1. Epidemiology and Global Health

Common genetic variations in telomere length genes and lung cancer: a Mendelian randomisation study and its novel application in lung tumour transcriptome

  1. Ricardo Cortez Cardoso Penha  Is a corresponding author
  2. Karl Smith-Byrne
  3. Joshua R Atkins
  4. Philip C Haycock
  5. Siddhartha Kar
  6. Veryan Codd
  7. Nilesh J Samani
  8. Christopher Nelson
  9. Maja Milojevic
  10. Aurélie AG Gabriel
  11. Christopher Amos
  12. Paul Brennan
  13. Rayjean J Hung
  14. Linda Kachuri
  15. James D Mckay  Is a corresponding author
  1. Genomic Epidemiology branch, International Agency for Research on Cancer/World Health Organization (IARC/WHO), France
  2. Cancer Epidemiology Unit, University of Oxford, United Kingdom
  3. MRC Integrative Epidemiology Unit, Bristol Population Health Science Institute, Bristol Medical School (PHS), United Kingdom
  4. Department of Cardiovascular Sciences, University of Leicester, United Kingdom
  5. NIHR Leicester Biomedical Research Centre, Glenfield Hospital, United Kingdom
  6. Ludwig Lausanne Branch, Faculty of Biology and Medicine, Switzerland
  7. Institute for Clinical and Translational Research, Baylor College of Medicine, United States
  8. Lunenfeld-Tanenbaum Research Institute, Sinai Health, Canada
  9. Departament of Epidemiology and Population Health, Stanford University, United States
https://doi.org/10.7554/eLife.83118
Share this article
Cite this article
  1. Ricardo Cortez Cardoso Penha
  2. Karl Smith-Byrne
  3. Joshua R Atkins
  4. Philip C Haycock
  5. Siddhartha Kar
  6. Veryan Codd
  7. Nilesh J Samani
  8. Christopher Nelson
  9. Maja Milojevic
  10. Aurélie AG Gabriel
  11. Christopher Amos
  12. Paul Brennan
  13. Rayjean J Hung
  14. Linda Kachuri
  15. James D Mckay
(2023)
Common genetic variations in telomere length genes and lung cancer: a Mendelian randomisation study and its novel application in lung tumour transcriptome
eLife 12:e83118.
https://doi.org/10.7554/eLife.83118
  2. Download BibTeX
  3. Download .RIS

Abstract

Background:

Genome-wide association studies (GWASs) have identified genetic susceptibility variants for both leukocyte telomere length (LTL) and lung cancer susceptibility. Our study aims to explore the shared genetic basis between these traits and investigate their impact on somatic environment of lung tumours.

Methods:

We performed genetic correlation, Mendelian randomisation (MR), and colocalisation analyses using the largest available GWASs summary statistics of LTL (N=464,716) and lung cancer (N=29,239 cases and 56,450 controls). Principal components analysis based on RNA-sequencing data was used to summarise gene expression profile in lung adenocarcinoma cases from TCGA (N=343).

Results:

Although there was no genome-wide genetic correlation between LTL and lung cancer risk, longer LTL conferred an increased risk of lung cancer regardless of smoking status in the MR analyses, particularly for lung adenocarcinoma. Of the 144 LTL genetic instruments, 12 colocalised with lung adenocarcinoma risk and revealed novel susceptibility loci, including MPHOSPH6, PRPF6, and POLI. The polygenic risk score for LTL was associated with a specific gene expression profile (PC2) in lung adenocarcinoma tumours. The aspect of PC2 associated with longer LTL was also associated with being female, never smokers, and earlier tumour stages. PC2 was strongly associated with cell proliferation score and genomic features related to genome stability, including copy number changes and telomerase activity.

Conclusions:

This study identified an association between longer genetically predicted LTL and lung cancer and sheds light on the potential molecular mechanisms related to LTL in lung adenocarcinomas.

Funding:

Institut National du Cancer (GeniLuc2017-1-TABAC-03-CIRC-1-TABAC17‐022), INTEGRAL/NIH (5U19CA203654-03), CRUK (C18281/A29019), and Agence Nationale pour la Recherche (ANR-10-INBS-09).

Editor's evaluation

This study is of interest to epidemiologists and geneticists studying the association between telomere length and lung cancer risk. This work provides useful insight into risk factors for lung cancer. Overall, the results of this study are solid, as the genetic instrument used here is better powered and the battery of Mendelian randomization analysis makes this broad set of results convincing compared to previous work.

https://doi.org/10.7554/eLife.83118.sa0

Introduction

Telomeres are a complex of repetitive TTAGGG sequences and nucleoproteins located at the end of chromosomes and have an essential role in sustaining cell proliferation and preserving genome integrity (de Lange, 2009). Telomere length progressively shortens with age in proliferative somatic cells due to incomplete telomeric regions replication (Watson, 1972) and low activity of the telomerase TERT in adult cells. The shortening of the telomere length results in cell cycle arrest, cellular senescence, and apoptosis in somatic cells (Harley et al., 1990). The maintenance of telomere length, which allows cancer cells to escape the telomere-mediated cell death pathways, is one feature related to the hallmarks of cancer (Hanahan and Weinberg, 2011).

Telomere length appears to vary between individuals and has been studied in relation to many diseases. In observational studies, telomere length is measured as the average length of telomeric sequences in a given tissue (Montpetit et al., 2014). Telomere length appears correlated across tissue types (Demanelis et al., 2020), and as such, leukocyte telomere length (LTL) is generally measured in epidemiologic studies as a proxy for telomere length in other tissues. Recently, LTL has been measured in 472,174 individuals from the UK Biobank (UKBB; Codd et al., 2021), and LTL was associated with multiple biomedical traits (i.e. pulmonary and cardiovascular diseases, haematological traits, lymphomas, kidney cancer, and other cancer types). Genetic analysis of LTL also revealed 138 genetic loci linked to LTL across a variety of different genes involved in telomere biology and DNA repair (Codd et al., 2021).

In the context of lung cancer, genetic variants at several loci have been associated with both LTL and lung cancer risk, including variants near the TERT, TERC, OBFC1, and RTEL1 genes, fundamental to telomere length maintenance (McKay et al., 2008; Wang et al., 2008; Rafnar et al., 2009; Kachuri et al., 2016; McKay et al., 2017). The effects of the telomere-related variants appear more relevant to lung adenocarcinoma risk than other histologic subtypes (McKay et al., 2017; Landi et al., 2009). Accordingly, a causal relationship between LTL and susceptibility to lung cancer was observed using Mendelian randomisation (MR) approaches (Zhang et al., 2015; Haycock et al., 2017; Kachuri et al., 2019) as well as in observational studies that have associated directly measured telomere length with risk of lung cancer (Sanchez-Espiridion et al., 2014; Zhang et al., 2017).

The aim of the current work was to investigate the relationship between genetically predicted LTL and lung cancer, including lung cancer histological subtypes and smoking status. To this end, we conducted genome-wide correlations, MR, and colocalisation analyses to explore the relationship between LTL and lung cancer. We additionally undertook polygenic risk score (PRS) analysis using the LTL genetic instrument to explore the influence of LTL on the demographic, clinical, and molecular features of lung adenocarcinoma tumours.

Materials and methods

Reporting guidelines

Request a detailed protocol

The current study has been reported according to the STROBE-MR guidelines (Reporting Standards Document).

Data

Genome-wide association studies (GWASs) summary statistics for lung cancer (29,239 cases and 56,450 controls) and stratified by histological subtype (squamous cell carcinoma, small-cell carcinoma, and adenocarcinoma) and smoking status (ever and never smokers) were obtained from the International Lung Cancer Consortium (ILCCO; McKay et al., 2017). All analyses of LTL requiring summary statistics used results from a GWAS of LTL in 464,716 individuals of European ancestry from the UKBB (Codd et al., 2021). Downstream analyses considered additional lung cancer risk factors, such as lung function and cigarette smoking. We obtained GWAS summary statistics for forced expiratory volume in 1 s (FEV1) and forced vital capacity (FVC) from a published UKBB analysis (Kachuri et al., 2020). For smoking behaviour traits, we used results from the GWAS and Sequencing Consortium of Alcohol and Nicotine use (GSCAN) consortium meta-analysis of cigarettes per day (continuous), smoking initiation (ever versus never), smoking cessation (successfully quit versus continuing), and age at smoking initiation (continuous; Liu et al., 2019) excluding the UKBB participants. For the obesity-related traits (continuous), we used the results from the UKBB and GIANT meta-analysis of BMI (Pulit et al., 2019) and waist-to-hip ratio (WHR; Pulit et al., 2019), or OpenGWAS data using UKBB participants (Elsworth et al., 2020) for high-density lipoprotein (HDL), triglycerides, and systolic and diastolic blood pressure. For the alcohol behaviour trait, we obtained the results from GSCAN phase 2 of drinks per week (continuous; Saunders et al., 2022). Colocalisation analyses of gene expression used lung tissue expression quantitative trait loci (eQTL) summary statistics from the Genotype-Tissue Expression (GTEx) data version 8.

Analyses of molecular phenotypes were performed using 343 lung adenocarcinoma samples of European ancestry from The Cancer Genome Atlas (TCGA) cohort with germline profile, RNA-sequencing, and epidemiological data available. Genotyping and imputation of germline variants have been described elsewhere (Gabriel et al., 2022). The total somatic mutation burden of TCGA samples was obtained from Ellrott et al., 2018, and DNA mutational signatures were extracted and attributed, as previously described (Gabriel et al., 2022). RNA-sequencing data were obtained from TCGA data portal using TCGA biolinks package in R (version 2.22.3; Colaprico et al., 2016). Telomere length measurement by whole-genome sequencing (WGS-measured TL, 655 samples across cancer sites) was retrieved from Barthel et al., 2017.

Tumour genomic characteristics were defined by the analyses of the TCGA data, including gene expression-based scores of telomerase activity (Barthel et al., 2017) and cellular proliferation (Thorsson et al., 2018), as well as the observed frequency of somatic homologous recombination-related events (represented as a homologous recombination repair deficiency score), and the average number of somatic copy number alteration within the tumours (Knijnenburg et al., 2018).

Linkage disequilibrium score regression

Request a detailed protocol

Genetic correlations across traits were calculated using linkage disequilibrium score regression (LDSC) by the LDSC package (v1.0.0; Bulik-Sullivan et al., 2015). Linkage disequilibrium (LD) scores were generated on the 1000 Genomes Project Phase 3 reference panel with the HLA region excluded as provided by the package due to long range LD patterns. The genome-wide correlations that passed Bonferroni correction (adjusted p-values<0.05) were considered statically significant.

Mendelian randomisation

Request a detailed protocol

MR is a method for interrogating relationships between putative risk factors and health outcomes by using genetic variants associated with the exposure of interest, typically obtained from GWAS, as instrumental variables. Assuming that fundamental MR assumptions are satisfied, this approach can be said to identify unbiased causal estimates. The genetic instrument for LTL was defined as the set of 144 genetic variants that were genome-wide significant (p<5x10–08) but not in linkage disequilibrium with each other (r2<0.01) and restricted to common genetic variation (minor allele frequency >1%) in European populations. Proxy variants in LD (r2>0.8, whenever possible) were chosen when a genetic variant was not available in the lung cancer GWAS. Primary MR analyses were conducted using the inverse-variance method with multiplicative random-effects (Yavorska and Burgess, 2017). Sensitivity analyses to horizontal pleiotropy and other violations of MR assumptions were performed using other MR estimation methods, such as weighted median, MR-Egger, contamination mixture model, MR-PRESSO, and MR-RAPS (Yavorska and Burgess, 2017; Sanderson et al., 2022). Multivariable MR (MVMR) methods included the inverse-variance weighted, MR-Egger, and least absolute shrinkage and selection operator (LASSO)-based methods (Yavorska and Burgess, 2017; Sanderson et al., 2019).

Colocalisation methods

Request a detailed protocol

Unlike MR, where the goal is to assess the evidence for a causal effect of an exposure on an outcome, colocalisation is agnostic with respect to direction of effect and only assesses the probability that the two traits are affected by the same genetic variants at a given locus. Colocalisation can be viewed as a complementary approach for evaluating MR assumptions within specific genes or regions since strong evidence of colocalisation indicates overlap in genetic mechanisms affecting LTL and lung cancer. We used COLOC (v5.1.0; Wallace, 2020) to estimate the posterior probability for two traits sharing the same causal variant (PP4) in a 150 kb LD window, with PP4 >0.70 corresponding to strong evidence of colocalisation, as previously suggested (Wallace, 2020; Lopes et al., 2022). Priors chosen for the colocalisation analyses were p1=10–3, p2=10–4, and p12=10–5, or approximately, a 75% prior belief that a signal will only be observed in the LTL GWAS and less than 0.01% prior belief in favour of colocalisation between the two traits at a given locus (Giambartolomei et al., 2014). Conditioning and masking colocalisation methods were also used as they may identify putative shared causal variants in the presence of multiple causal variants present in a defined LD window (Wallace, 2021). We present the average PP4 from all methods as our posterior belief in favour of colocalisation between LTL and lung cancer risk. Multi-trait colocalisation based on a clustering algorithm was also performed using HyPrColoc (v1.0) to identify shared genetic signals with other lung cancer-related traits (Foley et al., 2021).

Principal component analysis based on RNA-sequencing data

Request a detailed protocol

Read counts of RNA-sequencing data were normalised within (GC-content and gene length) and between (sequencing depth) lane procedures by EDASeq R package (version 2.28.0; Risso et al., 2011) and excluding low read counts. Principal component analysis was applied using singular value decomposition method, after excluding extreme outliers. Pathway analyses were conducted using Gene Set Enrichment Analysis software (GSEA, version 4.2.3; Subramanian et al., 2005) on gene annotations from Gene Ontology database. Pathway analyses were restricted to the top 500 genes positively and negatively correlated with each principal component that passed multiple-testing correction (Bonferroni-adjusted p-value<0.05 for 74,465 tests), which is the maximum number of genes supported by the online software.

The PRS for LTL was composed of the same 144 variants used in the MR analysis and was computed as the sum of the individual’s beta-weighted genotypes using PRSice-2 software (Choi and O’Reilly, 2019). Associations were estimated per SD increase in the PRS, which was normalised to have a mean of zero across lung adenocarcinoma samples of European ancestry within the TCGA cohort. The associations between the eigenvalues of the gene expression principal components (outcome) and demographic, clinical, and genomic features related to genome stability (predictors derived from TCGA published papers and TCGA data portal, except for the DNA mutational signatures [Gabriel et al., 2022]) were calculated using a multivariate linear regression model.

Inferring PC2 gene expression signature based on RNA-sequencing data

Request a detailed protocol

The TCGA-lung adenocarcinoma (LUAD) tumour samples were split into training (70%, N=255) and validation (30%, N=108) datasets. Principal components analysis based on RNA-sequencing data was performed to summarise the gene expression profiles of lung adenocarcinoma tumours into five principal components in the training and validation datasets separately, as previously described (see ‘Principal component analysis based on RNA-sequencing data’ section in methods). Subsequently, we applied the partial least squares-based method called rigid transformation (Hubert and Branden, 2003) to align the first five principal components in both datasets. This method compares embeddings, low-dimensional representations in both datasets, performing a slightly rotation of principal components in order to translate and match them in both training and validation datasets. To select the most informative genes of PC2 in the training dataset, RNA levels of the genes correlated with PC2 (N=3914 out of 14,893 genes, FDR <0.05 for 14,893 tests) were selected as variables for the LASSO regression models. The LASSO tune parameters were chosen by resampling the training dataset (1000 bootstraps: root mean of SE=0.12 ± 0.0004, Lambda = 10 x 10–10, r2=0.99 ± 0.00007) using the tidymodels metapackage in R (v1.0.0; wrapper of glmnet). The 10 genes selected by the LASSO model were used to infer the gene expression signature of PC2 by adding up the scaled values of the log-normalised read counts of each gene multiplied by the respective LASSO regression coefficients. For validation purpose, the inferred PC2 signature was calculated in the validation dataset and compared with the observed principal components. After validation in the subset of the TCGA-LUAD cohort, the inferred PC2 signature was calculated in TCGA-LUSC dataset to compare differences between lung cancer histological subtypes.

Results

Genome-wide genetic correlations

The design of the study is represented in Figure 1—figure supplement 1. We first assessed the shared genetic basis of telomere length, lung cancer risk, and other putative lung cancer risk factors, such as smoking behaviours (age start smoking, smoking cessation, smoking initiation, and cigarettes per day) and lung function (FEV1 and FVC) using genome-wide correlations (Figure 1A). There was little evidence for genetic correlations by LDSC between LTL variants and lung cancer (rg = −0.01, p=0.88) or when stratified by histologic subtypes (Figure 1A). Increasing LTL was genetically correlated with older age at smoking initiation (rg = 0.13, p=3.0 × 10–3), and negatively correlated with smoking cessation: (rg = −0.21, p=6.9 × 10–09), smoking initiation (rg = −0.16, p=1.3 × 10–10), and cigarettes per day (rg = −0.19, p=2.1 × 10–08). Longer LTL was genetically correlated with improved lung function, as indicated by increasing values of FEV1 (rg = 0.09, p=5.1 × 10–07) and FVC (rg = 0.09, p=1.1 × 10–05). To better understand the absence of genome-wide correlations between LTL and lung cancer, we visualised the Z-scores for each trait for approximately 1.2 million variants included in the LDSC analyses (Figure 1B). A subgroup of variants associated with longer LTL was correlated with increased lung adenocarcinoma risk, while the subgroup of smoking-behaviour associated variants, which also conferred an increased risk of lung adenocarcinoma, tended to have lower LTL.

Figure 1 with 1 supplement see all
Download asset Open asset
Genetic correlations between leukocyte telomere length (LTL) and lung cancer (LC) related traits.

(A) Heatmap representing the genetic correlation analyses (rg) for LTL across LC, histological subtypes (lung adenocarcinoma [ADE], squamous cell carcinoma [SQC], and small-cell carcinoma [SCC]), smoking propensity (cigarettes per day [CPD], smoking cessation [SmkCes], Smoking initiation [SmkInit], and age of smoking initiation [AgeSmk]), and lung function related (forced vital capacity [FVC] and forced expiratory volume [FEV1]) traits. The black star indicates correlations that passed Bonferroni correction (p<4x10–04). Heritability (h2) as the proportion of the phenotypic variance caused by SNPs. (B) Plot of Z-scores (ADE versus LTL), restricting to the Hapmap SNPs (~1.2 million) but excluding HLA region. Genome-wide significant SNPs (p<5x10–08) for each trait were coloured (CPD in red, SmkInit in dark red, LTL in dark blue, AgeSmk in blue, SmkCes in lightblue, and not genome-wide hits for LTL or any other selected trait in white). Linear regression line was coloured in red.

MR analyses

From the 490 genetic instruments associated with LTL at genome-wide significance (p<5x10–08), 144 LTL genetic instruments, that explained ~3.5% of the variance in LTL, and were in low-linkage disequilibrium (r2<0.01) were used in MR analysis (Supplementary file 1a). As a sensitivity analysis, a PRS composed of these genetic instruments was associated with TL estimated from WGS in blood samples across TCGA cohorts (Beta = 0.03, 95%CI = 0.01–0.05, p=0.001) but was not associated with TL in tumour material from the same patients (Figure 2—figure supplement 1).

MR analyses demonstrated that longer genetically predicted LTL was associated with increased lung cancer risk (OR = 1.62, 95% CI = 1.44–1.84, p=9.91 × 10–15) (Figure 2; Supplementary file 1). Longer LTL conferred the largest increase in risk for lung adenocarcinoma tumours (OR = 2.43, 95% CI = 2.02–2.92, p=3.76 × 10–21), but there was limited evidence of a causal relationship for other histologic subtypes, such as squamous cell carcinoma (OR = 1.00, 95% CI = 0.84–1.19, p=0.98) and small-cell carcinoma (OR = 1.13, 95% CI = 0.87–1.45, p=0.34; Figure 2, Supplementary file 1). However, our study was underpowered to detect an association between lung small-cell carcinoma and LTL at OR of 1.13 and considering alpha type-1 error rate of 5% (Figure 2—figure supplement 1). When stratifying the analyses by smoking status, LTL was associated with lung cancer risk in both never (OR = 2.02, 95% CI = 1.45–2.83, p=3.78 × 10–05) and ever smokers (OR = 1.54, 95% CI = 1.34–1.76, p=7.75 × 10–10; Figure 2, Supplementary file 1). Evidence for negative pleiotropy (supplementary file 1c) and heterogeneity (supplementary file 1d) was observed for all lung cancer outcomes except for squamous cell carcinoma. However, a significant association for LTL and lung cancer risk was found for methods robust to the significant directional pleiotropy (MR-Egger: lung cancer [OR = 2.35, p=3.37 × 10–13]; lung adenocarcinoma [OR = 4.48, p=7.30 × 10–17]; never smokers [OR = 6.84, p=2.07 × 10–10]; supplementary file 1b). Leave-one-out analyses detected only one outlier, rs7705526 in TERT, resulting in >10% change in MR effect size for associated lung cancer subtypes (supplementary file 1e). MVMR analyses considering instruments related to LTL and WHR, HDL, total triglycerides, systolic blood pressure, smoking, and alcohol intake, as well as multiple traits combined, suggested that the association between LTL and lung adenocarcinoma risk is independent of smoking propensity, obesity-related, and alcohol intake-related traits (supplementary file 1f, 1g).

Figure 2 with 1 supplement see all
Download asset Open asset
Genetically predicted leukocyte telomere length (LTL) association with lung cancer.

Lung cancer (by histology or by smoking status) risk associations with the LTL instrument from the inverse-variance-weighted MR analyses are expressed as OR per SD increase in genetically predicted LTL. Statistically significant associations with p-values<0.05 (red square). Heterogeneity is estimated by the statistic I2, tau variance of subgroups (τ2), and p-values for Cochran’s Q heterogeneity measure.

Colocalisation analyses

We investigated whether there was evidence of shared genetic signals between LTL and lung adenocarcinoma at loci centred on the 144 genetic instruments used in MR analyses using colocalisation methods (Figure 3A, supplementary file 1h). Loci with evidence of colocalisation between LTL and lung adenocarcinoma tended to be near genes that encode telomerase subunits and its associated complex, including genetic variants at TERT (5p15.33; rs116539972, rs7705526, rs61748181, rs71593392, and rs140648021), TERC (3q26.2; rs12638862 and rs146546514), and OBFC1 (10q24.33; rs9419958 and rs139122544). Several colocalised loci mapped to genes that have not been previously linked to lung cancer risk at genome-wide significant level: MPHOSPH6 (16q23.3; rs2303262), PRPF6 (20q13.33; rs80150989), and POLI (18q21.2; rs2276182). Other telomere maintenance genes showed limited evidence of colocalisation with lung adenocarcinoma (i.e. TERF1 and PIF1). For instance, while the RTEL1 locus (20q13.33: rs117238689, rs115610405, rs35640778, and rs35902944) harboured variants associated with both LTL and lung adenocarcinoma (Figure 3—figure supplement 1), these signals appeared to be distinct and independent of each other (avg_PP3=0.999, avg_PP4=0.001; Figure 3A, supplementary file 1h).

Figure 3 with 1 supplement see all
Download asset Open asset
Colocalisation analyses for the genetic loci defined by the 144 leukocyte telomere length (LTL) variants.

(A) Distribution of the average posterior probability for shared genetic loci between LTL and lung adenocarcinoma, highlighting in orange the telomere maintenance loci that colocalised (avg_PP4≥0.70) and in blue the ones where there was limited evidence for colocalisation (avg_PP4<0.70). Dashed red line represents the arbitrary avg_PP4 cutoff of 0.70. Representative stack plots for the multi-trait colocalisation results within (B) MPHOSPH6 and (C) OBFC1 loci, centred on a 150 kb LD window of rs2303262 and rs9419958 variants, respectively. Left Y-axis represents the –log10(p-values) of the association in the respective genome-wide association study for a given trait. The right Y-axis represents the recombination rate for the genetic loci. The X-axis represents the chromosome position. SNPs are coloured by the linkage disequilibrium correlation threshold (r2) with the query labelled SNP in European population. Sentinel SNPs within the defined LD window were labelled in each trait.

We further evaluated whether the loci colocalised between LTL and lung adenocarcinoma also shared genetic signals with other traits related to lung cancer susceptibility (supplementary file 1i). Multi-trait analyses at the 16q23.3 locus colocalised rs2303262 with MPHOSPH6 expression in lung tissue, FVC and FEV1, but not with any of the traits related to smoking behaviour (p=0.72; Figure 3B, supplementary file 1i). We additionally identified evidence of colocalisation (p=0.74) between lung adenocarcinoma, LTL, and gene expression in lung epithelial cells for two variants at the OBFC1 locus: rs139122544 and rs9419958 (Figure 3C, supplementary file 1i).

Genetically predicted LTL association with tumour features

We investigated the impact of genetically predicted LTL on lung adenocarcinoma tumour features by estimating molecular expression patterns within 343 lung adenocarcinomas tumours using principal component analysis in RNA-sequencing data. The first five components explained ~54% of the observed variance in the RNA-sequencing data (Figure 4, Figure 4—figure supplement 1). To explore the biological meaning of the five components, we performed pathway analyses for the top 500 genes with the highest loadings in each component (supplementary file 1j, supplementary file 1k). Overall, the genes correlated with each component tended to be enriched for specific cell signaling pathways (PC1: RNA processing; PC2: cell-cycle; PC3: metabolic processes; PC4: immune response; PC5: cellular response to stress and DNA damage; false discovery rate <5%; supplementary file 1l).

Figure 4 with 1 supplement see all
Download asset Open asset
Associations between molecular expression patterns of lung adenocarcinoma tumours, LTL PRS, and The Cancer Genome Atlas (TCGA) features.

(A) LTL PRS association with the first five principal components based on RNA-sequencing data of lung adenocarcinomas tumours (n=343). Results are expressed as beta estimate per SD increase in genetically predicted LTL. Linear regression model adjusted by sex, age, smoking status, and PC1-5 (genetic ancestry) covariates. Statistically significant associations with p-values<0.05 (red square). (B) Heatmap representing the correlations among PC2 and selected molecular features related to telomere length canonical roles. LTL = leukocyte telomere length; PRS = polygenic risk score; PC = principal component; TMB = tumour total mutation burden; HRD = homologous recombination deficiency, SBS (single base substitution DNA mutational signatures). SBS1 and SBS5 are DNA mutational signatures associated with age-related processes, and SBS4 is associated with tobacco smoking exposure. X-shaped marker to cross correlations with p-value>0.05.

We then tested the association between the PRS composed of the 144 genetic instruments selected for MR analysis and the five components of gene expression within lung adenocarcinoma tumours (Figure 4A). The LTL PRS was positively associated with the second component (PC2) of tumour expression (Beta = 0.17, 95% CI = 0.12–0.19, p=1.0 × 10–3; Figure 4A). In multivariate analysis, higher values of PC2 tended to be associated with patients older at diagnosis (p=0.001), female (p=0.005), being never smokers (p=0.04), and diagnosed with early-stage tumours (p=0.002; Table 1). PC2 was also highly correlated with gene expression-based measure of cell proliferation and several genomic features related to genomic stability (Figure 4B). In multivariate analysis, higher values of PC2 were associated with reduced tumour proliferation (p=3.7 × 10–30), lower somatic copy number alternations (p=1.6 × 10–05), and higher tumour telomerase activity scores (p=1.6 × 10–5). Multivariate analysis also indicated that LTL PRS remained an independent predictor of PC2 when considering these genomic features (p=0.009; Table 1). It is noteworthy only nominal associations between LTL PRS and above-mentioned features and none remained statistically significant after correction for multiple testing (supplementary file 1m). We next inferred the gene expression signature of PC2, based on 10 genes informative of this component selected by the LASSO regression models, in both lung adenocarcinoma (TCGA-LUAD) and squamous cell carcinoma (TCGA-LUSC) cohorts (Figure 5, Figure 5—figure supplement 1). The association between LTL PRS and inferred PC2 was observed in TCGA-LUAD (p=0.001) but not in TCGA-LUSC (p=0.729) cases (Figure 5A). The inferred PC2 signature levels were higher in TCGA-LUAD than in TCGA-LUSC (Figure 5B), while higher proliferation rate (Figure 5C, p=1.45 × 10–141) and TERT activity (Figure 5D, p=1.36 × 10–20) were observed in TCGA-LUSC than in TCGA-LUAD cases. Of note, the low RNA levels of the telomere-related genes (less than five read counts), such as TERT and TERC, in both TCGA-LUAD and TCGA-LUSC tumour samples limited the direct comparison of these genes between these cohorts.

Figure 5 with 1 supplement see all
Download asset Open asset
Comparing inferred PC2 gene expression signature by lung cancer histological subtypes.

(A) Leukocyte telomere length (LTL) polygenic risk score (PRS) association with the 10-gene expression signature of PC2 in lung adenocarcinoma (The Cancer Genome Atlas [TCGA]-LUAD, N=343) and squamous cell carcinoma (TCGA-LUSC, N=338) cases from TCGA dataset. Results are expressed as beta estimate per SD increase in genetically predicted LTL. Linear regression model adjusted by sex, age, and PC1-5 (genetic ancestry) covariates, PC2 signature as outcome. Statistically significant associations with -values<0.05. Values per SD of (B) PC2, (C) proliferation score, and (D) telomerase/TERT activity gene expression signatures by lung cancer histological subtypes (TCGA-LUAD and TCGA-LUSC). p-Values derived from Student’s t-tests.

Table 1
Association between PC2 (outcome) and lung adenocarcinoma tumour features in univariate and multivariate models (n=343).
Non-molecular features
PredictorsUnivariate modelMultivariate model
OR/Beta (SE)p-valueOR/Beta (SE)p-value
Age at diagnosis*0.17±0.050.0010.17±0.050.001
Gender (male)†0.73±0.110.0050.74±0.110.005
Smoking status (ever)†0.67±0.160.0130.72±0.150.035
Tumour stage (late)†0.67±0.130.0020.67±0.130.002
Molecular features
PredictorsUnivariate modelMultivariate model
Beta (SE)p-valueBeta (SE)p-value
LTL PRS ‡0.17±0.050.0010.10±0.040.009
Telomerase activity–0.37±0.059.34E-130.25±0.061.32E-05
Proliferation–0.69±0.043.30E-46–0.80±0.063.66E-30
Copy number alteration–0.41±0.056.36E-16–0.23±0.051.62E-05
Homologous recombination deficiency–0.4±0.058.32E-150.12±0.060.048
Tumour total mutation burden–0.28±0.051.37E-07–0.09±0.240.695
SBS1–0.18±0.050.0010.01±0.050.827
SBS4–0.24±0.056.36E-060.04±0.180.814
SBS5–0.27±0.054.84E-070.03±0.090.770

  1. SBS (single base substitution DNA mutational signatures); LTL = leukocyte telomere length; PC = principal component; PRS = polygenic risk score.

  2. *

    age of diagnosis represented as beta estimate per 1 unit of SD.

  3. †

    OR per 1 unit of SD.

  4. ‡

    LTL PRS is adjusted by first five PC of genetic ancestry in the univariate model.

Discussion

The maintenance of telomere length is one of the hallmarks of cancer, being critical for cell proliferation and genome integrity (Hanahan and Weinberg, 2011). Individual differences in telomere length, measured either directly or indirectly by germline determinates, have been linked with multiple diseases, including cancer susceptibility (Codd et al., 2021). The measurement of LTL within the UKBB has provided a resource for the development of a more powerful set of genetic instruments that capture a greater proportion of variation in LTL compared to previous studies (Codd et al., 2021). We applied genetic determinants of LTL to the largest GWAS of lung cancer to further characterise the role of telomere maintenance in lung cancer aetiology.

Using an MR analysis framework, we confirmed the previously reported relationship between genetically predicted longer LTL and increased risk of lung cancer. Our expanded genetic instrument detected systematic negative pleiotropy, which has not been observed in previous MR studies (Zhang et al., 2015; Haycock et al., 2017; Kachuri et al., 2019). Correcting for this pervasive directional bias resulted in substantially larger effects of LTL on risk of lung adenocarcinoma and lung cancer in never smokers, implying that LTL may even be more important to these phenotypes than previously estimated (Zhang et al., 2015; Haycock et al., 2017; Kachuri et al., 2019). Our observations in never smokers were also supported by multivariate MR analyses where adjustment for smoking did not attenuate the effect of LTL on lung cancer susceptibility. We used MVMR analysis (Sanderson et al., 2019) to assess the potential that factors, such as BMI, smoking, alcohol use, and other obesity related factors, may account for the association between LTL and lung cancer. While the influence of alternative unknown potential confounding factors cannot be excluded, the association of LTL with lung cancer risk was materially unaltered following adjustment for a range of potential confounders considered in our MVMR analyses.

Colocalisation analyses for the variants selected as the MR genetic instrument highlighted shared genetic signals between LTL and lung adenocarcinoma, including loci near genes related to telomere length maintenance (TERT, TERC, and OBFC1) and three genetic loci not previously linked to lung cancer susceptibility (POLI, PRPF6, and MPHOSPH6). The lung cancer risk allele of the MPHOSPH6 sentinel variant (rs2303262) was associated with longer LTL, reduced pulmonary function, and increased MPHOSPH6 gene expression in lung tissue. MPHOSPH6 encodes an enzyme associated with the RNA exosome complex where it modulates RNA binding activity. PRPF6 in 20q13.33 is involved in androgen binding and has been shown to promote colon tumour growth via preferential splicing of genes involved in proliferation (Adler et al., 2014). POLI is a member of the Y-family of DNA damage-tolerant polymerases involved in translesion synthesis (McIntyre, 2020). As part of its role in DNA repair and replication stress, POLI interacts with TP53 to bypass barriers during DNA replication, which may confer a pro-survival effect to stem cells and cancer cells (Guo et al., 2021).

Despite the limited evidence of colocalisation between several loci important for telomere maintenance and lung cancer, the MR analyses restricting to the non-colocalised LTL SNPs pointed out that these loci might lie in the same causal pathway of LTL and lung cancer, highlighting the heterogeneity in the genetic effects of LTL loci.

We additionally identified the relationship between genetic determinants of LTL and a specific gene expression component in lung adenocarcinoma tumours. The aspect of this component associated with longer LTL was also associated with demographic and clinical features, such as never smoking, female, and early-stage tumours compared with other lung adenocarcinoma patients, which is an interesting but not completely understood lung cancer strata. This expression component also tended to be related to genomic features related to genomically stable tumours and strikingly associated with cell proliferation score, implying that this component might be a proxy for this feature. These results appear consistent with the canonical role of telomere length in preserving genome stability and cell proliferation (de Lange, 2009).

A plausible explanation for why long LTL was associated with an increased risk of lung cancer might be that individuals with longer telomeres have lower rates of telomere attrition compared to individuals with shorter telomeres. Given a very large population of histologically normal cells, even a very small difference in telomere attrition would change the probability that a given cell is able to escape the telomere-mediated cell death pathways (Aviv et al., 2017). Such inter-individual differences could suffice to explain the modest lung cancer risk observed in our MR analyses. However, it is not clear why longer TL would be more relevant to lung adenocarcinoma compared to other lung cancer subtypes. A suggestion may come from our observation that longer LTL is related to genomically stable lung tumours (such as lung adenocarcinomas in never smokers and tumours with lower proliferation rates) but not genomically unstable lung tumours (such as heavy smoking related, highly proliferating lung squamous carcinomas). One possible hypothesis is that histologic normal cells exposed to highly genotoxic compounds, such as tobacco smoking, might require an intrinsic activation of telomere length maintenance at early stages of carcinogenesis that would allow them to survive, and therefore, genetic differences in telomere length are less relevant in these cells. By contrast, in more genomically stable lung tumours, where TL attrition rate is more modest, the hypothesis related to differences in TL length may be more relevant and potentially explain the heterogeneity in genetic effects between lung tumours. Alternately, we also note that the cell of origin may also differ, with lung adenocarcinoma postulated to be mostly derived from alveolar type 2 cells, the squamous cell carcinoma is from bronchiolar epithelium cells (Sainz de Aja et al., 2021), possibly suggesting that LTL might be more relevant to the former.

One surprising finding from the genetic analysis was that despite the robust and large effects of LTL on lung cancer risk observed in the MR analyses, the genetic correlation between LTL and lung cancer was effectively null. The LDSC approach considers genetic variants across the entire genome, whereas the MR approach preferentially selects variants based on their association with LTL, restricting to those that achieved genome-wide significance. One possibility for the lack of genetic correlation between LTL and lung cancer is that genetic variants may differ in the direction that they influence these traits, and we used the smoking behaviour traits to exemplify that. For example, the subgroup of genetic variants noted at genome-wide significance from LTL studies was associated with increased LTL and lung cancer risk. However, the subgroup related to smoking behaviours which, in turn, are linked with increased lung cancer risk, tends to decrease LTL. If such opposing effects were widespread across the genome, it could account for the lack of genetic correlation between LTL and lung cancer estimated by LDSC and highlights the complex nature of the genetic variants that determine LTL and lung cancer risk.

Some limitations of this study should be acknowledged. A limited sample size might have limited the detection of an association between lung small-cell carcinoma and LTL. Our colocalisation approach is generally more conservative and may fail to accurately determine the posterior probability for shared genetic signals in the presence of multiple independent associations in a given locus (Hukku et al., 2021), which may be a reasonable explanation for the lack of colocalisation observed at RTEL1 locus, and we stress that many of the variants that are COLOC negative are likely to be associated with lung cancer. Furthermore, the relatively small sample size of the lung adenocarcinoma cohort from TCGA may have reduced the power of our study, and larger cohorts of expression profiles tumours will be necessary to validate and explore some of our findings. The potential limitations such as collider bias within the lung adenocarcinoma case only study design should also be considered.

In conclusion, we describe an association between long genetically predicted LTL and lung cancer risk, which provides insights into how telomere length influences the genetic basis of lung cancer aetiology, including never smoking and female lung adenocarcinoma, which is an enigmatic subset of lung cancer. By using a novel framework to explore the biological implications of genetically complex traits, we unravel one gene expression component, highly correlated with proliferation rate score and other genomic stability-related features, associated with LTL in lung adenocarcinoma tumours. As far as we are aware, this is the first time an association between a PRS related to an aetiological factor, such as telomere length, and a particular expression component in the lung tumours is reported. Our findings suggest that lung adenocarcinoma patients with longer LTL might have more genomically stable tumours than the ones with shorter LTL, shedding some light on telomere biology in those tumours.

Appendix 1

Additional methods: Datasets

UK Biobank

The UK Biobank (UKBB) is a large cohort with more than 500,000 individuals from the United Kingdom who were recruited between 2006 and 2010, as previously described (Bycroft et al., 2018). At the time of recruitment, individuals provided electronic signed consent, answered questions regarding socio-demographic, lifestyle, health-related factors, and physical measures. The genetic data were obtained using the UK BiLEVE and the UKBB Axiom arrays. UKBB has ethical approval from the National Information Governance Board for Health and Social Care and the North-West Multicentre Research Ethics Committee (Ref: 11/NW/0382).

The latest genome-wide association study (GWAS) summary statistics of leukocyte telomere length (LTL; https://figshare.com/s/caa99dc0f76d62990195) were performed on 464,716 UKBB participants of European ancestry with 19.4 million imputed variants ( minor allele frequency≥ 1% and imputation quality score ≥0.3), adjusting by age, sex, array, and the first 10 principal components. LTL was measured from UKBB baseline samples, and values were log-transformed and Z-standardised.

The forced expiratory volume in 1 s (FEV1) and forced vital capacity (FVC) GWAS summary statistics were conducted on 372,750 and 370,638 individuals of European genetic ancestry, respectively, who had spirometry data and sample passed quality control of genetic data (Hardy–Weinberg equilibrium at p>1 × 10−5, MAF >0.5%, and imputation quality score ≥0.3). Linear regression models were applied to standardised Z-scores for FEV1 and FVC, adjusting by age, sex, genotyping array, and 15 principal components. These GWAS summary statistics are available by the corresponding authors of the original article upon reasonable request (see Kachuri et al., 2020).

International lung cancer consortium

The International Lung Cancer Consortium (ILCCO) is an international group of lung cancer researchers for sharing comparable data from ongoing lung cancer case-control and cohort studies. The lung cancer GWAS summary statistics used in the current work is derived from a collaborative effort between Transdisciplinary Research of Cancer in Lung of the ILCCO (TRICL-ILCCO) and the Lung Cancer Cohort Consortium (LC3). All participants in these studies signed an informed consent, approved by the local internal review board or ethics committee.

The lung cancer GWAS summary statistic (29,266 cases and 56,450 controls of European descent; McKay et al., 2017) is meta-analysis of the combined imputed genotypes from OncoArray series (14,803 cases and 12,262 controls) and previous lung cancer GWAS (IARC, MDACC, SLRI, ICR, Harvard, NCI, Germany, and deCODE studies; 44,188 controls and 14,436 cases). The meta-analysis only considered imputed variants with MAF >1%, imputation quality score ≥0.3, and with little evidence for heterogeneity in effect between the studies (Cochran’s Q statistic: p>0.05), adjusting by sex, age, and principal components. Additionally, analyses were conducted by lung cancer histological subtype (adenocarcinoma [11,273 cases and 55,483 controls], squamous cell carcinoma [7426 cases and 55,627 controls], and small-cell carcinoma [2664 cases and 21,444 controls]). Of note, histological subtype information was not available for all studies.

The GWAS and Sequencing Consortium of Alcohol and Nicotine use consortium

The GWAS and Sequencing Consortium of Alcohol and Nicotine use dataset is derived from an international genetic association meta-analysis consortium for tobacco smoking and alcohol consumption traits across 30 studies and up to 1.2 million individuals (Liu et al., 2019). For the GWAS summary statistics used in the current study, UKBB participants were excluded from the meta-analysis, and only genetic variants present in more than 19 studies, MAF >1%, and imputation quality score >0.3 were considered. The smoking traits were defined as follow: cigarettes per day (CPD; N=337,334; continuous variable defined as average number of cigarettes smoked per day, either as a current smoker or former smoker), smoking initiation (SmkInit; N=1,232,091; categorical variable defined as ever smokers versus never smokers), smoking cessation (SmkCes; N=547,219; categorical variable defined as current smokers versus former smokers), and age of initiation (AgeSmk; N=341,427; continuous variable defined as the age at which an individual first became a regular smoker). Age, sex, and genetic principal components were used as covariates in all analyses. GWAS summary statistics were downloaded from https://conservancy.umn.edu/handle/11299/201564.

The Genotype-Tissue Expression (GTEx) project data is a project that aims to investigate the effects of genetic variants in transcriptome and their impact on regulatory mechanisms of a traits and diseases (GTEx Consortium, 2020). The GTEx data contain 15,201 RNA-sequencing samples from 49 tissues of 838 postmortem donors. For the current study, we accessed version 8 of the GTEx program to obtain the effect estimates and risk alleles of eQTL genetic variants (in normal lung tissue) around the queried SNPs for the multi-trait colocalisation analyses.

The Cancer Genome Atlas

The Cancer Genome Atlas (TCGA) is currently the biggest resource of cancer genomics, with more than 20,000 primary cancer and matched normal samples molecularly characterised across 33 cancer types with available demographic and clinical data.

For the current study, the access for the germline data of TCGA dataset was obtained via TCGA project #2731 via dbGAP (phs000178.v11.p8). Genotyping and imputation from TCGA data were performed in-house and described elsewhere (Gabriel et al., 2022). Briefly, germline variants with low genotyping call rate (<97%), MAF <1%, strong deviation (p-value<10–8) from the Hardy–Weinberg equilibrium, and allele frequency differing more than 20% from 1000 Genomes (1000 G) within European ancestry group were removed. Phasing was performed using Eagle (v2.4.1; Gabriel and Lipinski, 2021) with the 1000 genomes phase 3 reference panel, and Minimac4 (v1.0.1; Gabriel and Lipinski, 2021) was used to perform the imputation (window size of 500 kb; Das et al., 2016). Samples with discrepancies between genetic inferred and self-reported sex as well as with high inter-sample relatedness (PI_HAT >0.185%) were removed.

Of the 480 lung adenocarcinoma cases with genotype data, 364 cases had available tumour RNA-sequencing data (more details in Data availability section). We excluded 20 cases due to missing information for smoking status (ever smokers versus never smokers) from the downstream analyses. After summarising the normalised log-transformed read counts for all genes in two dimensions, one sample was removed due to extreme gene expression profile in comparison with all the lung adenocarcinoma tumours. Thus, 343 lung adenocarcinoma tumour samples were selected for the principal component analysis and association with LTL polygenic risk score (PRS), molecular, clinic, and demographic variables of TCGA dataset.

Data availability

Lung cancer GWAS summary statistics obtained from ILCCO can be accessed by the database of Genotypes and Phenotypes (dbGAP) under accession phs000876.v1.p1. The GWAS summary statistics for tobacco-smoking behaviors (GSCAN: https://conservancy.umn.edu/handle/11299/201564), LTL (https://figshare.com/s/caa99dc0f76d62990195), and GTEx version 8 (downloaded via GTEx google cloud resource) are publicly available. Germline data of TCGA cohorts were accessed by dbGAP under accession number phs000178.v11.p8 and project application #2731. RNA-sequencing data from TCGA cohorts were retrieved from GDC open-access data portal (https://portal.gdc.cancer.gov/) using TCGAbiolinks package in R. TCGA-related data are publicly available as described in the data section. The code for LDSC analysis is available at: https://github.com/bulik/ldsc/wiki/Heritability-andGenetic-Correlation. The codes used in this study for two-sample MR, colocalisation, multi-trait colocalisation, and principal component analyses are available at https://github.com/ricardocortezcardoso/Telomere_Length_Code, (Penha, 2023a, copy archived at swh:1:rev:f365df300919c46bb99a96b4040d90576fc878e2). Plots were created using R packages ‘meta’ (v5.5, forest plots), ‘corrplot’ (v0.92, correlation matrix), and ‘ggplot2’ (v3.3.6). The R package to generate stackplots for visualisation of the multi-trait colocalisation results is available at https://github.com/jrs95/gassocplot, (Penha, 2023b, copy archived at swh:1:rev:ae6a59dff2e43d39eead3d483af1d50f151c3d5b).

References

    1. Haycock PC
    2. Burgess S
    3. Nounu A
    4. Zheng J
    5. Okoli GN
    6. Bowden J
    7. Wade KH
    8. Timpson NJ
    9. Evans DM
    10. Willeit P
    11. Aviv A
    12. Gaunt TR
    13. Hemani G
    14. Mangino M
    15. Ellis HP
    16. Kurian KM
    17. Pooley KA
    18. Eeles RA
    19. Lee JE
    20. Fang S
    21. Chen WV
    22. Law MH
    23. Bowdler LM
    24. Iles MM
    25. Yang Q
    26. Worrall BB
    27. Markus HS
    28. Hung RJ
    29. Amos CI
    30. Spurdle AB
    31. Thompson DJ
    32. O’Mara TA
    33. Wolpin B
    34. Amundadottir L
    35. Stolzenberg-Solomon R
    36. Trichopoulou A
    37. Onland-Moret NC
    38. Lund E
    39. Duell EJ
    40. Canzian F
    41. Severi G
    42. Overvad K
    43. Gunter MJ
    44. Tumino R
    45. Svenson U
    46. van Rij A
    47. Baas AF
    48. Bown MJ
    49. Samani NJ
    50. van t’Hof FNG
    51. Tromp G
    52. Jones GT
    53. Kuivaniemi H
    54. Elmore JR
    55. Johansson M
    56. Mckay J
    57. Scelo G
    58. Carreras-Torres R
    59. Gaborieau V
    60. Brennan P
    61. Bracci PM
    62. Neale RE
    63. Olson SH
    64. Gallinger S
    65. Li D
    66. Petersen GM
    67. Risch HA
    68. Klein AP
    69. Han J
    70. Abnet CC
    71. Freedman ND
    72. Taylor PR
    73. Maris JM
    74. Aben KK
    75. Kiemeney LA
    76. Vermeulen SH
    77. Wiencke JK
    78. Walsh KM
    79. Wrensch M
    80. Rice T
    81. Turnbull C
    82. Litchfield K
    83. Paternoster L
    84. Standl M
    85. Abecasis GR
    86. SanGiovanni JP
    87. Li Y
    88. Mijatovic V
    89. Sapkota Y
    90. Low S-K
    91. Zondervan KT
    92. Montgomery GW
    93. Nyholt DR
    94. van Heel DA
    95. Hunt K
    96. Arking DE
    97. Ashar FN
    98. Sotoodehnia N
    99. Woo D
    100. Rosand J
    101. Comeau ME
    102. Brown WM
    103. Silverman EK
    104. Hokanson JE
    105. Cho MH
    106. Hui J
    107. Ferreira MA
    108. Thompson PJ
    109. Morrison AC
    110. Felix JF
    111. Smith NL
    112. Christiano AM
    113. Petukhova L
    114. Betz RC
    115. Fan X
    116. Zhang X
    117. Zhu C
    118. Langefeld CD
    119. Thompson SD
    120. Wang F
    121. Lin X
    122. Schwartz DA
    123. Fingerlin T
    124. Rotter JI
    125. Cotch MF
    126. Jensen RA
    127. Munz M
    128. Dommisch H
    129. Schaefer AS
    130. Han F
    131. Ollila HM
    132. Hillary RP
    133. Albagha O
    134. Ralston SH
    135. Zeng C
    136. Zheng W
    137. Shu X-O
    138. Reis A
    139. Uebe S
    140. Hüffmeier U
    141. Kawamura Y
    142. Otowa T
    143. Sasaki T
    144. Hibberd ML
    145. Davila S
    146. Xie G
    147. Siminovitch K
    148. Bei J-X
    149. Zeng Y-X
    150. Försti A
    151. Chen B
    152. Landi S
    153. Franke A
    154. Fischer A
    155. Ellinghaus D
    156. Flores C
    157. Noth I
    158. Ma S-F
    159. Foo JN
    160. Liu J
    161. Kim J-W
    162. Cox DG
    163. Delattre O
    164. Mirabeau O
    165. Skibola CF
    166. Tang CS
    167. Garcia-Barcelo M
    168. Chang K-P
    169. Su W-H
    170. Chang Y-S
    171. Martin NG
    172. Gordon S
    173. Wade TD
    174. Lee C
    175. Kubo M
    176. Cha P-C
    177. Nakamura Y
    178. Levy D
    179. Kimura M
    180. Hwang S-J
    181. Hunt S
    182. Spector T
    183. Soranzo N
    184. Manichaikul AW
    185. Barr RG
    186. Kahali B
    187. Speliotes E
    188. Yerges-Armstrong LM
    189. Cheng C-Y
    190. Jonas JB
    191. Wong TY
    192. Fogh I
    193. Lin K
    194. Powell JF
    195. Rice K
    196. Relton CL
    197. Martin RM
    198. Davey Smith G
    199. Telomeres Mendelian Randomization Collaboration
    (2017) Association between telomere length and risk of cancer and non-neoplastic diseases: a mendelian randomization study
    JAMA Oncology 3:636–651.
    https://doi.org/10.1001/jamaoncol.2016.5945
    1. Liu M
    2. Jiang Y
    3. Wedow R
    4. Li Y
    5. Brazel DM
    6. Chen F
    7. Datta G
    8. Davila-Velderrain J
    9. McGuire D
    10. Tian C
    11. Zhan X
    12. Choquet H
    13. Docherty AR
    14. Faul JD
    15. Foerster JR
    16. Fritsche LG
    17. Gabrielsen ME
    18. Gordon SD
    19. Haessler J
    20. Hottenga J-J
    21. Huang H
    22. Jang S-K
    23. Jansen PR
    24. Ling Y
    25. Mägi R
    26. Matoba N
    27. McMahon G
    28. Mulas A
    29. Orrù V
    30. Palviainen T
    31. Pandit A
    32. Reginsson GW
    33. Skogholt AH
    34. Smith JA
    35. Taylor AE
    36. Turman C
    37. Willemsen G
    38. Young H
    39. Young KA
    40. Zajac GJM
    41. Zhao W
    42. Zhou W
    43. Bjornsdottir G
    44. Boardman JD
    45. Boehnke M
    46. Boomsma DI
    47. Chen C
    48. Cucca F
    49. Davies GE
    50. Eaton CB
    51. Ehringer MA
    52. Esko T
    53. Fiorillo E
    54. Gillespie NA
    55. Gudbjartsson DF
    56. Haller T
    57. Harris KM
    58. Heath AC
    59. Hewitt JK
    60. Hickie IB
    61. Hokanson JE
    62. Hopfer CJ
    63. Hunter DJ
    64. Iacono WG
    65. Johnson EO
    66. Kamatani Y
    67. Kardia SLR
    68. Keller MC
    69. Kellis M
    70. Kooperberg C
    71. Kraft P
    72. Krauter KS
    73. Laakso M
    74. Lind PA
    75. Loukola A
    76. Lutz SM
    77. Madden PAF
    78. Martin NG
    79. McGue M
    80. McQueen MB
    81. Medland SE
    82. Metspalu A
    83. Mohlke KL
    84. Nielsen JB
    85. Okada Y
    86. Peters U
    87. Polderman TJC
    88. Posthuma D
    89. Reiner AP
    90. Rice JP
    91. Rimm E
    92. Rose RJ
    93. Runarsdottir V
    94. Stallings MC
    95. Stančáková A
    96. Stefansson H
    97. Thai KK
    98. Tindle HA
    99. Tyrfingsson T
    100. Wall TL
    101. Weir DR
    102. Weisner C
    103. Whitfield JB
    104. Winsvold BS
    105. Yin J
    106. Zuccolo L
    107. Bierut LJ
    108. Hveem K
    109. Lee JJ
    110. Munafò MR
    111. Saccone NL
    112. Willer CJ
    113. Cornelis MC
    114. David SP
    115. Hinds DA
    116. Jorgenson E
    117. Kaprio J
    118. Stitzel JA
    119. Stefansson K
    120. Thorgeirsson TE
    121. Abecasis G
    122. Liu DJ
    123. Vrieze S
    124. 23andMe Research Team
    125. HUNT All-In Psychiatry
    (2019) Association studies of up to 1.2 million individuals yield new insights into the genetic etiology of tobacco and alcohol use
    Nature Genetics 51:237–244.
    https://doi.org/10.1038/s41588-018-0307-5
    1. McKay JD
    2. Hung RJ
    3. Han Y
    4. Zong X
    5. Carreras-Torres R
    6. Christiani DC
    7. Caporaso NE
    8. Johansson M
    9. Xiao X
    10. Li Y
    11. Byun J
    12. Dunning A
    13. Pooley KA
    14. Qian DC
    15. Ji X
    16. Liu G
    17. Timofeeva MN
    18. Bojesen SE
    19. Wu X
    20. Le Marchand L
    21. Albanes D
    22. Bickeböller H
    23. Aldrich MC
    24. Bush WS
    25. Tardon A
    26. Rennert G
    27. Teare MD
    28. Field JK
    29. Kiemeney LA
    30. Lazarus P
    31. Haugen A
    32. Lam S
    33. Schabath MB
    34. Andrew AS
    35. Shen H
    36. Hong Y-C
    37. Yuan J-M
    38. Bertazzi PA
    39. Pesatori AC
    40. Ye Y
    41. Diao N
    42. Su L
    43. Zhang R
    44. Brhane Y
    45. Leighl N
    46. Johansen JS
    47. Mellemgaard A
    48. Saliba W
    49. Haiman CA
    50. Wilkens LR
    51. Fernandez-Somoano A
    52. Fernandez-Tardon G
    53. van der Heijden HFM
    54. Kim JH
    55. Dai J
    56. Hu Z
    57. Davies MPA
    58. Marcus MW
    59. Brunnström H
    60. Manjer J
    61. Melander O
    62. Muller DC
    63. Overvad K
    64. Trichopoulou A
    65. Tumino R
    66. Doherty JA
    67. Barnett MP
    68. Chen C
    69. Goodman GE
    70. Cox A
    71. Taylor F
    72. Woll P
    73. Brüske I
    74. Wichmann H-E
    75. Manz J
    76. Muley TR
    77. Risch A
    78. Rosenberger A
    79. Grankvist K
    80. Johansson M
    81. Shepherd FA
    82. Tsao M-S
    83. Arnold SM
    84. Haura EB
    85. Bolca C
    86. Holcatova I
    87. Janout V
    88. Kontic M
    89. Lissowska J
    90. Mukeria A
    91. Ognjanovic S
    92. Orlowski TM
    93. Scelo G
    94. Swiatkowska B
    95. Zaridze D
    96. Bakke P
    97. Skaug V
    98. Zienolddiny S
    99. Duell EJ
    100. Butler LM
    101. Koh W-P
    102. Gao Y-T
    103. Houlston RS
    104. McLaughlin J
    105. Stevens VL
    106. Joubert P
    107. Lamontagne M
    108. Nickle DC
    109. Obeidat M
    110. Timens W
    111. Zhu B
    112. Song L
    113. Kachuri L
    114. Artigas MS
    115. Tobin MD
    116. Wain LV
    117. Rafnar T
    118. Thorgeirsson TE
    119. Reginsson GW
    120. Stefansson K
    121. Hancock DB
    122. Bierut LJ
    123. Spitz MR
    124. Gaddis NC
    125. Lutz SM
    126. Gu F
    127. Johnson EO
    128. Kamal A
    129. Pikielny C
    130. Zhu D
    131. Lindströem S
    132. Jiang X
    133. Tyndale RF
    134. Chenevix-Trench G
    135. Beesley J
    136. Bossé Y
    137. Chanock S
    138. Brennan P
    139. Landi MT
    140. Amos CI
    141. SpiroMeta Consortium
    (2017) Large-scale association analysis identifies new lung cancer susceptibility loci and heterogeneity in genetic susceptibility across histological subtypes
    Nature Genetics 49:1126–1132.
    https://doi.org/10.1038/ng.3892
    1. Saunders GRB
    2. Wang X
    3. Chen F
    4. Jang S-K
    5. Liu M
    6. Wang C
    7. Gao S
    8. Jiang Y
    9. Khunsriraksakul C
    10. Otto JM
    11. Addison C
    12. Akiyama M
    13. Albert CM
    14. Aliev F
    15. Alonso A
    16. Arnett DK
    17. Ashley-Koch AE
    18. Ashrani AA
    19. Barnes KC
    20. Barr RG
    21. Bartz TM
    22. Becker DM
    23. Bielak LF
    24. Benjamin EJ
    25. Bis JC
    26. Bjornsdottir G
    27. Blangero J
    28. Bleecker ER
    29. Boardman JD
    30. Boerwinkle E
    31. Boomsma DI
    32. Boorgula MP
    33. Bowden DW
    34. Brody JA
    35. Cade BE
    36. Chasman DI
    37. Chavan S
    38. Chen Y-DI
    39. Chen Z
    40. Cheng I
    41. Cho MH
    42. Choquet H
    43. Cole JW
    44. Cornelis MC
    45. Cucca F
    46. Curran JE
    47. de Andrade M
    48. Dick DM
    49. Docherty AR
    50. Duggirala R
    51. Eaton CB
    52. Ehringer MA
    53. Esko T
    54. Faul JD
    55. Fernandes Silva L
    56. Fiorillo E
    57. Fornage M
    58. Freedman BI
    59. Gabrielsen ME
    60. Garrett ME
    61. Gharib SA
    62. Gieger C
    63. Gillespie N
    64. Glahn DC
    65. Gordon SD
    66. Gu CC
    67. Gu D
    68. Gudbjartsson DF
    69. Guo X
    70. Haessler J
    71. Hall ME
    72. Haller T
    73. Harris KM
    74. He J
    75. Herd P
    76. Hewitt JK
    77. Hickie I
    78. Hidalgo B
    79. Hokanson JE
    80. Hopfer C
    81. Hottenga J
    82. Hou L
    83. Huang H
    84. Hung Y-J
    85. Hunter DJ
    86. Hveem K
    87. Hwang S-J
    88. Hwu C-M
    89. Iacono W
    90. Irvin MR
    91. Jee YH
    92. Johnson EO
    93. Joo YY
    94. Jorgenson E
    95. Justice AE
    96. Kamatani Y
    97. Kaplan RC
    98. Kaprio J
    99. Kardia SLR
    100. Keller MC
    101. Kelly TN
    102. Kooperberg C
    103. Korhonen T
    104. Kraft P
    105. Krauter K
    106. Kuusisto J
    107. Laakso M
    108. Lasky-Su J
    109. Lee W-J
    110. Lee JJ
    111. Levy D
    112. Li L
    113. Li K
    114. Li Y
    115. Lin K
    116. Lind PA
    117. Liu C
    118. Lloyd-Jones DM
    119. Lutz SM
    120. Ma J
    121. Mägi R
    122. Manichaikul A
    123. Martin NG
    124. Mathur R
    125. Matoba N
    126. McArdle PF
    127. McGue M
    128. McQueen MB
    129. Medland SE
    130. Metspalu A
    131. Meyers DA
    132. Millwood IY
    133. Mitchell BD
    134. Mohlke KL
    135. Moll M
    136. Montasser ME
    137. Morrison AC
    138. Mulas A
    139. Nielsen JB
    140. North KE
    141. Oelsner EC
    142. Okada Y
    143. Orrù V
    144. Palmer ND
    145. Palviainen T
    146. Pandit A
    147. Park SL
    148. Peters U
    149. Peters A
    150. Peyser PA
    151. Polderman TJC
    152. Rafaels N
    153. Redline S
    154. Reed RM
    155. Reiner AP
    156. Rice JP
    157. Rich SS
    158. Richmond NE
    159. Roan C
    160. Rotter JI
    161. Rueschman MN
    162. Runarsdottir V
    163. Saccone NL
    164. Schwartz DA
    165. Shadyab AH
    166. Shi J
    167. Shringarpure SS
    168. Sicinski K
    169. Skogholt AH
    170. Smith JA
    171. Smith NL
    172. Sotoodehnia N
    173. Stallings MC
    174. Stefansson H
    175. Stefansson K
    176. Stitzel JA
    177. Sun X
    178. Syed M
    179. Tal-Singer R
    180. Taylor AE
    181. Taylor KD
    182. Telen MJ
    183. Thai KK
    184. Tiwari H
    185. Turman C
    186. Tyrfingsson T
    187. Wall TL
    188. Walters RG
    189. Weir DR
    190. Weiss ST
    191. White WB
    192. Whitfield JB
    193. Wiggins KL
    194. Willemsen G
    195. Willer CJ
    196. Winsvold BS
    197. Xu H
    198. Yanek LR
    199. Yin J
    200. Young KL
    201. Young KA
    202. Yu B
    203. Zhao W
    204. Zhou W
    205. Zöllner S
    206. Zuccolo L
    207. Batini C
    208. Bergen AW
    209. Bierut LJ
    210. David SP
    211. Gagliano Taliun SA
    212. Hancock DB
    213. Jiang B
    214. Munafò MR
    215. Thorgeirsson TE
    216. Liu DJ
    217. Vrieze S
    218. 23andMe Research Team
    219. Biobank Japan Project
    (2022) Genetic diversity fuels gene discovery for tobacco and alcohol use
    Nature 612:720–724.
    https://doi.org/10.1038/s41586-022-05477-4

Decision letter

  1. Nicholas E Banovich
    Reviewing Editor; Translational Genomics Research Institute, United States
  2. Eduardo L Franco
    Senior Editor; McGill University, Canada
  3. Ben Voight
    Reviewer; University of Pennsylvania, United States

Our editorial process produces two outputs: (i) public reviews designed to be posted alongside the preprint for the benefit of readers; (ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "Common genetic variations in telomere length genes and lung cancer: a Mendelian Randomization study and its novel application in lung tumor transcriptome" for consideration by eLife. Your article has been reviewed by 2 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and a Senior Editor. The following individual involved in the review of your submission has agreed to reveal their identity: Ben Voight (Reviewer #2).

As is customary in eLife, the reviewers have discussed their critiques with one another. What follows below is the Reviewing Editor's edited compilation of the essential and ancillary points provided by reviewers in their critiques and in their interaction post-review. Please submit a revised version that addresses these concerns directly. Although we expect that you will address these comments in your response letter, we also need to see the corresponding revision clearly marked in the text of the manuscript. Some of the reviewers' comments may seem to be simple queries or challenges that do not prompt revisions to the text. Please keep in mind, however, that readers may have the same perspective as the reviewers. Therefore, it is essential that you attempt to amend or expand the text to clarify the narrative accordingly.

Essential revision:

The reviewers provided a set of critiques that are manageable in scale. Please respond to all of the reviewer comments.

Reviewer #1 (Recommendations for the authors):

Figures are difficult to read as they are poor resolution.

Reviewer #2 (Recommendations for the authors):

1. Resolution is not great on some of the figures – made reading them a little bit hard. I trust the authors will provide higher-resolution versions in the proofing phases!

2. Figure 1B – Not sure what this overall means/how to be interpreted – I think this is too much data plotted on the same figure for my taste.

3. I think the authors should create a supplementary table that enumerates the list of SNPs used for their instrument, along with annotated info about the SNP (pos, alleles, allele frequency), effect estimate, p-value report; then, for each SNP, the relevant outcome data of interest in the paper (effect sizes, se, P-value). That way, it is perfectly clear what SNPs are used and what estimated effects are used as input for MR experiments.

4. I'm a little confused about the colocalization analysis:

a. Suppl. Table 6: it looks to me like there are annotation mislabelling issues.

b. Also it isn't clear in this table which LTL loci have /any/ evidence of lung cancer association, i.e., PP1/2 vs. PP3.

For these, suggest checking the table and adding a column for the top lung cancer association (and SNP) in the interval tested by coloc.

c. But also confusing somewhat that there appear to be multiple TERT signals and those with colocalization but are basically on top of one another whilst still being LD independent (e.g., rs61748181 and rs33977403).

Overall here it looks like there may be some differences between genetic variants selected at this locus compared between Codd et al. and here. It could be these choices make sense and are reasonable, but might be worth looking back at differences to see if different choices drive the results they see.

5. In methods describing the Lung cancer data, are histologic data a complete subset of the overall data? It looks like there are shared controls (? a small limitation if so), but why is the number of controls in two strata greater than the overall? The cases totals are also less than overall, are 'lost' cases here simply because a subtype isn't specified? Perhaps the authors can clarify this a bit in methods.

6. Given the sample size differences, what effect sizes were the authors powered to observe in the Lung Squamous cell carcinoma and Lung Small Cell Carcinoma groups? Should this be interpreted as a false negative due to lack of power OR were the authors reasonably powered to observe (say) the estimated effect size overall in the univariate or epidemiological correlation data (but failed to see that effect), consistent with apparent heterogeneity in these strata?

7. As the author indicates, the MR Egger results (Table S2) suggest substantial 'negative' confounding. This is a little confusing – in my experience, I have seen negative confounding for sure previously, but this is probably some of the stronger effects that I've seen before. I don't have any great advice about how to look at this in detail, other than perhaps through MVMR, or by looking carefully at the IV regression plot to see if there are outliers that are somehow strongly influencing MR-Egger.

8. Inclusion of an epidemiological correlational data baseline between LTL >> lung cancer to compare to the MR estimates would also add value.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Common genetic variations in telomere length genes and lung cancer: a Mendelian Randomization study and its novel application in lung tumour transcriptome" for further consideration by eLife. Your revised article has been evaluated by a Senior Editor and a Reviewing Editor.

The manuscript has been improved but there are some remaining issues that need to be addressed, as outlined below:

New revisions required:

Reviewer #2 (Recommendations for the authors):

The authors have responded to most comments reasonably.

However, I still have some concerns about the given report. I know the collection of these points below are a little extra work -- I am honestly not trying to "add on" but do think these experiments are important to bring out the robustness of the author's claim, add value to the extant literature, and interpretation at these key points around the MR they describe.

1. First, focused around my original comment #2 (in the author's response, their point #6) on multivarible MR experiments. To my point that there are actually multiple potential confounders – including smoking, BMI, and waist-hip ratio that could potentially confound the interpretation that LTL explains the lung-cancer associations, the authors performed one experiment focused on BMI. They note that for each cancer trait there appears to be no impact on their primary association between LTL and Lung cancer data when considering BMI.

However, what they present is still a bit inadequate from my point of view, in the following ways.

1a. The authors perform a form of this MVMR which is a little 'less' good in my view for this -- they build a new instrument based on SNPs ascertained for association with BMI, and then include that in the MVMR regression analysis. This is not how I would do this -- this result describes a "joint" model for the effect of exposures (BMI, LTL) on outcome (Lung cancer). This version does not regress out any potential association of the SNPs *selected for association with LTL* has with the potential confounder. In this case, if the LTL variants had strong association with the confounder (e.g., BMI), I don't think that this would be fully accounted for.

Instead, the authors should take the n=144 SNPs selected for the LTL instrument, identify associations from extant data with the putative confounder of interest (e.g., BMI), create an instrument for the genetically mediated effects of that confounder, and include that as an instrument for covariate confounding.

I.e., "regress out" any effect that the LTL instrument might have on the confounder, and report adjusted effects of the LTL instrument on cancer risk.

Yes: this may NOT be a very strong instrument for the confounder -- but that's not the point here. The point is to determine the effect of genetically mediated LTL on cancer adjusting for any effect that instrument the authors built for it has on the confounder.

1b. Furthermore, considering a single confounder here is really suboptimal. As indicated in the previous comments and as the authors know, smoking behavior and other anthropomorphic measures (e.g., waist-hip ratio) are also correlated with everything here. So the authors should really consider an expanded list of confounders for this to be clearly convincing that LTL really does operate as a causal exposure independently of these other risk factors.

1c. In addition to experiments that are pairwise, they should also consider performing an experiment with all potential confounders included in the model

Cancer ~ LTL + BMI + WHR + SMOKING + …..

I.e., can the authors convincingly demonstrate that the observed relationship between LTL and these lung cancer outcomes is not explained by a battery of obvious confounders individually or as a collective

1d. As suggested earlier – I think value and novelty would be added if the authors really took this on and explored this space a bit more -- look hard and think carefully about what confounders really *COULD* explain this result and work hard to refute that those could explain these observations

I do think a robust set of experiments should be included in the primary report beyond smoking, either way (in the supplement). I think the authors report the BMI result in the rebuttal response to me but don't add those back to the paper. I'd honestly like to see a healthy version of that added to the paper, but that's my taste on this point.

2. On my previous point #4 (in the rebuttal, point #8), I asked the authors to consider a subset MR analysis which focused on those where the genes/signals map quite obviously on known telomere length biology and represent some of the strongest signals. The authors respond that they can't actually do this analysis, suggesting that using the data for co-localization somehow invalidates an MR sub-set analysis.

I thought about this a little bit and while I admit that I can be daft sometimes I honestly do not understand what basis – empirical or theoretical – this argument has to somehow drive positive or negative biases here, i.e., why they need an "out of sample" set to perform the MR after they used this data to perform colocalization. If there's specific literature or reasoning the authors would like to articulate this in more detail, that would certainly be appreciated!

My intuition is that this actually isn't much of an issue -- performing subset analyses for different reasons (biological or otherwise) is actually pretty reasonable and there are MR methods that actually try to cluster / subset instruments used.

This all goes to interpretation and as a sensitivity analysis -- if the result is driven entirely by variation at TERT, TERC, OBFC1 but not obviously by other variation that influences telomere length, then I personally think that is worth trying to understand and report. Analytically, this is a *trivial* experiment to perform in the UVMR space.

3. Regarding the rebuttal point #13, which involved my point about power for discovery in the subset given the sample size, the authors state that there are sample size difference but then write something about confidence intervals overlapping that I don't quite understand.

What I think the authors SHOULD do with respect to the tumor histology analysis is to report a credible set based on power calculation which determine what range of causal effect sizes the Lung Squamous CC and Lung Small CC were powered to discover (at an α type-1 error rate) of 5% and/or 1%, say, given the instruments used and sample sizes involved here.

This is actually a very easy thing to do analytically with some assumptions – back of envelope -- you could check out what formula I pulled together in PMID: 25165093 to get the distribution under the alternative; I believe there are also MR power calculators out there as web tools. You easily do this with some R calculations in a straight forward way.

I think the authors should be able to articulate is that effect sizes (say) of OR=1.6 or better would be discoverable in those lung cell subsets that basically catch the null hypothesis. Or, that those sets were well powered to discover ORs that were even smaller than that. This would then give you a quantitative assessment that this isn't simply a false-negative result, but that there is putative heterogeneity here. But if they are small enough that they AREN'T particularly well powered, then that's a pretty important interpretive point on this plot.

https://doi.org/10.7554/eLife.83118.sa1

Author response

Essential revision:

The reviewers provided a set of critiques that are manageable in scale. Please respond to all of the reviewer comments.

Reviewer #1 (Recommendations for the authors):

Figures are difficult to read as they are poor resolution.

We apologize for these lower resolution figures. All figures have been updated in higher resolution (600 dpi) in the revised manuscript.

Reviewer #2 (Recommendations for the authors):

1. Resolution is not great on some of the figures – made reading them a little bit hard. I trust the authors will provide higher-resolution versions in the proofing phases!

We apologize for these lower resolution figures. All figures have been updated in higher resolution (600 dpi) in the revised manuscript.

2. Figure 1B – Not sure what this overall means/how to be interpreted – I think this is too much data plotted on the same figure for my taste.

The figure 1B sought to provide insights on the possible reasons why there were not genome-wide genetic correlations observed between lung adenocarcinoma (or any lung cancer histological subtype) and genetically predicted leukocyte telomere length.

We are highlighting in this figure that only a small subset of LTL variants is associated with longer telomere length and higher risk of lung adenocarcinoma development on one hand, but variants also related to lung cancer risk such as smoking behavior traits are associated in the opposite direction. We speculated that a possibility for the lack of genetic correlation between LTL and lung cancer may be these opposing effects that may explain the lack of global genetic correlation between TL and lung adenocarcinoma.

We have re-worded the figure legend to make this figure clearer to the reader and further highlighted this aspect and link to this figure in the discussion.

3. I think the authors should create a supplementary table that enumerates the list of SNPs used for their instrument, along with annotated info about the SNP (pos, alleles, allele frequency), effect estimate, p-value report; then, for each SNP, the relevant outcome data of interest in the paper (effect sizes, se, P-value). That way, it is perfectly clear what SNPs are used and what estimated effects are used as input for MR experiments.

Thanks, this reviewer for the suggestion. We added supplementary file 1a listing all the SNPs used as LTL instruments in the current study in the revised manuscript.

4. I'm a little confused about the colocalization analysis:

a. Suppl. Table 6: it looks to me like there are annotation mislabelling issues.

We apologize for the annotation mislabelling issues, and the table was corrected in the revised manuscript. The genetic variants were annotated using ANNOVAR tool and the dbSNP150 database in the supplementary file 1g of the revised manuscript. Of the 144 SNPs from the LTL instrument, 11 were replaced by proxies (R2>0.8, whenever possible, in European population 1000 genomes) in the current study because they were missing in the lung adenocarcinoma GWAS summary statistic. A column was added to specify the proxies in the supplementary file 1g of the revised manuscript.

b. Also it isn't clear in this table which LTL loci have /any/ evidence of lung cancer association, i.e., PP1/2 vs. PP3.

For these, suggest checking the table and adding a column for the top lung cancer association (and SNP) in the interval tested by coloc.

We have included the columns with PP1/PP2/PP3 in the supplementary file 1g of the revised manuscript. Only one locus (RETL1) displayed high PP3 (0.99). We have added the information of this PP3 levels in our results describing the lack of colocalisation for the variants within RTEL1 locus (PP4=0) in the result section of the revised manuscript (page 12, lines 260-264).

c. But also confusing somewhat that there appear to be multiple TERT signals and those with colocalization but are basically on top of one another whilst still being LD independent (e.g., rs61748181 and rs33977403).

Again, we apologise for the error in the genetic variant annotation in the former supp table S7/ current supplementary file 1g noted above. This has complicated again the interpretation relative to this point. Supplementary file 1g has been corrected in the revised version of the manuscript. Indeed, there seems to be multiple independent hits at TERT locus when using our filtering (r2<0.01).

5. In methods describing the Lung cancer data, are histologic data a complete subset of the overall data? It looks like there are shared controls (? a small limitation if so), but why is the number of controls in two strata greater than the overall? The cases totals are also less than overall, are 'lost' cases here simply because a subtype isn't specified? Perhaps the authors can clarify this a bit in methods.

Indeed, histological subtype information was not available for all studies. We clarify this point in the revised manuscript (page 34, line 755).

6. Given the sample size differences, what effect sizes were the authors powered to observe in the Lung Squamous cell carcinoma and Lung Small Cell Carcinoma groups? Should this be interpreted as a false negative due to lack of power OR were the authors reasonably powered to observe (say) the estimated effect size overall in the univariate or epidemiological correlation data (but failed to see that effect), consistent with apparent heterogeneity in these strata?

Yes, there are differences in sample size and therefore the statistical power within lung cancer strata, nevertheless, the confidence intervals of squamous cell carcinoma and small cell carcinoma did not overlap, implying a distinct association rather than a lack of power.

7. Inclusion of an epidemiological correlational data baseline between LTL >> lung cancer to compare to the MR estimates would also add value.

Unfortunately, we do not have measured telomere length at baseline for most of the case series and we used the LTL MR instrument to infer it.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Reviewer #2 (Recommendations for the authors):

The authors have responded to most comments reasonably.

However, I still have some concerns about the given report. I know the collection of these points below are a little extra work -- I am honestly not trying to "add on" but do think these experiments are important to bring out the robustness of the author's claim, add value to the extant literature, and interpretation at these key points around the MR they describe.

1. First, focused around my original comment #2 (in the author's response, their point #6) on multivarible MR experiments. To my point that there are actually multiple potential confounders – including smoking, BMI, and waist-hip ratio that could potentially confound the interpretation that LTL explains the lung-cancer associations, the authors performed one experiment focused on BMI. They note that for each cancer trait there appears to be no impact on their primary association between LTL and Lung cancer data when considering BMI.

However, what they present is still a bit inadequate from my point of view, in the following ways.

1a. The authors perform a form of this MVMR which is a little 'less' good in my view for this -- they build a new instrument based on SNPs ascertained for association with BMI, and then include that in the MVMR regression analysis. This is not how I would do this -- this result describes a "joint" model for the effect of exposures (BMI, LTL) on outcome (Lung cancer). This version does not regress out any potential association of the SNPs *selected for association with LTL* has with the potential confounder. In this case, if the LTL variants had strong association with the confounder (e.g., BMI), I don't think that this would be fully accounted for.

Instead, the authors should take the n=144 SNPs selected for the LTL instrument, identify associations from extant data with the putative confounder of interest (e.g., BMI), create an instrument for the genetically mediated effects of that confounder, and include that as an instrument for covariate confounding. I.e., "regress out" any effect that the LTL instrument might have on the confounder, and report adjusted effects of the LTL instrument on cancer risk. Yes: this may NOT be a very strong instrument for the confounder -- but that's not the point here. The point is to determine the effect of genetically mediated LTL on cancer adjusting for any effect that instrument the authors built for it has on the confounder.

1b. Furthermore, considering a single confounder here is really suboptimal. As indicated in the previous comments and as the authors know, smoking behavior and other anthropomorphic measures (e.g., waist-hip ratio) are also correlated with everything here. So the authors should really consider an expanded list of confounders for this to be clearly convincing that LTL really does operate as a causal exposure independently of these other risk factors.

1c. In addition to experiments that are pairwise, they should also consider performing an experiment with all potential confounders included in the model Cancer ~ LTL + BMI + WHR + SMOKING + … i.e., can the authors convincingly demonstrate that the observed relationship between LTL and these lung cancer outcomes is not explained by a battery of obvious confounders individually or as a collective.

1d. As suggested earlier – I think value and novelty would be added if the authors really took this on and explored this space a bit more -- look hard and think carefully about what confounders really *COULD* explain this result and work hard to refute that those could explain these observations.

I do think a robust set of experiments should be included in the primary report beyond smoking, either way (in the supplement). I think the authors report the BMI result in the rebuttal response to me but don't add those back to the paper. I'd honestly like to see a healthy version of that added to the paper, but that's my taste on this point.

We thank the reviewer for their thorough and detailed review of our methods. We would like to take this opportunity to clarify that the standard MVMR analysis does consider multiple instrumental variables. The analyses that we have performed indeed account for the associations of LTL instruments with other variables. These established methods are reviewed in detail here (PMID: 30535378) and we have added this citation to the methods for clarity.

For the BMI example, the inputs for the MVMR analyses were as follows:

Therefore, the output of the MVMR analysis is a reasonable test of the effect of LTL on lung cancer that conditions on the genetic associations between LTL instruments and BMI.

We focused our initial response using BMI as an MVMR analysis for an adiposity-related trait was specifically requested by the reviewer and we felt BMI was reasonable in that context.

We have now expanded this to other traits. These include MVMR analyses for LTL on lung cancer adjusting for waist-to-hip-ratio (WHR), HDL, total triglycerides, blood pressure, smoking, and alcohol intake. These expanded analyses did not result in any material change to our conclusions. We similarly fitted a multivariate model including multiple traits (LC~ LTL+BMI+CPD+drinksperWeek), which similarly noted no material difference to our conclusion that LTL is associated with lung cancer risk. We acknowledge the reviewers point that these results should be included in the manuscript. They were included as supplementary file 1g in the revised manuscript.

We similarly acknowledge the reviewers point that we cannot exclude a potential effect of an unknown confounder. As such, we included the following sentence in the Discussion section of the revised manuscript: “We used MVMR analysis (citation pubmed PMID: 30535378) to assess the potential that factors, such as BMI, smoking, alcohol use and other obesity related factors, may account for the association between LTL and lung cancer. While the influence of alternative unknown potential confounding factors cannot be excluded, the association of LTL with lung cancer risk was materially unaltered following adjustment for a range of potential confounders considered in our MVMR analyses.”

2. On my previous point #4 (in the rebuttal, point #8), I asked the authors to consider a subset MR analysis which focused on those where the genes/signals map quite obviously on known telomere length biology and represent some of the strongest signals. The authors respond that they can't actually do this analysis, suggesting that using the data for co-localization somehow invalidates an MR sub-set analysis.

I thought about this a little bit and while I admit that I can be daft sometimes I honestly do not understand what basis – empirical or theoretical – this argument has to somehow drive positive or negative biases here, i.e., why they need an "out of sample" set to perform the MR after they used this data to perform colocalization. If there's specific literature or reasoning the authors would like to articulate this in more detail, that would certainly be appreciated!

My intuition is that this actually isn't much of an issue -- performing subset analyses for different reasons (biological or otherwise) is actually pretty reasonable and there are MR methods that actually try to cluster / subset instruments used.

This all goes to interpretation and as a sensitivity analysis -- if the result is driven entirely by variation at TERT, TERC, OBFC1 but not obviously by other variation that influences telomere length, then I personally think that is worth trying to understand and report. Analytically, this is a *trivial* experiment to perform in the UVMR space.

We apologise if our initial response was not sufficiently clear. While MR approaches that cluster variants within genetic instruments can be used to identify biological relevant subgroups of genetic predictors (DOI:10.1093/bioinformatics/btaa778), the results generated by this method can be challenging to interpret.

In our study, the genetic variants included in the MR analyses that was used to estimate the association of LTL with lung cancer risk have been associated with LTL (at least to genome wide significance) in an independent large GWAS (Codd et al. NG Nat Genet 2021). We subsequently used COLOC to highlight loci that have a particularly strong evidence of a shared causal association between LTL and lung cancer risk to enrich our understanding of the association of LTL with lung cancer. Our study has not sought to identify a subset of loci that predict LTL that are not associated with lung cancer risk.

We feel that MR analysis based on the stratified MR instrument (COLOC+ vs. COLOC-) would greatly complicates the interpretation of the results. Our concern is that to successfully estimate colocalisation, the variants of interest should be associated with both traits (i.e. LTL and lung cancer), particularly when conservative priors are used. By creating an MR instrument based on variants that are COLOC+, we are therefore selecting variants that are strongly associated with lung cancer in this dataset. As such, if that MR instrument was then re-applied into the same dataset, we expect that the instrument will return an estimate biased upwards due to how we selected the variants. We have created a MR instrument based on COLOC+ variants (n=12, variance in LTL explained of 1.05%) and applied that COLOC+ instrument in an MR analysis within our lung cancer dataset. This analysis is presented in the table below. The MR estimates using the COLOC+ instrument had a larger association with lung cancer risk (OR = 3.48 for lung adenocarcinoma (table below)) compared to MR instrument using all variants associated with LTL (OR = 2.43 for lung adenocarcinoma (figure 2)). However, in line with our concerns above, it is unclear to us if this increase in magnitude of association is driven by anything other than the expected greater strength of association for contributing variants and lung cancer risk provoked by the in-sample selection of these variants. We feel that an independent sample would be required to distinguish potential bias from a true biological difference from the effects of the COLOC+ MR instrument, which did not fall within the remit of our study.

Author response table 1
ExposureOutcomeMethodPORLCIUCI
Telemore Length (Colocalised SNPs)Lung CancerInverse variance weighted3.93E-062.211.583.10
Telemore Length (Colocalised SNPs)Lung AdenocarcinomaInverse variance weighted2.91E-073.482.165.60
Telemore Length (Colocalised SNPs)Lung Squamous Cell CarcinomaInverse variance weighted0.4271.250.722.19
Telemore Length (Colocalised SNPs)Lung Small Cell CarcinomaInverse variance weighted0.6910.8220.312.19
Telemore Length (Colocalised SNPs)Lung Ever SmokersInverse variance weighted0.0081.791.162.74
Telemore Length (Colocalised SNPs)Lung Never SmokersInverse variance weighted0.1223.010.7412.20
Telemore Length (Non-Colocalised SNPs)Lung CancerInverse variance weighted2.36E-081.511.311.74
Telemore Length (Non-Colocalised SNPs)Lung AdenocarcinomaInverse variance weighted5.85E-152.381.912.96
Telemore Length (Non-Colocalised SNPs)Lung Squamous Cell CarcinomaInverse variance weighted0.2930.890.721.10
Telemore Length (Non-Colocalised SNPs)Lung Small Cell CarcinomaInverse variance weighted0.8521.030.761.39
Telemore Length (Non-Colocalised SNPs)Lung Ever SmokersInverse variance weighted8.35E-061.441.231.69
Telemore Length (Non-Colocalised SNPs)Lung Never SmokersInverse variance weighted1.03E-031.851.282.66

As suggested by the reviewer, an alternate approach could be the stratification of variants based on biological rational, limiting to genes encoding proteins described in telomere maintenance, for example. However, in this instance, we have performed the COLOC analysis and based on those COLOC results, we have already noted that telomere maintenance genes are those that tend to colocalise (Pg 15 lines 328-330). Therefore, a MR instrument based on the variants located at or near genes involved in telomere maintenance may also be potentially biased because of the observations we have already made.

We also have concerns about variants in the COLOC(-) strata. Here, we might expect a result to be estimates biased downwards (i.e. we could be selecting within this data, a set of variants that are less strongly associated with lung cancer). However, this isn’t that straight forward as COLOC is a generally conservative method and, as stated previously, we employed relatively strict priors. Additionally, COLOC can be a less efficient estimator of shared causal loci where multiple causal variants exist in a region but where only a subset are shared between two or more traits. As such, many loci associated with LTL but without strong evidence for colocalisation in our study may be associated with lung cancer. Moreover, a large proportion of variation in LTL is explained by COLOC(-) strata variants (n=132, variance explained of 2,45%). MR analyses using the COLOC(-) strata variants described in table below do estimate a strong association between LTL ad lung cancer. For these reasons, we feel that interpreting results based on genetic instruments stratified by COLOC status is very challenging for a reader.

We appreciate this reviewer’s point of view and thank them for their input. However, for the complexities related to how we have employed COLOC subsequent to our MR analysis outlined above, we prefer to be conservative and limit our analysis to MR instruments selected from the large LTL from a large GWAS of telomere length.

We highlight the limitations of the COLOC approach (discussion lines 328-329) and have adapted this text further for clarity.

“Our colocalisation approach is generally more conservative and may fail to accurately determine the posterior probability for shared genetic signals in the presence of multiple independent associations in a given locus (45) and we stress that many of the variants that are COLOC negative are likely to associated with lung cancer. Indeed, which may be a reasonable explanation for the lack of colocalisation observed at RTEL1 locus”.

3. Regarding the rebuttal point #13, which involved my point about power for discovery in the subset given the sample size, the authors state that there are sample size difference but then write something about confidence intervals overlapping that I don't quite understand.

What I think the authors SHOULD do with respect to the tumor histology analysis is to report a credible set based on power calculation which determine what range of causal effect sizes the Lung Squamous CC and Lung Small CC were powered to discover (at an α type-1 error rate) of 5% and/or 1%, say, given the instruments used and sample sizes involved here.

This is actually a very easy thing to do analytically with some assumptions – back of envelope -- you could check out what formula I pulled together in PMID: 25165093 to get the distribution under the alternative; I believe there are also MR power calculators out there as web tools. You easily do this with some R calculations in a straight forward way.

I think the authors should be able to articulate is that effect sizes (say) of OR=1.6 or better would be discoverable in those lung cell subsets that basically catch the null hypothesis. Or, that those sets were well powered to discover ORs that were even smaller than that. This would then give you a quantitative assessment that this isn't simply a false-negative result, but that there is putative heterogeneity here. But if they are small enough that they AREN'T particularly well powered, then that's a pretty important interpretive point on this plot.

Thank the reviewer for this fair suggestion. We performed power calculations to determine the range of LTL causal effects on lung cancer that we are adequately powered to detect, and report these results in Supplementary Figure 2—figure supplement 1 (panel B) and on pages 11 and 18 of the revised manuscript. As seen in Figure 2—figure supplement 1b, our study has >80% power to detect rather modest effects (OR~1.20) for lung cancer overall, adenocarcinoma, and squamous cell carcinoma. For less common subtypes, such as small cell carcinoma and lung cancer in never smokers, we 80% power to detect OR~1.40 and OR~1.50, respectively.

https://doi.org/10.7554/eLife.83118.sa2

Article and author information

Author details

  1. Ricardo Cortez Cardoso Penha

    Genomic Epidemiology branch, International Agency for Research on Cancer/World Health Organization (IARC/WHO), Lyon, France
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing – original draft, Writing – review and editing
    For correspondence
    cortezr@iarc.who.int
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-2847-1993

  2. Karl Smith-Byrne

    Cancer Epidemiology Unit, University of Oxford, Oxford, United Kingdom
    Contribution
    Data curation, Formal analysis, Investigation, Methodology, Writing – original draft, Writing – review and editing
    Competing interests
    No competing interests declared

  3. Joshua R Atkins

    Genomic Epidemiology branch, International Agency for Research on Cancer/World Health Organization (IARC/WHO), Lyon, France
    Contribution
    Formal analysis, Methodology, Writing – original draft, Writing – review and editing
    Competing interests
    No competing interests declared

  4. Philip C Haycock

    MRC Integrative Epidemiology Unit, Bristol Population Health Science Institute, Bristol Medical School (PHS), Bristol, United Kingdom
    Contribution
    Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-5001-3350

  5. Siddhartha Kar

    MRC Integrative Epidemiology Unit, Bristol Population Health Science Institute, Bristol Medical School (PHS), Bristol, United Kingdom
    Contribution
    Writing – review and editing
    Competing interests
    No competing interests declared

  6. Veryan Codd

    1. Department of Cardiovascular Sciences, University of Leicester, Leicester, United Kingdom
    2. NIHR Leicester Biomedical Research Centre, Glenfield Hospital, Leicester, United Kingdom
    Contribution
    Data curation, Writing – review and editing
    Competing interests
    No competing interests declared

  7. Nilesh J Samani

    1. Department of Cardiovascular Sciences, University of Leicester, Leicester, United Kingdom
    2. NIHR Leicester Biomedical Research Centre, Glenfield Hospital, Leicester, United Kingdom
    Contribution
    Data curation, Writing – review and editing
    Competing interests
    No competing interests declared

  8. Christopher Nelson

    1. Department of Cardiovascular Sciences, University of Leicester, Leicester, United Kingdom
    2. NIHR Leicester Biomedical Research Centre, Glenfield Hospital, Leicester, United Kingdom
    Contribution
    Data curation, Writing – review and editing
    Competing interests
    No competing interests declared

  9. Maja Milojevic

    Genomic Epidemiology branch, International Agency for Research on Cancer/World Health Organization (IARC/WHO), Lyon, France
    Contribution
    Data curation, Writing – review and editing
    Competing interests
    No competing interests declared

  10. Aurélie AG Gabriel

    Ludwig Lausanne Branch, Faculty of Biology and Medicine, Lausanne, Switzerland
    Contribution
    Data curation, Writing – review and editing
    Competing interests
    No competing interests declared

  11. Christopher Amos

    Institute for Clinical and Translational Research, Baylor College of Medicine, Houston, United States
    Contribution
    Funding acquisition, Writing – original draft, Writing – review and editing
    Competing interests
    No competing interests declared

  12. Paul Brennan

    Genomic Epidemiology branch, International Agency for Research on Cancer/World Health Organization (IARC/WHO), Lyon, France
    Contribution
    Writing – review and editing
    Competing interests
    No competing interests declared

  13. Rayjean J Hung

    Lunenfeld-Tanenbaum Research Institute, Sinai Health, Toronto, Canada
    Contribution
    Writing – review and editing
    Competing interests
    No competing interests declared

  14. Linda Kachuri

    Departament of Epidemiology and Population Health, Stanford University, Stanford, United States
    Contribution
    Formal analysis, Investigation, Writing – original draft, Writing – review and editing
    Competing interests
    No competing interests declared

  15. James D Mckay

    Genomic Epidemiology branch, International Agency for Research on Cancer/World Health Organization (IARC/WHO), Lyon, France
    Contribution
    Conceptualization, Formal analysis, Supervision, Funding acquisition, Investigation, Methodology, Writing – original draft, Project administration, Writing – review and editing
    For correspondence
    mckayj@iarc.who.int
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-1787-3874

Funding

Terry Fox Foundation (IARC Postdoctoral Fellowship)

  • Ricardo Cortez Cardoso Penha

National Institutes of Health (K99CA246076)

  • Linda Kachuri

Cancer Prevention and Research Institute of Texas (RR170048)

  • Christopher Amos

Cancer Research UK (K99CA246076)

  • Philip C Haycock

Cancer Research UK (C18281/A29019)

  • Philip C Haycock

Institut National Du Cancer (GeniLuc 2017–1-TABAC-03-CIRC-1 - [TABAC 17‐022])

  • Philip C Haycock

National Cancer Institute (INTEGRAL NIH 5U19CA203654-03)

  • Philip C Haycock

Agence Nationale pour la Recherche (ANR-10-INBS-09)

  • Philip C Haycock

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We would like to acknowledge the TCGA Research Network (https://www.cancer.gov/tcga) and the contribution of specimen donors and research groups involved in this resource. We also would like to acknowledge the ILCCO consortium, the participants of the UK biobank and GTEx project and the supporting bodies (https://commonfund.nih.gov/GTEx), specimen donors, and research groups. This work was supported by the Institut National du Cancer (INCa) (GeniLuc 2017–1-TABAC-03-CIRC-1 - [TABAC 17‐022]), NIH/NCI, INTEGRAL NIH 5U19CA203654-03, Cancer Research UK (grant number C18281/A29019), the France Génomique National infrastructure, funded as part of the « Investissements d’Avenir » program managed by the Agence Nationale pour la Recherche (contract ANR-10-INBS-09). Christopher Amos is a Research Scholar of the Cancer Prevention Institute of Texas ( RR170048). The work of Ricardo Cortez Cardoso Penha reported in this paper was undertaken during the tenure of an IARC Postdoctoral Fellowship at the International Agency for Research on Cancer. Linda Kachuri is supported by funding from the National Institutes of Health (K99CA246076). Philip Haycock is supported by funding from Cancer Research UK (C18281/A29019).

Ethics

Human subjects: Mendelian Randomization analyses did not require ethical approval as it used secondary, genome-wide association data from studies that obtained informed consent from all participants and ethical approval from review boards and/or ethics committees. Individual level data, accessed under accession number phs000178.v11.p8 and project application #2731, was subject to the ethnical policies that govern the Cancer Genome Atlas. These policies can be found at https://www.cancer.gov/about-nci/organization/ccg/research/structural-genomics/tcga/history/policies.

Senior Editor

  1. Eduardo L Franco, McGill University, Canada

Reviewing Editor

  1. Nicholas E Banovich, Translational Genomics Research Institute, United States

Reviewer

  1. Ben Voight, University of Pennsylvania, United States

Version history

  1. Preprint posted: August 24, 2022 (view preprint)
  2. Received: August 31, 2022
  3. Accepted: April 3, 2023
  4. Version of Record published: April 20, 2023 (version 1)

Copyright

© 2023, Cortez Cardoso Penha et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Ricardo Cortez Cardoso Penha
  2. Karl Smith-Byrne
  3. Joshua R Atkins
  4. Philip C Haycock
  5. Siddhartha Kar
  6. Veryan Codd
  7. Nilesh J Samani
  8. Christopher Nelson
  9. Maja Milojevic
  10. Aurélie AG Gabriel
  11. Christopher Amos
  12. Paul Brennan
  13. Rayjean J Hung
  14. Linda Kachuri
  15. James D Mckay
(2023)
Common genetic variations in telomere length genes and lung cancer: a Mendelian randomisation study and its novel application in lung tumour transcriptome
eLife 12:e83118.
https://doi.org/10.7554/eLife.83118

Categories and tags

Research organism