activation, and self-efficiency.
On the basis of the evidence in the literature, health education
programs are beneficial to patients with CKD. Health literacy
can influence self-care behaviors and renal function in these patients. However, there is a lack of rigorous research on the impact of health literacy on the psychology of patients with
CKD. Therefore, the aim of this study was to investigate patients
with Stage 3–4 CKD (the largest group in Taiwan's Pre-End
Stage Renal Disease Patients Program), develop a health literacy
education program (HLEP), and evaluate the effect of this program on participants' mental health and renal function. This
study addressed the following research hypotheses:
1. Patients with CKD who participate in the HLEP will have increased health literacy compared their nonparticipant peers.
2. Patients with CKD who participate in the HLEP will have
improved depression compared their nonparticipant peers.
3. Patients with CKD who participate in the HLEP will have improved renal function compared their nonparticipant peers.

Methods
Design
The study design was a single-center, two-group, single-blinded,
randomized controlled trial with a repeated-measures design.

HLEP Improves Depression and Slow Renal Function

Downloaded from http://journals.lww.com/jnr-twna by BhDMf5ePHKav1zEoum1tQfN4a+kJLhEZgbsIHo4XMi0hCywC
X1AWnYQp/IlQrHD3i3D0OdRyi7TvSFl4Cf3VC1y0abggQZXdgGj2MwlZLeI= on 10/29/2024

The participants were randomized into either the HLEP with
multidisciplinary care (experimental) group (EG) or the multidisciplinary care (control) group (CG). Block randomization with
1:1 allocation was conducted using a computer-generated sequence and was performed by one of the authors not involved in screening, patient recruitment, clinical care, or data
collection using a random number generator. Sequentially
numbered, opaque, sealed envelopes were used to conceal the
sequence until the interventions were assigned at an outpatient
nephrology clinic. Patients were followed for 6 months. Data
were collected before health education (T1) and at 3 months
(T2) and 6 months (T3) after completion of the HLEP (Figure 1).

Setting
The participants in this study were conveniently sampled
from the nephrology outpatient clinic of a 988-bed regional
educational hospital in northern Taiwan.

VOL. 32, NO. 1, FEBRUARY 2024

to communicate in Mandarin or Taiwanese, were diagnosed
by a nephrologist with Stage 3 or 4 CKD, and had received
less than 1 year of comprehensive care. The exclusion criteria
included having a cognitive disorder or mental illness (severe
depression, schizophrenia), being on routine hemodialysis,
or current hospitalization.

Sample Size
Following Wang et al. (2018), minimal sample size was calculated using G*Power V3.1 statistical software with eGFR
as the primary efficacy variable (EG: eGFR = 0.072 ± 8.212,
n = 118; CG: eGFR = −2.978 ± 8.680, n = 117). The effect
size was estimated as 0.36, α was set at .05, and power
was set at 0.95. The participants were divided into two
groups with three measurements each. The minimum sample
size was calculated as 70. Assuming a follow-up loss of 20%,
the final sample size was set as 84 (42 per group).

Participants

Experimental Intervention and the
Control Group

Patients who met the selection criteria were recruited. The inclusion criteria were patients aged ≥ 20 years who were able

The main components of the HLEP are shown in Table 1.
The HLEP included a self-management health education

Figure 1
Research Design Flowchart

Note. T1: demographic characteristics, Mandarin Multidimensional Health Literacy Questionnaire (MMHLQ), Beck Depression Inventory-II
(BDI-II), and estimated glomerular filtration rate (eGFR) were collected. T2 and T3: MMHLQ, BDI-II, and eGFR were collected.

3

The Journal of Nursing Research

Hsiao-Ling HUANG et al.

Downloaded from http://journals.lww.com/jnr-twna by BhDMf5ePHKav1zEoum1tQfN4a+kJLhEZgbsIHo4XMi0hCywC
X1AWnYQp/IlQrHD3i3D0OdRyi7TvSFl4Cf3VC1y0abggQZXdgGj2MwlZLeI= on 10/29/2024

manual and a dietary health education video designed for patients with Stage 3–4 CKD. The health education manual
was developed by the researchers based on the Health Literacy Concept and Material Preparation Guide (National
Health Insurance Administration, Ministry of Health and
Welfare, Taiwan, ROC, 2020) and frequently asked questions from patients and family members. The preliminary review focused on the content and format of the draft, with
subsequent revisions made based on comments (Devellis,
2016). Five experts were invited to assess the content validity, with the content validity index assessed in terms of “appropriateness,” “accuracy,” and “readability” as .98, .88,
and .95, respectively, on a 5-point Likert scale, with an overall content validity index of .93. The content of the health education manual was edited based on the experts' comments
to create the final version. In the interests of portability, the
size of the health education manual was designed as
145 mm in length and 210 mm in width with 15 pages. As
most of the participants in this study were older adults, the
dietary principles were presented in video format. To maximize learning outcomes, an attending physician from the department of nephrology and one of the researchers personally introduced the dietary principles for patients with Stage
3–4 CKD based on the health education manual. Media professionals were hired to produce the video and sound recordings using PhotoImpact and Adobe Audition for conversion
to MP3. The video was designed with a minimum of text and

used simple words, pictures, cartoon figures, large fonts, and
interactive images.
In the EG, members received one-on-one health education
from a study team member with 6 years of experience in kidney disease nursing. The HLEP was delivered using a health
education manual in the nephrology outpatient health education classroom. After each session, the participants and their
families watched a health education video and were encouraged to ask questions until they fully understood the concepts.
In addition, EG participants were taught how to access the
videos via their smartphones on YouTube or by scanning a
QR code on the cover of the health education manual.
CG participants received routine one-on-one health education from a case manager at the participating hospital
who explained the blood analysis results and precautions
and distributed an A4-sized health education leaflet.

Data Collection
Data for this study were collected from March 25 to
December 18, 2021. The data were collected at three time
points: before HLEP implementation (T1) and at 3 months
(T2) and 6 months (T3) after HLEP. For participants who
were illiterate or had difficulty reading and thus not able to
complete the questionnaire independently, a designated staff
member explained the questionnaire and assisted them to
complete it based on their answers. EG participants received

Table 1

Main Contents of the Health Literacy Education Program
Time
Day 1

Main Issue

Main Content

Chronic kidney disease self-management health
education manual (one-on-one health education:
20 minutes)

▪ Introduction of kidney function
▪ Chronic kidney disease and stage
▪ Importance of regular medication
▪ Importance of regular exercise
▪ Blood pressure control
▪ Dietary principles for Stage 3–4 CKD

Stage 3–4 CKD dietary health education video
(10 minutes)

▪ Low protein, enough calories
▪ Dietary principles of potassium restriction
▪ Low-sodium diet
▪ The strategies of blood sugar control and
glycated hemoglobin (HbA1c)
▪ Prevent hyperlipidemia
▪ Dietary limitation
▪ Examples of three-meal recipes and substitutions,
types of low-protein starches

Week 5, 9, 13, 17, 21 Telephone consultation (5–10 minutes)

▪ Remind patients to watch health education video
per month
▪ Discuss video content and dietary principles
▪ Encourage patients to ask questions

Week 12, 24

▪ Discuss health education content and
blood test value
▪ Clarify patient concerns

Routine outpatient follow-up

Note. CKD = chronic kidney disease.

4

HLEP Improves Depression and Slow Renal Function

multidisciplinary care, participated in the HLEP, and conducted monthly phone discussions with the researcher about
the program's content on the first Monday of each month.
CG participants received multidisciplinary care, and their data
were collected at the same time points as EG participants.
Downloaded from http://journals.lww.com/jnr-twna by BhDMf5ePHKav1zEoum1tQfN4a+kJLhEZgbsIHo4XMi0hCywC
X1AWnYQp/IlQrHD3i3D0OdRyi7TvSFl4Cf3VC1y0abggQZXdgGj2MwlZLeI= on 10/29/2024

Ethical Considerations
This study was approved by the research ethics committee of
National Taiwan University Hospital Hsinchu Branch (Approval No. 91T-27-0026) before initiation. The investigator
explained the study purpose and procedure to the participants before they signed informed consent. The participants
were informed they were free to withdraw at any time during
the study and that their withdrawal would not affect their
treatment or cause any negative impact. The study data were
coded and analyzed anonymously.

VOL. 32, NO. 1, FEBRUARY 2024

used to determine depression status. The BDI-II has excellent
validity (H. Y. Chen, 2000), and the internal consistency
Cronbach's α value in this study was .82.

Renal function
In this study, eGFR was used to monitor kidney function.
Data were collected from medical records, and eGFR was
calculated using the Modification of Diet in Renal Disease
simplified equation developed by the Modification of Diet
in Renal Disease Study Group (Levey et al., 2007). The participating hospital scheduled visits for the patients every
3 months, and one blood sample was collected based on
the guidelines of the comprehensive care program for kidney
disease. Blood samples were collected from the patient during the week before the scheduled visit date.

Data Analysis
Instruments
Demographic and disease characteristics
The patient characteristics considered in this study included
age, gender, educational level, marital status, monthly income, CKD stage, chronic disease history, and duration of
treatment in the nephrology department.
Mandarin Multidimensional Health Literacy
Questionnaire
The MMHLQ developed specifically for adults in Taiwan by
Wei et al. (2017) was used in this study. The 20 items of the
MMHLQ assess health information, health information
comprehension, health information appraisal, health information application, communication, and interaction using
a 4-point scale: 1 = very difficult, 2 = difficult, 3 = easy, and
4 = very easy. Total and subscale scores are converted to a
0–50 range using the equation (Mean − 1)  (50/3), with
0–25 indicating insufficient, 25–33 indicating limited,
33–42 indicating sufficient, and 42–50 indicating excellent
level of health literacy. Higher scores on the MMHLQ indicate better health literacy. The internal consistency reliability
analysis revealed good internal consistency (Wei et al., 2017),
with a Cronbach's alpha of .92 in this study.
Beck Depression Inventory
The 21-item Beck Depression Inventory-II (BDI-II) Chinese
version was used in this study to measure depression. Items
are scored on a 4-point Likert scale, with 0 = no, 1 = mild,
2 = moderate, and 3 = severe symptoms. The participant
chooses the statement in each item that best describes how
they felt over the past 2 weeks (including the day of the examination). The total score ranges between 0 and 63, with
0–13 indicating normal emotion, 14–19 indicating mild depression, 20–28 indicating moderate depression, and 29–63
indicating severe depression. The BDI-II is aligned with the
diagnostic principles of depression in the Diagnostic and Statistical Manual Disorders, Fourth Edition and thus may be

IBM SPSS Statistics for Windows 20.0 (IBM Inc., Armonk,
NY, USA) was used for data archiving and statistical analysis, with results presented as frequency, percentage, mean,
and standard deviation. The demographic and disease characteristics were compared between the groups using the w2
test, independent samples t test, and paired samples t test.
The outcome variables, including health literacy, depression,
and eGFR, were compared between the two groups using a
generalized estimating equation with repeated measures.

Results
Eighty-four participants completed the study (0% attrition),
with 42 each in the EG and CG. Age ranged from 30 to
87 years and averaged 65.39 (SD = 11.39) years. No significant differences between the two groups were found in terms
of demographic and disease characteristics (Table 2).
As shown in Table 3, no significant differences between
the two groups were found in terms of health literacy, depression, or eGFR before the HLEP intervention. The average
MMHLQ score of the participants before health education
was categorized as “limited.” The highest scores for both
groups were reported in the domain of “understanding
health information,” followed by “communication and interaction,” “applying health information,” “appraising
health information,” and “accessing health information.”
In terms of depression, only 16 (19.1%) participants reported depression before the intervention, including 15
(17.9%) with mild and one (1.2%) with severe depression.
There were 31, 10, 0, and 1 participant in the EG and 37,
5, 0, and 0 participants in the CG with normal, mild, moderate, and severe depression, respectively, with no significant
between-group difference in depression noted (w2 = 10.129,
p = .928). The MMHLQ, depression, and eGFR scores over
time for the two groups are presented in Figure 2, and a summary of the GEE results for MMHLQ, depression, and eGFR
is shown in Table 4. A model with an exchangeable correlation matrix and model-based estimates of variance was used.
5

The Journal of Nursing Research

Hsiao-Ling HUANG et al.

Table 2

Homogeneity Test of Demographic and Clinical Characteristics (N = 84)
Variable

Overall (N = 84)

EG (n = 42)

CG (n = 42)

Downloaded from http://journals.lww.com/jnr-twna by BhDMf5ePHKav1zEoum1tQfN4a+kJLhEZgbsIHo4XMi0hCywC
X1AWnYQp/IlQrHD3i3D0OdRyi7TvSFl4Cf3VC1y0abggQZXdgGj2MwlZLeI= on 10/29/2024

n

%

n

%

n

%

Gender
Female
Male

17
67

20.2
79.8

10
32

23.8
76.2

7
35

16.7
83.3

Age, years
≤ 64
65–74
≥ 75

35
33
16

41.7
39.3
19.0

18
16
8

42.9
38.1
19.0

17
17
8

40.5
40.5
19.0

Education
High school and below
Junior college
University and above
Marital status
Not married
Married

55
9
20
7
77

65.5
10.7
23.8
8.3
91.7

24
6
12
2
40

57.1
14.3
28.6
4.8
95.2

31
3
8
5
37

51.2
14.3
14.3
4.8
9.5
5.9

17
5
7
4
6
3

40.5
11.9
16.7
9.5
14.3
7.1

26
7
5
0
2
2

61.9
16.6
11.9
0.0
4.8
4.8

CKD stage
3a
3b
4

18
38
28

21.4
45.2
33.4

8
17
17

19.0
40.4
40.5

10
21
11

23.8
50.0
26.2

25.0
42.9
32.1

13
17
12

31.0
40.4
28.6

8
19
15

0.66

.588

2.30

.512

2.69

.260 a

1.40

.433 a

0.75

.119 a

1.93

.381

1.64

.442 a

0.66

.415 a

1.20

.274

0.00

1.000

17.71

.314

80.00

.447

11.9
88.1

43
12
12
4
8
5

21
36
27

p

73.8
7.1
19.1

Monthly income (NT$)
No
≤ 20,000
20,001–39,999
40,000–49,999
50,000–59,999
≥ 60,000

Chronic disease history
1 category
2 categories
≥ 3 categories

x2

19.1
45.2
35.7

Hypertension
No
Yes

17
67

20.2
79.8

10
32

23.8
76.2

7
35

16.7
83.3

Diabetes
No
Yes

39
45

46.4
53.6

22
20

52.4
47.6

17
25

40.5
59.5

Gout
No
Yes

54
30

64.3
35.7

27
15

64.3
35.7

27
15

64.3
35.7

Albumin (g/dl)
Normal (3.5–5.7)
Abnormal (< 3.5, > 5.7)

81
3

96.4
3.6

41
1

97.6
2.4

40
2

95.2
4.8

Proteinuria (g/dl)
Normal (≤ 150)
Abnormal (> 150)

23
61

27.6
72.4

9
33

21.6
78.4

14
28

33.6
66.4

Note. EG = experimental group; CG = control group; CKD = chronic kidney disease.
a
Fisher's exact test.

After adjusting for age and gender, relationships between
health literacy and, respectively, time, group, and Time 
Group interaction were explored. The model showed the time
6

effect as more significant for T2 and T3 compared with T1.
Trend differences (interactions between time and group) revealed significant differences in health literacy at T2,

HLEP Improves Depression and Slow Renal Function

VOL. 32, NO. 1, FEBRUARY 2024

Table 3

Homogeneity Test of MMHLQ, Depression, and eGFR (N = 84)
Variable

Downloaded from http://journals.lww.com/jnr-twna by BhDMf5ePHKav1zEoum1tQfN4a+kJLhEZgbsIHo4XMi0hCywC
X1AWnYQp/IlQrHD3i3D0OdRyi7TvSFl4Cf3VC1y0abggQZXdgGj2MwlZLeI= on 10/29/2024

MMHLQ/overall
Accessing health information
Understanding health information
Appraising health information
Applying health information
Communication and interaction

EG (n = 42)

CG (n = 42)

t

p

Mean

SD

Mean

SD

30.06
25.60
35.12
27.27
29.66
32.64

5.78
11.55
6.96
6.96
7.71
6.03

28.06
23.61
34.92
24.70
26.69
30.36

7.27
12.47
8.73
8.60
10.62
6.39

−1.40
−0.76
−0.12
−1.51
−1.47
−1.68

.166
.452
.909
.135
.145
.096

Depression

9.29

6.99

7.12

4.72

−1.67

.100

eGFR (ml/min per 1.73 m2)

34.76

12.69

37.16

10.27

0.88

.384

Note. MMHLQ = Mandarin Multidimensional Health Literacy Questionnaire; eGFR = estimated glomerular filtration rate; EG = experimental group; CG = control
group.

indicating time-dependent growth effects. Trend differences
revealed significant improvements in depression scores for
the EG at T2 and T3, indicating time-dependent growth effects. In terms of eGFR, the time effect was more significant
at T3 compared with T1. Trend differences revealed a significant increase in eGFR for the EG at T3, indicating
time-dependent growth effects. The overall mean eGFR at
pretest (T1) was 35.96 (SD = 11.53) ml/min per 1.73 m2
and, at T3, had increased in the EG from 34.96 ml/min per
1.73 m2 to 38.83 ml/min per 1.73 m2 and decreased in the
CG from 37.16 ml/min per 1.73 m2 to 35.35 ml/min per
1.73 m2.

Discussion
The participant characteristics align with the general population with Stage 3–4 CKD in Taiwan, with a predominantly
male (79.8%) and Stage 3 CKD (66.6%) study group, averaging 65.39 years old (SD = 11.39). According to statistics
from the 2016 to 2018 kidney disease in Taiwan annual report, CKD is most prevalent in older adult men, with Stage
3 being the most commonly observed stage (National
Health Research Institutes & Taiwan Social of Nephrology,
2020). The top three concomitant chronic diseases reported
by the participants were hypertension (79.8%), diabetes
(53.0%), and gout (35.7%). The largest percentage (40.5%)
had two concomitant diseases, 32.1% had three, and 25%
had one. These findings are consistent with previous statistical
data on patients with CKD, who frequently report one to two
concomitant chronic diseases (Elliott et al., 2020). According
to the 2020 kidney disease in Taiwan annual report, the top
three concomitant diseases reported in the previous year by
new dialysis patients were hypertension, heart disease, and diabetes (National Health Research Institutes & Taiwan Social
of Nephrology, 2020). These data were based on the kidney
biopsy results presented in the kidney disease in Taiwan annual report, in which nephrotic syndrome and proteinuria of
unknown cause accounted for 46.4% of the cases, and

20.2% of acute renal injury was caused by the improper use
of nonsteroidal anti-inflammatory drugs. Nonsteroidal
anti-inflammatory drugs are the most used drugs by patients
with gout (National Health Research Institutes & Taiwan
Social of Nephrology, 2020).
The health literacy of the participants before the intervention was “limited,” which is similar to the result reported by
the National Health Research Forum (2020). Up to 50% of
older patients lack sufficient health information for health
decision making because of an insufficient level of health literacy. The MMHLQ used in this study returned the highest
score for both groups in the “understanding health information” domain, followed by “appraising health information”
and “accessing health information.” Only the “understanding health information” domain achieved a “sufficient”
level, whereas the other domains were “limited.” This finding differs from the results of several previous studies (Wei
et al., 2017; C. L. Wu et al., 2020). Wei et al. reported the
highest mean MMHLQ score in the “understanding health
information” domain, followed by the “accessing health information,” “communication and interaction,” “applying
health information,” and “appraising health information”
domains. C. L. Wu et al., investigating the health literacy of
458 patients from eight patient support groups, reported
the highest mean MMHLQ score in the “understanding
health information” domain, followed by the “communication and interaction,” “accessing health information,” “applying health information,” “understanding health information,” and “appraising health information” domains. The
difference in findings may be attributed to differences in
study purposes and samples. Nevertheless, health education
materials and approaches for patients with CKD should emphasize these four domains of MMHLQ, with particular focus given to the health information domain.
Health literacy improved significantly in the EG between
T1 and T2 and remained relatively unchanged between T2
and T3. One possible reason for this outcome is that
65.5% of the participants had an educational level below
7

The Journal of Nursing Research

Figure 2
Group Comparisons of Outcomes at Baseline,
Month 3, and Month 6

Downloaded from http://journals.lww.com/jnr-twna by BhDMf5ePHKav1zEoum1tQfN4a+kJLhEZgbsIHo4XMi0hCywC
X1AWnYQp/IlQrHD3i3D0OdRyi7TvSFl4Cf3VC1y0abggQZXdgGj2MwlZLeI= on 10/29/2024

Note. MMHLQ = Mandarin Multidimensional Health Literacy
Questionnaire; eGFR = estimated glomerular filtration rate.

the college level and were predominantly older individuals
who likely required more time to learn. The level of health
literacy before health education was “limited” for both
groups and had improved to “sufficient” for the EG at T2
and T3 and to “sufficient” for the CG only at T3. The EG
showed a more significant growth trend in health literacy
compared with the CG, indicating that an even longer
follow-up period is necessary to determine the long-term effects of the intervention.
8

Hsiao-Ling HUANG et al.

Before the intervention, only 19.1% of the participants reported experiencing depression, with 17.9% having mild depression and 1.2% having severe depression. This differs
from Loosman et al. (2015), which found a 34% prevalence
of depression among patients with Stage 3b–4 CKD and a
mean eGFR of 20.4 (SD = 6.3) ml/min per 1.73 m2 at baseline. Their findings suggest that the proportion of patients
with depression rises as renal function declines. The prevalence of depression among the participants in this study
was relatively low, which may be explained by most participants (65.5%) being in Stage 3 CKD and the overall mean
eGFR at baseline being 35.96 ± 11.53 ml/min per 1.73 m2,
indicating better renal function. At T2 and T3, depression
had improved significantly in the EG, which is similar to S.
F. V. Wu et al. (2018), who also employed an innovative
self-management program on a group of patients with Stage
3b–5 CKD. In their study, depression decreased significantly
at 3 months after the intervention, which included individual
health education every 3 months during the study period and
a visit and phone interview on the first week of each month.
Regular contact and establishing excellent rapport helped
make subjects feel concern for their physical and mental issues, which further improved their depression.
The participants in this study included 65.5% in Stage 3
and 34.5% in Stage 4 CKD. According to the CKD prevalence in adults in a cohort study conducted from 2015 to
2018, 5.8% patients had Stage 3, 0.4% had Stage 4, and
0.1% had Stage 5 CKD, with Stage 3 comprising the highest
proportion (U.S. Renal Data System, 2021). In our study, after the intervention, eGFR increased consistently with time in
the EG and decreased significantly with time in the CG until
T3, at which time the eGFR was significantly higher in the
EG than the CG. This result supports the recommendations
of S. F. V. Wu et al. (2018). In their study, the EG was enrolled in an innovative self-care program, and no significant
improvement was observed in eGFR after 3 months of follow-up. The authors suggested that the follow-up period
should be extended to 6–12 months to better detect the effect. Wang et al. (2018) reported similar results in their study
investigating patients with Stage 1–5 CKD. In patients receiving more than 1 year of comprehensive CKD care, the
healthcare program group attained a 2.83-fold higher likelihood of experiencing a slower deterioration of kidney function than the non-healthcare program group. eGFR increased by 3.87 ml/min per 1.73 m2 on average in the EG
and decreased by 1.81 ml/min per 1.73 m2 on average in
the CG at 6 months after the intervention. On the basis of
their findings, HLEP was deployed in this study to enhance
kidney function.

Limitations of the Study
This study was affected by several limitations. First, recruitment in this study was limited to patients with Stage 3–4
CKD at a single district teaching hospital in northern
Taiwan. Although their characteristics were similar to the

HLEP Improves Depression and Slow Renal Function

VOL. 32, NO. 1, FEBRUARY 2024

Table 4

Results of Generalized Estimating Equations for MMHLQ, Depression, and eGFR Score (N = 84)
Parameter

MMHLQ

Downloaded from http://journals.lww.com/jnr-twna by BhDMf5ePHKav1zEoum1tQfN4a+kJLhEZgbsIHo4XMi0hCywC
X1AWnYQp/IlQrHD3i3D0OdRyi7TvSFl4Cf3VC1y0abggQZXdgGj2MwlZLeI= on 10/29/2024

Intercept
Experimental group

a

Month 3 (T2) b

Depression

eGFR

β

95% CI

p

β

95% CI

p

β

95% CI

p

42.48

[32.92, 52.05]

< .001

4.01

[−1.70, 9.73]

.170

74.29

[65.90, 82.68]

< .001

2.14

[−0.28, 4.56]

.082

1.86

[−0.60, 4.32]

.138

0.28

[−1.62, 2.18]

.770

5.95

[3.85, 8.06]

< .001

−1.07

[−2.23, 0.09]

.070

1.58

[−0.10, 3.26]

.060

b

5.79

[4.04, 7.55]

< .001

−0.64

[−2.14, 0.85]

.400

−1.80

[−3.40, −0.20]

.030

Experimental Group  T2 c

−3.37

[−5.68, −1.06]

.004

−2.24

[−4.11, −0.37]

.019

−0.25

[−3.15, 2.65]

.870

Experimental Group  T3

0.99

[−1.42, 3.40]

.419

−4.36

[−6.60, −2.12]

< .001

5.87

[1.35, 10.38]

.011

Month 6 (T3)

c

Note. Adjusted: age and gender. MMHLQ = Mandarin Multidimensional Health Literacy Questionnaire; eGFR = estimated glomerular filtration rate.
a
Reference group: control group. b Reference group: baseline (T1). c Reference group: Control Group  Baseline.

general population of patients with Stage 3–4 CKD in
Taiwan, the results may not be generalizable to patients with
CKD nationwide. Future studies should include larger sample sizes from multiple hospitals with different stages of
CKD. Second, this study followed the patients for a postintervention period of 6 months only, which may not capture
the long-term effects on health literacy of the HLEP program.
Therefore, the study period should be extended in future
studies to evaluate long-term efficacy. Third, the participants
in this study were patients who had participated in a multidisciplinary care program for less than a year. Future studies
should include patients with no prior multidisciplinary care
program experience to determine the effectiveness of the
self-management health education program on this patient
group. Finally, the study was conducted in a single clinic,
and the intervention and data collection were performed by
the same researcher, which may have introduced a Hawthorne effect. To minimize the potential for this bias, future
studies should recruit from multiple clinics and use different
researchers for intervention and data collection tasks.

Conclusions and Implications for Practice
The results of this study indicate the developed HLEP program significantly and positively affects the health literacy
of patients with CKD within 3 months of program completion. Moreover, at 6 months posttest, the severity of depression decreased and kidney function improved significantly in
the EG. Furthermore, the mean eGFR increased in the EG by
3.87 ml/min per 1.73 m2 at 6 months posttest, whereas
eGFR decreased by 1.81 ml/min per 1.73 m2 in the CG.
These findings suggest the HLEP may be an effective tool
for improving health literacy and clinical outcomes in patients with CKD. This study may serve as a reference for nephrology case managers responsible for educating patients
with Stage 3–4 CKD using the HLEP. In practice, the HLEP
developed in this study may be accessed by patients at home
on their computer/cellphone or via the provided QR code.
The provided health education manual is small and easy to

carry and may be used as a reference when dining out. Health
education manuals are a more convenient format than traditional health education leaflets. A good health education tool
can enhance case manager performance, increase patient confidence in controlling their kidney disease, and delay disease
progression. Therefore, this tool is worth considering in practice. Overall, the results of this study suggest providing tailored health education to patients with Stage 3–4 CKD using
the HLEP can improve health literacy and clinical outcomes
and benefit both patients and healthcare providers.

Author Contributions
Study conception and design: YCC, HLH
Data collection: YHH
Data analysis and interpretation: YCC, YHH, CWY, MFH
Drafting of the article: YCC, YHH
Critical revision of the article: YCC, HLH
Received: October 1, 2022; Accepted: May 19, 2023
*Address correspondence to: Yu-Chu CHUNG, PhD, RN, Department of
Nursing, Yuanpei University of Medical Technology, No. 306, Yuanpei
Street, Hsinchu City 30015, Taiwan, ROC. Tel: +886-3-6102309; E-mail:
yuchu@mail.ypu.edu.tw
The authors declare no conflicts of interest.
Cite this article as:
Huang, H.-L., Hsu, Y.-H., Yang, C.-W., Hsu, M.-F., & Chung, Y.-C. (2024).
Effects of a health literacy education program on mental health and
renal function in patients with chronic kidney disease: A randomized
controlled trial. The Journal of Nursing Research, 32(1), Article e310.
https://doi.org/10.1097/jnr.0000000000000595

References
Chen, H. Y. (2000). Manual for the Beck Depression Inventory-II.
Chinese Behavioral Science. (Original work published in Chinese)
Chen, Y.-C., Chang, L.-C., Liu, C.-Y., Ho, Y.-F., Weng, S.-C., & Tsai,
T.-I. (2018). The roles of social support and health literacy in
self-management among patients with chronic kidney disease.
Journal of Nursing Scholarship, 50(3), 265–275. https://doi.org/
10.1111/jnu.12377

9

The Journal of Nursing Research

Chu, C. D., McCulloch, C. E., Banerjee, T., Pavkov, M. E., Burrows,
N. R., Gillespie, B. W., Saran, R., Shlipak, M. G., Powe, N. R.,
Tuot, D. S., & Centers for Disease Control and Prevention
Chronic Kidney Disease Surveillance Team. (2020). CKD
awareness among US adults by future risk of kidney failure.
American Journal of Kidney Diseases, 76(2), 174–183. https://
doi.org/10.1053/j.ajkd.2020.01.007
Downloaded from http://journals.lww.com/jnr-twna by BhDMf5ePHKav1zEoum1tQfN4a+kJLhEZgbsIHo4XMi0hCywC
X1AWnYQp/IlQrHD3i3D0OdRyi7TvSFl4Cf3VC1y0abggQZXdgGj2MwlZLeI= on 10/29/2024

Devellis, R. F. (2016). Scale development: Theory and applications
(4th ed.). Sage Publications.
Elliott, M. J., Love, S., Donald, M., Manns, B., Donald, T., Premji,
Z., Hemmelgarn, B. R., Grinman, M., Lang, E., & Ronksley, P.
E. (2020). Outpatient interventions for managing acute complications of chronic diseases: A scoping review and implications
for patients with CKD. American Journal of Kidney Diseases,
76(6), 794–805. https://doi.org/10.1053/j.ajkd.2020.04.006
Hanpaiboon, K., & Pratoomsoot, C. (2019). Factors influencing patient health behaviors for delaying the progress in stage 3–4
chronic kidney disease patients at Khlongkhlung hospital,
Khampangphet province. Thai Pharmaceutical and Health Science Journal, 14(2), 53–61.
Kunwar, D., Kunwar, R., Shrestha, B., Amatya, R., & Risal, A.
(2020). Depression and quality of life among the chronic kidney disease patients. Journal of Nepal Health Research Council, 18(3), 459–465. https://doi.org/10.33314/jnhrc.v18i3.2556
Levey, A. S., Coresh, J., Greene, T., Marsh, J., Stevens, L. A.,
Kusek, J. W., & Van Lente, F.; Chronic Kidney Disease Epidemiology Collaboration. (2007). Expressing the modification of
diet in renal disease study equation for estimating glomerular
filtration rate with standardized serum creatinine values. Clinical Chemistry, 53(4), 766–772. https://doi.org/10.1373/
clinchem.2006.077180
Lin, M. Y., Cheng, S. F., Hou, W. H., Lin, P. C., Chen, C. M., & Tsai, P. S.
(2021). Mechanisms and effects of health coaching in patients
with early-stage chronic kidney disease: A randomized controlled
trial. Journal of Nursing Scholarship, 53(2), 154–160. https://doi.
org/10.1111/jnu.12623
Loosman, W. L., Rottier, M. A., Honig, A., & Siegert, C. E. (2015).
Association of depressive and anxiety symptoms with adverse events in Dutch chronic kidney disease patients: A prospective cohort study. BMC Nephrology, 16(1), Article No.
155. https://doi.org/10.1186/s12882-015-0149-7
Machida, S., Shibagaki, Y., & Sakurada, T. (2019). An inpatient educational program for chronic kidney disease. Clinical and Experimental Nephrology, 23(4), 493–500. https://doi.org/10.
1007s10157-018-1660-5
National Health Insurance Administration, Ministry of Health and
Welfare, Taiwan, ROC. (2020). Pre-ESRD patient care and
health education program. https://www.nhi.gov.tw/Content_
List.aspx?n=74FB9F36D1234D73&topn=5FE8C9FEAE863B46
(Original work published in Chinese)
National Health Research Forum. (2020). The determinates of
health literacy and related health outcomes among elders in
Taiwan. https://forum.nhri.edu.tw/wp-content/uploads (Original work published in Chinese)
National Health Research Institutes & Taiwan Social of Nephrology.
(2020). 2020 kidney disease in Taiwan annual report. Authors.

10

Hsiao-Ling HUANG et al.

National Kidney Foundation. (2022). Chronic kidney disease (CKD)
symptoms and causes. https://www.kidney.org/atoz/content/
about-chronic-kidney-disease#chronic-kidney-disease-covid-19
Palmer, S. C., Vecchio, M., Craig, J. C., Tonelli, M., Johnson, D. W.,
Nicolucci, A., Pellegrini, F., Saglimbene, V., Logroscino, G.,
Hedayati, S. S., & Strippoli, G. F. (2013). Association between
depression and death in people with CKD: A meta-analysis of
cohort studies. American Journal of Kidney Disease, 62(3),
493–505. https://doi.org/10.1053/j.ajkd.2013.02.369
Quaglio, G., Sørensen, K., Rübig, P., Bertinato, L., Brand, H.,
Karapiperis, T., Dinca, I., Peetso, T., Kadenbach, K., & Dario,
C. (2017). Accelerating the health literacy agenda in Europe.
Health Promotion International, 32(6), 1074–1080. https://doi.
org/10.1093/heapro/daw028
Schrauben, S. J., Cavanaugh, K. L., Fagerlin, A., Ikizler, T. A., Ricardo,
A. C., Eneanya, N. D., & Nunes, J. W. (2020). The relationship of
disease-specific knowledge and health literacy with the uptake
of self-care behaviors in CKD. Kidney International Reports,
5(1), 48–57. https://doi.org/10.1016/j.ekir.2019.10.004
Sørensen, K., Van den Broucke, S., Fullam, J., Doyle, G., Pelikan,
J., Slonska, Z., Brand, H., & (HLS-EU) Consortium Health Literacy Project European. (2012). Health literacy and public health:
A systematic review and integration of definitions and models.
BMC Public Health, 12(1), Article No. 80. https://doi.org/10.
1186/1471-2458-12-80
U.S. Renal Data System. (2021). 2021 Annual data report. https://
adr.usrds.org/2021/chronic-kidney-disease
Wang, S.-L., Chen, T.-H., Kung, L.-F., Chiou, C.-L., & Lin, M.-Y.
(2018). The effect of multidisciplinary integrated care program
on self-care behavior and renal function in patient with chronic
kidney disease. Chang Gung Nursing, 29(1), 1–15. https://doi.
org/10.6386/CGN.201803_29(1).0001 (Original work published
in Chinese)
Wei, M.-H., Wang, Y.-W., Chang, M.-C., & Hsieh, J.-G. (2017). Development of Mandarin Multidimensional Health Literacy
Questionnaire (MMHLQ). Taiwan Journal of Public Health,
36(6), 556–570. https://doi.org/10.6288/TJPH201736106061
(Original work published in Chinese)
Wong, K. K., Velasquez, A., Powe, N. R., & Tuot, D. S. (2018). Association between health literacy and self-care behaviors among
patients with chronic kidney disease. BMC Nephrology, 19(1),
Article No. 196. https://doi.org/10.1186/s12882-018-0988-0
Wu, C.-L., Liou, C.-H., Liu, S.-A., Sheu, W.-H., & Tsai, S.-F. (2020).
Mandarin Multidimensional Health Literacy Questionnaire
for patient supporting groups: A quality improvement article.
Medicine, 99(45), Article e23182. https://doi.org/10.1097/MD.
0000000000023182
Wu, S. F. V., Lee, M. C., Hsieh, N. C., Lu, K. C., Tseng, H. L., & Lin, L.
J. (2018). Effectiveness of an innovative self-management intervention on the physiology, psychology, and management
of patients with pre-end-stage renal disease in Taiwan: A randomized, controlled trial. Japan Journal of Nursing Science,
15(4), 272–284. https://doi.org/10.1111/jjns.12198
Xu, Z.-C. (2015). Taiwan chronic kidney disease clinical guidelines
—2015. National Health Research Institutes. (Original work
published in Chinese)

Gut Microbiota and Probiotics in Health and Disease
This webinar took place on 23rd September 2021,
as part of the 5th Global Microbiota Summit

Speakers:

Mary Ellen Sanders,1 Ana Teresa Abreu,2 Karine Clément3
1. Dairy and Food Culture Technologies, Centennial, Colorado, USA
2. Hospital Ángeles del Pedregal, Mexico City, Mexico
3. Inserm/Sorbonne University, Pitié-Salpêtrière Hospital, Paris, France

Disclosure:

Sanders has received consultancy and speaker fees from Bayer, Bloom
Pharmaceuticals, The Chronic Disease Research Foundation (CDRF), Church
& Dwight, PepsiCo, Kerry, Associated British Foods, Mead Johnson, Fairlife,
GlaxoSmithKline, Trouw Nutrition, Allergosan OMNi-BiOTiC, Probi, Sanofi, Cargill,
Danone North America, Danone Research, Winclove Probiotics, and Yakult. Abreu has
received consultancy, speaker, or research fees from Sanofi, AB Biotics, Axon Pharma,
Mayoly Spindler, Biocodex, Alfasigma, Tecnoquímicas, MD Pharma, Medix Healthcare,
Menarini, Ferrer, Takeda, Carnot (Mexico), Adare Pharmaceuticals, Abbott, Faes
Farma, Falk Institute, Instituto de Nutrición y Salud Kellogg’s, and Danone Institute.
Clément has received consultancy, speaker, or research fees from Danone Research,
Ysopia, Confo Therapeutics, and Sanofi Consumer HealthCare.

Acknowledgements:

Writing assistance was provided by Nicola Humphry, Nottingham, UK.

Support:

The symposium and publication of this article were funded by Sanofi CHC. The
views and opinions expressed are exclusively those of the speakers. The content was
reviewed by Sanofi CHC for medical accuracy.

Citation:

EMJ Gastroenterol. 2021;10[1]:42-50.

Meeting Summary
Mary Ellen Sanders opened the webinar by defining and differentiating the ‘biotic’ family, including
probiotics, prebiotics, synbiotics, and postbiotics. She discussed the need for improved labels on
commercial products in the biotics family and emphasised the research gaps in this field Ana Teresa
Abreu expanded on a specific probiotic, Bacillus clausii, describing the evidence for health benefits
associated with this bacterium and the potential mechanisms through which it might achieve these
effects. Finally, Karine Clément discussed the role of the gut microbiome in cardiometabolic disease,
suggesting that gut microbiota may represent a missing link between the environmental and genetic
factors that impact these diseases. Clément described the evidence for a dysbiosis of gut microbiota
in metabolic diseases and posited that a personalised approach to gut microbiome therapy might be
the best way to leverage this association.

42

GASTROENTEROLOGY • November 2021

EMJ

The Science of Probiotics
and Related Biotics: How to
Understand and Use Them
Mary Ellen Sanders
Mary Ellen Sanders introduced ‘biotics’ as a
family of four microbiome-targeted substances:
probiotics, prebiotics, synbiotics, and postbiotics.
Each type of biotic has the potential to impact the
resident microbes of a host, which have diverse
physiological functions, including promotion
of fat storage and angiogenesis, immune
development, synthesis of vitamins and amino
acids, drug metabolism, modification of the
nervous system, breakdown of food, resistance
to pathogens, protection against epithelial injury,
and modulation of bone-mass density.1
Many human diseases and disorders are
associated with an altered microbiome, including
irritable bowel disease, colon cancer, diabetes,
obesity, rheumatoid arthritis, and liver disease.1–3
However, Sanders emphasised that it is not yet
clear whether the altered microbiome is a cause
or a result of these conditions.1 This raises the
question of whether restoring the microbiota
in individuals with these conditions, to match
that of healthy individuals, would affect the
condition itself.
Biotics are intended to influence colonising
microbiota to improve health, but understanding

of what constitutes a healthy microbiome is still
quite limited. Sanders explained that rather than
focusing on the specific microbes present in the
microbiome, a healthy microbiome may be better
characterised by a high diversity of taxonomic
units, high resilience (the ability to recover from
perturbations such as antibiotic exposure), and
functional redundancy (more than one ecosystem
member can perform the same function).4
Sanders feels that although studying the
microbiome is helpful to understand the
mechanisms of biotics, the evidence of health
benefits is more important. For example,
probiotics have been shown to benefit health
for various clinical endpoints, across the human
lifespan, and in different organ systems, such as
preventing antibiotic-associated or traveller’s
diarrhoea, treating ulcerative colitis, and reducing
the incidence of infection gastrointestinal
disease.2 For most of these benefits, a
microbiome-mediated mechanism has not been
demonstrated yet.4
The International Scientific Association for
Probiotics and Prebiotics (ISAPP) has published
statements that include clear definition for
each of member of the biotic family, based on
consensus panels (Table 1). Importantly, these
definitions are deliberately broad enough to
support innovation; they do not restrict these
substances by host (e.g., human, agricultural
animals, etc), regulatory category (e.g., food,

Table 1: ISAPP Definitions of biotic substances.

Biotic substance

Definition

Probiotic

Live microorganisms that, when administered in adequate amounts, confer a health
benefit on the host5

Prebiotic

A substrate that is selectively utilised by host microorganisms, conferring a health
benefit on the host6

Synbiotic

A mixture comprising live microorganisms and substrate(s) selectively utilised by host
microorganisms that confers a health benefit on the host7

Postbiotic

Preparation of inanimate microorganisms and/or their components that confers a
health benefit on the host8

Definitions are concise; for full understanding, see the full statements. All substances must be safe for their
intended use.
ISAPP: International Scientific Association for Probiotics and Prebiotics.

Creative Commons Attribution-Non Commercial 4.0

November 2021 • GASTROENTEROLOGY

43

drug, or supplements), site of action (e.g., gut,
vaginal tract, skin, etc), or mechanism of action.5-8

Probiotics
A number of different microbes are used as
probiotics, many of which are members of
the Lactobacillaceae family or are species of
Bifidobacterium, Bacillus, or Saccharomyces.9 The
range of probiotic species is rapidly expanding10
as more is learnt about the microbes that reside
in the healthy human body. Sanders emphasised
the importance of recognising that probiotics
are a heterogenous group; two products which
contain the same microbial genus and species but
differ by microbial strain may differ in function.
To be defined as a probiotic, a substance must
be a properly identified (both sequenced
and named). The microbe must be alive
when administered, and studies need to have
demonstrated a health benefit for a specific
target host at the specific dose delivered by
the product. In addition, the microbial strain
and manufacturing process must be safe for the
intended use, and the product must be correctly
labelled with the strain and colony forming
units (CFU) expected at the end of its shelf life.5
Ideally, probiotic product labels should detail
health benefits (supported by evidence),
suggested serving size, proper storage
conditions, and contact details for consumer
information.11 However, a survey of refrigerated
probiotic foods in grocery stores in the USA
found that only one-half (22 of 45) of products
listed the constituent microbial strains. Those
that did, could be linked to evidence of health
benefits, tended to contain fewer strains, and
had a lower CFU per serving compared to other
products.12 A survey of probiotic supplements
found similar results: most products could not be
linked to evidence; 45% did not list constituent
microbial strains; and 45% did not provide CFU
at end of shelf life.13 Sanders emphasised that
similar problems exist outside of the USA.
Neither probiotics nor postbiotics are required
to target the microbiome directly, whereas
prebiotics and synbiotics should do so as part
of their mechanism of action.5–8 Despite these
distinctions, Sanders explained that there is a
common belief among both scientists and the
general public that probiotics have an important
impact on the gut microbiome. This belief is not

44

GASTROENTEROLOGY • November 2021

fully substantiated by the available research data;
a systematic review of clinical trials showed that
probiotics did not have a global impact on the
faecal microbial communities in healthy subjects.14
Sanders suggested that this does not prove that
probiotics have no effect; their effects may be
limited to minor components of the microbiota,
not evident in faecal samples or in healthy
subjects, or only evident in the metabolites rather
than the microbiome composition. However,
she stressed that the evidence to date indicates
that the effects of current probiotics on the
microbiome are likely to be quite subtle.
In summary, Sanders reiterated that the healthy
gut microbiome has not yet been defined by
researchers, but that for probiotics, effects on
the microbiome are probably less important
than health benefits. There is a clear need for
improved labels on commercial products in the
biotics family so that healthcare practitioners
and consumers know what they are buying,
and the terms probiotics, prebiotics, synbiotics,
and postbiotics should only be used when the
scientifically accepted criteria are met. She
emphasised the research gaps in this arena,
including defining a healthy microbiome,
robust trials to confirm health benefits, and
identifying the best strains and doses for specific
applications. Finally, Sanders emphasised that it
will be important to understand the mechanisms
that drive the clinical benefits of biotics in order
to optimise these substances for future use.

Bacillus clausii: Mechanisms
as Spore Probiotics in
Gastrointestinal Disorders
Ana Teresa Abreu
Bacillus is one of the most studied bacterial
genera15 and its species can be found in
soil, water, food, and in the human gut.16
These aerobic bacteria can differentiate
into a dormant endospore, allowing them to
survive in stomach acid and bile salts in the
gastrointestinal system.16,17
Most Bacilli are not pathogenic to humans
or animals, and in the case of B. clausii
(Figure 1),¹⁸ an endosymbiotic relationship, where
one organism lives inside the other, between

EMJ

8 μm

Figure 1: B. clausii (combined antibiotic resistant strains: O/C, SIN, N/R, T).
N/R: novobiocin and rifampicin; O/C: chloramphenicol; SIN: streptomycin and neomycin; T: tetracycline.

species and their hosts has been suggested.17,19,20
Four strains of B. clausii are resistant to
antibiotics, a property considered advantageous
to restoring a healthy gut, and are named for their
predominant antibiotic resistance: novobiocin
and rifampicin (strain N/R), chloramphenicol
(strain O/C), streptomycin and neomycin (strain
SIN), and tetracycline (strain T). 19
There are several properties of B. clausii that
contribute to its probiotic effects:
> B. clausii spores can survive the hostile
environment of the gastrointestinal tract and
multiply to colonise the intestine.21-23
> The pan-genome of B. clausii (O/C, SIN, N/R,
T) includes genes involved in carbohydrate
metabolism,19 and one strain, SKAL 16, has
been shown to excrete butyrate in in vitro
conditions.24 Butyrate serves as the major
energy source for enterocytes, exerts antiinflammatory effects, and enhances gut
barrier function.25
> The antibiotic resistance genes of B. clausii
are stable and cannot be transferred to
other bacteria.26 Many strains of B. clausii are
recommended for use along with antibiotics,
and Abreu emphasised that it is important
for clinicians to match probiotic strains to the
prescribed antibiotic therapy.
> Some strains of B. clausii, particularly SIN and
T, produce the essential vitamin riboflavin
(vitamin B2) in vitro, suggesting that B. clausii

Creative Commons Attribution-Non Commercial 4.0

has the potential to compensate for host
deficits in riboflavin that can occur in clinical
contexts such as chemotherapy.27
> Bacillus species produce a wide range of
antimicrobial substances, including lantibiotics
(post-translationally modified peptides) which
are active against gram-positive bacteria such
as Clostridium difficile.20,28 One such lantibiotic,
clausin, has been isolated from B. clausii
and interacts with lipid intermediates in the
bacterial envelope biosynthesis pathways,29
suggesting that it could help to manipulate the
constituents of the intestinal microbiota.

Immunomodulation
B.
clausii
has
been
shown
to
have
immunomodulatory properties in preclinical
studies. In a human enterocyte model of rotavirus
infection, B. clausii strains (O/C, SIN, N/R, and T)
induced the synthesis of bacteriocins, reduced
enterocyte cell death, and inhibited the release of
pro-inflammatory cytokines. They also increased
mucin production and the synthesis of tight
junction proteins, both important for the integrity
of the gut mucosal barrier.30 In addition, a small in
vivo experiment has shown that B. clausii modifies
the gene expression profile in the intestine in
patients with mild oesophagitis, including genes
involved in immunity and inflammation.31 Finally,
in an animal model of asthma, B. clausii reduced
the numbers of eosinophils, neutrophils, and
lymphocytes, and lowered IL-4 and IL-5 levels,

November 2021 • GASTROENTEROLOGY

45

suggesting a potential use in reducing airway
inflammation in clinical settings.32
Abreu explained that one potential mechanism
for the immunomodulatory capacity of
probiotic B. clausii strains could be the
expression of extracellular compounds and/or
immunostimulation via the cell wall. In murine
cell lines, B. clausii MTC 8326 was shown to
activate metabolic activity and innate immune
responses in macrophages,33 and B. clausii (O/C,
N/R, SIN, and T) was also shown to stimulate
the production of nitrite in peritoneal cells, IFN-γ
in spleen cells, and CD4+ T-cell proliferation.20
One route through which B. clausii may induce
these immunomodulatory effects is through the
secretion of lipoteichoic acid.34

Gut Homeostasis
Other studies have suggested that B. clausii
contributes to gut homeostasis. In an in vitro
simulation of the human gastrointestinal tract, B.
clausii SC 109 spores (along with other probiotic
bacteria and prebiotic ingredients) were shown
to increase microbiome production of butyrate,
and the overall diversity of gut microbiota.35 The
presence of B. clausii in patients with pancreatic
adenocarcinoma has been associated with
longer survival times,36 and treatment with
B. clausii UBBC07 has been shown to reduce
serum urea levels in rats with acetaminopheninduced renal failure, suggesting a novel clinical
use for probiotics in chronic kidney disease.37

Antimicrobial Properties
Abreu explained that B. clausii can produce
antimicrobial peptides, including lantibiotics,
that inhibit the growth of pathogenic bacteria in
vitro.20 This characteristic means that probiotics
can be supportive when delivered alongside
antibiotic therapy. B. clausii (O/C, N/R, SIN, and
T) appears to be protective during Escherichia
coli infection in mice, increasing protective mucus
secretion and resulting in minimal mucosal
damage and less sloughing of villus tips.38,39
Infection with C. difficile can result in symptoms
ranging from diarrhoea to pseudomembranous
colitis,40 and infection with B. cereus can cause
vomiting, diarrhoea, and haemorrhage.41 B. clausii
strain O/C has been shown to secrete a serine
protease capable of inhibiting the cytotoxic
effects of both C. difficile and B. cereus in vitro.41

46

GASTROENTEROLOGY • November 2021

Abreau explained that B. clausii has been
efficaciously and safely used in humans for
several decades. For example, in patients
with dietary endotoxemia, believed to be
caused by disruptions in gut permeability,
administration of probiotic strains including
B. clausii was associated with a 42% reduction in
post-prandial serum endotoxin and reductions
in pro-inflammatory markers.42 In patients
with recurrent aphthous stomatitis, a disease
of the oral mucosa that results in ulcers and
pain, local adjunct application of B. clausii,
alongside glucocorticoid treatment, reduced
oral pain and ulcer severity compared to
glucocorticoid alone.43
In
summary,
Abreu
reiterated
that
the
physiological,
antimicrobial,
and
immunomodulatory properties of B. clausii
have been demonstrated both in vitro and
in vivo; and antimicrobial activity against
enteropathogens such as C. difficile and
B. cereus has been demonstrated, providing one
potential mode of action for the efficacy of this
probiotic in gastrointestinal disorders. Further
clinical studies using specific strains in targeted
medical conditions are needed to validate these
findings, and to increase the scientific credibility
of B. clausii.

Gut Microbiota in
Cardiometabolic Diseases
Karine Clément
Clément began by emphasising that there is a
heavy societal burden from cardiometabolic
and nutrition-related diseases and that the gut
microbiota can be considered a ‘super-integrator’
for many of the risk factors for mortality.44
Obesity, the fourth highest risk factor for
mortality in Western Europe,44 is associated
with altered inter-organ cross-talk involving the
intestinal tract, brain, adipose tissue, muscles,
and others (PRIEST 2019). In the adipose
tissue, obesity is connected to perturbed
endocrine secretions, immune or inflammatory
imbalances, altered angiogenesis, organelle
dysfunction, altered extracellular matrix, and
adipocyte hypertrophy.45–47 The development
of obesity involves the pathogenic remodelling

EMJ

of white adipose tissue, which may lead to the
development of obesity-related cardiometabolic
disease and compromised response to obesity
treatment.48 There is substantial heterogeneity
in the clinical trajectory of subjects with obesity
and their weight loss responses, for which
gut microbiota-derived elements may be
contributing factors.49
Clément explained that the role of the gut
microbiota genomes in host biology should
be considered: while it is accepted that both
environmental and genetic factors play a role
in the development of metabolic disease, gut
microbiota may represent the missing link
between them.
The key functions of the gut microbiota are in
the digestion of food and the production of
metabolites, the development and integrity of
intestinal structure, immune system development,
metabolism of toxic compounds, and synthesis
of vitamins K and B.50 However, several studies
have suggested that gut microbiota also play
a role in energy balance and our capacity to
store fat. Clément described ground-breaking
pre-clinical experiments that showed that
germ-free rodents have decreased adiposity
and are resistant to diet-induced weight gain,
compared to conventionally raised rodents.51
In addition, transplanting gut microbiota from

Obese twin

Microbiota
transplant

mouse models of obesity into germ-free mice can
partially transfer the obesity phenotype.51 Similar
experiments have been conducted to transfer
microbiota from humans to mice, and these have
shown that the receipt of gut microbiota from an
obese human can result in increased adiposity in
a mouse, even when a healthy diet is followed.
In parallel, the receipt of gut microbiota from a
lean individual (a twin of the obese individual)
results in a lean mouse when a healthy diet is
followed52,53 (Figure 2).
Clément then discussed the importance of
diversity in the gut microbiome in healthy
individuals. Subjects living in westernised
countries such as the USA have been shown to
have a lower diversity of gut microbiota from
an early age, compared to populations that are
more isolated or live with an ancestral mode, such
as Malawians or Amerindians.54 Some studies
have attempted to stratify individuals by their
microbiotic gene richness. Across these studies,
20–30% of subjects were considered to have low
gene richness, and this group was characterised
by increased overall adiposity, dysmetabolism,
and
a
more
pronounced
inflammatory
phenotype than individuals with high gene
richness.55,56 Approximately 75% of patients with
severe obesity (candidates for bariatric surgery)
can be classified as having low gene richness.

Recipient mice

Lean twin

Increased adiposity

Lean

Figure 2: The protective role of gut microbiota from a lean donor in the presence of a healthy diet.
Reproduced with permission, Walker and Parkhill.52

Creative Commons Attribution-Non Commercial 4.0

November 2021 • GASTROENTEROLOGY

47

This is important because a low gene count is
associated with enrichment of pro-inflammatory
bacteria, whereas a high gene count is associated
with enrichment of anti-inflammatory bacteria.55
One of the important characteristics of ‘healthy’
gut microbiota is the production of short-chain
fatty acids (SCFAs), including butyrate.6,8,25,57
SCFAs act on enterocytes to stimulate the
production of certain hormones, improving insulin
sensitivity and glucose tolerance and modifying
lipid metabolism.8,25 Clément emphasised that
there is considerable research effort focused on
understanding the imbalance between the gut
microbiota in healthy individuals and those with
disease. Gut microbiota may also contribute to
the health of the intestinal barrier in metabolic
diseases.58 For example, studies have shown
that modification of the gut microbiota affects
the thickness of the mucus barrier.58
The effects of gut microbiota on the host can be
classified as metabolism-independent pathways,
driven by components of the bacterial membrane
such as lipopolysaccharide or peptidoglycan and
impacting low-grade inflammation processes
or modifying host biology; or metabolismdependent pathways driven by microbial
metabolites such as imidazole propionate, SCFAs,
secondary bile acids, or trimethylamine. 59,60
Clément described several studies that have
attempted to stratify gut microbiomes into
groups based on their genome. In a European
study, Arumugam et al., described three distinct
clusters of microbiomes, termed enterotypes,
each characterised by a dominant gut microbial
species: Type 1, enriched in Bacteroides; Type
2, enriched in Prevotella; and Type 3, enriched
in Ruminococcus.61 Subsequent studies have
identified a subset of the Type 2 microbiome
with a low proportion of Faecalibacterium and
low microbial cell density, named Bact2, which
is more prevalent in patients with inflammatory
bowel disease versus the general population
(78% versus 13%, respectively).62 The prevalence
of Bact2 also correlates with higher BMI and
with low-grade systemic inflammation in the
MetaCardis European cohort.63
Clément explained that interventions to increase
microbial diversity, increase beneficial microbes,

48

GASTROENTEROLOGY • November 2021

and change metabolite concentrations are
intended to improve metabolism and the immune
response, potentially reducing the burden of
complications. Potential mechanisms to modify
the gut microbiome include dietary changes,
selective enrichment of gut bacteria, faecal
transplant, and bariatric surgery.
One example of such an intervention is
diet-induced weight loss in patients with obesity
or overweight, which improved gut microbiotic
diversity and clinical phenotypes in patients with
a low microbial gene count at baseline.56 Bariatric
surgery also appears to increase microbial gene
richness one-year post-surgery.63 Administration
of Akkermansia muciniphila to mouse models
of obesity or Type 2 diabetes resulted in a
reduction in fat mass, insulin resistance, and
low-grade inflammation,64 and A. muciniphila is
associated with healthier metabolic status and
greater insulin sensitivity in human subjects with
obesity or overweight.65 Finally, a study of faecal
transfer from healthy individuals to patients
with obesity and metabolic syndrome showed
an improvement in insulin sensitivity, however,
the effect was transient and mainly observed
in patients with a low gut microbiota diversity
at baseline.66
Clément concluded that there is evidence for a
dysbiosis of gut microbiota in metabolic diseases,
and that a personalised approach to the gut
microbiome may be the best way to leverage this
association. However, she stressed that further
research is needed as the links between the
changes in the gut microbiota and the expected
clinical effects have yet to be fully elucidated.

Summary
In summary, the evidence to date supports the
hypothesis that both probiotics and the gut
microbiome have an impact on the health of
humans and other animals. However, though
potential mechanisms of action have been
suggested experimentally, further research
including well-designed trials is needed to fully
understand how probiotics manipulate the gut
microbiota to benefit the host.
MAT-GLB-2104926

EMJ

References
1.

Laukens D et al. Heterogeneity of the
gut microbiome in mice: guidelines
for optimizing experimental design.
FEMS Microbiol Rev. 2016;40(1):117–
132.

2.

Merenstein DJ et al. Probiotics as a Tx
resource in primary care. J Fam Pract.
2020;69(3):E1–E10.

3.

Wang B et al. The human microbiota
in health and disease. Engineering.
2017;3(1):71–82.

4.

McBurney MI et al. Establishing what
constitutes a healthy human gut
microbiome: state of the science,
regulatory considerations, and future
directions. J Nutr. 2019;149(11):1882–
95.

5.

Hill C et al. The International Scientific
Association for Probiotics and
Prebiotics consensus statement on
the scope and appropriate use of the
term probiotic. Nat Rev Gastroenterol
Hepatol. 2014;11(8):506–14.

6.

7.

8.

9.

Gibson GR et al. Expert consensus
document: The International
Scientific Association for Probiotics
and Prebiotics (ISAPP) consensus
statement on the definition and scope
of prebiotics. Nat Rev Gastroenterol
Hepatol. 2017;14(8):491–502.
Swanson KS et al. The International
Scientific Association for Probiotics
and Prebi-otics (ISAPP) consensus
statement on the definition and scope
of synbiotics. Nat Rev Gastroenterol
Hepatol. 2020;17(11):687–701.
Salminen S et al. The International
Scientific Association of Probiotics
and Prebiotics (ISAPP) consensus
statement on the definition and scope
of postbiotics. Nat Rev Gastroenterol
Hepatol. 2021;18(9):649–67.
McFarland LV et al. Strain-specificity
and disease-specificity of probiotic
efficacy: a systematic review and
meta-analysis. Front Med (Lausanne).
2018;5:124.

10. International Scientific Association
for Probiotics and Prebiotics. ISAPP
position statement on minimum
criteria for harmonizing global
regulatory approaches for probiotics
in foods and supplements. 2018.
Available at: https://isappscience.
org/minimum-criteria-probiotics. Last
accessed: October 2021.
11.

World Health Organisation. Probiotics
in food Health and nutritional
properties and guidelines for
evaluation. 2006. Available at: http://
www.fao.org/3/a0512e/a0512e.pdf.
Last accessed: October 2021.

2019;34(12):2735–7.

Microbiol. 2003;95(6):1255–60.

14. Kristensen NB et al. Alterations
in fecal microbiota composition
by probiotic supplementation in
healthy adults: a systematic review of
randomized controlled trials. Genome
Medicine. 2016;8(1):52.
15.

Khurana H et al. Genomic insights
into the phylogeny of Bacillus strains
and elucidation of their secondary
metabolic potential. Genomics.
2020;112(5):3191–3200.

28. Ahire JJ et al. Survival and
Germination of Bacillus clausii
UBBC07 Spores in in vitro Human
Gastrointestinal Tract Simulation
Model and Evaluation of Clausin
Produc-tion. Front Microbiol.
2020;11:1010.
29. Bouhss A et al. Specific interactions
of clausin, a new lantibiotic, with lipid
precursors of the bacterial cell wall.
Biophys J. 2009;97(5):1390–7.

16.

Elshaghabee FMF et al. Bacillus as
potential probiotics: status, concerns,
and future perspectives. Front
Microbiol. 2017;8:1490.

30. Paparo L et al. Protective action of
Bacillus clausii probiotic strains in an
in vitro model of Rotavirus infection.
Sci Rep. 2020;10(1):12636.

17.

Cutting SM et al. Bacterial sporeformers: friends and foes. FEMS
Microbiol Lett. 2014;358(2):107–9.

31.

18.

Bacillus clausii Enterogermina.png.
Wikimedia Commons. 2018. Available
at: https://commons.wikimedia.
org/wiki/File:Bacillus_clausii_
Enterogermina.png. Last accessed:
October 2021

19.

Khatri I et al. Composite genome
sequence of Bacillus clausii, a
probiotic commer-cially available
as Enterogermina®, and insights
into its probiotic properties. BMC
Microbiology. 2019;19(1):307.

Kolacek S et al. Commercial probiotic
products: a call for improved
quality control. a position paper by
the ESPGHAN working group for
probiotics and prebiotics. J Pediatr
Gastroenterol Nutr. 2017;65(1):117–24.

22. Senesi S et al. Molecular
characterization and identification
of Bacillus clausii strains marketed
for use in oral bacteriotherapy. Appl
Environ Microbiol. 2001;67(2):834–9.
23. Vecchione A et al. Compositional
quality and potential gastrointestinal
behaviour of probiotic products
commercialized in Italy. Front Med
(Lausanne). 2018;5:59.
24. Lee SH et al. Isolation and
physiological characterization
of Bacillus clausii SKAL-16i from
wastewater. J Microbiol Biotechnol.
2008;18(12):1908–14.
25. Cantu-Jungles TM et al. Potential
of prebiotic butyrogenic fibers in
parkinson’s dis-ease. Front Neurol.
2019;10:663.

12.

Dailey Z et al. Retail refrigerated
probiotic foods and their association
with evidence of health benefits.
Benef Microbes. 2020;11(2):131–3.

26. Lakshmi SG et al. Safety assesment
of Bacillus clausii UBBC07, a spore
forming probiotic. Toxicol Rep.
2017;4:61-71.

13.

Merenstein D et al. More information
needed on probiotic supplement
product labels. J Gen Intern Med.

27. Salvetti S et al. Rapid determination
of vitamin B2 secretion by bacteria
growing on solid media. J Appl

Creative Commons Attribution-Non Commercial 4.0

32. Park H et al. Bacillus clausii, a
foreshore-derived probiotic,
attenuates allergic airway
inflammation through downregulation
of hypoxia signaling. J Rhinol.
2020;27(2):108–116.
33. Pradhan B et al. Comparative analysis
of the effects of two probiotic
bacterial strains on metabolism
and innate immunity in the raw
264.7 murine macrophage cell line.
Probiotics Antimicrob Proteins.
2016;8(2):73-84.

20. Urdaci MC et al. Bacillus clausii
probiotic strains antimicrobial
and immunomodulatory activities.
J Clin Gastroenterol. 2004;38(6
Suppl):S86–S90.
21.

Di Caro S et al. Bacillus clausii effect
on gene expression pattern in small
bowel mucosa using DNA microarray
analysis. Eur J Gastroenterol Hepatol.
2005;17(9):951–60.

34. Villéger R et al. Characterization of
lipoteichoic acid structures from
three probiotic Bacillus strains:
involvement of D-alanine in their
biological activity. Antonie Van
Leeuwenhoek. 2014;106(4):693–706.
35. Duysburgh C et al. A synbiotic
concept containing spore-forming
Bacillus strains and a prebiotic
fiber blend consistently enhanced
metabolic activity by modulation
of the gut microbiome in vitro. Int J
Pharm X. 2019;1:100021.
36. Riquelme E et al. Tumor microbiome
diversity and composition influence
pancreatic cancer outcomes. Cell.
2019;178(4):795–806.
37. Patel C et al. Therapeutic prospective
of a spore-forming probioticBacillus clausii UBBC07 against
acetaminophen-induced uremia in
rats. Probiotics Antimicrob Pro-teins.
2020;12(1):253–8.
38. Yu MG et al. Histomorphologic
effects of Bacillus clausii spores in
enteropathogenic E. coli O127:H21infected mice: A Pilot Study. Phillipine
J Int Med. 2016;54:1–7.
39. De Castro JA et al. Bacillus clausii
as adjunctive treatment for acute
community-acquired diarrhea
among Filipino children: a largescale, multicenter, open-label study
(CODDLE). Tropical Dis Travel Med
Vacc. 2019;5:14.

November 2021 • GASTROENTEROLOGY

49

40. Mills JP et al. Probiotics for
prevention of clostridium difficile
infection. Curr Opin Gastroenterol.
2018;34(1):3–10.
41.

the pathological relevance of
extracellular matrix in human obesity.
Genome Biol. 2009:9(1):R14.

Ripert G et al. Secreted compounds
of the probiotic Bacillus clausii strain
O/C inhibit the cytotoxic effects
induced by Clostridium difficile and
Bacillus cereus toxins. Antimicrob
Agent Chemother. 2016;60(6):3445–
54.

42. McFarlin BK et al. Reversing mealassociated gastrointestinal gut
permeability issues: potential
treatment target for spore-based
probiotics? Am J Gastroenterol.
2017;112:S658–59.

50. Fouhy F et al. Composition of
the early intestinal microbiota:
knowledge, knowledge gaps and the
use of high-throughput sequencing
to address these gaps. Gut Microbes.
2012;3(3):203-20.
51.

43. Cheng B et al. The efficacy of
probiotics in management of
recurrent aphthous stomatitis: a
systematic review and meta‑analysis.
Sci Rep. 2020;10(1):21181.
44. IHME, Global Burden of Disease
(2017). Number of deaths by
risk factor, Western Europe.
2017. Available at: https://
ourworldindata.org/grapher/
number-of-deaths-by-riskfactor?country=~Western+Europe.
Last accessed: October 2021.

Xiao H et al. The role of the gut
microbiome in energy balance with a
focus on the gut-adipose tissue axis.
Front Genet. 2020;11:297.

52. Ridaura VK et al. Cultured gut
microbiota from twins discordant
for obesity modulate adiposity
and metabolic phenotypes in mice.
Science. 2013;341(6150):1241214.
53. Walker AW and Parkhill J. Fighting
obesity with bacteria. Science.
2013:341(6150):1069–70.

45. Kasinska MA et al. Epigenetic
modifications in adipose tissue –
relation to obesity and diabetes. Arch
Med Sci. 2016;12(6):1293–1301.
46. Nijhawans P et al. Angiogenesis
in obesity. Biomed Pharmacother.
2020;126:110103.
47. Xiao Y et al. Chronic stress,
epigenetics, and adipose tissue
metabolism in the obese state. Nutr
Metab (Lond). 2020:17:88.
48. Henegar C et al. Adipose tissue
transcriptomic signature highlights

49. Hung TKW et al. Understanding the
heterogeneity of obesity and the
relationship to the brain-gut axis.
Nutrients. 2020;12(12):3701.

54. Yatsunenko T et al. Human gut
microbiome viewed across
age and geography. Na-ture.
2012;486(7402):222–7.
55. Le Chatelier E et al. Richness of
human gut microbiome correlates
with metabolic markers. Nature.
500(7464):541–546.
56. Cotillard A et al. Dietary
intervention impact on gut
microbial gene richness. Nature.
2013;500(7464):585–8.
57. Montassier E et al. Chemotherapydriven dysbiosis in the intestinal
microbiome. Aliment Pharmacol Ther.
2015;42(5):515–28.

58. Cani PD et al. Changes in gut
microbiota control inflammation in
obese mice through a mechanism
involving GLP-2-driven improvement
of gut permeability. Gut.
2009:58(8):1091–103.
59. Brown JM et al. The gut microbial
endocrine organ: bacterially-derived
signals driving cardiometabolic
diseases. Annu Rev Med.
2015;66:343–59.
60. Koh A et al. Microbially produced
imidazole propionate impairs insulin
signaling through mTORC1. Cell.
2018;175(4):947–61.e17.
61.

Arumugam M et al. Enterotypes of
the human gut microbiome. Nature.
2011;473(7346):174–80.

62. Vieira-Silva S et al. Statin therapy
is associated with lower prevalence
of gut microbiota dysbiosis. Nature.
2020;581(7808):310–5.
63. Aron-Wisnewsky J et al. Major
microbiota dysbiosis in severe
obesity: fate after bariatric surgery.
Gut. 2019;68(1):70–82.
64. Everard A et al. Cross-talk between
Akkermansia muciniphila and
intestinal epithelium controls dietinduced obesity. Proc Natl Acad Sci
USA. 2013;110(22):9066-71.
65. Dao MC et al. Akkermansia
muciniphila and improved metabolic
health during a dietary intervention
in obesity: relationship with gut
microbiome richness and ecology.
Gut. 2016;65(3):426–36.
66. Kootte RS et al. Improvement of
insulin sensitivity after lean donor
feces in metabolic syndrome
is driven by baseline intestinal
microbiota composition. Cell Metab.
2017;26(4):611–9.e6.

FOR REPRINT QUERIES PLEASE CONTACT: INFO@EMJREVIEWS.COM

50

GASTROENTEROLOGY • November 2021

EMJ