Small-cell lung cancer (SCLC) is an aggressive neuroendocrine lung cancer. Oncogenic MYC amplifications drive SCLC heterogeneity, but the genetic mechanisms of MYC amplification and phenotypic plasticity, characterized by neuroendocrine and nonneuroendocrine cell states, are not known. Here, we integrate whole-genome sequencing, long-range optical mapping, single-cell DNA sequencing, and fluorescence in situ hybridization to find extrachromosomal DNA (ecDNA) as a primary source of SCLC oncogene amplifications and driver fusions. ecDNAs bring to proximity enhancer elements and oncogenes, creating SCLC transcription-amplifying units, driving exceptionally high MYC gene dosage. We demonstrate that cell-free nucleosome profiling can noninvasively detect ecDNA amplifications in plasma, facilitating its genome-wide interrogation in SCLC and other cancers. Altogether, our work provides the first comprehensive map of SCLC ecDNA and describes a new mechanism that governs MYC-driven SCLC heterogeneity. ecDNA-enabled transcriptional flexibility may explain the significantly worse survival outcomes of SCLC harboring complex ecDNA amplifications.

Significance:

MYC drives SCLC progression, but the genetic basis of MYC-driven SCLC evolution is unknown. Using SCLC as a paradigm, we report how ecDNA amplifications function as MYC-amplifying units, fostering tumor plasticity and a high degree of tumor heterogeneity.

This article is highlighted in the In This Issue feature, p. 799

Cancer cells from the same tumor often exhibit substantial heterogeneity in morphology, genetics, epigenetics, and gene expression. The categories, origins, and organization of tumor heterogeneity are open questions of fundamental importance to cancer biology. Moreover, a better understanding of the driving forces of tumor heterogeneity might facilitate the development of more effective approaches to combat heterogeneity and its consequences.

Small cell lung cancer (SCLC) is a model disease to study tumor heterogeneity, and its clinical implications including treatment resistance and metastases. It is a high-grade neuroendocrine lung carcinoma notable for exquisite chemosensitivity at diagnosis and a dramatic shift toward chemoresistance at relapse (1). SCLC consists of tumor cells with neuroendocrine (NE) and nonneuroendocrine (non-NE) features (2–7), further defined by differential expression of the three lineage-defining transcription factors ASCL1, NEUROD1, and POU2F3. A fourth subgroup has been variously characterized by YAP1 expression (2, 8–10) or low expression of all three transcription factors accompanied by an inflamed gene signature (11). SCLC heterogeneity increases over the course of treatment, with the expansion of the non-NE cell population associated with chemoresistance (11–13). In genetically engineered mouse models (GEMM), SCLC can arise from different cells of origin, profoundly influencing their gene expression, chromatin accessibility, and metastatic characteristics (14–17). Induction of Notch signaling, generally suppressed in NE SCLC, can induce transition from an NE to non-NE cell state (7, 18, 19). Recent studies demonstrate that MYC can activate Notch signaling to promote the temporal evolution of SCLC sequentially from an ASCL1 to a NEUROD1 to a non-NE state from an NE cell of origin (19–21). However, the genetic mechanisms that govern MYC-driven SCLC heterogeneity and transcriptional remodeling are poorly understood.

Extrachromosomal DNA (ecDNA) are extrachromosomal chromatin elements that frequently contain oncogenes (22–24). In contrast to chromosomal DNA, ecDNAs lack centromeres that dictate the organization of the mitotic spindle (22, 25), and as such, they are randomly distributed across daughter cells during mitosis, leading to a rapid increase of genomic heterogeneity (26). The high DNA template level generated by ecDNA amplification, coupled with its highly accessible chromatin, permits massive oncogene transcription (27). ecDNA-based oncogene amplifications have been described across multiple cancers (22, 26, 28). SCLC was one of the first tumor types wherein ecDNA and chromothripsis, the catastrophic shattering of chromosomes and a major driver of ecDNA amplification, were described (29–31). However, SCLC ecDNAs have not been characterized further, and recent pan-cancer analyses of ecDNA amplifications did not include SCLC (28). As such, the oncogenes transcribed by these structures and their role in SCLC plasticity are not known.

Although the majority of patients with SCLC present with and succumb to metastatic disease, the current understanding of SCLC heterogeneity is largely derived from mouse models (15, 32). Decoding the mechanistic underpinnings of the high degree of phenotypic heterogeneity, plasticity, and rapid adaptability in human tumors is crucial (32) and may inform efforts to combat heterogeneity and its consequences. We hypothesized that ecDNAs may underlie SCLC heterogeneity and sought to systematically investigate the spectrum and impact of ecDNA in human SCLC. We also hypothesized that ecDNA may be detectable noninvasively from circulating free DNA (cfDNA), serving as highly informative biomarkers.

ecDNAs Are a Major Source of SCLC Oncogene Amplifications and Intertumor Heterogeneity

To catalog oncogene amplifications in SCLC, we analyzed SCLC cell line (33–35) and primary tumor (1) whole-genome sequences (WGS) for high-level gene amplifications. MYC family genes, MYCL, MYCN, and MYC, were the most commonly amplified genes in SCLC. High-level focal genomic amplifications (copy number >5) of the MYC paralogs were observed in 23 (47.9%) cell lines and 14 (14.74%) tumor samples, with more than 50 copies noted in some cases (Fig. 1A, summarized in Supplementary Table S1), and involved MYCL and MYC more often than MYCN (Fig. 1B). Focal amplifications with high copy numbers also involved other oncogenes including FGFR1 and KRAS (Supplementary Fig. S1A summarizes the COSMIC database tier 1 genes) but were less frequent. Focal amplifications covered tiny portions of the genome, usually between 50,000 and 1 million base pairs (Mb) in contrast to chromosome-level amplifications that covered multiple Mb (Supplementary Fig. S1B). Focal amplifications were of monoallelic origin, with enrichment of one germline allele and not the other (Supplementary Fig. S1C). Consistent with their known role as key drivers of SCLC transcriptional subtypes (19, 36), focal MYCL amplification was predominantly observed in ASCL1 and NEUROD1-high SCLCs (SCLC-A and SCLC-N, respectively), whereas MYC was commonly amplified in the NEUROD1-high SCLC-N subtype (Fig. 1C; Supplementary Table S1). Maximal MYC expression levels were observed particularly in SCLC-N cells with MYC focal amplifications (Fig. 1D). Of non-NE SCLCs, the YAP1 cells had lower MYC expression than NEUROD1 cells and harbored no focal amplifications. POU2F3 cells included two MYC, and one MYCN highly amplified cell lines, paired with elevated MYC expression in most cell lines. Across all SCLC subtypes, MYC copy gains were highly correlated with MYC mRNA expression (Supplementary Fig. S1D).

Figure 1.

Gene amplification landscape and transcriptional subtypes of SCLC. A, Normalized sequencing depth quantified from WGS of highly amplified oncogenes and tumor suppressor genes in SCLC cell lines (left) and tumors (right). CCLE, Cancer Cell Line Encyclopedia. B, Number of samples with high amplification (>5 normalized signal) of MYC and other oncogenes in cell lines (red) and tumors (gray). Fraction of samples in each cohort is shown on top of the bars. ecDNA-positive cases are shaded with light red and light gray for cell lines and tumor samples, respectively. C, Expression heat map of MYC genes and lineage-specific transcription factors differentiating SCLC subtypes. Top bars denote subtypes and sample source (tumor or cell line). The second panel shows samples with high MYC/MYCL/MYCN amplification and classification based on AmpliconArchitect defined into complex ecDNA (red), simple ecDNA (gray), break–fusion–bridge (BFB, blue), highly amplified (green), and no high amplification (white) categories. Bottom, expression patterns of SCLC subtype–differentiating genes. D, Expression of MYC in the four main SCLC subtypes with focal MYC amplification (red dots) or no focal amplification (black dots). E, Distinct rearrangement patterns at oncogenes in SCLC tumors. Top, a complex ecDNA with multiple focal amplifications connected at break sites marked with red arches. Expected diploid levels of sequencing depths are marked by purple lines. The MYCN oncogene position is marked with a red box. Middle demonstrates a simple ecDNA pattern, in which the two break sites of a single focal amplification are assembled. Bottom, a genome-integrated BFB, in which distinct break sites are not connected together but loop back based on sequencing data. F, Kaplan–Meier plot showing survival differences between patients with or without complex rearrangements in SCLC.

Figure 1.

Gene amplification landscape and transcriptional subtypes of SCLC. A, Normalized sequencing depth quantified from WGS of highly amplified oncogenes and tumor suppressor genes in SCLC cell lines (left) and tumors (right). CCLE, Cancer Cell Line Encyclopedia. B, Number of samples with high amplification (>5 normalized signal) of MYC and other oncogenes in cell lines (red) and tumors (gray). Fraction of samples in each cohort is shown on top of the bars. ecDNA-positive cases are shaded with light red and light gray for cell lines and tumor samples, respectively. C, Expression heat map of MYC genes and lineage-specific transcription factors differentiating SCLC subtypes. Top bars denote subtypes and sample source (tumor or cell line). The second panel shows samples with high MYC/MYCL/MYCN amplification and classification based on AmpliconArchitect defined into complex ecDNA (red), simple ecDNA (gray), break–fusion–bridge (BFB, blue), highly amplified (green), and no high amplification (white) categories. Bottom, expression patterns of SCLC subtype–differentiating genes. D, Expression of MYC in the four main SCLC subtypes with focal MYC amplification (red dots) or no focal amplification (black dots). E, Distinct rearrangement patterns at oncogenes in SCLC tumors. Top, a complex ecDNA with multiple focal amplifications connected at break sites marked with red arches. Expected diploid levels of sequencing depths are marked by purple lines. The MYCN oncogene position is marked with a red box. Middle demonstrates a simple ecDNA pattern, in which the two break sites of a single focal amplification are assembled. Bottom, a genome-integrated BFB, in which distinct break sites are not connected together but loop back based on sequencing data. F, Kaplan–Meier plot showing survival differences between patients with or without complex rearrangements in SCLC.

Close modal

Oncogene amplification can present as ecDNA or intrachromosomal DNA elements. We applied AmpliconArchitect, an algorithm that computationally detects ecDNA, on primary SCLC WGS, followed by manual curation of results (MYC paralogs seen in Fig. 1C, summary in Supplementary Table S1; Fig. 1 extended supplement; refs. 1, 37). The algorithm identifies rearrangement patterns at focal amplifications, followed by assembly into continuous sequences based on orientation and alignment localization of read pairs. The assembled sequences are then classified as ecDNA or reintegrated events such as break–fusion–bridges (BFB) based on the absence or presence of “self-looping” break sites (Fig. 1E). Remarkably, most high focal amplifications in SCLC genomes coincided with regions assembled as ecDNA (Fig. 1E). We found that 18 of 96 (18.8%) tumors carried one or more ecDNA amplicons. In most cases, ecDNAs were assembled from multiple genomic segments that were amplified at distinct levels and interrupted by nonamplified or deleted regions, a profile reminiscent of complex or “seismic” amplification (38). An example can be seen in Fig. 1E, top, in which the MYCN oncogene is highlighted in red and the assembly of break sites is summarized with red arches. In five cases (5.2%), circular DNA was assembled from a single continuous locus, such as Fig. 1E (middle), which harbors the MYCL oncogene with one single assembly arch connecting the two edges of amplification. Linear amplicons with a stepwise genomic amplification pattern were observed in 11 tumors (11.5%, Fig. 1E, bottom), characteristic of BFB cycles, in which paired-end reads did not support any connections between the break sites but rather arched back to themselves. Importantly, patients with SCLC whose tumors contained complex ecDNAs, regardless of the specific oncogene amplified, had significantly worse overall survival compared with patients whose tumors harbored simple ecDNA, non-ecDNA amplifications, or no amplifications (HR = 2.8, P = 8.6E−03; Fig. 1F; Supplementary Fig. S1E). We fitted a Cox proportional hazards model that tested survival after controlling for clinical variables, including age, sex, and stage, and found that patients with tumors carrying complex circular amplicons had significantly higher HRs for death (HR = 2.7, P = 0.021; Supplementary Table S1). Together, we find that most highly amplified MYC paralog genes are encoded on ecDNA (Fig. 1B) and that tumors harboring complex ecDNA amplifications have significantly worse outcomes.

Assembly of SCLC ecDNA Structures and Epigenetic Characterization

Next, we examined SCLC cell lines for ecDNA-based oncogene amplifications using the same approach described above for patient tumors (33–35). Complex circular amplicons were more frequent in cell lines than in tumors, possibly reflecting the selective advantage afforded by ecDNA amplifications (Supplementary Table S1). Of the 32 cell lines with high focal amplifications, 28 were positive for ecDNA, 3 had BFB events, and 1 had a nonfocal amplification (NCI-H211) based on the AmpliconArchitect analysis (ref. 37; Supplementary Table S1; Fig. 2 extended supplement). Circular head-to-tail junctions were predicted computationally in NCI-H889, encoding the MYCL oncogene, and in NCI-H524, encoding the MYC oncogene. We used PCR and Sanger sequencing to confirm the junctional reads in NCI-H889 and NCI-H524 (blue arch in Fig. 2A; Supplementary Fig. S2A).

Figure 2.

ecDNAs in SCLC and their epigenetic regulation. A, Assembly of ecDNA from WGS of SCLC cell lines NCI-H889 and NCI-H524. Sequencing coverage is shown with gray bars, and assembled segments are connected with red arches. The validated assembly (arch) is highlighted in blue for the two cell lines. B, Assembly of ecDNA structure using long-read optical mapping. Blue segments represent the reads, the orange segments represent the genome used for mapping, and the light and dark blue segments represent the assembled ecDNA sequence for NCI-H889 and NCI-H524, respectively. C, FISH validation of ecDNA using MYCL probes (green) in NCI-H889 and MYC probes (light blue) in NCI-H524. DMS-114 showed chromosomal amplification of MYC (light blue). D and E, Enhancer interaction landscape in NCI-H889 (D) and NCI-H524 (E). The upper track shows the normalized interaction heat map derived from Hi-ChIP, with oncogenic enhancer interactions highlighted in blue circles (proximal) and a green oval (distal). Sequencing coverage tracks are shown for H3K27ac Hi-ChIP (purple), H3K27ac ChIP-seq (blue), ATAC-seq (yellow, only for NCI-H524), and genome sequencing (dark gray). F, Comparison of interaction strengths of enhancers within the ecDNA, at housekeeping genes (GAPDH, B2M, TOP2A, ACTB, and PGK1), and SCLC genes (ASCL1, NEUROD1, INSM1, SYP, NOTCH1, REST, and YAP1) and their proximal enhancers. ***, P < 1e-16 [calculated with R using the wilcoxon.test() function].

Figure 2.

ecDNAs in SCLC and their epigenetic regulation. A, Assembly of ecDNA from WGS of SCLC cell lines NCI-H889 and NCI-H524. Sequencing coverage is shown with gray bars, and assembled segments are connected with red arches. The validated assembly (arch) is highlighted in blue for the two cell lines. B, Assembly of ecDNA structure using long-read optical mapping. Blue segments represent the reads, the orange segments represent the genome used for mapping, and the light and dark blue segments represent the assembled ecDNA sequence for NCI-H889 and NCI-H524, respectively. C, FISH validation of ecDNA using MYCL probes (green) in NCI-H889 and MYC probes (light blue) in NCI-H524. DMS-114 showed chromosomal amplification of MYC (light blue). D and E, Enhancer interaction landscape in NCI-H889 (D) and NCI-H524 (E). The upper track shows the normalized interaction heat map derived from Hi-ChIP, with oncogenic enhancer interactions highlighted in blue circles (proximal) and a green oval (distal). Sequencing coverage tracks are shown for H3K27ac Hi-ChIP (purple), H3K27ac ChIP-seq (blue), ATAC-seq (yellow, only for NCI-H524), and genome sequencing (dark gray). F, Comparison of interaction strengths of enhancers within the ecDNA, at housekeeping genes (GAPDH, B2M, TOP2A, ACTB, and PGK1), and SCLC genes (ASCL1, NEUROD1, INSM1, SYP, NOTCH1, REST, and YAP1) and their proximal enhancers. ***, P < 1e-16 [calculated with R using the wilcoxon.test() function].

Close modal

Because there were multiple break sites, ecDNA structures were confirmed using optical mapping, which enabled the development of a physical map based on long contiguous pieces of DNA (ref. 39; Fig. 2B). For example, MYCL and RLF genes are ∼250 kb apart in the genome but positioned in close proximity on ecDNA based on WGS assembly and optical mapping contigs in NCI-H889. Optical maps also reconstructed complex break sites within PVT1, a known regulator of MYC expression in NCI-H524. The PVT1 promoter was positioned close to MYC on ecDNA, whereas a PVT1 intronic segment was more distally positioned on ecDNA (Supplementary Fig. S2B; ref. 34). Next, we examined localization of MYC, MYCL, and MYCN in metaphase nuclei using fluorescence in situ hybridization (FISH) probes that target these oncogenes in multiple cell lines. Both NCI-H889 and NCI-H524 (Fig. 2C) demonstrated several copies of MYCL and MYC located outside the chromosome in each cell examined (median count of 27 and 45 ecDNA/cell in NCI-H524 and NCI-H889, respectively). We probed DMS-114 for MYC as a negative control, which showed chromosomal amplification (Fig. 2C). We probed ecDNA in additional cell lines MYC (DMS-273 and NCI-H446), MYCN (NCI-H69), or MYCL (NCI-H1963 and NCI-H1092) to confirm ecDNA predicted by WGS (Supplementary Fig. S2C). We identified extrachromosomal signals in all cases except NCI-H446 (Supplementary Fig. S2C), which showed homoge­neous staining regions (HSR), likely representing chromosomal reintegration of ecDNA (40).

EcDNA amplification is associated with increased oncogene expression (Fig. 1C; Supplementary Table S1). To understand the epigenetic regulation of SCLC ecDNA, we characterized the enhancer landscape of several cell lines using Hi-ChIP, which examines chromosome conformation, targeting the enhancer-associated mark histone H3 lysine 27 acetylation. The genomic signal of H3K27ac derived from Hi-ChIP closely mirrored previously published chromatin immunoprecipitation sequencing (ChIP-seq) signals, which served as a validation (41, 42). In the case of the MYCL ecDNA-positive NCI-H889, the enhancer signal of the MYCL promoter was strong but substantially weaker than the enhancer loci at TRIT1 or RLF, localized respectively downstream and upstream of MYCL (Fig. 2D, blue circle), suggesting that the enhancers of these neighboring genes may be key regulators. In the case of the MYC ecDNA-positive NCI-H524, we observed strong enhancer–enhancer interactions between MYC and the neighboring PVT1 genes (Fig. 2E, blue circle), with high chromatin accessibility defined from assay for transposase-accessible chromatin with sequencing (ATAC-seq; refs. 21, 43). We also observed ecDNA-based interactions between enhancers that would be multiple megabases apart in the genome (Fig. 2E, green oval). Enhancers on ecDNA had exceptionally strong interaction signals in comparison with enhancer–enhancer pairs at housekeeping genes or key SCLC genes and their neighboring enhancers (Fig. 2F, left). However, after normalizing for DNA copy number, these enhancer–enhancer interaction signals on the ecDNA were more comparable with non-ecDNA enhancers at housekeeping genes and SCLC genes (Fig. 2F, right; Supplementary Table S2). These data illustrate how SCLC ecDNAs function as transcription-amplifying elements (44), enabling enhancers that would otherwise be distant in the genome to be positioned in close proximity to oncogenes, which, coupled with the multicopy nature of ecDNA, drives massive oncogene transcription.

Next, we examined DNA methylation, another key regulator of gene expression that is also associated with chromatin organization, in ecDNA and non-ecDNA regions. Because DNA methylation assays can predict copy number (45), we first tested whether ecDNA was detectable from methylation data. Indeed, ecDNA focal amplification patterns were detectable from genome-wide methylation platforms including 850K methylation array and reduced-representation bisulfite sequencing (RRBS; Supplementary Fig. S2D), suggesting that ecDNAs and HSRs are subject to methylation. Next, we focused on a shared region of ecDNA among three MYCL ecDNA-positive cell lines, comparing their distribution of methylation probes to three ecDNA-negative cell lines (ref. 34; Supplementary Fig. S2E). The ecDNA-positive cell lines had a similar distribution of methylation probes as the ecDNA-negative cell lines, although methylation levels were slightly lower (Supplementary Fig. S2F). Pairwise comparison of the common methylation probes displayed a high correlation between all cell lines, suggesting that much like chromosomal DNA, ecDNA may also be regulated by methylation (Supplementary Fig. S2G). A similar analysis for MYC and MYCN ecDNA is summarized in Supplementary Fig. S2H and S2I, and data are summarized in Supplementary Table S3.

Together, these analyses integrate WGS, long-range optical mapping, and DNA FISH to resolve SCLC ecDNA structures. Beyond serving as vehicles for oncogene amplifications, ecDNA enables the juxtaposition of enhancers proximal to SCLC oncogenes regulating their expression, with distinct transcriptional consequences based on genes involved.

ecDNAs Are a Major Source of SCLC Oncogenic Fusions

In addition to driving high levels of oncogene transcription, ecDNAs contain significantly more gene fusions compared with noncircular amplicons (28). Seeking evidence of genomic fusions at SCLC ecDNA loci, we first examined the gene fusion landscape of SCLC cell lines. We identified 5 recurrent fusions (Supplementary Fig. S3A) involving 12 genes (Supplementary Fig. S3B) in which the number of junction-supporting reads was above 100× (Supplementary Table S4). Among them was RLFMYCL (n = 4 cell lines), the most frequent oncogenic fusion event in SCLC primary tumors (46, 47). To further characterize the RLFMYCL fusion, we compared the expression of the first two exons of the RLF gene, previously described as boundaries of the fusion. Five cell lines had high RLF exon 1 expression paired with an elevated expression ratio of exon 1 to exon 2 (Fig. 3A; Supplementary Table S4), of which four had detectable rearrangements between RLF and MYCL in the transcriptome (Fig. 3B). In the fifth cell line (HCC33), RLF was part of the ecDNA without fusions to MYCL. For further analysis, we focused on three RLFMYCL fusion–positive cell lines with available WGS data (NCI-H889, NCI-H1092, and NCI-H1963). All three cell lines had several MYCL-positive ecDNAs (median of 21, 28, and 45 ecDNAs/cell in NCI-H1092, NCI-H1963, and NCI-H889, respectively) based on MYCL metaphase spread FISH (Fig. 2C; Supplementary Fig. S2C).

Figure 3.

Hijacking the promoter of RLF through the RLF–MYCL fusion. A, Gene expression analysis of RLF exon 1 and RLF exon 1/exon 2 ratios identifies 5 cells with elevated RLF exon 1 expression. Cell lines marked in red are positive for RLF–MYCL fusion. TPM, transcript per million. B, Summary of read counts that splice from RLF exon 1 to either MYCL exon 2 (red) or MYCL exon 3 (blue) based on RNA sequencing (RNA-seq) data. C, Differential RLF and MYCL exon expression in three fusion-positive cell lines (red) and three fusion-negative cell lines. D, Rearrangement patterns at the RLF-MYCL locus in three fusion-positive cell lines. NCI-H1963 has rearrangements within the ecDNA sequence. NCI-H1092 and NCI-H889 both have amplifications at the RLF exon 1 that rearranges upstream of MYCL. E, RLF–MYCL fusion representation based on rearrangements on ecDNA. The number of supporting reads is shown in purple. F, Validation of the RLF–MYCL fusion using FISH. The RLF probe is colored in red, and the MYCL probe in green. G, Validation of the RLF–MYCL fusion RNA using qPCR. H, RLF–MYCL is translated to protein identified using Western blot. Cell lines expressing RLF–MYCL (based on qPCR) and MYCL (based on RNA-seq) are marked with “+” on top of the graph, and negative cells are marked as “−.”

Figure 3.

Hijacking the promoter of RLF through the RLF–MYCL fusion. A, Gene expression analysis of RLF exon 1 and RLF exon 1/exon 2 ratios identifies 5 cells with elevated RLF exon 1 expression. Cell lines marked in red are positive for RLF–MYCL fusion. TPM, transcript per million. B, Summary of read counts that splice from RLF exon 1 to either MYCL exon 2 (red) or MYCL exon 3 (blue) based on RNA sequencing (RNA-seq) data. C, Differential RLF and MYCL exon expression in three fusion-positive cell lines (red) and three fusion-negative cell lines. D, Rearrangement patterns at the RLF-MYCL locus in three fusion-positive cell lines. NCI-H1963 has rearrangements within the ecDNA sequence. NCI-H1092 and NCI-H889 both have amplifications at the RLF exon 1 that rearranges upstream of MYCL. E, RLF–MYCL fusion representation based on rearrangements on ecDNA. The number of supporting reads is shown in purple. F, Validation of the RLF–MYCL fusion using FISH. The RLF probe is colored in red, and the MYCL probe in green. G, Validation of the RLF–MYCL fusion RNA using qPCR. H, RLF–MYCL is translated to protein identified using Western blot. Cell lines expressing RLF–MYCL (based on qPCR) and MYCL (based on RNA-seq) are marked with “+” on top of the graph, and negative cells are marked as “−.”

Close modal

Transcriptome sequencing tracings of the RLF exons further demonstrated the magnitude of differential expression of the first two exons in the ecDNA-positive cells (Fig. 3C; Supplementary Fig. S3C), ranging from 1.5-fold (NCI-H1963) to nearly 20- and 100-fold (NCI-1092 and NCI-H889). The three cell lines had diverse ecDNA reconstructions based on WGS data. In the case of NCI-H1963, the entire RLF gene was on the ecDNA, with multiple rearrangements within the ecDNA between the RLF first intron and upstream of the MYCL promoter (Fig. 3D; Supplementary Fig. S3D). NCI-H889 and NCI-H1092 had small focal amplifications at the promoter of RLF, where the breakpoint site within the first intron of RLF was assembled to a focal amplification break site slightly upstream of MYCL (Fig. 3D). Even though the RLF and MYCL genes are ∼259 kb apart and on different strands of the genome, on the ecDNAs, the RLF promoter and first exon strand “switched” orientation and fused a few kilobases upstream of the MYCL promoter (Fig. 3E), enabling RNA to splice between the two genes bringing exon 1 of RLF in frame with MYCL1 exon 2 or 3.

The fusion was detectable on long-read optical mapping in NCI-H889 and NCI-H1092 (Fig. 2B; Supplementary Fig. S3E) but was not observed for NCI-H1963, likely due to the high complexity of the rearrangement. Colocalization of RLF and MYCL was confirmed by imaging of metaphase nuclei by FISH using targeted probes for the RLF promoter area and MYCL, with extrachromosomal colocalization of probes seen in all three cell lines (Fig. 3F; Supplementary Fig. S3F). We also confirmed the presence of in-frame RLFMYCL fusion transcript through RNA qPCR. Splicing between RLF exon 1 and MYCL exon 2 or 3 was over 1,000× higher than between RLF exon 1 and RLF exon 2 (Fig. 3G; Supplementary Fig. S3G; Supplementary Table S4). On Western blot, the RLF–MYCL fusion protein expression was substantially higher than endogenous MYCL expression of fusion-positive cell lines (NCI-H889, NCI-H1092, and NCI-H1963) and cell lines with chromosomal MYCL amplification (DMS-273 and CORL88; Fig. 3H). Presence of the fusion explains why the MYCL enhancer is weaker than the RLF enhancer in the NCI-H889 (Fig. 2D), because the fusion transcript serves as the main target for MYCL expression. In murine models, RLFMYCL1 fusion accelerates SCLC tumorigenesis while also altering the nature of cancer progression, enhancing the frequency and diversity of metastases (48). Our studies reveal ecDNA as a structural basis for RLFMYCL fusion in SCLC.

ecDNA and Plasticity of SCLC NE Cell States

To understand the genetic diversity afforded by ecDNA and its role in SCLC plasticity, we focused first on DMS-273, which was predicted to contain two distinct circular amplicons by WGS: an MYC amplicon derived from chromosome 8 and an MYCL amplicon derived from chromosome 1. Two-color DNA FISH confirmed the coincident occurrence of numerous MYC and several MYCL ecDNAs in each cell (Fig. 4A, quantified in Supplementary Fig. S4A) with an overall positive correlation between MYC and MYCL ecDNA/cell. Based on ecDNA reconstructions, the MYCL ecDNA was assembled from two main segments, whereas the MYC ecDNA had a more complex pattern (Fig. 4B), and a broader range of sequencing depths (Supplementary Fig. S4B). To characterize the genetic diversity that defines cell populations with distinct ecDNA types, we performed single-cell DNA sequencing of recurrently amplified SCLC genomic loci using primers containing unique barcodes against 196 amplicons spanning 12 genes and several ecDNA areas (Supplementary Table S5). MYC and MYCL ecDNA spanning amplicons had significantly higher sequencing depths in all cells compared with nonamplified genomic regions (Supplementary Fig. S4C), further validating the nonuniform amplification status of the MYC-positive ecDNA (Supplementary Fig. S4D). Surprisingly, the MYC and MYCL copy numbers in the individual cells ranged over an order of magnitude (5.1–59.7 and 1.9–14.9 normalized depth range between upper and lower 10% for MYC and MYCL, respectively: normalized data in Supplementary Table S5), suggesting large variation of copy-number (CN) profiles of ecDNA. Due to the high level of heterogeneity, in some cell clusters, MYC and MYCL amplicons had comparable levels (Fig. 4C; Supplementary Fig. S4E). Comparable MYC- and MYCL-positive ecDNA counts/cell from FISH analysis but with higher overall MYC copy numbers based on sequencing suggest that there may be several copies of MYC on the ecDNAs. Even though the MYC and MYCL oncogenes are part of distinct ecDNAs, there was still an overall strong correlation between the sequencing depths (Spearman correlation mean ∼0.7 ranging from 0.4 to 0.99) of MYC and MYCL targeting amplicons, suggesting that some cells had overall more or fewer copies of both types of ecDNAs (Supplementary Fig. S4F). Previous studies in other cancer types have described intermolecular regulation between ecDNAs involving distinct oncogenes, providing a selective advantage for cells carrying multiple ecDNAs (49). In contrast, our enhancer interaction analysis using bulk H3K27ac Hi-ChIP revealed little to no intermolecular contacts between MYCL and MYC ecDNAs in DMS-273 (Fig. 4D; Supplementary Fig. S4G; Supplementary Table S6).

Figure 4.

Heterogeneity of SCLC and ecDNA. A, Identification of MYC- and MYCL-positive ecDNA in the DMS-273 cell line using FISH. B, Assembly of MYC and MYCL ecDNAs in DMS-273 from WGS data. Gray bars show normalized sequencing depth, and red arches represent assembly paths. C, Single-cell copy-number (CN) analysis of targeted amplicons shows differential levels of MYC and MYCL copy numbers based on normalized read depths. The heat map shows normalized sequencing depths for five amplicons covering MYCL and six amplicons covering MYC in 2,000 cells. scCNV, single-cell copy-number variation. D, Enhancer–enhancer interaction analysis using Hi-ChIP data reveals little to no interaction between the MYC and MYCL ecDNAs (blue box) in the DMS-273 cell line. E, Selection of suspension (NCI-H524S) and adherent (NCI-H524A) cells from the parental NCI-H524 cell line. F, Microscope images of suspension and adherent NCI-H524 cells. G, Western blot of MYC, NE, and non-NE genes shows non-NE differentiation in the adherent NCI-H524 cell line compared with the suspension counterpart. H, MYC FISH analysis identified ecDNA in both suspension and adherent NCI-H524 cells. Quantification of MYC-positive ecDNA foci is summarized at bottom. Statistical comparison was performed using the Wilcoxon rank sum test. I, Differential amplification landscape at the MYCL locus in the adherent and suspension NCI-H524 cells. Top track shows the H3K27ac Hi-ChIP signal for the parental cell line. Lower tracks show the genomic DNA sequencing for the suspension (H524S) and adherent (H524A) cells. J, The PVT1 promoter adjacent to MYC in the parental and suspension cells is lost from the ecDNA population of adherent H524A cells. Top track shows the H3K27ac Hi-ChIP signal from the parental NCI-H524 cell line. Lower tracks show the genomic DNA sequencing for the suspension (H524S) and adherent (H524A) cells, respectively.

Figure 4.

Heterogeneity of SCLC and ecDNA. A, Identification of MYC- and MYCL-positive ecDNA in the DMS-273 cell line using FISH. B, Assembly of MYC and MYCL ecDNAs in DMS-273 from WGS data. Gray bars show normalized sequencing depth, and red arches represent assembly paths. C, Single-cell copy-number (CN) analysis of targeted amplicons shows differential levels of MYC and MYCL copy numbers based on normalized read depths. The heat map shows normalized sequencing depths for five amplicons covering MYCL and six amplicons covering MYC in 2,000 cells. scCNV, single-cell copy-number variation. D, Enhancer–enhancer interaction analysis using Hi-ChIP data reveals little to no interaction between the MYC and MYCL ecDNAs (blue box) in the DMS-273 cell line. E, Selection of suspension (NCI-H524S) and adherent (NCI-H524A) cells from the parental NCI-H524 cell line. F, Microscope images of suspension and adherent NCI-H524 cells. G, Western blot of MYC, NE, and non-NE genes shows non-NE differentiation in the adherent NCI-H524 cell line compared with the suspension counterpart. H, MYC FISH analysis identified ecDNA in both suspension and adherent NCI-H524 cells. Quantification of MYC-positive ecDNA foci is summarized at bottom. Statistical comparison was performed using the Wilcoxon rank sum test. I, Differential amplification landscape at the MYCL locus in the adherent and suspension NCI-H524 cells. Top track shows the H3K27ac Hi-ChIP signal for the parental cell line. Lower tracks show the genomic DNA sequencing for the suspension (H524S) and adherent (H524A) cells. J, The PVT1 promoter adjacent to MYC in the parental and suspension cells is lost from the ecDNA population of adherent H524A cells. Top track shows the H3K27ac Hi-ChIP signal from the parental NCI-H524 cell line. Lower tracks show the genomic DNA sequencing for the suspension (H524S) and adherent (H524A) cells, respectively.

Close modal

Next, we sought to examine the role of ecDNA in SCLC cell states and plasticity. SCLC cells with NE features tend to remain in suspension and grow as cell aggregates. By contrast, non-NE cells show mesenchymal or spindle morphology and exhibit varying degrees of adherence (4). Some SCLC cell lines such as NCI-H524 have both suspension and adherent cell populations (Fig. 4E). We isolated subpopulations of adherent (H524A) and suspension (H524S) cells from the NCI-H524 cell line and studied them in relation to one another (Fig. 4F). Suspension cells exhibited high expression of NE proteins and MYC, potentially representing a transition between NE and non-NE states (19), whereas the adherent cells showed high YAP1 expression (Fig. 4G). We observed similar MYC protein expression differences when comparing suspension and adherent cells from the DMS-273 and NCI-H446 cell lines (Supplementary Fig. S4H and S4I). Intriguingly, despite differential MYC protein expression, both the suspension and adherent cells harbored MYC ecDNAs in similar numbers by FISH (P = 0.087, Wilcoxon rank sum test; median 20 and 24 for H524A and H524S, respectively; Fig. 4H; Supplementary Fig. S4J). However, adherent cells demonstrated notably diminished MYC copy numbers compared with suspension cells (Fig. 4I; ∼1.7× fold higher sequencing depth in suspension cells using linear regression analysis; Supplementary Fig. S4K), suggesting multiple MYC copies in ecDNA of suspension cells.

In addition, the adherent cells also exhibited loss of ecDNA regions that encompass the PVT1 promoter, which has an active enhancer signal in the parental cell line that interacts with the MYC enhancer (Fig. 4J). The two enhancers have strong interactions in both NCI-H524 and NCI-H446 (Supplementary Fig. S4L), uncovering a potential regulatory relationship. Although the PVT1 promoter is included in many MYC-positive ecDNAs (5/8 cell lines and 1/3 tumor samples), the amplifications only partially cover PVT1, which may result in an inactive (truncated) transcript (Supplementary Fig. S4M). Taken together with previous studies that have demonstrated the PVT1 promoter, located downstream of MYC on chromosome 8q24.21 region, as a key regulator of MYC expression (50), these data suggest a possible role of ecDNA as regulatory elements in remodeling SCLC plasticity between the suspension NE and the adherent non-NE cell states.

Plasma Detection of ecDNA and Tumor Heterogeneity in Patients

ecDNA has tens to hundred more copies compared with regular chromosomes. We sought to determine whether ecDNA was detectable using ChIP-seq of the H3K4me3 histone modification from plasma cell-free nucleosomes (cfChIP-seq), an approach described in more detail in a recent study (bioRxiv 2022.06.24.497386). Among 82 patients with extensive stage SCLC, we detected elevated MYC and MYCL enhancer signals in plasma cfChIP-seq from 6.1% (n = 5) of patients with SCLC (summarized in Supplementary Table S7). All five cases also demonstrated ecDNA-like signal traces on the sequencing depth profile outside of the promoter H3K4me3 histone modification signal. In all five cases, we found remarkably concordant ecDNA-like amplification patterns and assemblies between the tumor WGS and plasma (Fig. 5A and B; Supplementary Fig. S5A). Overall, these data underscore the potential utility of cfDNA-based sequencing methods to provide noninvasive detection of tumor ecDNA and HSR amplifications.

Figure 5.

Identification of ecDNA using cfChIP-seq and ecDNA heterogeneity across metastatic sites. A, Overview of the cfChIP-seq experiment and establishment of patient-derived cell lines from metastatic sites. B, Examples of ecDNA-like focal amplifications are detectable using cfChIP-seq from patient plasma samples, validated with tumor genome sequencing. C, Comparison of normalized sequencing depth of cell lines and tumor sample (top) and ecDNA assembly, and H3K4me3 signal derived from the cfChIP-seq from plasma samples. The top track shows normalized sequencing depths of all cell lines and tumor WGS samples, with assembly arches shown in purple. The middle track shows the H3K4me3 cfChIP signal from the patient. The lower track shows the background sequencing depth from the cfChIP-seq sample. D, Differential CN validation of the MYC locus in the cell lines. LN, lymph node. E, Differential ecDNA and HSR status in the cell lines identified by FISH probing of the MYC gene. Scale bars in the bottom right corner represent 5 μm. F, Number of ecDNA- or HSR-positive cells in the patient-derived cell lines. G, MYC CN analysis using single-cell CN analysis demonstrated higher MYC CN heterogeneity in the ecDNA-positive cell line compared with the HSR-positive cell line. ***, Wilcoxon test P values below < 1e-16. H, Enhancer–enhancer interactions at the MYC–PVT1 locus. Highest interactions were observed in the lung tumor–derived cell line. I, Quantified enhancer signal at the MYC and PVT1 promoters using the TMM (trimmed mean of M values)-normalized FPKM (fragments per kilobase of exon per million mapped) values. J, ASCL1 and NEUROD1 protein expression levels in the patient-derived cell lines.

Figure 5.

Identification of ecDNA using cfChIP-seq and ecDNA heterogeneity across metastatic sites. A, Overview of the cfChIP-seq experiment and establishment of patient-derived cell lines from metastatic sites. B, Examples of ecDNA-like focal amplifications are detectable using cfChIP-seq from patient plasma samples, validated with tumor genome sequencing. C, Comparison of normalized sequencing depth of cell lines and tumor sample (top) and ecDNA assembly, and H3K4me3 signal derived from the cfChIP-seq from plasma samples. The top track shows normalized sequencing depths of all cell lines and tumor WGS samples, with assembly arches shown in purple. The middle track shows the H3K4me3 cfChIP signal from the patient. The lower track shows the background sequencing depth from the cfChIP-seq sample. D, Differential CN validation of the MYC locus in the cell lines. LN, lymph node. E, Differential ecDNA and HSR status in the cell lines identified by FISH probing of the MYC gene. Scale bars in the bottom right corner represent 5 μm. F, Number of ecDNA- or HSR-positive cells in the patient-derived cell lines. G, MYC CN analysis using single-cell CN analysis demonstrated higher MYC CN heterogeneity in the ecDNA-positive cell line compared with the HSR-positive cell line. ***, Wilcoxon test P values below < 1e-16. H, Enhancer–enhancer interactions at the MYC–PVT1 locus. Highest interactions were observed in the lung tumor–derived cell line. I, Quantified enhancer signal at the MYC and PVT1 promoters using the TMM (trimmed mean of M values)-normalized FPKM (fragments per kilobase of exon per million mapped) values. J, ASCL1 and NEUROD1 protein expression levels in the patient-derived cell lines.

Close modal

To understand the impact of ecDNA-mediated oncogene amplification on tumor heterogeneity, we profiled metastatic sites and tumor-derived cell lines obtained during rapid autopsy from a patient with SCLC whose plasma cfChIP-seq showed MYC ecDNA amplification. The patient was a 65-year-old man with a 39-pack-year smoking history who was diagnosed with extensive-stage SCLC involving the upper lobe of the right lung, mediastinal and right hilar lymph nodes. He was treated with a combination of cisplatin and etoposide for two cycles, but the disease was refractory to treatment, with the appearance of new chest wall and adrenal tumors after only two cycles. The patient underwent palliative radiotherapy to the chest wall tumor and then received immune-checkpoint inhibitor combination of ipilimumab and nivolumab, followed by two sequential investigational therapeutic combinations that produced no tumor response. The patient died of progressive disease 11 months after diagnosis (Supplementary Fig. S5B). A rapid autopsy was performed 2 hours postmortem. Cell lines generated from cervical lymph node, liver, lung, and adrenal as well as the parent tumors were profiled using WGS, transcriptome, long-read optical mapping, and H3K27ac Hi-ChIP.

Cell lines obtained at rapid autopsy had similar amplification patterns at the MYC locus compared with the parent lung tumor sample, but the cell lines had variable MYC copy-number states between them (Fig. 5C). For example, normalized MYC sequencing depth varied between 15× in the case of the liver and cervical lymph node to 45× for the adrenal gland–derived cell line. MYC copy number was validated using qPCR, with the adrenal- and lung-derived cell lines exhibiting the highest MYC levels (Fig. 5D) in agreement with WGS data (Supplementary Fig. S5C; Supplementary Table S8). Next, we probed MYC in metaphase nuclei by FISH to find marked heterogeneity of MYC ecDNA and HSRs across the metastatic sites. Specifically, lung- and adrenal-derived cells were predominantly MYC ecDNA-positive whereas cervical lymph node– and liver-derived cells exhibited MYC HSRs (Fig. 5E and F; Supplementary Table S8). These cell lines were an optimal model to examine the extent of uneven segregation of ecDNA during mitosis. Single-cell CN analysis of amplicons targeting MYC demonstrated that the ecDNA-positive adrenal gland–derived cell line had a wide range of ecDNA copies. In comparison, the HSR-positive cervical lymph node–derived cell line displayed more stable levels of copies among the single cells due to the MYC locus being integrated into the chromosome (Fig. 5G; Supplementary Table S5). These analyses demonstrate the high levels of heterogeneity arising from ecDNAs both at the single-cell level and across metastatic sites from a single patient.

Next, we sought to understand the regulatory basis of ecDNA heterogeneity across the metastatic sites. Long-read optical mapping demonstrated multiple possible assemblies in all cell lines, with some common events: full-length MYC gene with a partial PVT1 promoter downstream of MYC [without any gene fusion based on RNA sequencing (RNA-seq)] and a more distal localization of part of the PVT1 intronic segment (Supplementary Fig. S5D), consistent with the rearrangement patterns observed in the WGS data. To understand extrachromosomal PVT1MYC coregulation, we performed H3K27ac Hi-ChIP. Enhancer connectome analysis revealed high PVT1–MYC promoter contact frequency in 3D within a small but dense topologically associated domain (Fig. 5H), but with differential interaction strengths across the cell lines. The strongest enhancer signals were observed in the ecDNA-positive lung- and adrenal-derived cell lines, which were more than 2- to 3-fold greater than HSR-positive cervical lymph node and liver-derived cells (Fig. 5I). MYC protein expression tracked the MYC–PVT1 enhancer interaction strength (Fig. 5H), rather than the MYC enhancer strength itself (Fig. 5I), with the lung tumor–derived cell line having the highest MYC expression (Fig. 5J; Supplementary Fig. S5E).

Finally, we examined the association of ecDNA MYC amplification on SCLC phenotypes. All four cell lines had positive neuroendocrine scores, calculated using a 50-gene-set signature (4), consistent with the elevated expression of NE-specific lineage transcription factors ASCL1 and NEUROD1 (Supplementary Fig. S5F and S5G). However, immunostaining and Western blot experiments revealed marked heterogeneity of ASCL1 expression across cell lines from the different metastatic sites. The high-MYC-expressing lung tumor–derived cell line also notably expressed NEUROD1, potentially representing an MYC-driven cell-state transition between ASCL1 and NEUROD1 (Supplementary Fig. S5H–S5J).

Although these data warrant functional studies modulating MYC ecDNA and HSR, the heterogeneity of NE cell states associated with distinct MYC amplification patterns is consistent with the current understanding of SCLC plasticity, with MYC being sufficient to promote temporal evolution of SCLC from an ASCL1-high NE cell state to a NEUROD1-high less-NE cell state (19), and suggests that specific mechanisms of MYC amplification, ecDNA or HSR, and its regulatory interactions with distal enhancers on ecDNA may underlie the dynamic evolution of cell states between metastatic sites.

Phenotypic plasticity endows tumor cells with the capacity to convert between distinct cell identities—allowing acquisition of metastatic capacity, drug resistance, and cancer progression (51, 52). Although SCLC heterogeneity and plasticity are correlated with, and likely drive therapy resistance, the genetic mechanisms underlying tumor heterogeneity and phenotypic switching between cell states are still largely unknown (53). In this study, we provide the first detailed analysis of ecDNA in SCLC patient and cell line samples. We find that ecDNAs are a major source of high focal amplifications in key SCLC oncogenes and drivers of fusions. Beyond being vehicles of oncogene amplification and fusion, ecDNAs bring distal DNA elements to close proximity, allowing chromatin contacts between enhancers and oncogenes that cannot normally occur in chromosomal DNA, thus functioning as SCLC transcription amplifying units. The transcriptional flexibility afforded by ecDNAs may underlie the rapid expansion of SCLC intratumor heterogeneity, treatment resistance, and metastases and the significantly worse survival outcomes of SCLCs harboring complex ecDNA amplifications.

Our study introduces ecDNA as a major contributor to SCLC heterogeneity (Fig. 6). We observe distinct mechanisms of MYC amplification across multiple metastatic tumors from the same patient, with some metastatic sites harboring MYC ecDNA and others MYC HSR. Tumor cells with ecDNA-amplified MYC exhibit marked heterogeneity of MYC copy states compared with MYC HSR–harboring cells, further validating previous observations that ecDNAs are unevenly segregated across the daughter cells during mitosis (54), leading to a rapid increase of genomic heterogeneity (22, 26). Our results also reveal ecDNA as a structural basis of PVT1 amplifications (55) and suggest the PVT1 enhancer as a key regulator of MYC expression (56, 57). Previous studies have shown that PVT1 inhibits MYC by competing promoter–enhancer interactions during development (58). Cancer cells may escape this inhibitory regulation by depleting PVT1 through the amplification of MYC or fusion of PVT1 promoter with MYC (49). In SCLC, we did not identify PVT1–MYC fusions but rather MYC amplifications that occasionally harbored the PVT1 promoter area, which leads to truncated transcripts of PVT1. MYC protein expression levels and enhancer–enhancer interaction analyses revealed that the strength of MYC–PVT1 interactions was more predictive of MYC protein expression levels than simply the MYC enhancer signal strength. We find that the RLFMYCL fusion, the most common gene rearrangement in human SCLC, can be a consequence of ecDNA assemblies. MYCL oncogene hijacks the RLF promoter through ecDNA-driven structural rearrangements creating a fusion protein, further elevating MYCL expression and possibly also altering the nature of MYCL signaling. Together, these findings suggest a model wherein oncogene amplifications and fusions on ecDNA shape SCLC heterogeneity and plasticity by hijacking regulatory dependencies in the noncoding genome.

Figure 6.

Summary of the study.

Figure 6.

Summary of the study.

Close modal

Our findings raise a number of fundamental issues for follow-up investigation, including the biology of ecDNA formation in SCLC, the determinants of transitions between ecDNA and HSRs including the impact of drug treatments, and therapeutic approaches to overcome the resulting plasticity. The impact of ecDNA on other SCLC phenotypes, including its metastatic proclivity, also warrants further study. We find that the DMS-273 cell line, known for its remarkable ability to metastasize (59), harbored both MYC and MYCL amplifications on ecDNA. We recently found that ecDNA-amplified SCLC cell lines have elevated levels of phosphorylated histone H2AX (60), suggesting high levels of intrinsic DNA damage, instability, and replication stress (6, 60, 61). Recent studies have demonstrated that DNA circularity confers highly potent immunostimulant activity for ecDNA (62). Analysis of ecDNA regulatory sequences would be valuable for the design of enhancer-based SCLC treatment strategies.

When considering the clinical impact of ecDNA, it is important to detect and stratify ecDNA-driven tumors. To do so, we provide proof-of-concept of a cell-free DNA (cfDNA) based noninvasive approach, leveraging ChIP-seq of histone modifications, to decipher ecDNA and HSR amplifications genome-wide from patient plasma samples. The lower frequency of ecDNA+ cases detected from plasma (6.1%) in our cohort of metastatic and relapsed patients compared with SCLC primary tumors (19.9%) may be explained by the overall worse survival of ecDNA-positive patients. Although our approach awaits further validation in independent datasets, detection of ecDNA and HSRs from plasma opens a wide range of diagnostic opportunities in SCLC and other cancers.

While introducing MYC ecDNA as a source of SCLC heterogeneity, the current study has not established the causality between the ecDNA and HSR MYC amplifications and transcriptional subtype states. However, recent studies demonstrate conclusively that MYC is sufficient to promote the conversion of SCLC sequentially from an ASCL1 to a NEUROD1 to a YAP1 state from an NE cell of origin (19–21). Further studies are also needed to better understand how transitions between ecDNA and HSRs are affected by drug treatments, and for developing therapeutic approaches to overcome the resulting plasticity. Such studies will be challenging given the need to manipulate ecDNA copy numbers and regulatory sequences precisely in the intact nuclei. We also recognize that gene mutations and amplifications are but one mechanism that underlies the biologically distinct cellular states and phenotypic plasticity of SCLC. Epigenetic regulators such as the histone methyltransferase EZH2 have been implicated in controlling NE cell fate and driving chemoresistance (63). Increasing evidence also supports the key role of the tumor microenvironment in shaping cell SCLC states (2). Further studies are needed to decipher the interplay between genetic mechanisms revealed here and the nongenetic mechanisms that together make SCLC a shapeshifting tumor.

Analysis of Published WGS Data

WGS data were obtained from the Sequence Read Archive (SRA) for the Cancer Cell Line Encyclopedia (CCLE) cell lines (PRJNA523380) and through the European Genome-phenome Archive (EGA) for the patient tumor samples (EGAS00001000925). The data of the tumor WGS samples were aligned to the hg19 genome version. The CCLE cell line reads were aligned to the hg19 human genome using the BWA–MEM aligner (arXiv:1303.3997v2). Alignments in SAM format were converted to BAM format, followed by sorting using the samtools package (64). Duplicate reads were marked using the picard-tools package (https://broadinstitute.github.io/picard/). Normalized sequencing depth tracks were generated using the BAMscale program (65), setting the bin size to 100 bp. Normalized gene signal was calculated from the normalized bigwig files using the rtracklayer (66) and GenomicRanges (67) package, where the mean normalized sequencing depth of each gene was calculated using the TxDb.Hsapiens.UCSC.hg19.knownGene gene annotation summarized by the genes() function in R.

ecDNA was assembled from the WGS data using the AmpliconArchitect (37) program. For each sample, reads were down-sampled to 20% of the original size to reduce run time due to the high sequencing depths. Focal amplified regions were identified from the normalized bigwig files using the rtracklayer and GenomicRanges packages in R by selecting regions with over five normalized signals, and merging segments within 10 kb distance. These regions were used as putative intervals during the ecDNA reconstruction with AmpliconArchitect. Assembled ecDNAs were classified as circular or BFBs using the AmpliconClassifier (28) tool. Circular ecDNAs were manually curated and categorized as complex if the assembly derived from multiple genomic focal amplifications, whereas single-locus ecDNAs were categorized as simple ecDNAs. Survival analysis was performed using the manually curated results from the analysis using the Survival R package.

MYC, MYCL, and MYCN FISH

Metaphase chromosomes were prepared by incubating cells for 2 hours with 0.02 mg/mL Colcemid (Invitrogen). The harvested cells were then incubated in hypotonic solution (0.075 mol/L KCl), followed by methanol/acetic acid (3:1) fixation. Metaphases were then prepared in a controlled humidity chamber (Thermotron) before being aged at 37°C before hybridization. FISH probes were purchased from Empire Genomics. We followed our standard FISH protocol (https://ccr.cancer.gov/staff-directory/thomas-ried under Resources) for hybridization and detection of the probes. The hybridized slide was imaged using the Leica Thunder Imaging microscope (Leica) equipped with custom optical filters and a 60× objective. All slides were counterstained with 4′,6-diamidino-2-phenylindole.

Hi-ChIP Sample Preparation and Analysis

H3k27ac Hi-ChiP was performed following the previously published Hi-ChIP (68). Briefly, cells grown at 70% confluency were detached, cross-linked with formaldehyde for 10 minutes, and quenched with 125 mmol/L glycine. Here, cells were lysed in preparation for in situ contact generation. Isolated nuclei were permeabilized, and restriction digestion was carried out for 2 hours at 37°C with MboI (New England Biolabs). Restriction sites were filled with dNTPs using Biotin-14-dATP (Jena Biosciences) for 1 hour at 37°C. The filled ends were then ligated together using a T4 ligase at room temperature for 4 hours before nuclei were lysed and sonicated (using Covaris), and then H3k27ac (Abcam, ab4729) immunoprecipitation was carried out overnight using antibodies. The morning after, 30 μL of beads was added to collect the chromatin–antibody complex; then the ChIP DNA was collected and washed, and cross-links were reversed overnight using proteinase K. ChIP DNA was eluted, and samples were purified using the DNA Cleanup Kit. The DNA was quantified using a Qubit before biotin ligation junction capture using Streptavidin C-1 beads. Samples were washed and taken forward for Tn5 Tagmentation. Tagmentation and PCR amplification were performed as described (68). Libraries were size selected to 200 to 700 bp and sequenced on the HiSeq using 2  ×  150 bp.

Sequenced reads were processed using the HicPro (69) pipeline (v3.1.0), aligning the data to the hg19 genome version and MboI motif for the analysis. The resulting valid read pairs were converted to a contact matrix “.hic” format with the hicpro2juicebox.sh tool from the Juicebox (70) suite. Contact quantifications from the resulting “.hic” were calculated using the straw R functions from the Juicebox suite. Hi-ChIP contact matrices and genome tracks were plotted using the gtrack package (https://github.com/mskilab/gTrack) in R. Published Hi-ChIP data [ref. 71; Gene Expression Omnibus (GEO) accession GSE151001] that served as validation were processed with the same pipeline.

Analysis of Published ChIP-seq and ATAC-seq Data

ChIP-seq for the enhancer marker H3K27ac was obtained from previous studies (refs. 41, 71; GEO accessions GSE151001 and GSE115123). ATAC-seq data were obtained from a previous study (ref. 21; GEO accession GSE156334). Raw data were trimmed using trimmomatic (72), followed by alignment to the hg19 genome using the BWA ME algorithm (arXiv:1303.3997v2). The BAM files were sorted with samtools (64), followed by duplicate marking using picard-tools. Normalized sequencing depth tracks were created using BAMscale (65). Peaks were called using MACS (73) peak caller, setting the FDR value to 0.05.

DNA Methylation Analysis (850K EPIC Array + RRBS)

DNA methylation for SCLC cell lines was obtained for the 850K EPIC array and RRBS platforms (33).

The 850K EPIC array data for the SCLC cell lines were processed using the minfi (74) package. Normalized values were calculated using the preprocessIllumina() function, from which the signal depth was calculated as the sum intensity of the methylated and unmethylated signal for each probe. Probes were mapped to the genome using the mapToGenome() function.

Methylation probe beta values were calculated using the getBeta() function. Methylation beta values and sum intensities were exported to bigwig format using the rtracklayer (66) package, using the BSgenome.Hsapiens.UCSC.hg19 package as reference for sequence lengths.

RRBS data were aligned to the hg19 genome using the Bismark (75) aligner (v0.22.3). Bisulfite-converted sequencing reads mapped to the genome were converted to methylation states using the bismark_methylation_extractor tool from the Bismark tools. The resulting coverage files were filtered to include sites with at least 20× coverage, filtering for sites either on C or G nucleotides and exporting to bedGraph format. The resulting bedGraph format files were converted to bigwig using the rtracklayer package, using the BSgenome.Hsapiens.UCSC.hg19 package as reference for sequence lengths.

RNA-seq Fusion Analysis and Expression Heat Map

The transcriptome data for the CCLE SCLC cell lines were obtained from the SRA under accession PRJNA523380. Fusions were identified using the STARfusion tool (bioRxiv120295), analyzing sequenced reads against the prebuild reference (GRCh37_gencode_v19_CTAT_lib_Oct012019.plug-n-play). Gene expression levels were quantified using the TPMcalculator (76) tool, combining unique exon reads for each gene, followed by normalization with DESeq2 (77), to obtain scaling factors for each sample. Normalized sequencing depth tracks were created using BAMscale, using the scaling factors determined by DESeq2.

The heat map demonstrating the expression of NAPY genes (ASCL1, NEUROD1, POU2F3, and YAP1) and MYC paralogs (MYC, MYCL, and MYCN) were created using gene expression data from the SCLC-CellMinerCDB (34) in case of cell lines and cBioPortal (78) for the tumor samples. Normalized RNA-seq values were Z-scored for the two data sources, followed by merging of data for the heat map visualization. Heat maps were prepared using the ComplexHeatmap (79) package in R.

RLF–MYCL FISH Experiment

All cells were incubated under standard conditions. Metaphase cells were enriched by treating cells with Karyomax (15210-040, Gibco, Invitrogen) at a final concentration of 0.1 μg/mL. Cells were collected, washed in phosphate-buffered saline (PBS), and resuspended in 75 mmol/L KCl (6858-04, Macron Chemical) for ∼15 minutes at 37°C. Cells were fixed by the addition of an equal volume of Carnoy's fixative (3:1 methanol: glacial acetic acid, JT Baker). Cells were washed three additional times in Carnoy's fixative and dropped onto humidified glass slides.

Using whole-genome sequencing data, probes were designed using the UCSC database using both hg19 and hg38 because the probes used in this experiment were custom builds (Empire Genomics). Build 1 was used for line NCI-H1092, chr1: 40,351,718–40,377,194 MYCL, chr1: 40,611,000–40,640,000 RFL exon 1 area and chr1: 41,136,523–41,805,548 for control region. Build 2 was used for lines NCI-H889 and NCI-H1963, chr1: 40,351,718–40,377,194 MYCL, chr1: 40,596,190–40,634,743 RLF exon2, chr1: 41,136,523–41,805,548 for control region. MYCL was labeled in green-dUTP (5-Fluorescein), RFL was labeled with red-dUTP (5-Carboxy-x-rhodamine), and the control was labeled gold-dUTP (carboxyrhodamine 6G).

Metaphase spreads were equilibrated in 2× SSC (30 mmol/L sodium citrate, 300 mmol/L NaCl, pH 7) for ∼5 minutes. They were dehydrated using successive washes of 70%, 85%, and 100% ethanol for 2 minutes each and allowed to dry. FISH probes were diluted in hybridization buffer (Empire Genomics), and 10 μL was added to metaphase spreads on slides, along with 22-mm2 coverslips. Samples were placed on ThermoBrite (Leica) for codenaturation and hybridization, and the program was set to denature at 75°C for 3 minutes and hybridize 37°C overnight in humid conditions (ThermoBrite Leica). Samples were washed successively in 0.4× SSC/0.3% Tween20 at 73°C for 2 minutes and 2× SSC with 0.1% Tween-20 at room temperature for 2 minutes. Samples were briefly rinsed with H2O. Samples were mounted with ProLong Gold Antifade with DAPI (P36935, Invitrogen, Life Technologies), #1.5 coverslips.

Fluorescent microscope images of the hybridized metaphases were acquired using a Hyper Spectral Imaging system (Applied Spectral Imaging Inc.) mounted on top of an epifluorescence microscope (Imager Z2, Zeiss). Images were analyzed using acquisition software (HiFISHv8.2.2, Genasis, Applied Spectral Imaging Inc.). An average of 5 to 25 mitoses of comparable staining intensity and quality were examined per cell line. Images were subsequently analyzed in paint3D to perform cropping and any additional adjustments to image brightness.

Cell Culture

Cell lines were cultured in RPMI 1640 (Thermo Fisher Scientific, cat. #11875-119) supplemented with 10% heat-inactivated fetal bovine serum (Gemini Bio-products, cat. #100-106-500) and 1% penicillin–streptomycin (Thermo Fisher Scientific, cat. #15140122) in a 37°C/5% CO2 humidified incubator. Cell lines were tested as negative for Mycoplasma (Lonza, cat. #LT07-418). Adherent NCI-H524 cells were selected by removing three quarters of the floating cells at regular intervals to avoid the medium from turning yellow. Adherent NCI-H524 cells were first collected by trypsinization and transferred to new flasks when ∼50% confluency was achieved. Suspension cells for NCI-H446 (adherent) and DMS273 (loosely adherent) were created by serially transferring the nonadherent cells to new flasks. Around 8 and 15 passages were necessary to obtain suspension populations for H446 and DMS273, respectively. Suspension cell cultures were maintained by transferring suspension cells only to new flasks during cell passage.

Cell Lines

NCI-H889, NCI-H524, NCI-H1092, NCI-H69, NCI-H1963, DMS-114, and NCI-H446 were obtained from ATCC (https://www.atcc.org/), whereas DMS-273 and CORL88 were obtained from Sigma-Aldrich (https://www.sigmaaldrich.com/). DMS-273, NCI-H446, NCI-H524, COR-L88, and NCI-H69 were tested by genotyping (Laragen Sequencing and Genotyping, Laragen Inc.). Cell lines were tested for Mycoplasma using a MycoAlert Mycoplasma Detection Kit (Lonza, cat. #LT07-418) following the manufacturer's instructions. Cell lines were routinely tested before freezing and after thawing. Cells were used after at least five passages from thawing.

Western Blotting

Cell pellets containing 1 million cells were harvested by centrifugation at 2,700 × g for 5 minutes at 4°C, washed twice with 1 × PBS (pH 7.4; Thermo Fisher Scientific, cat. #10010-072) at 18,000  ×  g for 5 minutes at 4°C, and resuspended in 200 μL 2 × Laemmli sample buffer (Bio-Rad, cat. #1610737) supplemented with 5% 2-mercaptoethanol (Sigma Life Sciences, cat. #M3148-100ML). The resulting total cell lysates were vortexed, boiled for 5 minutes, and then centrifuged at 11,000 ×  g for 5 minutes at 4°C. Lysate supernatants (7 μL) were separated by SDS–polyacrylamide gel electrophoresis (PAGE; Thermo Fisher Scientific, cat. #XP04205BOX) and then transferred onto PVDF membranes (Millipore-Sigma, cat. #IPVH304F0 prior to immunodetection). Protein bands from Western blot images were quantified using ImageJ software (https://imagej.nih.gov/ij/index.html).

The following primary antibodies were used: anti-ASCL1 (Santa Cruz Biotechnology, cat. #SC-374104), anti–c-MYC (Cell Signaling Technology, cat. #5605S), anti–L-MYC (Cell Signaling Technology, cat. #76266S), anti–N-MYC (Cell Signaling Technology, cat. #51705S), anti–α-tubulin (Millipore-Sigma, cat. #T9026), anti-NEUROD1 (Cell Signaling Technology, cat. #7019S), anti-YAP1 (Cell Signaling Technology, cat. #4912S), and anti-INSM1 (Santa Cruz Biotechnology, cat. #SC-271408; 1:1,000 dilutions). The following secondary antibodies were used: HRP-linked anti-mouse IgG (Cell Signaling Technology, 7076S), HRP-linked anti-rabbit IgG (Cell Signaling Technology, 7074S), Alexa 647–conjugated anti-mouse IgG (Invitrogen, cat. #A28181), and Alexa 488–conjugated anti-rabbit IgG (Invitrogen, cat. #A11034; 1:2,000 dilutions).

Immunofluorescence Analysis

Cells collected by centrifugation or after trypsinization were fixed in 2% paraformaldehyde (PFA) and washed twice in PBS; 200 μL cell solution (1.5–3 million cells/mL) was then spotted on slides by cytospin (Thermo Shandon Cytospin3, 800 RPM, 4 minutes). After cytospin, slides were dried quickly, placed in PBS for 15 minutes, and then permeabilized with prechilled ethanol 70% overnight. Slides were then washed twice with PBS (total wash of 15 minutes). For blocking and antibody incubations, PBS was replaced with PBS containing 0.5% Tween-20 and 0.1% Triton X-100 for blocking and antibody incubations (PBS-TT). Cells were blocked with 5% bovine serum albumin (BSA; Sigma-Aldrich, cat. #A9647-100G) in PBS-TT for 1 hour. After a 5-minute wash in PBS, primary antibodies were incubated for 2 hours (1:500 in 1% BSA in PBS-TT), followed by three washes with PBS 95 minutes each and 1-hour incubations with Alexa-conjugated IgG antibodies. After a 5-minute wash with PBS, samples were incubated with DAPI for 10 minutes (2 μL of a 20 μg/mL DAPI solution per 1 mL PBS). After a quick wash with PBS, samples were mounted with PI-containing mounting medium containing (Vectashield, Vector Laboratories, cat. #H-1200) and sealed with nail polish. Slides were imaged using a ZEISS LSM780 microscope.

The following primary antibodies were used: anti-ASCL1 (Santa Cruz Biotechnology, cat. #SC-374104), anti–c-MYC (Cell Signaling Technology, cat. #5605S), and anti-NEUROD1 (Cell Signaling Technology, cat. #7019S; 1:500 dilutions). The following secondary antibodies were used: Alexa 568–conjugated anti-mouse IgG (Invitrogen, cat. #A11004), and Alexa 488–conjugated anti-rabbit IgG (Invitrogen, cat. #A11034; 1:500 dilutions).

ecDNA Junction Sequencing

Genomic DNA (gDNA) samples from 5 million cells were prepared using the Qiagen DNeasy blood and tissue kit (69504). gDNA was diluted to 100 ng/μL prior to its use for PCR. Nested PCR, involving two sequential amplification reactions, was used to generate DNA fragments with ecDNA junctions. Primers used for PCR reactions were as follows: PCR1 (5′ GGATTAAGTCGTCCGCAGTC 3′ and 5′AGGAACCTGAAGCTCTATGTGA 3′) and PCR2 (5′ TGGCTGTCTGTTCTCTCCAG 3′ and 5′ CTTGGCCAAGCACAGAACTC 3′). The settings for PCR1 reaction were as follows: 1. Initial denaturation 95°C/5 minutes, 2. Denaturation 94°C/40 seconds, 3. Annealing 60°C/40 seconds, 4. Elongation 72°C 1 minute, 39 repeat for steps 2 to 4 and final extension for 10 minutes at 72°C. The reaction volume was 25 μL. The settings for PCR2 reaction were as follows: 1. Initial denaturation 95°C/5 minutes, 2. Denaturation 94°C/40 seconds, 3. Annealing 59°C/40 seconds, 4. Elongation 72°C 1 minutes 10 seconds, 39 repeat steps for 2 to 4 and final extension for 10 minutes at 72°C. The reaction volume was 50 μL. Fragment from PCR2 was isolated and purified from agarose gel using a gel extraction kit (New England Biolabs, cat. #T1020S), following the manufacturer's instructions. Sanger sequencing of the junction was performed using the primers 5′ ATGCAGAATCCAGCAGCCCC 3′ and 5′ AACAGTGATTGGCCCAGGGG 3′.

RLF–MYCL Validation Using qPCR and Western Blot

qPCR.

Cell lines were plated at 1 million cells in a 10-cm dish and collected 24 hours later. RNA was collected using the Qiagen RNeasy Mini Kit per the manufacturer's instructions. cDNA was generated using the Applied Biosystems High-Capacity cDNA Reverse Transcription Kit and a MiniAmp thermal cycler per the manufacturer's instructions. qPCR was performed using Applied Biosystems Fast SYBYR Green Master Mix and a Quant Studio 5 per the manufacturer's instructions. We utilized the following probes:

  • RLF-E2: TGGACGTGCACTGGCAAAGC

  • RLF-E1b: GTCTCCCTTTCCGTCCGCC

  • RLF-E1a: ACTACTGCCGGAGCTTCTGC

  • MYCL-E1a: GGCAGCCAGCACACACGCAC

  • MYCL-E1b: GTGCGTGTGTGCTGGCTGC

  • MYCL-E2a: CGAGTCGTAGTCCATGTCCG

  • MYCL-E2b: CAAGCGACTCGGGTAAGGAC

  • MYCL-E3a: GCTTCATGCAGGGATCCAGG

  • MYCL-E3b: CGGGCAGCATAGTTGTGCTG

Western Blot.

Cells were washed twice with PBS and then lysed with RIPA buffer containing protease inhibitor cocktail (Thermo Fisher Scientific) and micrococcal nuclease (Thermo Fisher Scientific). The resulting mixtures were incubated on ice for 30 minutes and then centrifuged for 20 minutes to get the supernatants. After adding Tris-Glycine SDS sample buffer including 5% 2-mercaptoethanol, the lysates were boiled for 10 minutes, analyzed by SDS-PAGE, and immunoblotted with RLF-L-MYC antibody (Cell Signaling Technology, cat. #76266). The immunoblotted membrane was stripped and reprobed with an antibody against tubulin as a loading control.

Long-Read Optical Mapping Data

DNA Preparation.

Ultra-high-molecular-weight (UHMW) gDNA was extracted from several human small-cell lung cancer lines with the Bionano Prep SP Blood and Cell DNA Isolation Kit (Bionano Genomics, 80030) as described in the Bionano Prep SP Frozen Cell Pellet DNA Isolation Protocol (Bionano Genomics). Briefly, between 1.3 and 2 million frozen pelleted cells were thawed in a 37°C bath and resuspended in DNA stabilizing buffer (Bionano Genomics). After that, the cells were lysed in the presence of detergents, proteinase K, and RNase A, and the UHMW gDNA was bound to a silica Nanobind Disk (Bionano Genomics), washed, and eluted. The extracted gDNA was equilibrated for 3 days at room temperature to homogenize and quantified with the Qubit dsDNA BR assay kit (Thermo Fisher Scientific, Q32850).

DNA Labeling and Data Collection.

The isolated UHMW gDNA was fluorescently labeled using the Bionano Prep DNA Labeling Kit-DLS (Bionano Genomics, 80005) according to the Prep Direct Label and Stain (DLS) Protocol (Bionano Genomics). In short, Direct Label Enzyme (DLE-1) and DL-green fluorophores were used to label 750 ng of purified gDNA at a specific sequence motif. After a cleanup of the fluorophores excess, DNA backbone was counterstained overnight before quantitation with Qubit dsDNA HS assay kit (Thermo Fisher Scientific, Q32851). Finally, the labeled UHMW gDNA molecules were loaded on a Saphyr G2.3 chip for sequential imaging across nanochannels on the Saphyr instrument (Bionano Genomics).

Data Analysis.

Genome analysis was performed using Bionano Solve-v 3.7, Access 1.7 provided by Bionano Genomics. For genome assembly and variant analysis, subsampling was performed for each sample to obtain 70 to 90× coverage of the reference genome. The Bionano De Novo Assembly pipeline was run to assembly a genome from the molecule files (*bnx) for each sample with a human GRCh38 reference provided. The Bionano assembly consensus genome map (CMAP) file was generated for each sample. CMAPs were assembled from clustered molecules that identified the same variant. Structural variants (SV) were identified based on discrepant alignment between molecules of each sample and reference sequence. CMAPs were realigned to reference genome sequence, with SV confirmed by consensus to generate the final SV call set. Finally, CN analysis was performed from the alignment of molecules and labels against GRCh38 using alignmolvref. The molecule label coverage was normalized against relative coverage from normal human control, and baseline CN state was estimated by calculating mode of coverage of all labels. CN states of segmented genomic intervals were assessed for significant increase/decrease from the baseline. The corresponding duplication and deletion CN variant calls were output as result. For cell lines in which optical mapping was performed, reads were realigned to the hg38 genome, followed by ecDNA assembly using AmpliconArchitect. The reference sequence of ecDNAs was prepared using the AmpliconReconstructor (79) tool with default settings. Inputs used for the analysis were the de novo assembly results from the Bionano analysis tools and the amplicons calculated from the whole-genome sequencing data calculated by the AmpliconArchitect tool.

Single-Cell CNV Analysis Using Targeted Sequencing Data

Targeted amplicons were designed to cover amplified ecDNA regions for the DMS-273, NCI-H889, and the patient-derived RA022 cell lines, as well as certain tumor suppressor genes (TP53, RB1, and FHIT) and background sites. The design covered a total of 199 sites, summarized in Supplementary Table S5.

Cryopreserved cells were quickly thawed by gently swirling the vials in the 37°C water bath. Cells were suspended and washed with 1× DPBS at room temperature. Cells were then resuspended in Mission Bio cell buffer at a concentration of 3,000 to 4,000 cells/μL. Afterward, the cells were microfluidically encapsulated, lysed, and barcoded in the Tapestri platform. Barcoded samples were then simultaneously PCR-amplified in droplets with a custom 199-amplicon panel covering 16 genes. Amplified products were purified and ligated with Illumina-specific P5 and P7 indices. In the next step, libraries were quantified with an Agilent Bioanalyzer for quality control, pooled, and sequenced on an Illumina NovaSeq platform.

Pooled libraries were sequenced on NovaSeq 6000 SP with paired-end sequencing (2 × 151 cycles). The sequencing run was demultiplexed using Bcl2fastq v2.20. The fastq files were processed through Mission Bio's Tapestri pipeline v2.0.1 with default parameters. Briefly, the adapter sequences were trimmed from raw reads using Cutadapt (79) v1.18. Reads that were too short with less than 30 bp were discarded. Trimmed reads were mapped to the reference genome hg19 using BWA-MEM (arXiv:1303.3997v2) v0.7.17. The CellFinder module extracted the barcodes from the mapped reads to identify captured cells. Barcodes that passed a total reads cutoff (the number of amplicons in the panel × 8 reads) were selected from at least 80% of good-performing amplicons (with the number of reads more than 0.2 × the mean of all amplicon reads). In the next step, the GATK HaplotypeCaller was used to call variants individually on each cell, and then GATK GenotypeGVCFs was applied to jointly genotype all cells using genotype likelihoods from the previous step based on the GATK Best Practices. The unfiltered VCF was parsed into an HDF5 file, which contains single-cell variants and per-amplicon read count matrices compatible with downstream analysis.

Raw amplicon quantifications for the cell lines were analyzed using R. Data were normalized using the Seurat (80) package, using the NormalizeData() function, setting the method to “RC,” minimum number of cells to 3, and minimum features to 25. Because many of the amplicons targeted ecDNA regions, normalization was first performed on background amplicons and amplicons targeting tumor suppressor genes, from which the scaling factors were obtained. The normalized read counts for all amplicons were calculated using the obtained scaling factors. Normalization was performed for each of the cell lines separately (DMS-273, NCI-H889, and the patient-derived cells together), to exclude any variation in copy number of the non-ecDNA regions.

Sample Preparation and Analysis of Adherent and Suspension Cells

NCI-H524 cell lines were obtained from ATCC (www.atcc.org) and tested for Mycoplasma (Lonza, LT07-418). Human H524 cells were grown at 37°C in a 5% CO2 atmosphere in RPMI 1640 medium (Themo Fisher, 11875119), supplemented with 10% fetal calf serum (GeminiBio, 100-106). Cells were passaged by splitting cells 1:3 to 1:4, keeping cells below 0.5 × 105/mL. Adherent NCI-H524 cells were selected by removing three quarters of the floating cells at regular intervals to avoid the medium from turning yellow. Adherent NCI-H524 cells were first collected by trypsinization and transferred to new flasks when ∼50% confluency was achieved.

Western blots were done with whole-cell lysates by adding 200 μL SDS loading buffer (Bio-Rad, 1610737) per 1 million cell pellets. Samples were heated at 100°C for 5 minutes and centrifuged, and the supernatant was used for Western blot. Immunoblotting signals were detected using the ChemiDoc MP imaging system (Bio-Rad). The following primary antibodies were used: anti–c-Myc (Cell Signaling Technology, 5605S), anti-NEUROD1 (Cell Signaling Technology, 7019S), anti-YAP1 (Cell Signaling Technology, 4912S), anti-INSM1 (Santa Cruz, SC-271408), and anti–α-tubulin (Millipore-Sigma, T9026).

gDNA samples from 5 million cells were prepared using the Qiagen DNeasy blood and tissue kit (69504). DNA samples were sequenced using the Novaseq S1 sequencing system (150 bp, paired-end reads) to obtain more than 300 million total reads (319,758,170 for H524A and 431,577,608 for H524).

Raw data were trimmed using trimmomatic (72), followed by alignment to the hg19 genome using the BWA-MEM (arXiv:1303.3997v2) algorithm. The BAM files were sorted with samtools (64), followed by duplicate marking using picard-tools. Normalized sequencing depth tracks were created using BAMscale (65). ecDNA assembly was performed using the AmpliconArchitect tool.

cfChIP Samples and Data Analysis

For cfChiP-seq, plasma samples were collected from patients with small cell lung cancer who received care at the NCI. Patients were enrolled in therapeutic clinical trials (ClinicalTrials.gov #NCT02484404; NCI protocol #15-C-0145; ClinicalTrials.gov #NCT02487095; NCI protocol #15-C-0150; ClinicalTrials.gov #NCT02769962; NCI protocol #16-C-0107; ClinicalTrials.gov #NCT03554473; NCI protocol #18-C-0110; and ClinicalTrials.gov #NCT03896503; NCI protocol #20-C-0009). We also collected samples from patients with small cell lung cancer who were enrolled in the NCI thoracic malignancies natural history protocol (ClinicalTrials.gov #NCT02146170; NCI protocol #14-C-0105). Rapid autopsy was performed to collect metastatic tumor samples from a patient with small cell lung cancer under NCI protocol #13-C-0131. The human subjects committee at NCI approved the studies; all patients provided written informed consent per the Declaration of Helsinki principles. Sample preparation and sequence analysis were performed similarly to a previous study (81). Highly amplified oncogenes were detected based on the normalized genic signal from regions outside of detected H3K3me3 peaks.

Patient-Derived Cell Lines

Samples were obtained from five patients diagnosed with thoracic malignancies who underwent rapid autopsy. Informed consent for rapid autopsy was obtained under an institutional review board–approved protocol 13-C-0131 (NCT01851395) entitled “A Pilot Study of Inpatient Hospice with Procurement of Tissue on Expiration in Thoracic Malignancies.” Patients previously treated at the NCI and with a life expectancy of less than 3 months were offered inpatient hospice treatment at the Clinical Center of the NIH, and upon death, autopsies were initiated within 3 hours. Clinical information, including the sex and gender of each patient, is available in Supplementary Table S1. One patient, RA003, elected to receive end-of-life care at home and was subsequently transported to the NIH Clinical Center postmortem. Prioritization of lesions removed at autopsy was based on a CT scan performed within 1 month before death. All tumors within each patient were removed by an experienced pathologist and macrodissected to remove surrounding nonneoplastic tissue. Punch biopsy needles were used to obtain spatially distinct cores from each tumor. One third of each tissue core sample was fixed in 10% buffered formalin, one third in optimal cutting temperature compound, and the remaining tissue was immediately flash frozen in liquid nitrogen and stored at −80°C. Part of each fresh tumor tissue was stored in RPMI. Within 1 hour, tumors were mechanically minced into 1-mm3 pieces and enzymatically digested with PBS containing collagenase and hyaluronidase for 1 hour at 37°C. Single-cell suspensions were passed through a 70 μmol/L cell strainer, and red blood cell lysis was performed for 5 minutes. Cells were washed with PBS and then placed in a 25-cm3 flask containing RPMI with 10% FBS. Media were changed weekly until adequate cell growth. For each tissue sample, a 5-μm section was taken to create a hematoxylin and eosin slide to visualize neoplastic cellularity using a microscope.

MYC CNV

Cell lines were plated at 1 million cells in a 10-cm dish and were collected after 24 hours. DNA was collected using the Qiagen DNeasy blood and tissue kit per the manufacturer's instructions. To assess CN state, we utilized Thermo Fisher Scientific TaqMan Genotyping Master Mix along with TaqMan Copy-Number Assay MYC and TaqMan Copy-Number Reference Assay RNase P and a Quant Studio 5 per the manufacturer's instructions.

Western Blot Results for RLF–MYCL and Patient-Derived Cell Lines

Cells were washed twice with PBS and then lysed with RIPA buffer containing protease inhibitor cocktail (Thermo Fisher Scientific) and micrococcal nuclease (Thermo Fisher Scientific). The resulting mixtures were incubated on ice for 30 minutes and then centrifuged for 20 minutes to get the supernatants. After adding Tris-Glycine SDS sample buffer including 5% 2-mercaptoethanol, the lysates were boiled for 10 minutes, analyzed by SDS-PAGE, and immunoblotted with the antibodies against to RLF-L-MYC (Cell Signaling Technology, cat. #76266), c-MYC (Santa Cruz, cat. #sc-40), ASCL1 (Santa Cruz, cat. #sc-390794), and NEUROD1 (Cell Signaling Technology, cat. #4373S). The immunoblotted membrane was stripped and reprobed with antibody against α-tubulin (Sigma-Aldrich, cat. #T9026) as a loading control.

Data Availability

The sequencing data of H3K27ac Hi-ChIP, scCNV, and whole genome for previously characterized cell lines were deposited at GEO under accession number GSE206354 (H3K27ac Hi-ChIP: GSE206351; scCNV: GSE206352; whole genome: GSE206353). The long-read optical mapping data for previously characterized cell lines are available at the GenBank repository (NCI-H889: PRJNA849929; NCI-H524: PRJNA849955; NCI-H1092: PRJNA849961; NCI-H1963: PRJNA849963). The newly generated patient-derived cell line H3K27ac Hi-ChIP data, RNA-seq, genome sequencing data and long-read optical mapping data, as well as patient-derived genome sequencing data are available at phs003190.v1.p1.

Ethics Reporting

All patients provided written informed consent per the Declaration of Helsinki principles.

L. Rinaldi reports other support from Delfi Diagnostics and an EMBO long-term fellowship outside the submitted work. J. Gutin is a cofounder of Senseera. N. Roper reports grants from Taiho Pharmaceuticals and ADC Therapeutics outside the submitted work. N. Friedman reports grants from the European Research Council and the Israel Science Foundation during the conduct of the study; personal fees from Senseera outside the submitted work; a patent for diagnostic use of cell-free DNA chromatin immunoprecipitation pending to The Hebrew University of Jerusalem; and is cofounder of Senseera. A. Thomas reports grants from EMD Serono Research and Development Institute, AstraZeneca, Tarveda Therapeutics, Immunomedics, and Prolynx during the conduct of the study. No disclosures were reported by the other authors.

L.S. Pongor: Conceptualization, data curation, software, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. C.W. Schultz: Formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. L. Rinaldi: Investigation, methodology, writing–review and editing. D. Wangsa: Formal analysis, investigation, visualization, methodology, writing–review and editing. C.E. Redon: Formal analysis, investigation, methodology, writing–review and editing. N. Takahashi: Data curation, formal analysis, investigation, writing–review and editing. G. Fialkoff: Formal analysis, investigation, methodology, writing–review and editing. P. Desai: Data curation, investigation, writing–review and editing. Y. Zhang: Validation, investigation, writing–review and editing. S. Burkett: Formal analysis, investigation, methodology, writing–review and editing. N. Hermoni: Investigation, writing–review and editing. N. Vilk: Investigation, writing–review and editing. J. Gutin: Investigation, writing–review and editing. R. Gergely: Investigation, writing–review and editing. Y. Zhao: Data curation, formal analysis, investigation, writing–review and editing. S. Nichols: Investigation, writing–review and editing. R. Vilimas: Investigation, writing–review and editing. L. Sciuto: Investigation, writing–review and editing. C. Graham: Investigation, writing–review and editing. J.M. Caravaca: Investigation, writing–review and editing. S. Turan: Investigation, writing–review and editing. T. Shen: Investigation, writing–review and editing. V.N. Rajapakse: Investigation, visualization, writing–review and editing. R. Kumar: Investigation, writing–review and editing. D. Upadhyay: Investigation, writing–review and editing. S. Kumar: Investigation, writing–review and editing. Y.S. Kim: Investigation, writing–review and editing. N. Roper: Investigation, writing–review and editing. B. Tran: Investigation, writing–review and editing. S.M. Hewitt: Validation. D.E. Kleiner: Investigation, writing–review and editing. M.I. Aladjem: Resources, investigation, writing–review and editing. N. Friedman: Resources, supervision, investigation, writing–original draft, writing–review and editing. G.L. Hager: Investigation, writing–review and editing. Y. Pommier: Investigation, writing–review and editing. T. Ried: Investigation, writing–review and editing. A. Thomas: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, visualization, writing–original draft, writing–review and editing.

We gratefully acknowledge the contributions of our patients who participated in the study and their families. This work utilized the computational resources of the NIH HPC Biowulf cluster (http://hpc.nih.gov). This work was supported by the intramural programs of the Center for Cancer Research, National Cancer Institute (ZIA BC 011793). We are grateful to Michael Kruhlak, Langston Lim, and Andy Tran [Confocal Microscopy Core Facility, Center for Cancer Research (CCR), NCI, NIH], Liz Conner, Madeline Wong, and Steven Shema (CCR Genomics Core), and Jyoti Shetty (CCR core sequencing facility).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.

Note:Supplementary data for this article are available at Cancer Discovery Online (http://cancerdiscovery.aacrjournals.org/).

1.
George
J
,
Lim
JS
,
Jang
SJ
,
Cun
Y
,
Ozretic
L
,
Kong
G
, et al
.
Comprehensive genomic profiles of small cell lung cancer
.
Nature
2015
;
524
:
47
53
.
2.
Lissa
D
,
Takahashi
N
,
Desai
P
,
Manukyan
I
,
Schultz
CW
,
Rajapakse
V
, et al
.
Heterogeneity of neuroendocrine transcriptional states in metastatic small cell lung cancers and patient-derived models
.
Nat Commun
2022
;
13
:
2023
.
3.
Gazdar
AF
,
Carney
DN
,
Nau
MM
,
Minna
JD
.
Characterization of variant subclasses of cell-lines derived from small cell lung-cancer having distinctive biochemical, morphological, and growth properties
.
Cancer Res
1985
;
45
:
2924
30
.
4.
Zhang
W
,
Girard
L
,
Zhang
YA
,
Haruki
T
,
Papari-Zareei
M
,
Stastny
V
, et al
.
Small cell lung cancer tumors and preclinical models display heterogeneity of neuroendocrine phenotypes
.
Transl Lung Cancer Res
2018
;
7
:
32
49
.
5.
Calbo
J
,
van Montfort
E
,
Proost
N
,
van Drunen
E
,
Beverloo
HB
,
Meuwissen
R
, et al
.
A functional role for tumor cell heterogeneity in a mouse model of small cell lung cancer
.
Cancer Cell
2011
;
19
:
244
56
.
6.
Thomas
A
,
Takahashi
N
,
Rajapakse
VN
,
Zhang
X
,
Sun
Y
,
Ceribelli
M
, et al
.
Therapeutic targeting of ATR yields durable regressions in small cell lung cancers with high replication stress
.
Cancer Cell
2021
;
39
:
566
79
.
7.
Roper
N
,
Velez
MJ
,
Chiappori
A
,
Kim
YS
,
Wei
JS
,
Sindiri
S
, et al
.
Notch signaling and efficacy of PD-1/PD-L1 blockade in relapsed small cell lung cancer
.
Nat Commun
2021
;
12
:
3880
.
8.
McColl
K
,
Wildey
G
,
Sakre
N
,
Lipka
MB
,
Behtaj
M
,
Kresak
A
, et al
.
Reciprocal expression of INSM1 and YAP1 defines subgroups in small cell lung cancer
.
Oncotarget
2017
;
8
:
73745
56
.
9.
Qu
S
,
Fetsch
P
,
Thomas
A
,
Pommier
Y
,
Schrump
DS
,
Miettinen
MM
, et al
.
Molecular subtypes of primary SCLC tumors and their associations with neuroendocrine and therapeutic markers
.
J Thorac Oncol
2022
;
17
:
141
53
.
10.
Owonikoko
TK
,
Dwivedi
B
,
Chen
Z
,
Zhang
C
,
Barwick
B
,
Ernani
V
, et al
.
YAP1 expression in SCLC defines a distinct subtype with T-cell-inflamed phenotype
.
J Thorac Oncol
2021
;
16
:
464
76
.
11.
Gay
CM
,
Stewart
CA
,
Park
EM
,
Diao
LX
,
Groves
SM
,
Heeke
S
, et al
.
Patterns of transcription factor programs and immune pathway activation define four major subtypes of SCLC with distinct therapeutic vulnerabilities
.
Cancer Cell
2021
;
39
:
346
–60.
12.
Stewart
CA
,
Gay
CM
,
Xi
Y
,
Sivajothi
S
,
Sivakamasundari
V
,
Fujimoto
J
, et al
.
Single-cell analyses reveal increased intratumoral heterogeneity after the onset of therapy resistance in small-cell lung cancer
.
Nat Cancer
2020
;
1
:
423
36
.
13.
Wagner
AH
,
Devarakonda
S
,
Skidmore
ZL
,
Krysiak
K
,
Ramu
A
,
Trani
L
, et al
.
Recurrent WNT pathway alterations are frequent in relapsed small cell lung cancer
.
Nat Commun
2018
;
9
:
3787
.
14.
Yang
D
,
Denny
SK
,
Greenside
PG
,
Chaikovsky
AC
,
Brady
JJ
,
Ouadah
Y
, et al
.
Intertumoral heterogeneity in SCLC is influenced by the cell type of origin
.
Cancer Discov
2018
;
8
:
1316
31
.
15.
Ferone
G
,
Lee
MC
,
Sage
J
,
Berns
A
.
Cells of origin of lung cancers: lessons from mouse studies
.
Genes Dev
2020
;
34
:
1017
32
.
16.
Olsen
RR
,
Ireland
AS
,
Kastner
DW
,
Groves
SM
,
Spainhower
KB
,
Pozo
K
, et al
.
ASCL1 represses a SOX9(+) neural crest stem-like state in small cell lung cancer
.
Genes Dev
2021
;
35
:
847
69
.
17.
Chen
J
,
Guanizo
A
,
Luong
Q
,
Jayasekara
WSN
,
Jayasinghe
D
,
Inampudi
C
, et al
.
Lineage-restricted neoplasia driven by Myc defaults to small cell lung cancer when combined with loss of p53 and Rb in the airway epithelium
.
Oncogene
2022
;
41
:
138
45
.
18.
Lim
JS
,
Ibaseta
A
,
Fischer
MM
,
Cancilla
B
,
O'Young
G
,
Cristea
S
, et al
.
Intratumoural heterogeneity generated by Notch signalling promotes small-cell lung cancer
.
Nature
2017
;
545
:
360
–4.
19.
Ireland
AS
,
Micinski
AM
,
Kastner
DW
,
Guo
B
,
Wait
SJ
,
Spainhower
KB
, et al
.
MYC drives temporal evolution of small cell lung cancer subtypes by reprogramming neuroendocrine fate
.
Cancer Cell
2020
;
38
:
60
78
.
20.
Mollaoglu
G
,
Guthrie
MR
,
Bohm
S
,
Bragelmann
J
,
Can
I
,
Ballieu
PM
, et al
.
MYC drives progression of small cell lung cancer to a variant neuroendocrine subtype with vulnerability to aurora kinase inhibition
.
Cancer Cell
2017
;
31
:
270
85
.
21.
Patel
AS
,
Yoo
S
,
Kong
RR
,
Sato
T
,
Sinha
A
,
Karam
S
, et al
.
Prototypical oncogene family Myc defines unappreciated distinct lineage states of small cell lung cancer
.
Sci Adv
2021
;
7
:
eabc2578
.
22.
Turner
KM
,
Deshpande
V
,
Beyter
D
,
Koga
T
,
Rusert
J
,
Lee
C
, et al
.
Extrachromosomal oncogene amplification drives tumour evolution and genetic heterogeneity
.
Nature
2017
;
543
:
122
5
.
23.
Wu
S
,
Turner
KM
,
Nguyen
N
,
Raviram
R
,
Erb
M
,
Santini
J
, et al
.
Circular ecDNA promotes accessible chromatin and high oncogene expression
.
Nature
2019
;
575
:
699
703
.
24.
Verhaak
RGW
,
Bafna
V
,
Mischel
PS
.
Extrachromosomal oncogene amplification in tumour pathogenesis and evolution
.
Nat Rev Cancer
2019
;
19
:
283
8
.
25.
deCarvalho
AC
,
Kim
H
,
Poisson
LM
,
Winn
ME
,
Mueller
C
,
Cherba
D
, et al
.
Discordant inheritance of chromosomal and extrachromosomal DNA elements contributes to dynamic disease evolution in glioblastoma
.
Nat Genet
2018
;
50
:
708
17
.
26.
Nathanson
DA
,
Gini
B
,
Mottahedeh
J
,
Visnyei
K
,
Koga
T
,
Gomez
G
, et al
.
Targeted therapy resistance mediated by dynamic regulation of extrachromosomal mutant EGFR DNA
.
Science
2014
;
343
:
72
6
.
27.
Morton
AR
,
Dogan-Artun
N
,
Faber
ZJ
,
MacLeod
G
,
Bartels
CF
,
Piazza
MS
, et al
.
Functional enhancers shape extrachromosomal oncogene amplifications
.
Cell
2019
;
179
:
1330
–41.
28.
Kim
H
,
Nguyen
NP
,
Turner
K
,
Wu
SH
,
Gujar
AD
,
Luebeck
J
, et al
.
Extrachromosomal DNA is associated with oncogene amplification and poor outcome across multiple cancers
.
Nat Genet
2020
;
52
:
891
.
29.
Cox
D
,
Yuncken
C
,
Spriggs
AI
.
Minute chromatin bodies in malignant tumours of childhood
.
Lancet
1965
;
2
:
55
.
30.
Stephens
PJ
,
Greenman
CD
,
Fu
BY
,
Yang
FT
,
Bignell
GR
,
Mudie
LJ
, et al
.
Massive genomic rearrangement acquired in a single catastrophic event during cancer development
.
Cell
2011
;
144
:
27
40
.
31.
Shoshani
O
,
Brunner
SF
,
Yaeger
R
,
Ly
P
,
Nechemia-Arbely
Y
,
Kim
DH
, et al
.
Chromothripsis drives the evolution of gene amplification in cancer
.
Nature
2021
;
591
:
137
41
.
32.
Thomas
A
,
Desai
P
,
Takahashi
N
.
Translational research: A patient-centered approach to bridge the valley of death
.
Cancer Cell
2022
;
40
:
565
8
.
33.
Ghandi
M
,
Huang
FW
,
Jane-Valbuena
J
,
Kryukov
GV
,
Lo
CC
,
McDonald
ER
3rd
, et al
.
Next-generation characterization of the cancer cell line encyclopedia
.
Nature
2019
;
569
:
503
8
.
34.
Tlemsani
C
,
Pongor
L
,
Elloumi
F
,
Girard
L
,
Huffman
KE
,
Roper
N
, et al
.
SCLC-CellMiner: a resource for small cell lung cancer cell line genomics and pharmacology based on genomic signatures
.
Cell Rep
2020
;
33
:
108296
.
35.
Pongor
LS
,
Tlemsani
C
,
Elloumi
F
,
Arakawa
Y
,
Jo
U
,
Gross
JM
, et al
.
Integrative epigenomic analyses of small cell lung cancer cells demonstrates the clinical translational relevance of gene body methylation
.
iScience
2022
;
25
:
105338
.
36.
Kim
DW
,
Wu
N
,
Kim
YC
,
Cheng
PF
,
Basom
R
,
Kim
D
, et al
.
Genetic requirement for Mycl and efficacy of RNA Pol I inhibition in mouse models of small cell lung cancer
.
Genes Dev
2016
;
30
:
1289
99
.
37.
Deshpande
V
,
Luebeck
J
,
Nguyen
NPD
,
Bakhtiari
M
,
Turner
KM
,
Schwab
R
, et al
.
Exploring the landscape of focal amplifications in cancer using AmpliconArchitect
.
Nat Commun
2019
;
10
:
392
.
38.
Rosswog
C
,
Bartenhagen
C
,
Welte
A
,
Kahlert
Y
,
Hemstedt
N
,
Lorenz
W
, et al
.
Chromothripsis followed by circular recombination drives oncogene amplification in human cancer
.
Nat Genet
2021
;
53
:
1673
.
39.
Cao
HZ
,
Hastie
AR
,
Cao
DD
,
Lam
ET
,
Sun
YH
,
Huang
HD
, et al
.
Rapid detection of structural variation in a human genome using nanochannel-based genome mapping technology
.
Gigascience
2014
;
3
:
34
.
40.
Shimizu
N
,
Shingaki
K
,
Kaneko-Sasaguri
Y
,
Hashizume
T
,
Kanda
T
.
When, where and how the bridge breaks: anaphase bridge breakage plays a crucial role in gene amplification and HSR generation
.
Exp Cell Res
2005
;
302
:
233
43
.
41.
Huang
YH
,
Klingbeil
O
,
He
XY
,
Wu
XLS
,
Arun
G
,
Lu
B
, et al
.
POU2F3 is a master regulator of a tuft cell-like variant of small cell lung cancer
.
Gene Dev
2018
;
32
:
915
28
.
42.
Borromeo
MD
,
Savage
TK
,
Kollipara
RK
,
He
M
,
Augustyn
A
,
Osborne
JK
, et al
.
ASCL1 and NEUROD1 reveal heterogeneity in pulmonary neuroendocrine tumors and regulate distinct genetic programs
.
Cell Rep
2016
;
16
:
1259
72
.
43.
Kumar
P
,
Kiran
S
,
Saha
S
,
Su
Z
,
Paulsen
T
,
Chatrath
A
, et al
.
ATAC-seq identifies thousands of extrachromosomal circular DNA in cancer and cell lines
.
Sci Adv
2020
;
6
:
eaba2489
.
44.
Zhu
YF
,
Gujar
AD
,
Wong
CH
,
Tjong
H
,
Ngan
CY
,
Gong
L
, et al
.
Oncogenic extrachromosomal DNA functions as mobile enhancers to globally amplify chromosomal transcription
.
Cancer Cell
2021
;
39
:
694
.
45.
Feber
A
,
Guilhamon
P
,
Lechner
M
,
Fenton
T
,
Wilson
GA
,
Thirlwell
C
, et al
.
Using high-density DNA methylation arrays to profile copy number alterations
.
Genome Biol
2014
;
15
:
R30
.
46.
Vastrik
I
,
Makela
TP
,
Koskinen
PJ
,
Saksela
K
,
Alitalo
K. Myc
,
Max, and a novel rlf-L-myc fusion protein in small-cell lung-cancer
.
Princess Takamatsu Symp
1992
;
22
:
307
18
.
47.
Rudin
CM
,
Durinck
S
,
Stawiski
EW
,
Poirier
JT
,
Modrusan
Z
,
Shames
DS
, et al
.
Comprehensive genomic analysis identifies SOX2 as a frequently amplified gene in small-cell lung cancer
.
Nat Genet
2012
;
44
:
1111
.
48.
Ciampricotti
M
,
Karakousi
T
,
Richards
AL
,
Quintanal-Villalonga
A
,
Karatza
A
,
Caeser
R
, et al
.
Rlf-Mycl gene fusion drives tumorigenesis and metastasis in a mouse model of small cell lung cancer
.
Cancer Discov
2021
;
11
:
3214
29
.
49.
Hung
KL
,
Yost
KE
,
Xie
L
,
Shi
Q
,
Helmsauer
K
,
Luebeck
J
, et al
.
ecDNA hubs drive cooperative intermolecular oncogene expression
.
Nature
2021
;
600
:
731
6
.
50.
Tseng
YY
,
Moriarity
BS
,
Gong
W
,
Akiyama
R
,
Tiwari
A
,
Kawakami
H
, et al
.
PVT1 dependence in cancer with MYC copy-number increase
.
Nature
2014
;
512
:
82
6
.
51.
Gupta
PB
,
Pastushenko
I
,
Skibinski
A
,
Blanpain
C
,
Kuperwasser
C
.
Phenotypic plasticity: driver of cancer initiation, progression, and therapy resistance
.
Cell Stem Cell
2019
;
24
:
65
78
.
52.
Boumahdi
S
,
de Sauvage
FJ
.
The great escape: tumour cell plasticity in resistance to targeted therapy
.
Nat Rev Drug Discov
2020
;
19
:
39
56
.
53.
Sutherland
KD
,
Ireland
AS
,
Oliver
TG
.
Killing SCLC: insights into how to target a shapeshifting tumor
.
Genes Dev
2022
;
36
:
241
58
.
54.
Thakur
BL
,
Ray
A
,
Redon
CE
,
Aladjem
MI
.
Preventing excess replication origin activation to ensure genome stability
.
Trends Genet
2022
;
38
:
169
81
.
55.
Iwakawa
R
,
Takenaka
M
,
Kohno
T
,
Shimada
Y
,
Totoki
Y
,
Shibata
T
, et al
.
Genome-wide identification of genes with amplification and/or fusion in small cell lung cancer
.
Genes Chromosomes Cancer
2013
;
52
:
802
16
.
56.
Weischenfeldt
J
,
Dubash
T
,
Drainas
AP
,
Mardin
BR
,
Chen
Y
,
Stutz
AM
, et al
.
Pan-cancer analysis of somatic copy-number alterations implicates IRS4 and IGF2 in enhancer hijacking
.
Nat Genet
2017
;
49
:
65
74
.
57.
Spielmann
M
,
Lupianez
DG
,
Mundlos
S
.
Structural variation in the 3D genome
.
Nat Rev Genet
2018
;
19
:
453
67
.
58.
Cho
SW
,
Xu
J
,
Sun
R
,
Mumbach
MR
,
Carter
AC
,
Chen
YG
, et al
.
Promoter of lncRNA Gene PVT1 is a tumor-suppressor DNA boundary element
.
Cell
2018
;
173
:
1398
412
.
59.
Jin
X
,
Demere
Z
,
Nair
K
,
Ali
A
,
Ferraro
GB
,
Natoli
T
, et al
.
A metastasis map of human cancer cell lines
.
Cancer Res
2021
;
81
.
60.
Takahashi
N
,
Kim
S
,
Schultz
CW
,
Rajapakse
VN
,
Zhang
Y
,
Redon
CE
, et al
.
Replication stress defines distinct molecular subtypes across cancers
.
Cancer Res Commun
2022
;
2
:
503
17
.
61.
Thomas
A
,
Pommier
Y
.
Small cell lung cancer: time to revisit DNA-damaging chemotherapy
.
Sci Transl Med
2016
;
8
:
346fs12
.
62.
Wang
Y
,
Wang
M
,
Djekidel
MN
,
Chen
H
,
Liu
D
,
Alt
FW
, et al
.
eccDNAs are apoptotic products with high innate immunostimulatory activity
.
Nature
2021
;
599
:
308
14
.
63.
Gardner
EE
,
Lok
BH
,
Schneeberger
VE
,
Desmeules
P
,
Miles
LA
,
Arnold
PK
, et al
.
Chemosensitive relapse in small cell lung cancer proceeds through an EZH2-SLFN11 axis
.
Cancer Cell
2017
;
31
:
286
99
.
64.
Li
H
,
Handsaker
B
,
Wysoker
A
,
Fennell
T
,
Ruan
J
,
Homer
N
, et al
.
The sequence alignment/map format and SAMtools
.
Bioinformatics
2009
;
25
:
2078
9
.
65.
Pongor
LS
,
Gross
JM
,
Vera Alvarez
R
,
Murai
J
,
Jang
SM
,
Zhang
H
, et al
.
BAMscale: quantification of next-generation sequencing peaks and generation of scaled coverage tracks
.
Epigenetics Chromatin
2020
;
13
:
21
.
66.
Lawrence
M
,
Gentleman
R
,
Carey
V
.
rtracklayer: an R package for interfacing with genome browsers
.
Bioinformatics
2009
;
25
:
1841
2
.
67.
Lawrence
M
,
Huber
W
,
Pages
H
,
Aboyoun
P
,
Carlson
M
,
Gentleman
R
, et al
.
Software for computing and annotating genomic ranges
.
PLoS Comput Biol
2013
;
9
:
e1003118
.
68.
Mumbach
MR
,
Satpathy
AT
,
Boyle
EA
,
Dai
C
,
Gowen
BG
,
Cho
SW
, et al
.
Enhancer connectome in primary human cells identifies target genes of disease-associated DNA elements
.
Nat Genet
2017
;
49
:
1602
12
.
69.
Servant
N
,
Varoquaux
N
,
Lajoie
BR
,
Viara
E
,
Chen
CJ
,
Vert
JP
, et al
.
HiC-Pro: an optimized and flexible pipeline for Hi-C data processing
.
Genome Biol
2015
;
16
:
259
.
70.
Durand
NC
,
Robinson
JT
,
Shamim
MS
,
Machol
I
,
Mesirov
JP
,
Lander
ES
, et al
.
Juicebox provides a visualization system for Hi-C contact maps with unlimited zoom
.
Cell Syst
2016
;
3
:
99
101
.
71.
Pozo
K
,
Kollipara
RK
,
Kelenis
DP
,
Rodarte
KE
,
Ullrich
MS
,
Zhang
XY
, et al
.
ASCL1, NKX2–1, and PROX1 co-regulate subtype-specific genes in small-cell lung cancer
.
Iscience
2021
;
24
:
102953
.
72.
Bolger
AM
,
Lohse
M
,
Usadel
B
.
Trimmomatic: a flexible trimmer for Illumina sequence data
.
Bioinformatics
2014
;
30
:
2114
20
.
73.
Zhang
Y
,
Liu
T
,
Meyer
CA
,
Eeckhoute
J
,
Johnson
DS
,
Bernstein
BE
, et al
.
Model-based analysis of ChIP-Seq (MACS)
.
Genome Biol
2008
;
9
:
R137
.
74.
Aryee
MJ
,
Jaffe
AE
,
Corrada-Bravo
H
,
Ladd-Acosta
C
,
Feinberg
AP
,
Hansen
KD
, et al
.
Minfi: a flexible and comprehensive Bioconductor package for the analysis of Infinium DNA methylation microarrays
.
Bioinformatics
2014
;
30
:
1363
9
.
75.
Krueger
F
,
Andrews
SR
.
Bismark: a flexible aligner and methylation caller for bisulfite-seq applications
.
Bioinformatics
2011
;
27
:
1571
2
.
76.
Alvarez
RV
,
Pongor
LS
,
Marino-Ramirez
L
,
Landsman
D
.
TPMCalculator: one-step software to quantify mRNA abundance of genomic features
.
Bioinformatics
2019
;
35
:
1960
2
.
77.
Love
MI
,
Huber
W
,
Anders
S
.
Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2
.
Genome Biol
2014
;
15
:
550
.
78.
Gao
JJ
,
Aksoy
BA
,
Dogrusoz
U
,
Dresdner
G
,
Gross
B
,
Sumer
SO
, et al
.
Integrative analysis of complex cancer genomics and clinical profiles using the cBioPortal
.
Sci Signal
2013
;
6
:
pl1
.
79.
Gu
ZG
,
Eils
R
,
Schlesner
M
.
Complex heat maps reveal patterns and correlations in multidimensional genomic data
.
Bioinformatics
2016
;
32
:
2847
9
.
80.
Hao
YH
,
Hao
S
,
Andersen-Nissen
E
,
Mauck
WM
,
Zheng
SW
,
Butler
A
, et al
.
Integrated analysis of multimodal single-cell data
.
Cell
2021
;
184
:
3573
.
81.
Sadeh
R
,
Sharkia
I
,
Fialkoff
G
,
Rahat
A
,
Gutin
J
,
Chappleboim
A
, et al
.
ChIP-seq of plasma cell-free nucleosomes identifies gene expression programs of the cells of origin (vol 71, pg 515, 2021)
.
Nat Biotechnol
2021
;
39
:
642
.

Supplementary data