We used lentivirus vectors encoding doxycycline (DOX)-inducible mRNAs for wild-type (WT) and mutant (T58A) human MYC (see Materials and methods) to infect previously described hESCs from the RUES2 cell line carrying transgenes encoding DOX-inducible shRNAs complementary to TP53 and RB1. For simplicity of presentation, we refer to the latter RUES2 cells as RP cells, and the RP cells containing DOX-inducible WT or T58A mutant MYC as RPM and RPM (T58A) cells throughout the text and figures of this paper.

After the growth of infected cells and differentiation into lung cell progenitors (LPs) and mature lungcells (LCs) (Figure 1A), we induced PNECs by inhibition of NOTCH signaling with DAPT, as previously reported (Chen et al., 2019). The ability of the RPM and RPM (T58A) cells to produce MYC protein was documented by western blotting of cell extracts prepared 72 hr after addition of DOX (Figure 1B). As expected, both cell lines also contained dramatically reduced amounts of TP53 and RB1 proteins as a result of DOX induction of shRNAs specific for these tumor suppressor mRNAs (Figure 1B).

To determine whether production of WT or mutant MYC proteins renders PNECs derived from RPM lines more oncogenic, we sorted CGRP-positive RPM and RPM (T58A) cells from cultures of LCs treated with DAPT and injected the cells subcutaneously or into the renal capsule of immunodeficient mice (the NOG strain: NOD/Shi-scid/IL-2Rγnull) as depicted in Figure 1A. The mice were then maintained on diets with or without DOX (see Materials and methods). Three to four months later, subcutaneous growths larger than 1000 mm³ or renal engraftments greater than 100 mm³ and grossly metastatic lesions in the liver and lungs were tabulated, as summarized in Figure 1C and D. As previously reported, PNECs from cultures of RP lines of RUES2 cells produced small benign lesions subcutaneously without invasion or metastasis. Similar findings were observed when these cells were injected into the renal capsule (Figure 1D).

Thus far, our data suggest that at least some RUES2-derived, PNEC-like cells in which expression of two tumor suppressor genes, RB1 and TP53, is restricted, RP cells, can be transformed into aggressive, metastatic cells by the addition of transgenes encoding WT or mutant MYC protein (hereafter referred to as ‘RPM’ or ‘RPM (T58A)’ cells). However, we had not directly compared the relative growth rates of these engineered cell types.

We therefore engrafted immunocompromised mice subcutaneously with equivalent numbers of RP, RPM, or RPM (T58A) cells and monitored tumor growth rates. We noted that the volumes of RPM and RPM (T58A)-derived tumors quickly exceeded the volumes of RP-derived tumors (Figure 2A). The rapid growth of RPM tumors was dependent on continuous provision of DOX in the diet. If animals on a DOX-free diet were engrafted subcutaneously with RPM or RPM (T58A) cells, they failed to produce tumors (Figure 2B, left). However, tumor growth was observed in some mice engrafted with such cells if the animals were placed on a DOX-containing diet up to 2 months after injection (Figure 2B, right). These findings suggest that at least some of the RPM cells remained viable and responsive to the oncogenic effects of depleting RB1 and TP53 and producing MYC.

The rapid growth and morphological features of RPM and RPM (T58A) tumors in these experiments were histologically consistent with SCLC. Critically, subcutaneous injection did not produce metastatic disease to the liver (not shown), nor was there evidence of gross metastases to the lungs in any of the mice examined (Figure 2C). These findings suggest that SCLC-like tumors composed of RUES2-derived RPM cells grow rapidly after subcutaneous implantation, but may require a different microenvironment, such as the vascular-rich, highly perfused renal capsule (Bogden et al., 1979; Sobczuk et al., 2020), providing a more hospitable environment for metastatic seeding.

RPM tumors, as compared to RP tumors, were notable for increased vascularization, larger volume, and invasion into surrounding endothelium. RPM and RPM (T58A) PNECs injected into the renal capsule of immunocompromised mice also displayed frequent metastasis to the liver (Figure 3D). In animals administered DOX, histological examinations showed that approximately half developed metastases in distant organs, including the liver or lung. One of the mice developed metastatic tumors in both liver and lung sites (Figure 1D). No metastases were observed in the bone, brain, or lymph nodes. Metastatic lesions were characterized as having similar proliferative character and expression of MYC, as well as comparable expression of the neuronal markers NCAM (CD56) and NEUROD1 compared to their paired primary counterparts (Figure 3E and F). In contrast, ASCL1 expression was absent in RPM and RPM (T58A) tumors at both primary and metastatic sites (Figure 3F). Meanwhile, the majority of cells in RP tumors showed positive ASCL1 expression but lacked NEUROD1 expression (Figure 3G). These findings suggest that MYC expression may facilitate the transition of SCLC tumors from the classical ASCL1-driven subtype to the NEUROD1 variant. Thus, both the site of implantation of cells and the addition of a driver oncogene were required to produce vigorous growth and metastatic spread of SCLC-like tumors, and the metastatic lesions retained histological and immunohistochemical features of the primary tumors.

To better understand the characteristics of RUES2-derived RPM tumors, we performed single-cell RNA-seq (RNA-seq) on culture-derived (‘2D’) and tumor-derived RPM and RPM (T58A) viable EGFP+ cells. By integrating previously published RUES2-derived RP single-cell data (Chen et al., 2019), we observed three dominant cell-type groups in transcriptional space, using reference mapping, enrichment analysis, and manual curation to identify cell lineage likelihood: a basal-like cluster marked by expression of KRT19, a fibroblast-like cluster marked by expression of COL1A1, and the most prominent cluster of cells appearing to have features of neuroendocrine differentiation, marked by expression of the neuroendocrine lineage defining transcription factor ASCL1 (Figure 4A and Supplementary file 1a). Comparing the recently described SCLC subtype markers (Rudin et al., 2019) across these datasets showed that cells within the neuroendocrine differentiation cluster cells expressed ASCL1 and NEUROD1, whereas POU2F3 RNA was not detected. Levels of YAP1 RNA were variable and were not co-expressed with either ASCL1 or NEUROD1 (Figure 4B). YAP1 expression was mostly attributed to fibroblast- and basal-like cells that are often observed in long-term cultures of differentiated hESCs (Figure 4B; Huang et al., 2014; Chen et al., 2017; Dye et al., 2015; Rodrigues Toste de Carvalho et al., 2021).

Next, we asked whether the overexpression of MYC in RPM cells was associated with increased neuroendocrine differentiation when compared to cell lines that did not contain MYC transgenes (i.e. RP cells). One notable difference between our prior work and this study was our attempt to increase the tumor cell component of the RPM and RPM (T58A) samples by sorting cells for high EGFP expression. EGFP is expressed in cis with the RB1 targeting shRNA; however, EGFP is not expressed uniformly across all cells following the differentiation protocol, suggesting some cells may silence the integrated vector. We performed differential cluster abundance analysis after accounting for the fraction of cells that were EGFP+ during cell sorting. These results indicated that RPM cell lines (WT or T58A) were associated with a twofold increase in the neuroendocrine compartment when compared with RP cells (p<2.2 × 10^–16; Figure 4—figure supplement 1A and B). Similarly, RPM tumors had a threefold enrichment of cells expressing neuroendocrine markers compared to RP cell lines (p<2.2 × 10^–16). Notably, we found that RPM tumors and cells expressed MYC, not MYCN or MYCL at high levels (Figure 4—figure supplement 1C). Moreover, EGFP expression was strongly correlated with MYC (Pearson r=0.45, p<2.2 × 10^–16) and not MYCN (Pearson r=0.00043, p=0.97) nor MYCL (Pearson r=–0.031, p=0.002). Together, these results suggest that the overexpression of either WT or T58A MYC yielded a larger neuroendocrine compartment.

By sub-clustering the neuroendocrine differentiation group, we could clearly identify ASCL1-high and NEUROD1-high compartments, as well as several other groups of cycling cells that did not cleanly fit into any lung lineage cell identity (Figure 4D, Supplementary file 1). We conducted differential gene expression to further characterize the NEUROD1-high (abbreviated ‘NE-variant’) and ASCL1-high (abbreviated ‘NE’) clusters observed with RPM cells (from cell lines and tumors) and found that the NE-variant cluster was associated with increased expression of genes highly expressed in variant SCLC (NEUROD1, SCG3, IGFBPL1, SSTR2, and MYC) (Figure 4—figure supplement 1D, Supplementary file 1). In contrast, the NE (or ‘NE-classic’) cluster was linked to heightened expression of a variety of genes ranging from histones (HISTH1B, HISTH1D) to redox processes (MT1E, MT2A) (Figure 4E). Gene set enrichment analysis of these results in the NE-variant cluster revealed increased expression of genes associated with activation of epithelial-mesenchymal transition (p=6.81 × 10^–3) and pancreatic beta cell signaling (p=1.01 × 10^–3). However, this cluster was also linked to lower levels of oxidative phosphorylation (p=1.07 × 10^–25), possibly suggesting these subgroups of cells are biologically different (Figure 4—figure supplement 1E). Further examination of the ASCL1-high and NEUROD1-high clusters revealed that, across both clusters, RPM and RPM (T58A) cells were associated with higher expression of genes associated with cells in dividing phases of the cell cycle than were observed in RP cultures. We compared the proportion of G2M cells between culture-derived RPM and RP models. This revealed that RPM cells were associated with a 10- and 11-fold increase in the NE and NE-variant clusters, respectively (NE p=9.6 × 10^–6, NE-variant p=1.9 × 10^–7) with a relative reduction in cells in the G1 phase (Figure 4—figure supplement 1F). Similar trends were observed from data in human tumors (NE p<2.2 × 10^–16, NE-variant p=1.5 × 10^–13, Figure 4—figure supplement 2F).

Given the striking increase in proliferation in MYC-producing cells, we sought to understand the association between MYC and ASCL1 or NEUROD1 expression. We computed the cellular density of co-expression patterns of cells that express MYC in addition to either ASCL1 or NEUROD1. This identified that the joint density was highest for cells that co-express MYC and NEUROD1 and was localized to the NEUROD1-high cluster (Figure 4—figure supplement 1F). In comparison, MYC and ASCL1 co-expression patterns were heterogeneous and found in the ASCL1-high cluster as well as in cells that could not be annotated with confidence. Similar to what was previously reported by other groups, we found that ASCL1 and MYC expressions are negatively correlated (Pearson r=–0.11, p=1.5 × 10^–4). Moreover, a modest correlation between MYC and NEUROD1 expression was identified in our single-cell data (Pearson r=0.17, p=3.7 × 10^–4, Figure 4—figure supplement 1G). Given that single-cell sequencing experiments suffer from a high dropout rate, we attempted to validate MYC and NEUROD1 co-expression patterns in bulk primary and metastatic RPM tumors. Although we couldn't determine co-expression within the same cell using bulk gene expression data, we identified an extremely strong correlation between the expression of these two genes (Pearson r=0.93, p=0.02, Figure 4—figure supplement 1H). Taken together, we show that the overexpression of MYC (WT or T58A) may be linked to an SCLC-N phenotype.

Upon identification of the ASCL1- and NEUROD1-high clusters, we sought to further characterize how closely they resemble SCLC tumors transcriptionally using a two-pronged approach. First, we overlaid SCLC-A, SCLC-N, and SCLC-P signatures (Chan et al., 2021) from our single-cell data and calculated subtype enrichment scores. Enrichment patterns in RPM and RPM (T58A) xenografts revealed that both ASCL1- and NEUROD1-high clusters expressed SCLC-N-associated genes more frequently than SCLC-A and SCLC-P-associated counterparts (p<2.2 × 10^–16, Figure 4G). In contrast, RP cell cultures more closely resembled cells from tumors with an SCLC-A phenotype (p<2.2 × 10^–16, Figure 4—figure supplement 1I). Second, we identified the transcriptional identity of RPM xenografts and paired metastases by cellular deconvolution. Bulk gene expression of RPM tumors along with human SCLC biopsies (George et al., 2015) was deconvolved into SCLC subtypes. This identified relative cellular proportions of SCLC subtypes in bulk RNA-seq data. Overall, we found high concordance with published subtype annotations and those predicted by our analysis (Figure 4H; Rudin et al., 2019). Moreover, we found that all RPM xenografts strongly exhibited an SCLC-N transcriptional program. Taken together, these data suggest that overexpression of WT or mutant MYC alters the transcriptional identity of RP cells (Figure 4—figure supplement 2A–C), encouraging neuroendocrine differentiation, and may assist with the transformation from a SCLC-A to SCLC-N phenotype.

Protocols for maintenance of hESCs and generation of PNECs and other lung cells were modified from previous studies (Huang et al., 2015; Huang et al., 2014; Chen et al., 2019). Two hESC lines, RUES2 (Rockefeller University Embryonic Stem Cell Line 2, NIH approval number NIH hESC-09-0013, Registration number 0013; passages 7–10) and ES02 (HES2, NIH registry, WiCell Research Institute, Inc passages 3–7) were cultured and maintained in an undifferentiated state on irradiated mouse embryonic fibroblasts (Global Stem, cat. no. GSC-6001G). All cells were purchased from ATCC or WiCell in the past 3 years, and tests for mycoplasma were negative. The hESC lines were regularly checked for chromosome abnormalities and maintained with normal chromosome numbers. All embryonic stem cell studies were approved by the Institutional Review Board (IRB) at the University of Chicago, or by the Tri-Institutional ESCRO committee under protocols 2016-01-002 and 2017-10-004 (Weill Cornell Medicine, Memorial Sloan Kettering Cancer Center, and Rockefeller University).

hESC differentiation into endoderm was performed in serum-free differentiation (SFD) media of DMEM/F12 (3:1) (Life Technologies) supplemented with N2 (Life Technologies), B27 (Life Technologies), ascorbic acid (50 μg/ml, Sigma), GlutaMAX (2 mM, Life Technologies), monothioglycerol (0.4 μM, Sigma), 0.05% bovine serum albumin (BSA) (Life Technologies) at 37°C in a 5% CO₂/5% O₂/90% N₂ environment. hESCs were treated with Accutase (STEMCELL Technologies) and plated onto low attachment six-well plates (Corning Incorporated, Tewksbury MA, USA), resuspended in endoderm induction media containing Y-27632 (10 μM), human BMP4 (0.5 ng/ml), human bFGF (2.5 ng/ml), and human Activin A (100 ng/ml) for 72–84 hr dependent on the formation rates of endoderm cells. On day 3 or 3.5, the embryoid bodies were dissociated into single cells using 0.05% Trypsin/0.02% EDTA and plated onto fibronectin-coated (Sigma), 24-well tissue culture plates (~100,000–150,000 cells/well). For induction of anterior foregut endoderm, the endoderm cells were cultured in SFD medium supplemented with 1.5 μM dorsomorphin dihydrochloride and 10 μM SB431542 for 36–48 hr, and then switched to 36–48 hr of 10 μM SB431542 and 1 μM IWP2 treatment.

For induction of early-stage lung progenitor cells (days 6–15), the resulting anterior foregut endoderm was treated with CHIR99021 (3 μM), human FGF10 (10 ng/ml), human FGF7 (10 ng/ml), human BMP4 (10 ng/ml), and all-trans retinoic acid (ATRA; 50 nM) in SFD medium for 8–10 days. Day 10–15 cultures were maintained in a 5% CO₂/air environment. On days 15 and 16, the lung field progenitor cells were re-plated after brief 1 min trypsinization onto fibronectin-coated plates, in the presence of SFD containing either a combination of five factors: CHIR99021 (3 μM), human FGF10 (10 ng/ml), human FGF7 (10 ng/ml), human BMP4 (10 ng/ml), and ATRA (50 nM), or three factors: CHIR99021 (3 μM), human FGF10 (10 ng/ml), and human FGF7 (10 ng/ml) for days 14–16. Day 16–25 cultures of late-stage lung progenitor cells were maintained in SFD media containing CHIR99021 (3 μM), human FGF10 (10 ng/ml), and human FGF7 (10 ng/ml) in a 5% CO₂/air environment.

For differentiation of mature lung cells (days 25–55), cultures were re-plated after brief trypsinization onto 3.3% Matrigel-coated 24-well plates in SFD media containing maturation components CHIR99021 (3 μM), human FGF10 (10 ng/ml), human FGF7 (10 ng/ml), and DCI (50 nM dexamethasone, 0.1 mM 8-bromo-cAMP and 0.1 mM IBMX [3,7-dihydro-1-methyl-3-(2-methylpropyl)-1H-purine-2,6-dione]). DAPT (10 µM) was added to the maturation media for induction of PNECs. All growth factors and most small molecules used for hESC differentiation were purchased from R&D Systems. ATRA, 8-bromo-cAMP, IBMX, and DAPT were purchased from Sigma-Aldrich.

HSC lines were constructed to contain DOX-inducible cassettes encoding shRNAs specific for the TP53 and RB1 tumor suppressor genes (RP cells) and RP cells containing DOX-inducible cassettes encoding WT human MYC or a stable, mutant version of MYC (T58A), to generate RPM and RPM (T58A) cells. The lentiviral vector expressing tetracycline (TET)-inducible shRNA against human P53 was purchased from Gentarget. Inc (cat# LVP-343-RB-PBS). The lentiviral vectors expressing TET-inducible shRNAs against human RB1 construct (‘pSLIK sh human Rb 1534 hyg’ was a gift from Julien Sage; plasmid # 31500) (Conklin et al., 2012), TET-inducible WT MYC (FUW-tetO-hMYC was a gift from Rudolf Jaenisch; plasmid # 20723) or mutant MYC (T58A) tagged at the N-terminus with three copies of a hemagglutinin tag (3X-HA) (pLV-tetO-myc T58A was a gift from Konrad Hochedlinger; plasmid # 19763) were obtained from Addgene and sequence-verified prior to use. The shRNA target sequences are as follows: RB1 # 1: 5’-GGACATGTGAACTTATATA-3’, RB1 # 2: 5’-GAACGATTATCCATTCAAA-3’, p53: 5’-CACCATCCACTACAACTACAT-3’.

To generate lentiviral particles, the above plasmids were transfected into HEK293T cells with the PC-Pack2 lentiviral packaging mix (Cellecta, Inc), according to the manufacturer’s protocol. High titer viral particles were used to transduce hESCs in serum-free conditions, and antibiotic selection of transduced hESCs was performed without MEF feeder cells, using mTeSR1 stem cell media (STEMCELL Technologies). The efficiency of RB1 or P53 knockdown and production of MYC or MYC (T58A) was determined by western blot after antibiotic selection using the following antibodies: anti-RB1 (Cell Signaling, clone 4H1, Cat# 9309), anti-P53 (Santa Cruz, clone DO-1, Cat# Sc-126), anti-MYC (Cell Signaling, clone D84C12, Cat# 5605), anti-HA (Cell Signaling, clone 6E2, Cat# 2367), and anti-GAPDH (Cell Signaling, clone 14C10, Cat# 2118). All antibodies for western blot were used at a 1:500 working dilution.

Living cells in culture were directly fixed in 4% paraformaldehyde for 25 min, followed with 15 min permeabilization in 1% Triton X-100. Histology on tissues from mice was performed on paraffin-embedded or frozen sections from xenografted tumors and corresponding normal tissues, as previously described (Chen et al., 2015). Tissues were either fixed overnight in 10% buffered formalin and transferred to 70% ethanol, followed by paraffin embedding, or snap-frozen in OCT medium (Fisher Scientific, Pittsburgh, PA, USA) and fixed in 10% buffered formalin, followed by paraffin embedding. For immunofluorescence, cells or tissue sections were immunostained with antibodies and counterstained with 4,6-diamidino-2-phenylindole. Adjacent sections stained with H&E were used for comparison.

Antibodies used for immunostaining or western blot experiments are as follows: anti-CGRP (Sigma, Clone CD8, Cat# c9487), anti-ASCL1 (Sigma, clone 2D9, Cat# SAB1403577), anti-NCAM1/CD56 (R&D Systems, cat# AF2408), anti-Ki67 (Cell Signaling, clone D2H10, Cat# 9027), anti-MYC (Abcam, clone Y69, Cat# ab32072), and anti-NEUROD1 (Sigma-Aldrich, clone 3H8, Cat# WH004760M1). All antibodies were used at a 1:150 working dilution.

Fluorescent-activated cell sorting (FACS) to detect definitive endoderm cells was performed using APC anti-c-KIT (Invitrogen, Cat# CD11705), PE anti-human CXCR4 (BioLegend, Clone 12G5, Cat# 306506), and APC anti-EpCAM (Invitrogen, Clone G8.8, Cat# 17-5791-80) at 1:100 dilution. Cells were incubated with primary antibodies for 30 min at 4°C, then washed and suspended in 0.1% BSA/PBS buffer. PE and APC filters were then used to detect cells double positive for KIT and CXCR4, or EpCAM and CXCR4 by signal intensity gating. FACS with anti-CGRP antibody (Abcam, clone 4902, Cat# ab81887) was used to detect CGRP+ cells. Cells were first incubated with anti-human CGRP antibody for 30 min at room temperature, followed by incubation (1:1000) of secondary antibody conjugated with R-phycoerythrin (PE) for 30 min at room temperature. Then, cells were washed and suspended in 0.1% BSA/PBS buffer. PE filter was then used to separate cells into CGRP+ and CGRP- subgroups by signal intensity gating. Negative controls stained with isotype controls were performed with sample measurements. Cells were sorted on a BD FACSAria II, and data were analyzed in FlowJo.

0.3–0.5 × 10⁶ hESC-derived lung cells or PNECs (at day 55 with or without 30 days of prior exposure to 10 µM DAPT alone or to 10 µM DAPT and 1 µM DOX to reduce P53 and RB1, and also induce MYC expression) were injected subcutaneously or under the renal capsule of 6- to 8-week-old, female NOD.Cg-Prkdcscid^scid Il2rgtm1WjI^tm1WjI/SzJ (NSG) mice (Jackson Laboratory, Bar Harbor, ME) (Chen et al., 2012). DOX diet began the day after injection (Teklad; 625 ppm); tumor development was monitored at least two to three times weekly. When animals became moribund or tumor size reached protocol limitations, mice were sacrificed, necropsy performed, and tumors harvested for further histological or molecular study.

Single-cell capture, reverse transcription, cell lysis, and library preparation were carried out by the WCM Epigenomics Core Facility using Chromium Single Cell 3’ v3 kits according to the manufacturer’s protocol (10x Genomics, USA). Single-cell suspensions of RUES2-derived RPM and RPM (T58A) day 50 cultures were prepared by incubating wells in 0.05% Trypsin/0.02% EDTA for 10–15 min. Digestions were then quenched with excess 0.1% BSA/PBS buffer and passed through a 40 µm mesh strainer. EGFP+ singlets were sorted at the WCM Flow Cytometry Core Facility on a BD Influx or Sony MA900 cell sorter. Each RUES2-derived RPM and RPM (T58A) day 50 culture pools were isolated from three wells of a 12-well plate across two biological replicates, targeting ~50,000 viable, EGFP+ viable cells per replicate. Cell counts were then adjusted to 6000–9000 cells/50 µl in PBS by Trypan Blue exclusion to achieve an estimated capture of 4000–5000 cells. Sequencing was performed by the WCM Genomics Resources Core Facility on a HiSeq 2500 (Illumina, paired-end protocol with 26 base pairs for read 1 and 98 bases for read 2). Single-cell RNA-seq from RP cultures was obtained from previous experiments (Chen et al., 2019). Transcript abundance was quantified by mapping sequencing reads to a custom human transcriptome reference (GENCODE v42, hg38) that included EGFP and was performed using simpleaf/alevin-fry (He and Patro, 2023).

Single-cell analyses, including quality filtering, integration, clustering, differential gene expression, and reduced-dimensionality visualization, were performed using the Seurat package in R as described in the package SCTransform tutorial (version 4.3.0) (Hao et al., 2021). Briefly, cells to be included in the analysis were required to have at least 4000 and at most 45,000 unique molecular identifiers and expressed between 1000 and 7500 unique genes. In addition, cells were excluded if more than 10% of the determined RNA sequences mapped to mitochondrial genes or if they were classified as doublets by scds (Bais and Kostka, 2020). In total, 7728 cells from two RP culture samples, 1875 cells from RPM cultures, 1562 cells from RPM (T58A) cultures, 3401 cells from RPM tumors, and 3463 cells from RPM (T58A) tumors passed these filters for quality. Samples were integrated using Harmony with the top 15 principal components and were run with up to 50 iterations of the algorithm (Korsunsky et al., 2019). Dimensionality reduction was achieved by creating a Uniform Manifold Approximation and Projection of the first 15 dimensions of harmony integration. Clustering resolution was set at 0.1 for the integrated dataset, resulting in three clusters. Subclustering was conducted in a similar manner and involved re-integration of cells from the neuroendocrine cluster and used identical parameters. Cell-type annotation was called using the Azimuth functionality within the Seurat package, followed by manual inspection, and enrichment analysis was conducted using enrichR (Chen et al., 2013). A lung atlas containing 584,884 cells was obtained for Azimuth-mediated reference mapping from the HubMap portal (Bais and Kostka, 2020; Deprez et al., 2020; Goldfarbmuren et al., 2020; Habermann et al., 2020; Madissoon et al., 2019; Morse et al., 2019; Schupp et al., 2021; Travaglini et al., 2020; Vieira Braga et al., 2019).

Total RNA from the primary and metastatic tumors was isolated using Direct-zol RNA Miniprep kit (Zymo Research). Following RNA isolation, total RNA integrity was checked using a 2100 Bioanalyzer (Agilent Technologies). RNA concentrations were measured using the NanoDrop system (Thermo Fisher Scientific). Preparation of RNA sample library and RNA-seq were performed by the Genomics Core Laboratory at Weill Cornell Medicine or Northwestern University. Messenger RNA was prepared using the TruSeq Stranded mRNA Sample Library Preparation kit (Illumina), according to the manufacturer’s instructions. The normalized libraries were pooled and sequenced on Illumina NovaSeq 6000 sequencer with paired-end 50 cycles. Sequencing adapters were trimmed, and quality metrics were generated using fastp (Chen et al., 2018). Transcript abundance was quantified by mapping sequencing reads to a reference transcriptome (GENCODE v42, hg38) using Salmon (Patro et al., 2017).

Identification of differentially expressed genes involved adjusting for sample identity and MYC expression status when appropriate and was achieved using model-based analysis of single-cell transcriptomics (Finak et al., 2015). Gene set enrichment analysis was conducted by mapping differential expression results to hallmark pathways using clusterProfiler (version 4.0) (Liberzon et al., 2015; Wu et al., 2021). Cell-level enrichment scores for top 20 cluster markers were previously identified (Chan et al., 2021) using the AddModuleScore functionality of Seurat. Patterns of MYC and NEUROD1 co-expression were limited to cells in which at least one MYC and one NEUROD1 transcript were detected post-processing.

Processed single-cell RNA-seq data of primary SCLC were obtained from the cellxgene portal (Chan et al., 2021). Bulk RNA-seq data for human SCLC tumors (reads per kilobase per million mapped reads) were obtained from cBioPortal. Subtype annotations for 45 patients were obtained from previously published classifications (Rudin et al., 2019). Deconvolution of bulk RNA-seq data was achieved through BayesPrism (version 2.0) (Chu et al., 2022). Briefly, single-cell transcriptomes from SCLC-A, SCLC-N, SCLC-P, and lung adenocarcinoma (LUAD; ‘NSCLC’) samples were annotated as tumor cells. For deconvolution, aggregated tumor annotations were provided as cell-type labels, and specific tumor types (e.g. SCLC-A) were provided as tumor states. BayesPrism received inputs from single-cell annotations and associated count matrix, xenograft, and primary tumor bulk RNA-seq data and was run with outlier.cut=0.01 and outlier.fraction=0.1. Analysis of resultant cell-type fractions was limited to SCLC-A, SCLC-N, and SCLC-P. Estimated cell-type fractions were converted to proportion metrics within the three SCLC subtypes and visualized for concordance with published annotations.

Group sample sizes for all in vivo experiments were estimated based off of our previous studies, with at least five animals per experimental arm being included for purposes of statistical testing (Huang et al., 2015; Huang et al., 2014; Chen et al., 2019; Chen et al., 2016; Unni et al., 2015). For mouse experiments, animals were randomly assigned to each group, and no animals were excluded from the analyses. All statistical tests are two-sided. No adjustments were made for multiple comparisons. The relevant investigators (HJC, EEG, and KZ) were blinded to experimental allocations among different experimental arms. For all parametric statistical analyses, data were determined to be normally distributed by the D’Agostino-Pearson test. For comparison between experimental and control groups at a specific time point or tissue site in Figures 2 and 4, two-sided Student’s t-tests, one-way or two-way ANOVA tests, Fisher’s exact tests, and two-sided Kolmogorov-Smirnov tests were used in GraphPad Prism. Calculated p-values at or below 2.2×10⁻¹⁶ were reported as p<2.2 × 10⁻¹⁶, in line with statistical reporting in R.