Activation of PI3K/AKT Pathway Is a Potential Mechanism of Treatment Resistance in Small Cell Lung Cancer

Small cell lung cancer (SCLC) accounts for 13% to 15% of lung cancers and an estimated 250,000 cancer deaths worldwide (1). SCLC is known to be aggressive, owing to its rapid doubling time and high propensity to disseminate early, resulting in 80% to 85% of patients being diagnosed with extensive disease and a 5-year overall survival (OS) of 5% to 6% (2). Platinum doublet chemotherapy has been the main treatment modality for SCLC over the past three decades. Concurrent chemoradiotherapy (CCRT) remains the standard of care for limited-stage (LS)-SCLC (3). Recently, immune checkpoint inhibitors targeting PD-L1 have become part of standard of care in combination with chemotherapy as the first-line therapy for patients with extensive-stage (ES)-SCLC, although the added benefits are only modest (4, 5). Generally, platinum-based chemotherapy is the backbone of treatment and probably the primary driver of benefits in both LS- and ES-SCLC.

Although SCLC is highly sensitive to chemotherapy and radiotherapy initially, the development of treatment resistance is essentially universal. A lack of effective subsequent therapies after relapse has led to poor outcomes in patients with treatment failure. Despite the addition of radiotherapy for LS disease and checkpoint immunotherapy for ES disease, relapse is often unavoidable indicating common mechanisms underlying resistance to chemotherapy and CCRT. Consequently, there is an urgent need to elucidate the mechanisms involved in treatment resistance in SCLC.

High-throughput next-generation sequencing (NGS) has provided deep insight into the genomic landscape of many malignancies and demonstrated great potential for identifying genetic aberrations that can be used to match targeted drugs and monitoring-acquired genetic changes during treatment (6). In recent years, a greater understanding of the molecular alterations that occur in SCLC has been developed through comprehensive genomic analyses (7, 8). Comparing SCLC specimens before treatment and at recurrence may provide novel insights into the molecular features associated with therapeutic failure (9). However, because neither surgical resection nor repeat tumor biopsies are standard of care for relapsed SCLC, very few studies have been conducted to date that explore the molecular profiles of recurrent SCLC. Therefore, the molecular mechanisms that drive therapeutic resistance of SCLC remain poorly defined (10). Recently, a study using whole-exome sequencing (WES) of paired treatment-naïve and relapsed SCLC tumors from 12 patients, and unpaired relapsed SCLC tumors from 18 patients revealed WNT signaling activation as a mechanism of treatment failure in SCLC (11). To the best of our knowledge, there is no similar research in Asian patients to date.

In this study, we have investigated the mechanisms of treatment resistance in SCLC by focusing on comparing the genotype and phenotype in tumor samples of treatment-sensitive versus treatment-resistant SCLC. We conducted WES on paired tumor samples at initial diagnosis and relapse from 11 LS-SCLCs treated with CCRT. In addition, we performed targeted sequencing of 1,021 cancer-related genes on cell-free DNA (cfDNA) obtained at baseline and relapse from 9 additional patients with LS-SCLC treated with CCRT. Furthermore, we performed label-free mass spectrometry (MS)–based proteomics on formalin-fixed paraffin-embedded (FFPE) tumor samples from 28 chemo-resistant and 23 chemo-sensitive patients with ES-SCLC. Lastly, we validated our main findings with analyses of transcriptomic data of SCLC cell lines with distinct therapeutic sensitivities from a public database and in vitro study on chemo-sensitive versus chemo-resistant SCLC cell lines.

Treatment-naïve and paired recurrent tumor samples from 11 patients with LS-SCLC, which were treated with CCRT in Zhejiang Cancer Hospital between June 2015 and June 2018 (Supplementary Table S1), were collected for WES (the tests were performed by Nanjing Geneseeq Technology Inc.). Five milliliters of peripheral blood were collected from each patient and placed into EDTA-coated tubes (BD Biosciences). In addition, we performed global proteomics on baseline tissues of another cohort of patients with chemo-resistant (relapsed within 3 months after first-line chemotherapy; n = 28) and chemo-sensitive (relapsed after 6 months after first-line chemotherapy; n = 23) ES-SCLC (the age, gender, stage, smoking history, and treatment regimen are matched between the two groups; Supplementary Table S2; the tests were performed by Shanghai Genechem Co., Ltd.). FFPE blocks/sections were obtained from the hospitals and shipped to the central testing laboratory under required conditions. Diagnosis and tumor purity of the samples was confirmed by pathologists in the Cancer Hospital of the University of Chinese Academy of Sciences.

cfDNA was collected at baseline and relapse from another 9 patients with LS-SCLC treated with CCRT (Supplementary Table S3) and subjected to deep sequencing using a targeted panel containing 1,021 cancer-related genes (the tests were performed by Geneplus-Beijing Institute, Beijing, China). At least 20 mL of peripheral blood was drawn into the EDTA Vacutainer tube (BD Diagnostics) and processed within 2 hours. Plasma was separated by centrifugation at 1,600 × g for 10 minutes and transferred to new microcentrifuge tubes, then centrifuged at 16,000 × g for 10 minutes to remove remaining cell debris. Peripheral blood lymphocytes (PBL) from the first centrifugation were obtained for the extraction of germline DNA.

Study protocols were approved by the Ethical Review Committee in the Cancer Hospital of the University of Chinese Academy of Sciences. All patients provided written informed consent before study entry.

Genomic DNA was extracted from white blood cells using the DNeasy Blood & Tissue kit (QIAGEN). FFPE samples were deparaffinized with xylene, and genomic DNA was extracted using the QIAamp DNA FFPE Tissue Kit (QIAGEN). Purified genomic DNA was qualified by NanoDrop 2000 for A260/280 and A260/A230 ratios (Thermo Fisher Scientific). All DNA samples were quantified by Qubit 3.0 using the dsDNA High Sensitivity (HS) Assay Kit (Life Technologies) according to the manufacturer's recommendations.

cfDNA was isolated from 4.0–8.0 mL plasma using QIAamp Circulating Nucleic Acid Kit (QIAGEN), and PBL DNA was extracted using the QIAamp DNA Blood Mini Kit (QIAGEN) as previously described. DNA concentration was measured using the Qubit 3.0 fluorometer and the Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific Inc.). cfDNA samples were analyzed on the Agilent 2100 Bioanalyzer using the Agilent High Sensitivity DNA Kit (Agilent Technologies).

Sequencing libraries were prepared using KAPA Hyper Prep kit (KAPA Biosystems) with an optimized manufacturer's protocol. In brief, 1 to 2 μg of genomic DNA which was sheared into 350 bp fragments using Covaris M220 instrument (Covaris), underwent end-repairing, A-tailing, and ligation with indexed sequencing adapters sequentially, followed by size selection for genomic DNA libraries using Agencourt AMPure XP beads (Beckman Coulter). Finally, libraries were amplified by PCR and purified using Agencourt AMPure XP beads.

Exome capture was performed using the IDT xGen Exome Research Panel V1.0 (Integrated DNA Technologies). Enriched libraries were sequenced using the Illumina HiSeq 4000 platform to reach the mean coverage depth of approximately 60× for the PBL control and approximately 150× for the tumor samples.

Before library construction, 1.0 μg of PBL DNA was sheared to 200 to 250 bp fragments with a Covaris S2 ultrasonicator (Covaris). For cfDNA, 20 to 80 ng samples were used for library construction. Libraries were prepared using the NEBNext Ultra II DNA Library Prep Kit for Illumina (New England Biolabs). Hybridization capture of DNA libraries was employed with the custom-designed biotinylated oligonucleotide probes (xGen Lockdown Probes-standard, Integrated DNA Technologies, Inc.) covering approximately 1.0 Mbp coding region of genomic sequence of 1,021 genes frequently mutated in common solid tumors was designed. Captured libraries were measured using an Agilent 2100 Bioanalyzer and an Applied Biosystems 7500 real-time PCR system (Thermo Fisher Scientific Inc.). DNA sequencing was performed on the GeneSeq-2000 (Geneplus-Suzhou) with 2 × 100 bp paired-end reads.

Terminal adaptor sequences and low-quality reads were removed from raw sequencing data. Sequencing data were aligned to the reference human genome (build hg19) using BWA (bwa-mem; ref. 12) and further processed for deduplication, base quality recalibration, and indel (insertion and deletion) realignment with the Picard suite (http://picard.sourceforge.net/) and the Genome Analysis Toolkit (GATK; ref. 13). Single-nucleotide variation (SNV) and small indels were detected by MuTect (14). For tumor tissue, we required minimum variant allele frequency (VAF) equal to 5% or 1%, minimum variant supporting reads = 5 or 3, for nonhotspot and hotspot mutations (common mutated in SCLC), respectively. We recovered mutations that were identified in treatment-naïve samples but did not pass the cutoff in relapsed samples, and vice versa. For cfDNA samples, we required minimum VAF equal to 0.5%, minimum variant supporting reads = 5. SNVs and indels were annotated to genes by ANNOVAR software (15) to identify the mutated protein-coding position and filtered intronic and silent changes. Facets (16), a tumor purity–adjusted pipeline for analysis of allele-specific copy number, were used to detect copy-number variants (CNV). CNV was defined when more than a 90% section of the exon was deleted or amplified for each gene. GISTIC2 (17) was employed to evaluate cohort recurrence for significantly altered regions with CNV using default parameters.

Mutational signatures were extracted from catalogs of somatic mutations with the deconstructSigs R package (18). The signatures evaluated were the 22 signatures of mutational processes in human cancer (19) referred to in Alexandrov and colleagues.

PyClone (20), a hierarchical Bayesian model for inference of subclonal architecture, was used to estimate cancer cell fraction (CCF) and mutation clustering. The clusters with the greatest mean CCF of mutations or indels were defined as clonal or major clone, whereas the rest were subclonal or minor clone.

Data of somatic mutations for matched initial and recurrent human solid tumors in breast cancer, colorectal cancer, and non–small cell lung cancer (NSCLC) performed by WES were obtained from Supplementary Data of published literature (21). Expression data of 18,690 transcripts in 67 SCLC cell lines after adjustment for batch effects as measured using the Affymetrix GeneChip Human Exon 1.0 ST Array were publicly available at http://sclccelllines.cancer.gov (22).

To compare the different signaling pathways between two clusters (SCLC cell lines that were resistant to cisplatin but sensitive to multiple PI3K inhibitors versus SCLC cell lines with cisplatin sensitive but PI3K inhibitor resistant), we performed gene set enrichment analysis (GSEA) based on the gene sets from Kyoto Encyclopedia of Genes and Genomes (KEGG) database using “gseKEGG” function in clusterProfiler R package (v3.18.1; ref. 23). Pathway annotation based on hypergeometric distribution of the presupposed gene list was performed by “enrichKEGG” function in clusterProfiler R package.

The FFPE tissue samples were placed in a 1.5 mL extraction tube, 0.5 mL n-heptane was added and samples were vortexed for 10 seconds, incubated at room temperature for 1 hour, with 25 μL methanol was then added, and they were centrifuged at 9,000 × g for 2 minutes. The supernatant was discarded, and samples were removed from the paraffin floating on the surface and dried in the fume hood for 5 minutes. Next, we added 100 μL SDT buffer, vortexed, and sealed the extraction tubes. Samples were bathed in ice for 5 minutes and whirlpooled. Next, they were heated at 100°C for 20 minutes, then heated at 80°C for 2 hours and vibrated continuously (750 rpm). Samples were kept at 4°C for 1 minutes and centrifugation was performed at 4°C and 14,000 × g for 15 minutes. The supernatant was filtered with 0.22 μmol/L centrifuge tube to collect the filtrate. BCA and SDS-PAGE were used to detect the protein concentration. In order to prepare a large enough quantity of protein for MS analysis, the protein samples of the chemo-sensitive group or the chemo-resistant group were randomly mixed into three samples, respectively. A1, A2, and A3 represent chemo-resistant samples; B1, B2, and B3 represent chemo-sensitive samples.

Twenty microgram of proteins from each sample were mixed with 5× loading buffer respectively and boiled for 5 minutes. The proteins were separated on 10% SDS-PAGE gel (constant pressure 250 V, 40 minutes). Primary antibodies were used as follows: p70s6 Kinase (49D7; 2708, 1:1,000; CST), phospho-p70s6 Kinase (Thr389; 9205, 1:1,000; CST), AKT (pan; C67E7; 4691, 1:1,000; CST), phospho-AKT (Ser473; 4060, 1:1,000; CST), 4E-BP1 (53H11; 9644, 1:1,000; CST), phospho-4E-BP1 (Ser65; 9451, 1:1,000; CST), β-actin (66009, 1:3,000; Proteintech). Membranes were incubated with all primary antibodies at 4°C overnight. Protein bands were visualized by chemiluminescent HRP substrate kit (WBKLS0500, Millipore) with Azure biosystems c500.

Thirty-microliter protein solution for each sample was reduced with 100 mmol/L DTT (final concentration) for 5 minutes at 100°C, and then cooled to room temperature. Then the detergent, DTT and other low-molecular-weight components were removed using UA buffer (8 M Urea, 150 mmol/L Tris-HCl pH 8.5) by repeated ultrafiltration (Sartorius, 30 kD). Then 100 μL iodoacetamide (100 mmol/L IAA in UA buffer) was added to block reduced cysteine residues and the samples were incubated for 30 minutes in darkness. The filters were washed with 100 μL UA buffer three times and then 100 μL 40 mmol/L NH4HCO3 buffer three times. Finally, the protein suspensions were digested with 4 μg trypsin (Promega) in 40 μL 40 mmol/L NH4HCO3 buffer overnight at 37°C, and the resulting peptides were collected as a filtrate.

The peptide of each sample was desalted on C18 Cartridges, then concentrated by vacuum centrifugation and reconstituted in 40 μL of 0.1% (v/v) formic acid. The peptide content was estimated by UV light spectral density at 280 nm using an extinctions coefficient of 1.1 of 0.1% (g/L) solution that was calculated on the basis of the frequency of tryptophan and tyrosine in vertebrate proteins. LC-MS/MS analysis was performed on a Q Exactive Plus mass spectrometer (Thermo Fisher Scientific) that was coupled to Easy nLC (Thermo Fisher Scientific). Two microgram of peptide was loaded onto the C18-reversed phase analytical column (Thermo Fisher Scientific, Acclaim PepMap RSLC 50 μm X 15 cm, nano viper, P/N164943) in buffer A (0.1% formic acid) and separated with a linear gradient of buffer B (80% acetonitrile and 0.1% formic acid) at a flow rate of 300 nL/min. The linear gradient was as follows: 5% buffer B for 5 minutes, 5% to 28% buffer B for 90 minutes, 28% to 38% buffer B for 15 minutes, 38% to 100% buffer B for 5 minutes, hold in 100% buffer B for 5 minutes. MS data were acquired using a data-dependent top10 method dynamically choosing the most abundant precursor ions from the survey scan (350–1,800 m/z) for HCD fragmentation. MS1 scans were acquired at a resolution of 70,000 at m/z 200 with an AGC target of 3e6 and a maxIT of 50 ms. MS2 scans were acquired at a resolution of 17,500 at m/z 200 with an AGC target of 2e5 and a maxIT of 45 ms, and isolation width was 2 m/z. Only ions with a charge state between 2 and 6 and a minimum intensity of 2e3 were selected for fragmentation. Dynamic exclusion for selected ions was 30 seconds. Normalized collision energy was 27 eV.

MS raw files were processed with the MaxQuant software (version 1.6.14.0). The integrated Andromeda search engine was used for peptide and protein identification at a false discovery rate (FDR) of less than 1%. The human UniProtKB database was used as forward database and the automatically generated reverse database for the decoy search. An initial search was set at a precursor mass window of 6 ppm. The search followed an enzymatic cleavage rule of Trypsin/P and allowed maximal two missed cleavage sites and a mass tolerance of 20 ppm for fragment ions. Carbamidomethylation of cysteines was defined as fixed modification, whereas protein N-terminal acetylation and methionine oxidation were defined as variable modifications for database searching. The cutoff of global FDR for peptide and protein identification was set to 0.01. Protein abundance was calculated on the basis of the normalized spectral protein intensity (Label Free Quantitation, intensity).

Human cell line NCI-H69, which is sensitive to chemotherapy and carries multiple mutations in PI3K/AKT pathway and its derived chemo resistant cell line H69AR were obtained from the Chinese Tissue Culture Collections. DMS 53 and NCI-H446 cells were purchased from Cobioer Corporation. H69, DMS 53 and H446 cells were cultured in RPMI-1640 medium (Cat No. 11875093; Gibco) with 10% FBS (Cat No. F8687; Sigma). The multiple drug resistant cells H69AR were maintained in RPMI-1640 medium (Cat No. 11875093; Gibco) with 20% FBS (Cat No. F8687; Sigma). All cells were validated through short-tandem repeat profiling and incubated at 37°C with 5% CO₂ in a humidified incubator.

Total RNA was isolated using TRIzol reagent (Cat No. 3101-100; Shanghai Pufei Biotechnology). Reverse transcription of 2.0 μg of total RNA was performed using the M-MLV Reverse Transcriptase from Promega (Cat No. M1705). Quantitative real-time PCR was conducted using SYBR Master Mixture (Cat No. DRR041B; Takara) in a 12 μL reaction in triplicate on a Roche LightCycler 480 II Real-Time PCR machine. Primer sequences are available in Supplementary Table S4. Gene expression values were normalized to GAPDH levels. Western blot analyses were performed using anti–phospho-Akt (Ser473; Cat No. 4060; Cell Signaling Technology, CST), anti-Akt (pan; Cat No. 4691; CST), anti–phospho-p70 S6 Kinase (Thr389; Cat No. 9205; CST), anti-p70 S6 Kinase (49D7; Cat No. 2708; CST), anti–phospho-4E-BP1 (Ser65; Cat No. 9451; CST), anti–4E-BP1 (53H11; Cat No. 9644; CST), anti–HIF-1α (D2U3T; Cat No. 14179; CST), and anti–β-actin (2D4H5; Cat No. 66009-1-Ig; Proteintech) antibodies. Quantitative data were inputted into GraphPad Prism 5.0 (Graph Pad Software Inc.) for plotting.

SCLC cells were seeded in triplicate in 96-well plates at a density of 2 × 10⁴ per well and treated with increasing concentrations of cisplatin (Cat No. S1166; Selleck), or PI3K/AKT signaling inhibitors (BEZ235, Cat No. S1009; BKM120, Cat No. S2247; GSK2126458, Cat No. 2658), or cisplatin combined with small molecule inhibitors. After 48 h of treatment, cell proliferation was measured using cell viability assay (CCK-8 assay, Cat No. KGA317, KeyGene; AramarBlue cell viability kit, Cat No. DAL1100, Thermo Fisher Scientific). Normalized transformed dose response curves were generated and analyzed using GraphPad Prism (Graph Pad Software Inc.) to determine IC₅₀.

The statistical analyses were performed using R (v4.0.3). Wilcoxon signed rank test was used for comparison of values of paired treatment-naïve and relapsed tumors. The survival curves were evaluated by the Kaplan–Meier algorithm and compared by the log-rank tests using survminer and survival R packages. Proteins with P values (Student t test) < 0.05 were considered differentially expressed.

As shown in Supplementary Table S1, 11 patients with LS-SCLC were treated with CCRT. The best response was CR in 4 patients and PR in 7 patients. The median progression-free survival (PFS) was 10.1 months (4.9 months–20.9 months). Four of the 11 relapsed samples were collected from the in-field recurrence, whereas the remaining seven were collected from the distant metastases. At the last follow-up visit, 10 of the 11 patients had died with the median OS of 29.1 month (6.9 months–42.5 months).

A total of 4,824 nonsilent somatic mutations including SNVs and small indels were identified from these 11 paired tumors with a median of 216 variants (6.52 mutations/Mb) per sample (ranging from 59 to 516; Supplementary Table S5). All tumors harbored potentially functional (nonsynonymous or splice site) SNVs in at least one of the 1,169 cancer genes defined by COSMIC (https://cancer.sanger.ac.uk/cosmic) or OncoKB (https://www.oncokb.org/; Supplementary Table S6). Consistent with previous studies, TP53 mutations were identified in 8 out of 11 treatment-naïve and 8 out of 11 relapsed tumors. Nonsense or frameshift mutations of RB1 were observed in 2 of 11 treatment-naïve and 4 of 11 relapsed samples. Using the tumor purity-adjusted pipeline, Facets, for calling of copy number (16), copy-number loss of RB1 was observed in 7 of 11 treatment-naïve samples, including 5 samples without RB1 mutations, and in 8 of 11 relapse samples including 6 samples without RB1 mutations (Fig. 1). Together, 7 treatment-naïve and 8 relapsed SCLC tumors showed a mutation or CNV in RB1. These results indicated that alterations in TP53 and RB1 were characteristic features both in primary SCLCs and relapsed disease. For patients N05, N06, N07, N08, and N10, whose tumors did not have concomitant TP53/RB1 alterations, the pathologic diagnosis was confirmed by extensive pathologic review and immunohistochemistry (Supplementary Table S7).

Other frequently mutated cancer genes in SCLC (24) were identified in this cohort of SCLC including FAM135B (5 of treatment-naïve vs. 6 of relapsed samples), LRP1B (2 vs. 2), NOTCH1 (1 vs. 1), ZFHX3 (2 vs. 1), and LRRK2 (1 vs. 1). Copy-number variants (CNV) of multiple previously reported cancer genes were observed. In addition to RB1 mentioned above, another tumor-suppressor gene SETDB2 located in 13q14.2, the same chromosome sites harboring RB1, had high-frequency copy-number loss in both treatment-naïve and relapsed samples. Other copy-number losses including 13q13.1–13q14.1 (harboring FOXO1 and BRCA2), 3p11.1–3p12.3 (harboring ROBO1 and EPHA3), 3p14.2 (harboring FHIT), and 17p12–17p13.1 (MAP2K4 and TP53) were also frequent in both treatment-naïve and relapsed tumors. GISTIC analysis did not identify any significantly recurrent somatic copy-number amplification at q < 0.1 in relapsed and treatment-naïve SCLCs, probably due to the small sample size, but multiple large-segment copy-number loss referring to multiple genes were identified (Supplementary Fig. S1; Supplementary Table S8).

We also observed multiple mutated genes previously reported as playing a crucial role in mediating chemotherapy or radiotherapy resistance in SCLC cell lines or other malignancies (25–34). CSMD3 (3 vs. 3), MUC16 (3 vs. 3), RELN (3 vs. 3), BACH2 (2 vs. 2), CDH10 (2 vs. 2), FAT4 (2 vs. 2), KSR2 (2 vs. 2), TERT (2 vs. 2), WT1 (2 vs. 2), LRP1B (2 vs. 2), IKBKE (1 vs. 2), and POLD1 (1 vs. 2) were identified in both treatment-naïve and paired recurrent tumors, among which 34 out of 36 mutations had increased CCF in relapsed tumors. PCLO (0 vs. 2) was only identified in relapsed tumors. Copy-number gain of known oncogenes including TERT, IL7R, and RICTOR were recurrent genomic gain events. Copy-number gain of PIK3CA was identified only in relapsed tumors, which has been reported to be associated with chemotherapy resistance in patients with ovarian cancer (35).

DNA damage repair (DDR) genes and drug transporter genes have been reported to play crucial roles in therapeutic resistance across various malignancies (36–40). We next investigated alterations of these genes in this cohort of SCLCs. In addition to TP53 mutations, SNVs of DDR genes (Supplementary Table S9) were identified in 6 treatment-naïve and 8 relapsed tumors, whereas SNVs of drug transporter genes (Supplementary Table S9) were observed in 7 treatment-naïve and 10 relapsed tumors. Consistent with previous studies (11), these alterations were observed both in treatment-naïve and relapsed SCLCs. The frequent alterations of DDR and drug transporter genes highlighted the importance of these pathways in SCLC cancer biology and drug resistance.

We next analyzed the mutational spectrum and signatures in these SCLCs. C>A transversions were the most common nucleotide substitutions (Supplementary Fig. S2), and Signature 4 (associated with tobacco exposure) was the most or second predominant mutational signature in pretreatment samples (Supplementary Fig. S3) as expected, given all 11 patients were smokers or heavy second-hand smokers. We noted that signature 6, associated with DNA Mismatch repair (MMR) deficiency, was the predominant signature in relapsed tumor of patient N07 that harbored a missense mutation in MSH2 (Q130E). Signature 3, which is associated with failure of DNA double-strand break-repair by homologous recombination, became the predominant signature in recurrent samples of patient N09 and N11 carrying BRCA2 deletions.

The median somatic SNV and indel burden was 209 (ranging from 59 to 454) in treatment-naïve samples, which significantly increased to 220 (ranging from 98 to 520) in relapsed SCLC samples (P = 0.016, Wilcoxon signed rank test; Supplementary Fig. S4A). There was no significant difference in transition, transversion or indel rate between paired treatment-naïve and relapsed samples (Supplementary Fig. S4B–S4D). We next compared the genomic profiles of treatment-naïve versus relapsed SCLC tumors to identify genomic features associated with post-CCRT recurrence. Out of 2,827 mutations across all patients, 1,997 were shared between baseline and recurrent samples (Fig. 2A). The percentage of shared, treatment-naïve specific (present at initial diagnosis but not at relapse) and relapse-specific (present at relapse but absent at initial diagnosis) mutations for each patient were shown in Fig. 2B. We observed a median of 71% (ranging from 39% to 94%) of mutations shared between treatment-naïve and relapsed SCLC tumors, similar to that in NSCLC, but significantly higher than breast and colorectal cancers, (median 0.71 for SCLC vs. 0.81 for NSCLC with P = 0.724; vs. 0.32 for breast cancer with P < 0.001, vs. 0.47 for colorectal with P = 0.035; Fig. 2C; Supplementary Table S10; ref. 21). An average of 94% (73%–100%) of mutations in treatment-naïve tumors were also identified in relapsed tumors (Fig. 2D).

A total of 684 relapse-specific (or acquired) mutations in 630 genes were identified across the 11 patients (Fig. 2A) with a median of 74 (ranging from 1 to 132) acquired mutations in recurrent samples. The top relapse-specific mutated genes were RB1 (5 mutations in 3 patients), PCLO (2 mutations in 2 patients), DNAH5 (2 mutations in 2 patients), PRUNE2 (2 mutations in 2 patients), SYNE1 (2 mutations in 2 patients), and TACC2 (2 mutations in 2 patients). Among these genes, PCLO was reportedly associated with etoposide resistance in ES-SCLC (25) and TACC2 has been demonstrated to associate with therapeutic resistance in neuroblastoma (41), renal cell cancer (33), and prostate cancer (42). These acquired alterations involved multiple pathways, particularly PI3K/AKT pathway. Genes with acquired mutations were enriched in PI3K/AKT signaling pathway totaling 20 genes across 8 of the 11 patients, of which 16 of those genes were located upstream of this pathway (Fig. 3A–C). Of these 20 genes, 15 were mutated exclusively in relapsed tumors.

In addition to SNVs, relapsed tumors had significantly more genes with copy-number gains compared with treatment-naïve tumors (310 vs. 977, P = 0.024). When considering cancer genes only, copy-number gain burden remained significantly higher in relapsed tumors (19 vs. 78, P = 0.042). On the other hand, the copy-number loss burden was not significantly different between treatment-naïve and relapsed tumors. Furthermore, we identified large-scale acquisition of CNVs in 10,464 genes by comparing paired relapsed tumors versus treatment-naïve SCLCs. The top genes with acquired CNVs were MUC12 (copy-number loss in 5 patients), PIK3CA (copy-number gain in 4 patients), MECOM (copy-number gain in 4 patients), and EIF4A2 (copy-number gain in 4 patients). Among these genes, PIK3CA copy-number gain has been reported to associate with resistance to chemotherapy in patients with ovarian cancer (35). Of note, among genes with relapse-specific CNVs in more than 2 patients were also enriched in PI3K/AKT signaling pathway (Fig. 3D; Supplementary Fig. S5). Among these genes, copy-number gain of known oncogenes including PIK3CA (n = 4), MYB (n = 3), SGK1 (n = 3), AKT1 (n = 2), AKT3 (n = 2), BCL2 (n = 2), GSK3B (n = 2), PIK3CB (n = 2), and TCL1A (n = 2) was relapse-specific. In addition, acquired copy-number loss of tumor-suppressor gene TSC2 was identified in 2 patients. Furthermore, GISTIC analysis observed significant chromosome arm scale copy-number loss in cytoband 5q32 (harboring tumor-suppressor genes APC, PIK3R1, MSH3, and RAD50) and 9q22.1 (located by FANCC, TSC1, and TGFBR1) in relapsed tumors but not in treatment-naïve tumors. These genes have been previously reported to associate with therapeutic resistance in various cancers including breast cancer, myeloma, nasopharyngeal carcinoma, and colorectal cancer (43–46).

We next inferred the subclonal architecture of treatment-naïve and relapsed SCLC tumors by PyClone to depict the clonal evolution under chemoradiation. Interestingly, the majority of shared mutations (95%, 1,907/1,997) showed an increase in CCF in relapsed tumors, including 714 mutations with CCF increasing over 20%. In addition, the number of clones was also significantly higher in recurrent samples than that in treatment-naïve samples (13 vs. 12, P = 0.004), indicating new clones emerging after CCRT. We subsequently found that one to three new clones emerged in each of the patients. The relapsed tumors had a total of 684 acquired mutations composing new clones, of which 158 mutations had CCF > 0.6 in 6 relapsed tumors. We next compared subclonal architecture between matched treatment-naïve versus relapsed tumors for each patient. The clusters with the greatest mean CCF of mutations were defined as major clones, otherwise as minor clones or subclones. Overall, the major clones in treatment-naïve tumors were present as major or minor clones in relapsed tumors (Fig. 4). According to the changes of clonal architecture in treatment-naïve and relapsed tumors, we identified two evolutionary patterns. The first pattern showed sustaining major clones in both treatment-naïve and relapsed tumors (patients N04, N05, N06, N07, N09, and N10, Fig. 4A). The second pattern was characterized by the weakening of the major clones in treatment-naïve tumors, which then continued evolving into the two evolutionary patterns. Specifically, patient N01, N03, and N11 showed expansion of some minor clones in the treatment-naïve tumors into major clones in relapsed tumors (Fig. 4B). On the other hand, patients N02 and N08 showed emergence and rapid expansion of new clones as major clones in recurrent tumors (Fig. 4C). Interestingly, pattern 2 was associated with longer PFS than pattern 1 (P = 0.034, log-rank test; Fig. 4D). Of note, among the patients following clonal evolution pattern 1, three patients (N06, N09, and N10) had liver metastases at relapse. As liver metastasis is known to be associated with worse prognosis (47), an alternative explanation is that liver metastasis is associated with a worse prognosis and happens to correlate with clonal evolution pattern 1.

With particular interest in PI3K/AKT pathway activation associated with posttreatment relapse of SCLC, we sought to validate these findings in additional patient samples as well as publicly available SCLC datasets. Due to the scarcity and heterogeneous quality of SCLC tumors, particularly relapsed tumors, different approaches were attempted. First, targeted sequencing of 1,021 cancer-related genes was applied to cfDNA (48) at initial diagnosis and relapse from an independent cohort of 9 additional patients with LS-SCLC treated with CCRT. As shown in Supplementary Table S3, the best response was CR in 1 patients and PR in 8 patients. The median PFS was 8.7 months (4.6 months–19.5 months). In this cohort of patients, 56 genes carried acquired mutations or increased CCF in relapsed samples, which were also enriched in PI3K/AKT signaling pathway (Fig. 5A; Supplementary Table S11).

Next, we obtained a second independent set of 28 chemo-resistant and 23 chemo-sensitive ES-SCLC samples from patients who received standard platinum chemotherapy. Due to the relatively low sample quality, label-free MS-based proteomics was applied to identify the differentially expressed proteins between the chemo-resistant and chemo-sensitive SCLCs. MS analysis revealed 85 differentially expressed proteins including 36 upregulated and 49 downregulated genes in chemo-resistant SCLCs (Supplementary Fig. S6; Supplementary Table S12). Importantly, multiple genes have been reported to associate with aggressive cancer biology and therapeutic resistance across various cancers (Supplementary Fig. S6). For example, the expression of EIF4A2 was reported to be significantly higher in colorectal tumors and it has been demonstrated to promote metastasis and resistance to chemotherapy in colorectal cancer (49), whereas upregulation of RMND1 has been reported to play important roles not only in carcinogenesis but also in progression toward a more aggressive phenotype in breast cancers (50). CACYBP was reported to enhance multidrug resistance of pancreatic cancer cells by regulation of P-gp and Bcl-2 (51) and MTHFD1 was indicated in chemoresistance to gemcitabine in cholangiocarcinoma (52). Furthermore, KEGG pathway enrichment analysis revealed that genes with differentially upregulated proteins (fold change > 1.2) were enriched in WNT signaling pathway, RNA degradation, Purine metabolism and HIF-1 signaling pathway (Fig. 5B). As the PI3K/AKT pathway plays key roles in maintaining the transcription, translation and biological activity of HIF-1α in various malignancies (53–57), we reannotated the 62 PI3K/AKT signaling pathway genes with copy-number gains identified from the WES cohort. Interestingly, 7 genes (AKT1, AKT3, BCL2, IL6R, PIK3CA, PIK3CB, and RPS6) were involved in HIF-1 pathway with 6 of them located upstream of HIF-1 (Fig. 5C).

Moreover, we attempted to validate these findings in publicly available data from SCLCs with distinct therapeutic response. First, we analyzed the WES data from a pioneer study on 12 paired treatment-naïve SCLC tumors versus relapse tumors (11). The analysis identified 882 genes carrying acquired mutations in relapsed tumors, which were also enriched in PI3K/AKT signaling pathway (Supplementary Table S13). We next obtained transcriptomic data of SCLC cell lines with distinct therapeutic responses from a public database (22) for further validation at gene expression level. We identified 12 cell lines that are resistant to cisplatin but sensitive to PI3K inhibitors and 11 cell lines sensitive to cisplatin but resistant to PI3K inhibitors using DISARM (ref. 58; Supplementary Table S14). Differentially expressed pathway analysis by GSEA revealed that PI3K/AKT pathway and HIF-1 pathway were significantly upregulated in cells resistant to cisplatin but sensitive to PI3K inhibitors (Supplementary Fig. S7).

We further attempted to validate the role of PI3K/AKT pathway in development of therapeutic resistance of SCLC using a pair of chemo-sensitive (H69) and chemo-resistant (H69AR) SCLC cell lines. The H69 cell line harbors mutations in several PI3K/AKT pathway genes (n = 12; Supplementary Table S15), whereas H69AR displays cisplatin resistance compared with H69 cells (Fig. 6A). We assessed the expression of 14 genes carrying acquired CNV with copy-number ratio (relapsed/treatment-naïve tumors) > 1.5 for copy-number gains or < 0.5 for copy-number loss (Supplementary Table S4). Consistent with the CNV analysis, MYB, PI3KCA, HSP90AA1, and BCL2 with copy-number gains were upregulated, whereas the TP53 with copy-number loss was downregulated in chemo-resistant H69AR cells (Fig. 6B). Furthermore, we performed western blot to assess the total protein as well as the phosphorylated form of key genes in the PI3K/AKT pathway. As shown in Fig. 6C, phosphorylation of AKT and its downstream targets, 4EBP-1 and p70S6 were increased in chemo-resistant H69AR cells. Finally, we performed cell viability assay on H69AR, DMS53, and H446 cells treated with cisplatin with or without various PI3K inhibitors. The synergistic effect estimated by the Bliss Synergy Score (59) demonstrated synergistic effect between cisplatin with several PI3K inhibitors including BEZ235, BKM120 and GSK2126458 (Fig. 6D).

The rapid emergence of resistance to chemotherapy and CCRT in SCLC is a key contributor to treatment failure and poor survival. In our previous prospective clinical trial (60), nearly a third of the patients with LS-SCLC developed local/regional failure after receiving CCRT, suggesting intrinsic and/or acquired resistance by the residual cancer cells after chemotherapy and/or radiotherapy as a main reason for treatment failure. However, the mechanisms underlying treatment resistance in SCLC are unclear, largely due to the limited accessibility of patient tissues. There is an urgent need to understand the mechanisms underlying resistance to CCRT and facilitate developing novel strategies to improve outcomes of patients with SCLC. In this study, we revealed potential mechanisms leading to treatment resistance in SCLC by comparing the genotype and phenotype of treatment-sensitive versus treatment-resistant SCLC tumors collected through extensive efforts, analyzing publicly available data from well characterized SCLC tumors and cell lines and by in vitro functional validation.

Of the 11 paired tumor samples sequenced by WES, we observed frequently mutated genes previously reported to play a crucial role in mediating chemotherapy or radiotherapy resistance in SCLC cell lines or other malignancies (25–34). Several mutated genes were identified in both treatment-naïve and paired recurrent tumors, but 34 of 36 mutations had increased CCF in recurrent tumors suggesting expansion of treatment resistant subclones as major clones could be one of the mechanisms of tumor recurrence in SCLC. In addition, multiple CNVs previously reported to be associated with chemotherapy or radiotherapy resistance, as demonstrated by copy-number gain of PIK3CA (35) were observed in the current tumors.

Although a substantial proportion of mutations were shared when comparing treatment-naïve versus posttreatment relapsed tumors, relapse-specific mutations and CNVs were identified. Of particular interest, genes with acquired mutation or CNVs were both enriched in the PI3K/AKT signaling pathway, implying a potential role of this pathway in treatment failure for SCLC. This interesting finding was further supported by targeted cfDNA sequencing of an independent cohort of patients with LS-SCLC, reanalysis of a previously published cohort of paired treatment naïve and relapsed SCLC (11), analysis of the transcriptomic data of SCLC cell lines from a public database and in vitro functional validation. PI3K/AKT is one essential singling pathway involved in tumorigenesis, proliferation, apoptosis, angiogenesis, epithelial–mesenchymal transition, immune microenvironment, and treatment resistance (61). It is a frequently altered pathway across various malignancies (62), but its prevalence in SCLCs varies greatly in different studies. This high variability could be due to relatively small sample size-a major challenge for SCLC research, differences in patient population (e.g., Western versus Asian population), sequencing modalities (e.g., WES versus targeted sequencing) and definition of genetic alterations (all alterations versus targetable alterations) etc. In two landmark genomic profiling studies on Western patients with SCLC, 2% to 6% of treatment-naïve SCLC tumors were identified to carry mutations in PI3K/AKT pathway (7), whereas in a Japanese study, genetic alterations of PI3K/AKT/mTOR pathway were only detected in 36% of (18 of 51) treatment-naïve SCLC tumors (63). In another study on Japanese patients, targetable genetic alterations (total number of alterations was not reported) were identified by targeted sequencing in 53 of 9,300 patients with SCLC, most of whom have had prior chemotherapy (64).

In addition to the PI3K/AKT pathway, proteomic analysis of a cohort of chemo-sensitive versus chemo-resistant ES-SCLC revealed activation of HIF-1 signaling associated with treatment resistance. Interestingly, PI3K/AKT pathway is known to play key roles on maintaining the transcription, translation, and biological activity of HIF-1α in various cancers (53–57) via its downstream serine threonine kinases, AKT and mTOR, which subsequently mediate their actions via a cascade of downstream effectors including the translational regulatory proteins EIF-4E–binding protein 1 (4EBP1), p70 ribosomal protein S6 kinase (p70S6K), the ribosomal protein S6 (RPS6), and eukaryotic translation initiation factor 4E (EIF-4E), and eventually these actions result in enhanced translation of HIF-1α mRNA into protein (65). In this study, we also observed that the PI3K/AKT pathway was activated by the increase in phosphorylation of AKT and its downstream targets, 4EBP-1 and p70S6, in chemo-resistant H69AR cells. The caveat is that activation of the PI3K/AKT pathway was based on genomic sequencing of paired treatment-naïve and relapsed LS-SCLC, whereas HIF-1 pathway activation was identified by proteomics analysis of nonpaired chemo-sensitive versus chemo-resistant ES-SCLC. Therefore, further studies are warranted to delineate the mechanistic link between alterations of HIF-1 and PI3K/AKT pathways underlying chemoresistance in SCLC.

Interestingly, another pathway identified among the top enriched pathways was olfactory transduction pathway (Fig. 3). Of note, olfactory transduction pathway has been found significantly enriched across multiple malignancies such as esophageal squamous cell carcinoma, lung cancer, and glioblastoma (66–68). The role of this pathway during carcinogenesis is unknown. One possible explanation is that there are a huge number of genes related to this pathway, more than any other pathway. Therefore, it tends to be enriched using hypergeometric distribution test.

Finally, our subclonal analyses suggested that the different evolutionary patterns may be important for patients' outcomes. We identified two evolutionary patterns from WES data. Pattern 1 was characterized by sustained major clones in both treatment-naïve and relapsed tumors, whereas Pattern 2 demonstrated weakening of the major clones present in the treatment-naïve tumors accompanied by expansion of some minor clones to become the dominant major clones or establishment of rapidly growing new clones as major clones in the relapsed tumors. Evolutionary pattern 2 was associated with longer survival (Fig. 4D) which may be related to the increased numbers of clonal mutations, that is known to be associate with favorable prognosis of various cancers (69, 70), including SCLC following chemotherapy or concurrent CCRT (71, 72).

Two major limitations of this study are the small sample size and the caveat of combining ES-SCLC that is often treated with chemotherapy/immunotherapy and LS-SCLC that is usually treated with concurrent chemoradiation. Both limitations are rooted in the same challenge of studying SCLC: inadequate materials. The study of SCLC lags far behind other solid tumors, largely due to the lack of appropriate materials, especially paired treatment naïve and recurrent samples. Although different treatment modalities may be associated with different resistance mechanisms, relapse is often unavoidable for SCLC regardless of stage (LS- vs. ES-SCLC) or therapeutic approaches (radiation added to chemotherapy to LS-SCLC versus immunotherapy added to ES-SCLC) suggesting common resistance mechanisms may exist. Shared resistance mechanisms associated with chemotherapy and CCRT have been reported previously (11). Furthermore, consolidative chest radiotherapy has become a part of standard of care for appropriate patients with ES-SCLC. As a matter of fact, one third of the patients with ES-SCLC did receive consolidative chest radiotherapy after chemotherapy. Therefore, the findings in our study are likely relevant to the common mechanisms underlying resistance to chemotherapy and CCRT. Nevertheless, future studies are warranted on large cohorts of patients, ideally LS-SCLC and ES-SCLC being investigated separately with paired pretreatment and posttreatment specimens, which will likely require multi-institutional collaboration- as well as appropriate modeling systems such as patient-derived xenografts and/or genetically engineered mouse models to validate these intriguing findings and advance our understanding of mechanisms underlying treatment resistance to facilitate development of novel therapeutic strategies to improve survival of patients with SCLC.

C.M. Gay reports grants and personal fees from AstraZeneca and personal fees from Jazz Pharmaceuticals, Bristol Myers Squibb, and BeiGene outside the submitted work. J. Zhang reports grants from Merck; grants and personal fees from Johnson and Johnson and Novartis; and personal fees from Bristol Myers Squibb, AstraZeneca, GenePlus, Innovent, OrigMed, Innovent, and Roche outside the submitted work. No disclosures were reported by the other authors.

Y. Jin: Resources, data curation, software, formal analysis, methodology, writing–original draft, writing–review and editing. Y. Chen: Resources, data curation, writing–review and editing. H. Tang: Conceptualization, resources, writing–review and editing. X. Hu: Resources, data curation, writing–review and editing. S.M. Hubert: Formal analysis, writing–review and editing. Q. Li: Software, formal analysis, writing–review and editing. D. Su: Resources, writing–review and editing. H. Xu: Resources, writing–review and editing. Y. Fan: Resources, writing–review and editing. X. Yu: Resources, writing–review and editing. Q. Chen: Resources, writing–review and editing. J. Liu: Resources, writing–review and editing. W. Hong: Resources, writing–review and editing. Y. Xu: Resources, writing–review and editing. H. Deng: Resources, data curation, writing–review and editing. D. Zhu: Resources, writing–review and editing. P. Li: Software, formal analysis, writing–review and editing. Y. Gong: Software, formal analysis, writing–review and editing. X. Xia: Software, formal analysis. C.M. Gay: Formal analysis, supervision, writing–review and editing. J. Zhang: Conceptualization, supervision, writing–review and editing. M. Chen: Conceptualization, supervision, funding acquisition, writing–review and editing.

This study was supported by the National Natural Science Foundation of China (grant number 81672972) and Major Program of Provincial and Ministerial Co-construction, Ministry of Health Science Foundation (grant number WKJ-ZJ-1701).

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.