Circulating Tumor DNA Profiling in Small-Cell Lung Cancer Identifies Potentially Targetable Alterations

This article is featured in Highlights of This Issue, p. 6013

Small-cell lung cancer (SCLC) accounts for approximately 13% of all lung cancer diagnoses in the United States and is characterized by a tendency for early metastasis and an aggressive clinical course (1). Despite initial response to platinum-based chemotherapy, virtually all patients with extensive disease and the majority of those with limited stage disease develop tumor relapse, which is associated with treatment resistance and short survival. Although recent studies have identified potentially targetable alterations in SCLC, patients with relapsed disease generally deteriorate quickly and rarely undergo tissue biopsies at progression to facilitate molecular testing for biomarker-directed therapy (2–6). Challenges in obtaining tissue biopsies has also limited the ability to comprehensively characterize mechanisms of treatment resistance, which are of critical importance in aiding the discovery of novel therapies.

Circulating tumor DNA (ctDNA) may detect DNA from cancer cells shed into the bloodstream. Several studies have demonstrated the utility of ctDNA in directing therapy, examining mechanisms of treatment resistance, and prognosis in patients with non–small cell lung cancer (NSCLC; refs. 7–9). As ctDNA analysis can potentially capture spatial and temporal clonal heterogeneity of cancer without necessitating multiple invasive biopsies, it is an attractive modality for studying the genomic landscape of SCLC, especially at relapse (10–13). Here, we report the ctDNA landscape of SCLC using Guardant360, a targeted hybrid capture–based next-generation sequencing (NGS) platform focused on querying mutations in a set of cancer-related genes, and explore its utility and limitations in identifying alterations that potentially inform therapy.

We employed deidentified ctDNA test data from patients with SCLC who underwent ctDNA testing at a clinical laboratory (Guardant Health, Redwood City, CA) between September 2014 and 2017. The diagnosis of SCLC was confirmed by clinical information provided by physicians ordering the ctDNA test. Treatment information, date of diagnosis, age, gender, and sample collection details were collected from test requisition forms when available. The study was conducted in accordance with recognized ethical guidelines (e.g., CIOMS, Belmont report, and declaration of Helsinki). Clinical information for patients diagnosed at the Washington University in St. Louis (St. Louis, MO) and University of California in San Diego (La Jolla, CA) was verified by coinvestigators from those institutions after approval by the Institutional Review Board, and the requirement for consent was waived. Samples were classified as having been collected at “diagnosis” or “relapse” based on available clinical information. For samples lacking specific clinical information, treatment status was inferred on the basis of the dates of diagnosis and sample collection. Because ctDNA analysis is typically ordered by clinicians at the time of diagnosis or progression for clinical decision making, samples that were obtained within 3 weeks (duration corresponding to first chemotherapy cycle) of diagnosis were considered to have been collected “at diagnosis” and samples that were collected at least 84 days past date of diagnosis (corresponding to the approximate duration of first-line therapy) were considered to be collected “at relapse.”

Targeted NGS–based ctDNA detection was performed by Guardant Health, Inc. (Guardant360, www.guardanthealth.com/guardant360/), a Clinical Laboratory Improvement Amendment (CLIA)- certified and College of American Pathologists (CAP)-accredited, New York State Department of Health-approved clinical laboratory. The assay's gene panel was expanded from 54 to 73 genes over the study period. Overall 10, 86, 260, and 253 samples were sequenced using the 54, 68, 70, and 73 gene panels, respectively. Test characteristics differed between gene panels (Supplementary Table S1). The proportion of samples collected at diagnosis and relapse, which were sequenced on each panel was comparable (Supplementary Table S2). Single-nucleotide variants, insertion–deletion alterations (indels), fusions, and amplifications within various cancer-related genes were identified through analysis of cell-free (cf) DNA extracted from plasma collected from two 10 mL Streck blood collection tubes. Synonymous alterations were excluded from analysis. The ctDNA assay used in this study had high clinical sensitivity, detecting 85% or more of the variants present in tissue in patients with advanced cancer, and analytic specificity greater than 99.9999% (7, 14). Only somatic alterations were reported and analyzed in this study. SNVs and indels shed into circulation were quantitated using mutant allelic frequency (AF) by taking the number of mutated variants at that position, and dividing by the total number of reads at that position (including wild-type). Gene amplifications were reported as absolute gene copy number in plasma (14).

For the purpose of data analysis, genomic coordinates for each variant listed in the clinical test report were obtained using Transvar (15). This information was used to analyze data using the suite of functions, maftools (16). Amplification information was inputted as custom tables and overlaid onto co-mutation plots using the waterfall plot function where appropriate. Custom code was written in Python utilizing the PANDAS module to calculate alteration frequencies of tested genes (number of samples demonstrating mutation in a gene divided by the total number of samples). Alteration frequencies of tested genes between samples collected at diagnosis and relapse were compared using the mafcompare feature of maftools. The percentage at which various gene alterations were identified in individual samples or AF, was averaged to determine the mean allelic frequency (MAF). Lollipop and clonality plots were also generated using maftools. Clonality of an alteration in a sample was inferred on the basis of its AF ratio to the maximum AF in that sample (hereby referred to as “ctDNA clonality”), as described previously (7). A list of mutations known to be activating in RTK/RAS pathway genes such as KRAS, HRAS, NRAS, BRAF, EGFR, ERBB2 (HER2), MAP2K1, and MET, was curated on the basis of mutation hotspot information available for these genes from NSCLC in cBioportal (17, 18). Data from previously reported tissue sequencing studies in SCLC were obtained from publications or queried using the cBioportal interface (3, 4). GraphPad Prism was used for performing Student t tests and Fisher exact testing.

A total of 609 samples collected from 564 patients were analyzed (Table 1). The median turnaround time for test results was 14 days (range 6–32). Overall, 61 samples from 59 patients and 219 samples from 206 patients were classified as being collected at diagnosis or at relapse, respectively (Supplementary Table S2). Thirty patients underwent serial testing. At least one nonsynonymous mutation or amplification was detected in 552 of 609 samples (90.6%). One or more nonsynonymous mutations were detected in 531 samples [average of 3.54, 95% confidence interval (CI), 3.03–4.05]. Nonsynonymous mutations were identifiable in 87% of diagnosis (53/61) and relapse (191/219) samples, and amplifications in 43% (26) and 39% (85) of samples, respectively. While the maximum AFs (mean of 25.8% vs. 16.5%, P = 0.01) and MAFs (mean of 15.8% vs. 7.0%, P = 0.002) were significantly higher for samples collected at diagnosis compared with relapse, the number of nonsynonymous mutations or amplifications detected per sample did not differ significantly by treatment status (P = 0.2 and 0.1, respectively; Fig. 1). Among the 30 patients who underwent serial testing, nonsynonymous mutations and amplifications were detectable in more than one sample for 24 and 8 patients, respectively, without significant differences in the number of alterations between first and subsequent collections.

Nonsynonymous mutations in TP53 were the most frequent (72%), followed by alterations in RB1 (18%; Supplementary Figs. S1 and S2) across all samples, and this did not considerably differ by treatment status (Supplementary Tables S3 and S4). Following mutations in TP53 and RB1, mutations in ARID1A were the most frequent (14.5%) in samples collected at relapse, with approximately 33% of alterations consisting of nonsense mutations and indels (Fig. 2).

An analysis of differentially altered genes between diagnosis and relapse samples, demonstrated a higher frequency of alterations in the androgen receptor gene, AR, (14% vs. 2%, P < 0.05, Fisher's exact test) among relapsed samples (Fig. 2; Supplementary Table S4). AR mutations were observed in 26 samples from 25 patients at relapse, of which 21 were collected from females and 5 from males and all AR amplifications were only seen in samples collected from females. Of the nonsynonymous mutations identified in AR across all samples, 44% were in the N-terminal domain Activation Function 1 (NTD AF1) region, 15% in the DNA-Binding (DBD) and hinge Domain, and 21% in the Ligand-Binding Domain (LBD). While none of the identified mutations have been previously reported as activating, most activating AR mutations that have been reported, are typically found in these domains (Fig. 3; ref. 19). Mutations in APC, a negative regulator of β-catenin (CTNNB1) and WNT signaling, were also more prevalent among relapse samples (18 patients) compared with those obtained at diagnosis (1 patient; 11% vs. 2% of samples, P < 0.05, Fisher exact test; Figs. 2 and 3; Supplementary Fig. S3). In addition to alterations in APC, we also detected known hotspot mutations in CTNNB1 in 5 samples. While frequencies of alterations in multiple genes appeared to differ by treatment status, none reached statistical significance after correcting for multiple comparisons, possibly owing to inadequate number of samples in each treatment category.

SCLC is characterized by amplifications in SOX2 and MYC family genes including MYC, MYCL, and MYCN (3, 4). SOX2, which is not included in the Guardant360 panel, is located on the long arm of chromosome 3 (3q26) adjacent to PIK3CA. Apart from PIK3CA and MYC, amplifications in several other genes were observed (Supplementary Table S3). MYC amplifications, which may be potentially targetable by aurora kinase inhibitors, were observed in approximately 3.5% of the relapse samples (20). Despite data supporting a role for MYC in treatment-resistant SCLC, the frequency of MYC amplifications did not differ according to treatment status (5% at diagnosis vs. 3.5% at relapse; Supplementary Fig. S4; ref. 21). Notably, copy number assessment in ctDNA may have lower sensitivity than direct tumor analysis, which could explain the differences in frequency of amplification in genes like MYC between this study and other tissue sequencing studies.

Considering the emerging role of investigational agents targeting DNA repair pathways, genomic instability, and immunotherapies in relapsed SCLC, we examined alterations in DNA repair and damage response pathway genes (22, 23). Alterations in BRCA1 (9%), BRCA2 (5.5%), ATM (3%), and MLH1 (1%) were detectable in a subset of relapse samples, despite limited coverage of ATM and MLH1 exons. BRCA1, CCNE1, and FBXW7 alterations, which drive genome-wide instability (tandem-duplicator phenotype), were observed in 9%, 8%, and 5% of relapse samples, respectively (Fig. 4), and mutations in ARID1A, associated with responsiveness to PARP and EZH2 inhibitors, were observed in nearly 15% of patients at relapse (24).

Alterations in PI3K/mTOR signaling genes were observed in approximately a third of all SCLC samples (4). Although SCLC is not characterized by RAS/RTK pathway alterations, ctDNA profiling showed an unexpectedly high frequency of these alterations. Activating mutations in RTK/RAS/RAF pathway genes EGFR, KRAS, NRAS, HRAS, BRAF, MAP2K1, and ERBB2 were observed in 96 samples (18%) collected from 77 patients. Mutations in KRAS were the most frequent (n = 36), followed by EGFR (n = 34) and BRAF (n = 9). Amplifications in ERBB2 and MET were detectable in nearly 30% of patients at relapse. All samples demonstrating known activating EGFR mutations were collected at relapse, suggesting that these samples were possibly collected from patients with transformed SCLC following EGFR-directed therapy (25). We also observed oncogenic ALK and RET rearrangements in one sample each, but the treatment status of these samples was unknown.

TP53 alterations are essentially universal and typically clonal in SCLC (4). We analyzed the characteristics of TP53 mutations to obtain a better insight into the ability of ctDNA testing to facilitate disease monitoring. TP53 mutations were detectable in 72.5% of samples with detectable alterations, of which 31% had multiple TP53 mutations. Of the samples collected at relapse, 40% showed multiple TP53 mutations (vs. 31% at diagnosis, P = 0.35, Fisher exact test). TP53 mutations predominantly clustered in the DNA binding domain irrespective of their AF (Fig. 5; Supplementary Fig. S5). When only variants with the highest AFs in a sample were considered for each gene, mutations in TP53 and RB1 appeared to be clonal, as expected for SCLC (Fig. 5). Among samples with multiple TP53 mutations, the average ctDNA clonality value of TP53 mutations was approximately 4.5-fold higher for mutations with the highest AFs compared with mutations with lower AFs (0.89 vs. 0.22), suggesting a subclonal or hematopoetic source for low AF variants.

We examined the ctDNA profiles of nearly 600 patients with SCLC, making this the largest study to examine the genomic features of SCLC to date, whether in tissue or in plasma. This large sample size enabled us to make several important observations and investigate the utility of ctDNA in identifying potentially targetable alterations in SCLC. TP53 and RB1 alterations play an essential role in SCLC tumorigenesis, and are also retained at relapse, a feature that is consistently reproducible through ctDNA analysis (4, 26). The frequency of RB1 mutations was notably lower compared with their nearly universal presence in tumor biopsies. Of note, the test used herein cannot easily identify biallelic loss or loss of heterzygosity. Notably, the frequency of RB1 mutations in our data (12%) was comparable with previously published studies when only missense and nonsense variants were considered (18% to 35% in tissue sequencing studies), suggesting that these differences could be explained by differences in RB1 coverage (for type of mutation) between tissue sequencing in previous studies and earlier versions of the ctDNA test (3, 4, 17, 18).

The number of mutations detectable per sample, a surrogate for tumor mutation burden, did not differ by treatment status. The MAF of detected variants was significantly higher across samples collected at diagnosis compared with those collected at relapse. Previous studies suggest a correlation between MAF of variants detected on ctDNA and radiographic disease burden (27). Therefore, our observation may be explained by the fact that patients with SCLC often present with a higher burden of disease at diagnosis.

We also observed a higher percentage of alterations in APC and AR in relapse samples. Although identification of APC alterations in our study is consistent with a recent study highlighting the role of WNT activation in chemoresistant SCLC, a more systematic evaluation of the impact of these alterations in SCLC is needed (5). The identification of AR alterations is provocative because it may be potentially targetable. In SCLC, both testosterone and dihydrotestosterone stimulated growth of AR-positive SCLC cell lines, which was reversed by the addition of antiandrogens, including fludamide (28). In addition, AR manipulation has an established role in patients with AR-positive tumors, and there are anecdotal cases describing responses with this approach in other solid tumors (29, 30). Response to anti-AR therapy may also depend on whether AR copy number gain is focal or a result of aneuploidy. Although not addressable in this study, it may be possible to adopt approaches such as comparing the number of copies of ARAF and AR genes, as both reside on the X chromosome, to filter out aneuploidy. Further studies are warranted to explore the role of AR signaling in SCLC, especially considering the role this pathway plays in driving chemoresistance in prostate cancer (31).

The role of RTK/RAS signaling in SCLC is controversial (32). We observed known RTK/RAS pathway activating mutations in nearly 18% of our patients, including therapeutically targetable alterations in EGFR, BRAF, ERBB2, and MET. While it is possible that these findings are a consequence of misclassification of NSCLC or large-cell neuroendocrine cancers as SCLC, presence of mixed histology tumors, or identification of hematopoietic variants by ctDNA testing, alterations in RAS pathway genes, and/or pathway activation in SCLC have been previously reported (33, 34). Cancer initiation and progression are complex processes, and preclinical studies in murine models suggest that the final histology of lung cancer is likely to be influenced by the cell of origin and the combination and sequence in which these cells acquire specific mutations (35, 36). It is therefore possible that these findings reflect heterogeneity in the cell of origin, tumorigenesis, and inter- or intratumoral heterogeneity (37, 38). The higher ctDNA clonality values of some of the KRAS mutations support this hypothesis. The ability to target RAS signaling has been reported in SCLC with combination therapies (39). We also observed amplifications in ERBB2 and MET in 12% of the samples. Although there is no data in SCLC, studies have reported benefit from targeting these alterations in other solid tumors (40–45).

Mutations in genes involved in DNA repair and damage signaling were observed in a subset of relapse samples, and potentially predict for responsiveness to therapeutic strategies targeting genomic instability and immunotherapies (22, 24). In particular, we observed alterations in BRCA1, FBXW7, CCNE1, and ARID1A in relapse samples. Alterations in these genes may predict response to DNA-damaging therapies, PARP inhibitors, and/or immunotherapy raising the possibility for more targeted therapies for a subset of patients with SCLC (46–48).

Finally, nearly a third of samples in this analysis demonstrated multiple TP53 mutations. In many instances, these mutations were detectable at very low percentages in plasma, raising the possibility of them being subclonal or hematopoietic in origin. Recent data have shown the ability of ctDNA and tissue molecular testing to pick up variants from clonal hematopoiesis (CH; refs. 49, 50). Incorporation of strict quality control measures, a high level of clinician suspicion, and/or bioinformatic filtering is critical when calling mutations to ensure that they are related to the tumor of interest and not CH. Our results suggest that CH identified on ctDNA testing could potentially lead to the identification of spurious variants, incorrect estimation of tumor mutation burdens, or pose a challenge in interpreting presence of minimal residual disease. Delineating potential hematopoietic variants from ctDNA variants may require sequencing of matched PBMCs as shown by some groups (27). Utilizing ctDNA clonality values may aid in the identification of variants that are likely to have originated from the tumor compartment. Nevertheless, this approach will require further validation.

Given the novelty of ctDNA analysis in SCLC, it is important to recognize the limitations of our study. First, we were unable to perform independent verification of the histology or treatment status of a significant number of samples analyzed in this study. Second, samples collected at diagnosis and relapse were also not typically patient-matched. Therefore, we cannot rule out the possibility that the observed variant frequencies are a consequence of sample misclassification or lack of pairing. Third, samples in this study were sequenced using different versions of a commercial assay that differed in their abilities to identify mutations and amplifications in different genes and the extent to which some genes were sequenced (limited exons vs. full coding sequence). Fourth, treatment status of several samples was inferred on the basis of date of diagnosis and sample collection, which can be inaccurate. Some or all of these factors could have altered the frequency at which various mutations were identified in samples collected at diagnosis and at relapse. Fifth, information related to disease burden at time of collection, specific treatment received by individual patients and response to treatment were not available for a more informed clinical-correlative analysis. In addition, hematopoietic variants may have been misclassified as tumor derived in our analysis. However, this limitation is not exclusive to ctDNA analyses, because leukocyte infiltration of solid tumors can also lead to the detection of these variants with tissue-based sequencing (50).

Despite the noted limitations, our findings demonstrate that ctDNA profiling holds promise for examining mechanisms of treatment resistance and directing clinical care in SCLC by facilitating the rapid identification of potentially targetable alterations. Furthermore, unlike prior studies that predominantly sequenced surgically resected early-stage SCLC samples, xenografts, or cell lines, our study included samples collected from a large cohort of patients who most likely received treatment for advanced stage disease, shedding light on alterations that could play important roles in SCLC progression and serve as potential new candidates for therapy. Further studies of ctDNA in patients with SCLC may help accelerate the discovery and development of new drugs against this relentless disease.

S. Devarakonda is a consultant/advisory board member for Guardant 360. K.A. Gold is a consultant/advisory board member for AstraZeneca, Takeda, and Regeneron. R.B. Lanman is an employee of Biolase, Inc.; has ownership interests (including patents) at Guardant Health, Inc., Biolase, Inc., and Forward Medical, Inc.; and is a consultant/advisory board member for Forward Medical, Inc. A.A. Chaudhuri is an employee of Roche Sequencing, Geneoscopy, and Tempus Labs; reports receiving speakers bureau honoraria from Foundation Medicine, Roche Sequencing, and Varian Medical Systems; is a consultant/advisory board member for Roche Sequencing, Geneoscopy, and Tempus Labs; and reports receiving commercial research support from Roche Sequencing. T.K. Owonikoko has ownership interests (including patents) in Cambium Oncology; is a consultant/advisory board member for Novartis, Celgene, Lilly, Sandoz, AbbVie, Eisai, G1 Therapeutics, Takeda, Seattle Genetics, Bristol-Myers Squibb, MedImmune, BerGenBio, Lilly, Amgen, AstraZeneca, PharmaMar, Boehringer Ingelheim, EMD Serono, Xcovery, Bayer, Heron Pharmaceutical, and ARMO BioSciences; reports receiving commercial research grants in the form of payments of institution from Novartis, Astellas Pharma, Celgene, Bayer, Stem CentRx, Regeneron, AstraZeneca/MedImmune Abbvie, G1 Therapeutics, Bristol-Myers, Corvus Pharmaceuticals, United Therapeutics, Amgen, Loxo/Lilly, Fujifilm, Pfizer, Aeglea Biotherapeutics, Incyte, and Merck; and reports being a paid consultant as part of the independent data review committee for EMD Serono. B.T. Li is a consultant/advisory board member for Guardant Health and Thermo Fisher Scientific, and reports receiving commercial research grants from Guardant Health and Genentech. C.M. Rudin is a consultant/advisory board member for AbbVie, Ascentage, AstraZeneca, Bristol-Myers Squibb, Celgene, Daiichi Sankyo, Genentech/Roche, Ipsen, Loxo, Pharmamar, Vavotek, Bridge Medicines, Elucida Oncology, and Harpoon Therapeutics. D. Morgensztern is a consultant/advisory board member for Bristol-Myers Squibb, Takeda, PharmaMar, and AbbVie. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. Devarakonda, V.M. Raymond, R.B. Lanman, B.T. Li, R. Govindan, D. Morgensztern

Development of methodology: S. Devarakonda, V.M. Raymond, R.B. Lanman, T.K. Owonikoko, R. Govindan, D. Morgensztern

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Devarakonda, K.A. Gold, J.P. Ward, V.M. Raymond, R.B. Lanman, T.K. Owonikoko, C.M. Rudin, R. Govindan, D. Morgensztern

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Devarakonda, S. Sankararaman, B.H. Herzog, S.N. Waqar, J.P. Ward, V.M. Raymond, R.B. Lanman, A.A. Chaudhuri, T.K. Owonikoko, B.T. Li, J.T. Poirier, C.M. Rudin, D. Morgensztern

Writing, review, and/or revision of the manuscript: S. Devarakonda, B.H. Herzog, K.A. Gold, S.N. Waqar, J.P. Ward, V.M. Raymond, R.B. Lanman, A.A. Chaudhuri, T.K. Owonikoko, B.T. Li, J.T. Poirier, C.M. Rudin, D. Morgensztern

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B.H. Herzog, V.M. Raymond, R. Govindan, D. Morgensztern

Study supervision: S. Devarakonda, R. Govindan, D. Morgensztern

S. Devarakonda received research funding from the International Association for Study of Lung Cancer. J.P. Ward is supported by the NCI of the NIH Paul Calabresi Career Development Award in Clinical Oncology (K12CA167540). R. Govindan is supported by the NIH/NCI (1U01CA231844-01). C.M. Rudin and J.T. Poirier are supported by the NIH/NCI (R01 CA197936).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.