This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs
License, which permits use and distribution in any medium, provided the
original work is properly cited, the use is non-commercial and no
modifications or adaptations are made.
New genetic signatures associated with cancer cachexia as defined by low skeletal muscle index and weight lossNeil Johns1, Cynthia Stretch2, Benjamin H.L. Tan3, Tora S. Solheim4, Sveinung Sørhaug4, Nathan A. Stephens1, Ioannis Gioulbasanis5, Richard J.E. Skipworth1, D.A. Christopher Deans1,
Antonio Vigano6, James A. Ross1, Oliver F. Bathe7,8, Michel L. Tremblay6, Stein Kaasa4, Florian Strasser10, Bruno Gagnon9, Vickie E. Baracos2,†, Sambasivarao Damaraju11,†, Kenneth C.H. Fearon1,†,*
Version of Record online: 5 AUG 2016
How to Cite
2016) New genetic signatures associated with cancer cachexia as defined by low skeletal muscle index and weight loss. Journal of Cachexia, Sarcopenia and Muscle, doi: 10.1002/jcsm.12138., , , , , , , , , , , , , , , , , , and (
affects the majority with advanced cancer. Based on current demographic
and clinical factors, it is not possible to predict who will develop
cachexia or not. Such variation may, in part, be due to genotype. It has
recently been proposed to extend the diagnostic criteria for cachexia
to include a direct measure of low skeletal muscle index (LSMI) in
addition to weight loss (WL). We aimed to explore our panel of candidate
single nucleotide polymorphism (SNPs) for association with WL +/−
computerized tomography-defined LSMI. We also explored whether the
transcription in muscle of identified genes was altered according to
such cachexia phenotype
A retrospective cohort study design was used. Analysis explored associations of candidate SNPs with WL (n = 1276) and WL + LSMI (n = 943). Human muscle transcriptome (n = 134) was analysed using an Agilent platform.
nucleotide polymorphisms in the following genes showed association with
WL alone: GCKR, LEPR, SELP, ACVR2B, TLR4, FOXO3, IGF1, CPN1, APOE,
FOXO1, and GHRL. SNPs in LEPR, ACVR2B, TNF, and ACE were associated with
concurrent WL + LSMI. There was concordance between muscle-specific
expression for ACVR2B, FOXO1 and 3, LEPR, GCKR, and TLR4 genes and LSMI
and/or WL (P < 0.05).
rs1799964 in the TNF gene and rs4291 in the ACE gene are new
associations when the definition of cachexia is based on a combination
of WL and LSMI. These findings focus attention on pro-inflammatory
cytokines and the renin–angiotensin system as biomarkers/mediators of
muscle wasting in cachexia.
affects the majority of patients with advanced cancer and is associated
with a reduction in treatment tolerance, response to therapy, quality
of life, and duration of survival.
Cachexia is a complex multifactorial syndrome characterized by weight
loss (WL) and specific losses of muscle and/or adipose tissue.
Based on current knowledge, it is not possible to predict who will
develop cancer cachexia and who will not. Such variation may partly be
due to genotype. Knowledge of genotypic variation could contribute to
early identification of risk and allow institution of prophylaxis.
a candidate gene approach, research by our group identified cancer
cachexia with several single nucleotide polymorphisms (SNPs); among
these, a variant from the SELP gene (P-selectin) was investigated for functional significance. Since then, many new target genes have been reported[4-10];
these genes are involved in the key mechanisms thought to contribute to
cancer cachexia, and their transcripts have been shown to play
significant roles in the regulation of pathways such as muscle and
adipose tissue homeostasis We have recently published a review of
candidate genes and polymorphisms in cancer cachexia.[11, 12]
Although there is depletion of both adipose tissue and lean body mass in cancer cachexia, WL per se has long been used as the diagnostic criterion, and this remains in current classification systems.
skeletal muscle loss may have the greatest impact on patients' function
and quality of life. It has recently been possible to quantify muscle
mass in cancer patients' diagnostic computerized tomography (CT) scans,
and low skeletal muscle index (LSMI) so identified is associated with
One limitation is that due to the absence of pre-illness scans, it is
not possible to document active muscle loss but rather LSMI determined
by pre-determined cut-offs. In the present study, we use LSMI as
synonymous with sarcopenia defined by cut-offs related to excess
mortality.[16, 17] The combination of LSMI and WL has been suggested to combine a focus on muscle mass with a dynamic process of active loss. We used >2% WL because this is the minimal level associated with an increased risk of mortality.
Such a combined definition proved superior to its individual components
in identification of cancer patients with skeletal muscle fibre
utilized a candidate gene approach to explore our hypothesis that
inter-individual variations in susceptibility to cachexia are partly due
to inherited genetic variations (host); remaining phenotypic variance
may be ascribed to the tumour or other comorbidity. One limitation was
the lack of large bio-banks characterized for cachexia phenotypes. We
developed such a bio-bank with our primary objective to compare our
entire panel of candidate SNPs and their association with WL with and
without LSMI. We also investigated whether genes demonstrating
significant associations had altered transcript expression in muscle
from cancer patients with or without those phenotypes.
Materials and methods
Genotyped cancer patients: new and prior study cohorts
were recruited between 2004 and 2012 from the National Health Service
Lothian, UK; Cross Cancer Institute, Edmonton, Canada; McGill University
Health Centre, Montreal, Canada; Palliative Research Centre, Norwegian
University of Science and Technology, Norway; Cantonal Hospital, St
Gallen, Switzerland; and Department of Medical Oncology, University
Hospital of Larissa, Greece (Table 1).
All subjects participated in clinical or research studies on cancer
cachexia at the host institutions under ethically approved protocols
allowing for analysis of patients' DNA. Recruitment was on presentation
to surgical, oncology, or palliative care clinics. Recruitment was
sequential with the following exclusions: (i) <18 years; (ii)
cognitive impairment; (iii) underlying infection; and (v) on
corticosteroids. Overall, 1276 patients were included (Table 1).
More than 98% were of European descent. Information on patients
included date of birth, date of diagnosis, and type and stage of cancer.
Height and weight were measured upon recruitment (at time of diagnosis
of cancer). Pre-morbid weight was recalled and verified where possible
from the medical notes. WL was calculated and expressed as percentage of
pre-morbid body weight lost. The documentation of WL depends on
accurate recall. Studies in healthy populations suggest a strong
correlation between recalled and measured weight.
CT scans closest to the time of diagnosis (within 30 days on average)
were selected. About 943 patients were informative for cachexia
according to WL and LSMI. All patients provided written informed consent
for analysis of their DNA.Table 1. Patient demographics
Skeletal muscle transcriptome study
Patients who contributed to the muscle transcriptomic bio-bank have been described recently. Review of medical charts and CT images identified WL status and muscularity.
>5%, >10%, >15%. A range of WL was used to provide a subgroup
analysis to identify associations that would have been missed with a
single cut-off: the interest is to detect all potential associations in a
polygenic model where the variants are likely to be of lower penetrance
yet conferring finite effects.
- LSMI with
any degree of WL (>2%): analysis of CT scans allows classification
as LSMI or not. Cut-offs for LSMI were defined in relation to survival
duration of advanced cancer patients.
Computerized tomography analysis
stored CT images completed with a spiral CT were analysed as described
previously. Cross-sectional area for muscle was normalized for stature
(cm2/m2) and a lumbar skeletal muscle index (SMI) computed.[16, 21] SMI cut-offs for LSMI were based on a CT-based study of cancer patients by Martin et al..
Candidate gene and single nucleotide polymorphisms selection
Candidate genes and SNP selections were based on a systematic literature review.[11, 12] Candidate SNPs met the following criteria: previously published association with cancer cachexia,[22-24] statistically significant association with cancer cachexia in our prior study but still requiring validation, likely role in cancer cachexia based on functional or clinical relevance in more than one study, significant SNPs identified in a preliminary study,
and those SNPs that had been identified in relation to
pro-inflammatory/anti-inflammatory pathways, neuronal melanocortin
signalling pathways, energy regulation, appetite regulation, muscle, and
adipose tissue catabolic pathways since our prior study.
was performed on the Sequenom iPLEX Gold platform (San Diego, CA, USA)
or TaqMan assay (for rs4280262) using services from the McGill
University and Genome Quebec Innovation Centre, Montreal, Quebec,
Canada. Polymorphisms selected were validated for assay feasibility
using DNA from healthy Caucasians (n = 92) (Coriell Panel, Coriell Institute of Medicine, CA, USA).
Of the 148 SNPs selected initially (21 SNPs from a previous association
study and 127 newly selected SNPs for this study), for Sequenom
platform, 15 SNPs failed at the multiplex assay design stage, and 15
SNPs were non polymorphic, leaving 118 SNPs for genotyping. Assay
duplicates for 154 samples genotyped for all 118 SNPs; 100% concordance
for replicates was obtained. Of the 1452 patient samples, detailed
clinical annotations for the study end points were available for 1276
patients (Table 1).
Germline DNA isolated from buffy coat cells from these 1276 individuals
were interrogated for the 118 SNPs. SNP call rates >90% were
retained for all subsequent analysis (two SNPs did not meet this
criteria; rs4280262 and rs1544410: call rates of 80 and 86%,
respectively). Three SNPs showed a minor allele frequency <5%, and
these were excluded (rs1805086; rs2536; and rs16139), leaving 113 SNPs
from a total of 62 genes (Supporting Information Table S1). Deviations from Hardy–Weinberg equilibrium (HWE) were assessed in the Coriell panel of controls using the χ2 test with 1 degree of freedom; a P-value
of <0.001 was considered significant deviation from the HWE
proportions. None of the 118 SNPs considered for association analysis
showed deviations from HWE.
Microarray analysis was conducted as previously described.
The data used in this publication have been deposited in the US
National Centre for Biotechnology Information Gene Expression Omnibus25
and are accessible through GEO series accession number GSE41726.
calculations used Quanto. For the most prevalent cachexia phenotype
(i.e. >5% WL, 50% affected), the present study has 87% power to
detect an odds ratio of 1.5 for SNPs with a mean allele frequency of
>0.05. For the least prevalent cachexia phenotype (i.e. >15% WL,
16% affected), the present study has 35% power to detect an odds ratio
of 1.5 for SNPs with a mean allele frequency of >0.05.
Gene association study
Statistical analysis was conducted as previously described. Briefly, analyses were performed using PLINK (version 1.06).
Analyses were adjusted for covariates: age at diagnosis, sex,
pre-diagnosis body mass index, tumour type, and stage. Patients meeting
the criteria for each of the cachexia phenotypes were compared with
patients who had lost <5% body weight as control. To account for
multiple testing, permutation testing was performed using the adaptive
permutation test in PLINK within each phenotype. Finally, candidate
genes (and the SNPs in the corresponding gene regions) were grouped on
functional similarity according to gene ontology (AmiGO) (Supporting
Information Table S2).
The set-based test in PLINK was used to analyse association between
grouped SNPs and cachexia = phenotypes. The latter selects the best set
of SNPs whose mean of these single SNP statistics is significant after
permutation, which is particularly suited to large-scale candidate gene
studies. The empirical P-values were obtained by a permutation of 10 000 times of phenotype labels.
correlation analysis assessed the relationship between the phenotypes
independently (SMI or WL) with the expression of transcripts from select
candidate genes. t-test compared how SMI or WL values differed
with high vs. low expression for each of the candidate genes. The high
and low groups were determined by expression intensity and splitting
patients into three equal groups. The extremes were compared while
leaving out middle values. Cases considered for SMI and WL phenotypes
for gene expression were based on sorting of transcript expression in
all samples and binning based on extremes as described earlier. The
samples used for SNP studies and gene expression studies are from
non-matched cases as these two were independent studies.
Characteristics of the patient population are presented in Table 1.
Average age was 65 ± 13 years (mean ± SD). The majority was stage III
or IV. Average WL was 6 ± 9%. Of the patients with CT scans for the
assessment of muscularity, 47% had LSMI. There were no significant
differences in age, stage of disease, pre-diagnosis body mass index, and
percentage WL between patients who had CT scans suitable for the
measurement of muscularity and the entire cohort (Table 1).
Weight loss alone phenotype (n = 1276)
lists results for SNPs associated with cancer cachexia in patients
classified according to WL alone. Sixteen SNPs had significant
associations with various cachexia phenotypes based on increasing
severity of WL. Two SNPs (rs1935949 and rs4946935) found within
chromosome 6 in the Forkhead box O3 (FOXO3) gene associated with WL of
increasing severity (>5% and >10%) and one SNP (rs2297627) found
in the Forkhead box O1 (FOXO1) gene associated with WL > 10%.Table 2. Genes with variants significantly associated with cancer cachexia in patients classified according to weight loss alone
Weight loss plus low skeletal muscle index phenotype (n = 943)
lists all SNPs associated significantly with cancer cachexia classified
according to LSMI + WL >2% in all recruited patients. The analysis
compared those with the LSMI + WL >2% phenotype against those without
in the entire cohort. rs12409877 is in the leptin receptor (LEPR)
located on chromosome 3. rs2268757 is located in the activin receptor
type-2B (ACVR2B) gene on chromosome 3. SNPs in the tumour necrosis
factor (TNF) (rs1799964) and ACE (rs4291) genes also associated with the
phenotype.Table 3. Genes
with variants significantly associated with cancer cachexia in patients
classified according to weight loss >2% and low skeletal muscle
index compared with those who do not
Combining genes with functional similarity according to gene ontology
lists the phenotypes for candidate gene groups associated with specific
cancer cachexia phenotypes. SNPs in groups of genes involved in
appetite regulation, cell adhesion, cell membrane structure and
function, and signal transduction were associated with the phenotype WL
>10%. Only SNPs in the group of genes involved in cell adhesion were
significant with increasing WL. SNPs in groups of genes involved in
lipid metabolism, appetite regulation, signal transduction, and
glucocorticoid signalling were associated with the phenotype LSMI and WL
>2%. No SNPs in groups of genes were found to be significant with
all other phenotypes.Table 4. Candidate gene groups associated with cancer cachexia phenotypes
Table 5 lists the results from correlation and t-test
analysis between phenotypes and gene transcript level for the genes
that showed significant associations with any of the cachexia
phenotypes. Expression of ACVR2B, FOXO1 and 3, GCKR, LEPR, and TLR4
transcripts was significantly associated with different levels of SMI or
WL (P < 0.05). Specifically, these were all negatively
correlated with muscularity. FOXO1 and 3 and GCKR were the only genes
significantly correlated with WL; these were correlated negatively with
WL.Table 5. Results from correlation and t-test analysis between patient characteristics and rectus abdominus muscle gene transcripts for selected genesa
Associations with different cachexia phenotypes
In the present study, four SNPs are associated with WL + LSM (Table 3).
Two of these SNPs are associated with muscle metabolism in two genes
(ACVR2B and ACE), one with fat metabolism in one gene (LEPR) and one
with cytokine production in one gene (TNF). It would be attractive to
assign specific functional significance to the genetic signatures
identified. For example, ACVR2B decoy receptors abrogate muscle loss and
prolong survival in several murine models of cancer cachexia.
rs1799964 in the TNF gene and rs4291 in the ACE gene are new
associations (c.f. WL alone) when classification is based on WL + LSMI.
These findings focus attention on pro-inflammatory cytokines and the
renin–angiotensin system as biomarkers/mediators of muscle wasting in
cachexia. Replication of the present findings along with genome-wide
scans and an imputation approach to fine map the loci are needed in
parallel with functional studies (see the following) to resolve this
For the WL phenotype, sixteen candidate SNPs were identified (Table 2).
Seven of these SNPs are associated with muscle metabolism in five genes
(IGF1, CPN1, FOXO1, FOXO3, and ACVR2B), four are associated with
adipose tissue metabolism in two genes (LEPR and APOE), two with the
immune response in two genes (SELP and TLR4), two with corticosteroid
signalling in one gene (GCKR), and one with appetite regulation in one
gene (GHRL). Two polymorphisms (rs1935949 and rs4946935) in the gene
encoding for FOXO3 were consistently associated with WL of increasing
severity (>5% and >10%) (Table 2).
On the basis that WL is a continuum, the observation that both SELP and
FOXO3 associate with the highest degrees of WL suggests that these
signatures may be of particular significance. A recent study in a mouse
model of cancer cachexia demonstrated that FOXO-dependent transcription
is key in controlling diverse gene networks in skeletal muscle during
In keeping with our prior study, we confirmed in a larger validation cohort (Stage 2, n
= 545) that patients who carry the C allele of the rs6136 SNP in the
SELP gene are at a reduced risk of cachexia defined by WL (>5%,
>10%). This was confirmed recently in chemo-naïve patients with
locally advanced or metastatic pancreatic cancer.
Gene group analysis
The two dominant mechanisms of WL in cancer are anorexia /reduced food intake and abnormal metabolism. Appetite regulation was found to associate with the cachexia trait WL >10% (P = 0.0041). Regarding metabolism, lipid metabolism associated with LSMI and WL >2% (P = 0.0138). Fatty infiltration (myosteatosis) has been associated with cancer cachexia and reduced survival.[17, 31] The glucocorticoid signalling pathway also associate with LSMI and WL >2% (P = 0.0337). Glucocorticoids and associated signalling pathways accelerate protein degradation in muscle.
The muscle transcriptome is altered in the presence of cancer cachexia.[33, 34]
In the present study, there was concordance between a proportion of the
selected genes and either the level of WL or muscularity (Table 5). FOXO1 and FOXO3 are good examples: SNPs in both genes associated with the WL phenotype (Table 2) and transcript levels of both showed a correlation with WL (Table 5).
These transcription factors are not only key in the pro-inflammatory
driven up-regulation of the ubiquitin–proteasome pathway but also act as
negative regulators of the anabolic Akt-mTOR pathway.[8, 35]
present SNP analysis was not genome-wide, and therefore, other variants
with possible functional significance may have not have been examined.
Equally, the true functional significance of any individual SNP is
mostly unknown. It may be better to consider the genetic associations
identified as genetic signatures or biomarkers associated with the
cachexia syndrome. Interestingly, 17 of the 19 SNPs reported as showing
significant associations are in intronic, 3′, or 5′ Un translated
regions (UTRs). The purpose was to probe into the potential functional
impact of the loci as SNPs in this study are potentially proxy to the
causal variants (not yet captured in the region), which may also have an
influence on gene expression; as such, the probe position in the
expression array and the SNP position are not the same. Extrapolation to
an SNP under being an expression quantitative trait loci is premature.
The SNP identified may in some cases also affect gene expression
signatures not addressed herein (as in cis-acting and trans-acting
expression quantitative trait loci). The correlation pattern (albeit,
low to modest) observed is still encouraging because the trends reported
here for an SNP loci and gene expression are within the scope of known
cachexia literature. There is also a growing body of evidence that
microRNAs are involved in cancer cachexia,
and it may be that the newly discovered SNPs alter the gene transcripts
of these highlighted genes. Animal models may well be useful to look at
the biology of altering the transcripts from the genes where the SNPs
Equally, for those genes for which no strong
relationship was found between gene expression and patient
characteristics, it is important to consider that these may not be
transcriptionally regulated. For systemic mediators (e.g. cytokines), it
may be that circulating concentration is important rather than local
expression because tissue-specific expression may be transient, but the
activation of the signal transduction cascade could be the largest
It is important that the prevalence of LSMI is
in excess of that observed in the normal age-matched population. The
prevalence of LSMI/sarcopenia in age-matched subjects living in the
community varies according to the definition and methodology used but is
reported between 1 and 29%.
The prevalence of LSMI in this study was ~48%. Thus, the gene
associations with LSMI represent associations with a level of
muscularity at least partly independent of age, sex, or stature.
Clearly, there are other reasons why cancer patients may lose muscle
mass and weight apart from their tumour-related cachexia, e.g. severe
chronic obstructive pulmonary disease. Such co-morbidities were not
graded prospectively in the current study but should be considered for
the characterization of future cohorts.
gene SNP analysis offers the advantage that it is hypothesis-driven and
the associations are easily explained owing to compelling biological
rationale. However, the limitations are that the role of hitherto
unexplored genes and pathways that otherwise contribute to the trait
under investigation are missed. Issues surrounding phenotype complexity
are addressed in part in this study, and conducting a genome-wide
association study using high density of markers on the genome would help
relate the overlap of SNPs/pathways to the phenotypes of interest. The
consensus definitions for phenotypes may evolve in an iterative manner
from the cumulative wisdom from candidate SNPs, genome-wide association
study, and the current definitions available for cachexia. This could
potentially lead to the discovery of new SNPs depending on the phenotype
authors certify that they comply with the ethical guidelines for
authorship and publishing of the Journal of Cachexia, Sarcopenia and
Muscle: update 2015.
We thank Mr. Ashok Narasimhan for valuable discussions. Funding for
this project is from the Canadian Institutes of Health Research (CIHR)
operating grants awarded to S.D. and V.B. for cachexia network grant,
The Royal College of Surgeons of Edinburgh awarded to N.J., and this
research was partially supported by a grant from the Terry Fox Research
Institute, Canada. B Gagnon is a recipient of ‘Chercheur-clinicien
Boursier’ award from Fonds de recherche Québec Santé, Québec, Canada.
M.L.T. is the holder of the Jeanne and Jean-Louis Levesque Chair in
Cancer Research at McGill University.
Conflict of interest
No authors declare a conflict of interest.
N.J., B.T., J.R., V.B., S.D., and K.F. designed research.
N.J., C.S., and S.D. conducted research.
C.S., B.T., T.S., S.S., N.S., I.G., R.S., C.D., A.V., O.B., M.T., S.K.,
F.S., B.G., and V.B. provided essential reagents or provided essential
N.J. and C.S. analysed data and performed statistical analysis.
N.J., C.S., V.B., S.D., and K.F. wrote the paper.
N.J., V.B., S.D., and K.F. had primary responsibility for final content.
(angiotensin converting enzyme)
(activin receptor type-2B)
(protein kinase B)
(body mass index)
(cell adhesion molecule)
(Chronic obstructive pulmonary disease)
(carboxypeptidase N polypeptide 1)
(expression quantitative trait loci)
(insulin-like growth factor 1)
(low skeletal muscle index)
(mean allele frequency)
(mammalian target of rapamycin)
(National Centre for Biotechnology Information)
(nuclear factor kappa-light-chain-enhancer of activated B cells)
(Norwegian University of Science and Technology)
(peroxisome proliferator-activated receptor gamma)
(RNA integrity number)
(skeletal muscle index)
(single nucleotide polymorphism)
(toll like receptor 4)
(tumour necrosis factor receptor superfamily member 1A)
(WD Repeat Domain 20)