Tracing Lung Cancer Risk Factors Through Mutational Signatures in Never-Smokers

Abstract

Epidemiologic studies often rely on questionnaire data, exposure measurement tools, and/or biomarkers to identify risk factors and the underlying carcinogenic processes. An emerging and promising complementary approach to investigate cancer etiology is the study of somatic “mutational signatures” that endogenous and exogenous processes imprint on the cellular genome. These signatures can be identified from a complex web of somatic mutations thanks to advances in DNA sequencing technology and analytical algorithms. This approach is at the core of the Sherlock-Lung study (2018–ongoing), a retrospective case-only study of over 2,000 lung cancers in never-smokers (LCINS), using different patterns of mutations observed within LCINS tumors to trace back possible exposures or endogenous processes. Whole genome and transcriptome sequencing, genome-wide methylation, microbiome, and other analyses are integrated with data from histological and radiological imaging, lifestyle, demographic characteristics, environmental and occupational exposures, and medical records to classify LCINS into subtypes that could reveal distinct risk factors. To date, we have received samples and data from 1,370 LCINS cases from 17 study sites worldwide and whole-genome sequencing has been completed on 1,257 samples. Here, we present the Sherlock-Lung study design and analytical strategy, also illustrating some empirical challenges and the potential for this approach in future epidemiologic studies.

Abbreviations

CT

computed tomography

FFPE

formalin-fixed paraffin-embedded

H&E

hematoxylin and eosin

LCINS

lung cancer in never-smokers

GWAS

genome-wide association studies

QC

quality control

SBS

single-base substitution

WGS

whole-genome sequencing

Lung cancer in never-smokers (LCINS) accounts for 10%–25% of lung cancer cases (1) and ranks among the most common causes of cancer mortality (2, 3). Compared with former and current smokers with lung cancer, the predominant histology observed in never-smokers is adenocarcinoma (4). Geographic variability in lung cancer risk among never-smokers is also observed (5, 6), likely due to regional differences in lifestyle and in environmental and occupational exposures. Some established environmental risk factors associated with LCINS include exposure to secondhand smoke (5, 7–10), radon (11–15), outdoor (16, 17) and indoor (18–21) air pollution, and asbestos (22, 23), which have been reviewed extensively (Table 1). Other risk factors include history of respiratory diseases, such as tuberculosis (24–26), pneumonia (24, 25, 27), and asthma (27–29). However, most LCINS cases have no known risk factors, highlighting the critical need for etiological studies.

Table 1.

A Summary Table of Established Risk Factors for Lung Cancer in Never-Smokers

Exposure	Study Types	Exposure Assessment	Summary of Findings/Public Health Implications	Selected References
Secondhand tobacco smoke	Meta-analysis, pooled analysis, review	Questionnaire data, in-person interviews	Increased risk of LCINS for spousal secondhand smoke exposure as well as workplace exposure has been consistently observed regardless of study design, geographic region, etc. Associations between childhood exposures to secondhand smoke and LCINS have generally been weaker and not as consistent. Secondhand smoke exposure is an important risk factor in all geographical regions.	(6–10)
Outdoor pollution	Cohort, meta-analysis	Quantitative estimates of residential exposure to outdoor pollutants (e.g., inverse distance-weighted interpolation methods geocoded to baseline residential address, fixed site monitors, land use regression)	Different types of outdoor pollutants have been studied, including ozone, PM, and nitrogen oxides. Most consistently reported association with LCINS has been with PM2.5.	(16, 17, 93–96)
Indoor pollution	Case-control, cohort, meta-analysis, pooled analysis	Questionnaire data, in-person interviews on household air pollution	Different sources of indoor pollution have been studied. These include indoor cooking oil, coal fuel, waste and dung combustion, and biomass fuel, among others. The association between indoor pollution and LCINS has consistently been stronger among women, which is likely due to women having greater exposure to fuel combustion products at home than men. Approximately half of the world’s population is still exposed to indoor air pollution from domestic cooking and/or heating with solid fuels.	(18–21, 97–100)
Asbestos	Meta-analysis, pooled analysis	Questionnaire data, in-person interview of occupational exposure history, environmental monitoring/quantitative measurements of fibers linked to job title	Association between asbestos exposure and LCINS observed, generally stronger among men than women likely due to differences in exposure levels. Although most dangerous asbestos types are no longer used, other siliceous fibers and chrysotile are still incorporated into building projects in developing nations.	(22, 23, 101)
Radon	Meta-analysis, pooled analysis, review	Occupational exposure, residential exposure	Increased risk of LCINS is consistently seen in miners; level of association less clear with residential exposure. Population attributable fraction for residential radon exposure is higher in Europe.	(6, 11–13, 15, 102–104)
Pneumonia	Meta-analysis, pooled analysis	Self-reported history of pneumonia	Increased risk of LCINS observed with pneumonia. Substantial public health interest due to the large population diagnosed with pneumonia.	(24, 25, 27)
Tuberculosis	Meta-analysis, pooled analysis, review	Self-reported history of tuberculosis	Previous diagnosis of tuberculosis has been found to have an independent effect on LCINS. Although the incidence of tuberculosis is low in North America, it is common in low- and middle-income countries and affects millions; therefore, the possible association with lung cancer risk is of public health importance.	(6, 24–27)
Asthma	Case-control, meta-analysis, pooled analysis	Self-reported history of asthma, physician-diagnosed asthma	Some studies have reported an association between asthma and LCINS, but association is unclear and published literature is mixed.	(27–29, 105, 106)

Abbreviations: LCINS, lung cancers in never-smokers; PM, particulate matter.

Throughout life, somatic cells acquire mutations, many of which occur well before the development of cancer (30). Advances in sequencing technologies combined with the development of novel computational methods can decipher the characteristic patterns of somatic mutations, termed “mutational signatures,” imprinted by the activities of endogenous and exogenous mutational processes (31). Distinct mutational signatures can now be identified from the thousands of somatic substitutions, insertions/deletions, copy number alterations, and structural rearrangements (32) observed in cancerous or normal somatic genomes (30–32), using the sequence context of each alteration (33, 34). Mutational signatures can reveal established exposures (e.g., tobacco smoking in lung cancer (35), ultraviolet light in skin cancer (36), or alcohol (37) or aflatoxin (38) exposure in liver cancer). Mutational signatures can also identify failure of known endogenous processes (e.g., defective DNA mismatch repair (39), errors in homologous recombination repair pathways (40), or loss of both polymerase proofreading and mismatch repair function (41)). Similar “marks” are likely imprinted on the genomes of LCINS.

The objective of the Sherlock-Lung study is to identify mutational signatures and relate them to past exogenous and endogenous processes by analyzing the cancer genome of 2,000 ethnically diverse LCINS cases identified through previous research efforts. The analytical approach used in this retrospective case-only study design is novel in that the different patterns of mutations observed within tumors can be used to infer prior probable exposures, some occurring years before diagnosis, even in the absence of exposure data. The starting point is the identification of mutational signatures in tumor samples and linking them to potential exogenous exposures and endogenous biological processes in external databases (42). When information on environmental and lifestyle exposures is available for the cases, we can also estimate the magnitude of etiological heterogeneity by relating exposure data to tumor subtypes in case-only analyses (43). In contrast, more traditional analytical approaches typically analyze mutational data and relate it to exposures reported by the cases. Moreover, the approach we use here allows for the identification of potential new or unexpected risk factors for LCINS, as it happened, for example, with the identification of the plant-derived aristolochic acid, through its specific mutational patterns, as a risk factor for a subset of hepatocellular carcinoma (44) and clear cell renal cell carcinoma (45). We acknowledge that the case-only design does not allow estimation of relative risks for the association between specific exposures and the risk of developing LCINS, but the identification of potential heterogeneity of exposure–tumor mutation associations will lay the foundation for future cohort or case-control studies to obtain such relative risks. Notably, the analysis of genomic data can also lead to a greater understanding of the endogenous mutational mechanisms (e.g., deficient DNA repair (40) or apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like (APOBEC)-related mutations (46)) triggering or facilitating clonal outgrowth in the presence or absence of exogenous exposures.

Briefly, the 2 primary aims of the Sherlock-Lung study are to: 1) identify exogenous exposures and endogenous processes involved in LCINS through the analysis of mutational signatures and other molecular characteristics, and 2) develop an integrated molecular, histological, and radiological classification of LCINS (Figure 1). We also discuss secondary aims with their attendant challenges and opportunities.

Figure 1.

Aims of the Sherlock-Lung study, worldwide, 2018–ongoing.

METHODS

Study design

Sherlock-Lung will include 2,000 LCINS patients with treatment-naive fresh frozen tumor specimens and a source of germline DNA, of any histological type, but primarily adenocarcinoma, of all stages, ages, and sexes. Surgical samples will mostly come from stages I–IIIA (resectable lesions), while biopsies will also include more advanced cases. LCINS whose diagnosis was based only on imaging evaluation are excluded. A subgroup of patients (n ~ 500) will be sought who have known history of high exposure to established lung cancer risk factors (“special exposures populations”), while the remaining cases will be patients with no known exposures to established risk factors (“general populations”) (47) (Figure 1).

Study population and sample/data collection.

Retrospective collection of data and biospecimens from 2,000 LCINS cases with fresh frozen tumor specimens requires contacting many potential sources. Accordingly, we have established contact with institutions identified through publications, conferences, extensive web searches, and personal relationships; recruitment has been pursued by follow-up e-mail and phone/video calls. Sample collection began in 2019 and to date, LCINS cases have been drawn from tissue biobanks, hospital case series, population-based or hospital-based case-control studies, and clinical trials with fresh frozen lung tumor and a source of germline DNA samples. To capture a variety of exposures and genetic background across geographical regions, the study plan is to recruit at least 100 cases each from Asia, Africa, Central and South America, and the Middle East, in addition to those of European descent. Of note, germline data from this multiethnic population will increase the diversity of existing large-scale genome-wide association studies (GWAS) (>78% of participants in published GWAS are of European ancestry (48), and approximately 71.8% of these samples are collected from the United States, the United Kingdom, and Iceland (49)). Based on the power calculations from the “Mutographs of Cancer” whole-genome sequencing (WGS) data, 100 samples should be sufficient to detect 20% enrichment of mutations associated with a given exposure, whereas 2,000 samples can detect a 5% enrichment (50).

Subject prioritization.

Prioritization of subject selection is based on tumor sample requirements, exposure to specific risk factors, and availability/quality of data on exposure assessment, pathology, and imaging. The optimal sample requirements are listed in Web Figure 1 (available at https://doi.org/10.1093/aje/kwaa234). Overall, samples from cases with documented high exposures to established lung carcinogens (special exposures populations) have the highest priority as they have the most potential to identify a mutational signature associated with a distinct exogenous risk factor. Given the rarity and importance of these samples, we are willing to extend inclusion criteria and collect formalin-fixed paraffin-embedded (FFPE) samples for validation and pertinent analyses if frozen biospecimens are not available (e.g., for cases with high exposure to wood and coal smoke from Colombia, Table 2).

Table 2.

Lung Cancer Samples From Never-Smoker Patients Identified During the Initial 20 Months of Sherlock-Lung, Worldwide, 2018–Ongoing

Region	Cases With Data and Samples Received		Pending Cases a		Potential Additional Cases b
	General	Special Exposures	General	Special Exposures	General	Special Exposures
North America	353	0	55	180	138	0
Europe	424	0	149	47	9	125
Asia	434	28	300	200	200	90
Central/South America	0	131c	108	19	477	213
Africa	0	0	80	70	100	30
Middle East	0	0	72	0	40	0
Australia	0	0	0	5	0	40
Caribbean	0	0	50	0	30	0
All regions	1,211	159	814	521	994	498

^a Written commitments to provide data/samples has been received from the institutions.

^b Institutions have expressed interest in participating in Sherlock-Lung and efforts are underway to establish collaborations.

^c All samples received to date are fresh frozen specimens with the exception of 106 formalin-fixed paraffin-embedded samples from South America.

For cases without clear exposures (general populations), the minimum requirement for enrollment includes availability of data on environment or residence, lifestyle, demographics, histology, and >1 fresh frozen tissue sample paired with a source of germline DNA (i.e., whole blood, buffy coat, normal lung tissue, saliva, or buccal cells) per subject. Additional criteria for prioritization, in order of importance, include:

1. Ethnic diversity and geographic region.

2. Availability of multiple normal tissue samples (n = 4) to study the lifetime accumulation of mutations and presence of mutations in cancer driver genes.

3. Multiple hematoxylin and eosin (H&E) slides for histological classification.

4. Lung computed tomography (CT)-scan imaging data for radiological classification.

5. Availability of FFPE tissue blocks from the tumor center and periphery for tumor microenvironment analysis.

6. Availability of multiple tumor tissue samples (n = 4) for clonal evolution analysis.

7. Availability of plasma samples for circulating tumor DNA analysis.

Collection of exposure data.

We rely on 3 data sources to collect data on exposures to the major risk factors, including secondhand tobacco smoke, asbestos, radon, indoor and outdoor air pollution, and previous lung/respiratory tract diseases: 1) self-reported household, occupational, and lifestyle exposures; 2) medical records; 3) residential data (location of longest residence), which we plan to link to geogenic mapping and satellite data. The documented exposures must have occurred at least a decade prior to cancer diagnosis and must have lasted, cumulatively, for at least 1 year. Some case series will have extensive data on these risk factors while other case series will have limited data, depending on the approach used to recruit cases.

Code of ethics.

Because the National Cancer Institute is only receiving deidentified samples and data from collaborating centers, has no direct contact or interaction with study subjects, and does not use or generate identifiable private information, Sherlock-Lung has been determined to constitute “Not Human Subject Research (NHSR)” based on the Federal Common Rule (45 CFR 46; https://www.ecfr.gov/). Contributing cases are required to confirm collection under a local IRB-approved study.

Laboratory and analytical methods

Aim 1: Characterize the genomic landscape of LCINS and relationship with exposures and endogenous processes.

LCINS will be characterized using WGS, whole transcriptome, and genome-wide methylation analyses to describe the mutational burden of single nucleotide variants (SNVs), mutational signatures, alterations in major cancer driver genes and in lineage-specific surfactant genes, somatic copy number alterations (SCNAs), structural variants (SVs), germline variation, methylation patterns, gene expression, telomere length, presence of new tumor epitopes, viral sequences, lung microbiota, and additional genomic changes in LCINS (Table 3). Moreover, we plan to analyze germline genetic variants from lung GWAS (51–54) or polygenic risk scores in relation to quantifiable tumor genomic alterations.

Table 3.

Summary of the Experimental Plans to Characterize Lung Cancers in Sherlock-Lung, Worldwide, 2018–Ongoing

Experimental Plan	No. of Samples/Imaging Per Subject	Subjects
Aim 1: Characterize the genomic landscape of LCINS and relationship with exposures and endogenous processes
Whole genome sequencing	T coverage 80× = 1, Bcoverage 40× = 1	~2000
RNA sequencing + methylation array	T = 1, N = 1	~1,000
Cancer-driver gene targeted sequencing or WES + SNP arraya	N coverage 1,000× = 4, Bcoverage 400× = 1 or Ncoverage 200× = 4, Bcoverage 100x = 1	~500
16S ribosomal RNA	T = 1, N = 1	~2000
Aim 2: Develop an integrated molecular, histological, and radiological classification of LCINS.
H&E slides	Up to 6 tumor blocks per tumor	~2000
Lung CT–scan imaging	Diagnostic imaging (and before or after diagnosis if available)	~2000

Abbreviations: B, number of blood samples; CT, computed tomography; LCINS, lung cancers in never-smokers; N, number of normal lung tissue samples; SNP, single-nucleotide polymorphism; T, number of lung tumor samples; WES, whole-exome sequencing.

^a Ongoing pilot study in normal lung tissue samples.

Sample processing and quality control.

Samples from collaborating sites are received at a central laboratory for staging and preparation. Our central laboratory requires a minimum of 2 tissue samples of 40 mg each to extract DNA and RNA for genomic analyses. If collaborating sites send lung tissue specimens, these are first sent to a facility for validation of pathology and nucleotide extraction (DNA and RNA) with standard quality control (QC) metrics (Web Appendix 1). When already-extracted tumor DNA samples are provided, confirmatory quantification and QC are performed at a central laboratory following similar procedures. QC-approved DNA samples are then sent to a genomic center for whole genome sequencing, while methylation, microbiome, and RNA sequencing analyses are conducted at the central laboratory.

Whole-genome sequencing.

To date, studies of the genomic landscape of lung cancer have included mostly smokers and relied largely on targeted sequencing or whole-exome sequencing (55–60). However, the larger number of somatic mutations in whole-genome sequences, of which approximately 98% is not covered by whole-exome sequencing, provides increased power for signature decomposition. Moreover, WGS can better reveal structural rearrangements, detailed copy number profile, noncoding mutations, and other genomic changes not captured by whole-exome sequencing. Sherlock-Lung tumor tissue samples will be analyzed by WGS with an average coverage of 80×, using blood samples or other germline DNA sources (coverage of approximately 40×) as reference. For high-priority cases with uncertain sample quality, we will conduct deeper tumor sequencing (coverage of approximately 120×–200×) to increase the probability of detecting low-allele-fraction or subclonal mutations.

Using current bioinformatic tools, separation of clonal from subclonal mutations within each tumor enables inference of evolutionary trajectories of the mutational processes in tumors. In previous analyses of lung adenocarcinomas in smokers, we found that mutations assigned to the tobacco smoking signature (single-base substitution (SBS) 4) were predominantly clonal and therefore involved in tumor initiation (57). We will extend this analysis to normal tissue adjacent to the tumor tissue to improve our knowledge of early clonal expansion of cancer-driver mutations that are widely present in clinically normal tissues albeit in lower cell fractions (61–63). A pilot study is underway in normal tissue to identify the optimal approach to detect genomic changes in a small fraction of cells (Table 3).

RNA sequencing.

RNA sequencing analysis of tumor/normal tissue pairs is being carried out in approximately 1,000 cases, using 2 × 150 base pairs paired-end sequencing with a target depth of 100 million reads, to identify transcribed alterations and assess gene expression, gene splicing mutations, fusions, and tumor immune microenvironment.

Whole-genome DNA methylation analysis.

In contrast to stable genetic events, epigenetic states are reversible and responsive to environmental stressors (64). Understanding the epigenetic landscape that is specific to LCINS will help dissect the nongenetic factors contributing to the tumorigenesis through interaction with gene expression regulation and mutagenesis. To address this question and describe global DNA methylation patterns, their impact on gene expression, the presence of 5′—C—phosphate—G—3′ (CpG) island methylator phenotype (65), and methylation features associated with mutational signatures, DNA methylation analysis in tumor/normal tissue pairs from approximately 1,000 cases is being analyzed using Infinium EPIC arrays (approximately 850,000 probes; Illumina, Inc., San Diego, California).

Gene expression and methylation data will also be used to deconvolute the tumor microenvironment and its constituents (e.g., infiltrating immune cell populations) (66–68), in conjunction with H&E and immunofluorescent analyses. These data will also enable assessment of sample purity based on copy number alteration and single-nucleotide variant allele frequency data (69).

Microbiome analysis.

Recent studies have characterized the microbiota in both normal and tumor lung tissue. We observed that the proportion of microbiota species in normal lung tissue can be distinct from other organs and associated with environmental exposures related to lung cancer risk, including air pollution (70). Another study found a distinct lung microbiome in patients with lung cancer and particular genomic changes (71). It was also shown that local microbiota promote inflammation associated with lung adenocarcinoma via interaction with immune cells (72). The lung cancer microbiome was also characterized in The Cancer Genome Atlas (TCGA) data using a custom data analysis pipeline (73). In Sherlock-Lung, we will characterize the taxonomic and functional profiles of lung microbiota using 16S rRNA data to investigate the contribution to risk and progression of LCINS.

Aim 2: Develop an integrated molecular, histological, and radiological classification of LCINS.

Histological classification.

Integrated histological and molecular studies, largely based on smokers, have suggested that lung adenocarcinomas could be further classified into subtypes by combining histological features with genomics, transcriptomic, and epigenomic changes (74). Different histological subtypes (e.g., lepidic, acinar, papillary) suggest distinct biological pathways with implications for distinct etiological risk factors and clinical outcomes. In Sherlock-Lung, we are collecting, digitally scanning, annotating, and examining H&E slides from all available tissue blocks (up to 6 per tumor) to evaluate the tumor histological landscape. The original pathology report is collected for diagnostic review.

Radiological classification.

The widespread use of CT scans for lung cancer screening has resulted in a dramatic increase in the number of indeterminate ground-glass opacities (75), many with multifocal components, which makes judgment on the extent and benefit of surgical resection challenging. Collection of CT-scan images with reports enables integration of radiological imaging, including ground-glass opacities, with histological and molecular features. Histological and CT images are being archived for future studies, including application of deep learning/artificial intelligence–based algorithms to evaluate diagnostic and prognostic features.

Secondary aims

We plan further data collection and analyses (Table 4), including mining electronic medical records (e.g., the Clinical Practice Research Datalink, https://www.cprd.com/), to explore novel associations between LCINS and medical conditions or chronic medication use that can help inform genomic analyses and interpretation; lineage phylogenetic analysis (76–79) to reconstruct the tumor evolution using multiple tumor and normal tissue samples, possibly including single cell approaches; the analysis of density, colocalization, and spatial architectures of cells in the tumor microenvironment, to investigate cancer immunoediting (80, 81) using H&E-stained slides in conjunction with multiplex immunofluorescence staining of specific markers; the analysis of circulating tumor DNA (82) using low pass WGS; in-silico (46, 83, 84) and experimental (85, 86) validation of mutational signatures; and development of novel analytical approaches for the integrative analyses.

Table 4.

Secondary Aims, Sherlock-Lung, Worldwide, 2018–Ongoing

Objective	Experimental Plans
Electronic medical records	Analysis of medical conditions and long-term medication use in cases of LCINS
Tumor microenvironment	Quantification and spatial analysis of immune, endothelial, stromal cells
	H&E
	Multiplex immunofluorescent markers
	Digital imaging and spatial analysis using HALO imaging platforma
Liquid biopsy	Circulating tumor DNA (ctDNA)
	Deep target sequencing of driver genes + low-pass WGS
	Comparative analysis of WGS in T/ctDNA
Clonal evolution	Multiregion tumor and normal tissue samples
	Phylogenetic analysis
	Single cell analysis
Laboratory validation	Organoid/CRISPR/engineered cell lines
	Mutational signatures and other genomic changes
Development of algorithms	Novel approaches for mutational signature analyses
	Integrated analysis of -omics data and radiological and pathological imaging

Abbreviations: CRISPR, clustered regularly interspaced short palindromic repeats; H&E, hematoxylin and eosin; LCINS, lung cancers in never-smokers; WGS, whole-genome sequencing.

^a HALO image analysis platform (Indica labs, Albuquerque, New Mexico, USA).

Data sharing

The genomic data and digital imaging database from this study will be made available in accordance with National Institutes of Health policy through the National Cancer Institute’s Genomic Data Commons.

RESULTS

As of June 2, 2020, we have received samples and data from 1,370 LCINS cases from 17 collaborating institutions in North America, Europe, Asia, and Central and South America (Table 2). Of these, 159 cases have known high levels of special exposures (28 from Asia and 131 from Central and South America). The disproportionately low number of cases with high exposures to known risk factors underscores the challenge of collecting high-quality frozen specimens from cases with documented exposures at least a decade prior to cancer diagnosis, especially given that these high-level of exposures are primarily seen in cases from low-and middle-income countries.

We have received written commitment from institutions to provide additional fresh frozen and/or FFPE samples from 1,560 cases. In our experience, selected centers provided less than 50% of the promised frozen samples, and QC-related exclusions further decreased the number of samples available for WGS. Thus, if needed, we could collect more samples from 1,342 cases from other potential collaborators (Table 4). In addition to the geographical regions already represented, we anticipate samples from other regions, including Africa and the Middle East. We are prioritizing collection of samples from the regions outside Europe and the United States, especially those with documented exposures to known LCINS risk factors.

Frozen specimens collected from the 1,370 LCINS cases identified to date include 1,017 tumor and 820 nontumor lung-tissue specimens and 644 blood DNA samples. Out of a total of 1,837 lung tissue specimens received, 1,798 were shipped to the laboratory for DNA extraction. Of these, 210 failed tissue QC or pathology review (11.7%). The quality varied by region; Asia had the lowest proportion of tissues to fail QC (3.8%), followed by North America (11.6%) and Europe (14.1%). Samples from Central and Southern America are still under review. We found that if the lung tissue specimens passed QC or pathology review, extracted DNA was of high quality (94.9% of the extracted DNA passed QC for sequencing). Similarly, when collaborating sites sent existing extracted DNA, the quality was excellent, with 94.4% passing QC and only 8 samples out of 1,257 failing sequencing (Table 5).

Table 5.

Sample Collection, Processing, and Sequencing Attempts According to Geographical Region for Cases of Lung Cancer in Never-Smokers With Data/Samples Received as of June 2, 2020, Sherlock-Lung, Worldwide, 2018–Ongoing

		Frozen Lung Tissue				DNA Samples Received						WGS Attempted and Completed a
		Tumor (n = 1,017)		Nontumor Lung (n = 820)		Tumor (n = 385)		Nontumor Lung (n = 153)		Germline DNA (n = 644)		Tumor (n = 447)		Normal Lung (n = 196)		Germline DNA (n = 614)
Region (No. of Sites)	No of LCINS Subject (n = 1,370)	No.	% Extracted DNA Passed QC	No.	% Extracted DNA Passed QC	No.	% Passed DNA QC	No.	% Passed DNA QC	No.	% Passed DNA QC	No.	% Passed Sequencing	No.	% Passed Sequencing	No.	% Passed Sequencing
North America (8)	353	139	68.9	41	70.7	262	83.6	90	–b	336	95.8	207	98.1	12	100	312	98.7
Europe (4)	424	387	68.9	340	66.2	123	98.4	63	100	303	100	126	100	70	100	302	100
Asia (2)	462c	434	89.2	434	88.0							114	100	114	100
Central/South America (3)	131d	57	–b	5	–b					5	–b

Abbreviations: LCINS, lung cancers in never-smokers; QC, quality control; WGS, whole-genome sequencing.

^a Of total sequencing attempted and completed across all sample types (n = 1,257), 8 samples failed.

^b Data under review.

^c Twenty-eight out of 462 cases have known special exposures.

^d All 131 cases have known special exposures.

Notably, analysis of mutational signatures was important for excluding samples that were erroneously included in the study. For example, 2 samples were dominated by signature SBS7, attributed to ultraviolet exposure. Upon repeat review by 3 different pathologists, it was determined that the samples were not from primary lung tumors but from metastases of skin squamous cell carcinomas. Moreover, we identified 1 sample dominated by SBS4, attributed to tobacco smoking, and verified that the case was a current smoker, erroneously reported as a never-smoker. We also found 1 sample with a high level of signature SBS31 attributed to platinum-based chemotherapy. Upon retrieval of clinical data from previous hospitalizations, we found that this subject had a prior tumor treated with platinum and bevacizumab.

DISCUSSION

Here we have presented the framework of a large integrative study that uses mutational signatures and other genomic features to complement questionnaire and exposure assessment approaches to identify potential factors contributing to lung tumorigenesis in never-smokers. In tumorigenesis, endogenous factors, such as developmental and differentiation programs, or exogenous factors, such as mutagenic exposures, pathogens, and inflammation (32), can leave a “signature” on both tumor and adjacent nontumor tissue, which can be captured by the analysis of the mutations within trinucleotide or pentanucleotide context. To date, this approach has identified over 50 different mutational signatures across cancer types, many of which have mapped to one or more environmental or endogenous events (31, 32, 35). Moreover, mutational signatures can have potential clinical value as predictors of therapeutic response in cancer (87).

The second primary objective is to categorize LCINS based on molecular and clinical characteristics. According to The Cancer Genome Atlas analyses, lung adenocarcinomas (mostly from smokers) can be further classified into subtypes using a combination of histological features and transcriptomic, epigenomic, and genomic changes (74). Our study of nonsmoking cases might reveal pathways in adenocarcinomas (or other histological types) previously hidden by the strong effect of tobacco smoking, possibly allowing further tumor classification, which in turn might have treatment implications. The Cancer Genome Atlas has estimated that at least 1 in 10 cancer cases across cancer types might be classified and treated differently using a molecular taxonomy instead of current histopathology-based classification (88).

The use of WGS for the analysis of mutational signatures and tumor subclassification has several challenges and limitations, which extend beyond LCINS. Based on initial sample collection (Tables 2 and 5) and the analysis of mutational signatures in these samples, we have learned a series of lessons (Figure 2).

Figure 2.

Lessons learned from the Sherlock-Lung study, worldwide, 2018–ongoing.

Collecting frozen specimens can be challenging. Despite technological advancements, WGS analysis currently requires unfragmented DNA from fresh frozen tissue specimens. As noted above, we aim to gather samples from diverse racial/ethnic groups and geographical locations across 6 continents to ensure diversity of exposures and ancestry. However, in regions that lack resources and proximity to hospitals, have limited cancer screening programs, or often incur misdiagnoses (e.g., lung cancer can be initially misdiagnosed as tuberculosis), cancer diagnoses are often delayed, when surgery is a treatment option for a very small percentage of such cases. In the absence of surgical specimens, tumor biopsies have been collected. Although biopsies can be obtained from advanced tumors, balancing the stage distribution, which is skewed towards early stage in surgical cases, the materials have been either too small or necrotic. Additionally, only a subset of hospitals had the infrastructure to rapidly freeze, maintain, and ship adequate samples.

Collection of samples from high-income countries, although expected to be more feasible, is daunting. For example, 3 well-established institutions in North America initially identified 494 cases, but only 215 samples could be retrieved. Moreover, there has been variability in the provision of adequate samples—some centers effectively provided frozen tissue, others had extracted DNA, while others could provide FFPE blocks, tissue sections (with often different section numbers and thickness), H&E slides, or digitally scanned images, necessitating great flexibility from the central laboratory. Also, we encountered delays due to country-specific restrictions/policies on data and biospecimens sharing; accordingly, we developed a solution to conduct tumor profiling analyses locally with careful validation of comparable platforms, procedures, and QC, as in our central laboratory.

Collection of samples from individuals with high exposure to known risk factors for lung cancer remains a major challenge. Exposures to these risk factors is most relevant decades before the cancer diagnosis, when they potentially had an impact on tumor initiation. These exposures have been high in the past for many countries, including those where current exposure levels are low because of stringent occupational exposure limits and public health interventions (e.g., asbestos exposure). However, tumor samples from lung cancer cases with previous high levels of exposures, even drawn from occupational cohorts, exist only as FFPE archived tissue blocks, which currently limit their utility for genomic analyses. Sample collection in low- and middle-income countries, where exposures are still high, might not have had exposures in the most important etiological time window for lung cancer and might experience challenges described above. As an alternative approach to identify exposure-specific mutational signatures, we will utilize known experimental mutational signatures of environmental or microbial mutagens generated by exposing pluripotent stem cells, cell lines, or organoids to different dosages of mutagens (38, 85, 89, 90). We will test these mutagen signatures in our samples and estimate the proportion of mutations that can be explained by them. These analyses could shed light on potential exposures associated with the mutations, which will require further epidemiologic and experimental validation.

Many algorithms have been proposed to identify mutational signatures from a composite of genomic changes. However, they often lack consensus on analysis and result interpretation, their parameters can vary across tissue or cancer types (91), supervised fitting of signatures can lead to false results, and no recognized gold standard exists (92). Although these issues are less likely to be important for signatures with distinct patterns (e.g., APOBEC-related signatures SBS2 and SBS13), others, particularly the so called “flat” signatures (e.g., SBS3, SBS5, or SBS40) characterized by similar mutation distributions across trinucleotide contexts, cannot be always robustly separated. Moreover, there are several mutational signatures identified across cancer types whose origin is still unknown (32). The analytical plan envisions the use of multiple algorithms, the pentanucleotide context, and the verification of mutation enrichment and distributions to confirm signatures.

In conclusion, Sherlock-Lung is predicated on the examination of the genomic tumor landscape with careful clinical and exposure assessment to enable investigation of LCINS etiology. We have described the study design and objectives as well as opportunities and challenges of this integrative approach. With the reduction in cost for sequencing technologies and the progress of analytical tools, similar studies across different cancer types could become a viable approach for future epidemiologic investigation of risk factors for distinct cancers.