Abstract
Neuroendocrine to nonneuroendocrine plasticity supports small cell lung cancer (SCLC) tumorigenesis and promotes immunogenicity. Approximately 20% to 25% of SCLCs harbor loss-of-function (LOF) NOTCH mutations. Previous studies demonstrated that NOTCH functions as a SCLC tumor suppressor, but can also drive nonneuroendocrine plasticity to support SCLC growth. Given the dual functionality of NOTCH, it is not understood why SCLCs select for LOF NOTCH mutations and how these mutations affect SCLC tumorigenesis. In a CRISPR-based genetically engineered mouse model of SCLC, genetic loss of Notch1 or Notch2 modestly accelerated SCLC tumorigenesis. Interestingly, Notch-mutant SCLCs still formed nonneuroendocrine subpopulations, and these Notch-independent, nonneuroendocrine subpopulations were driven by Runx2-mediated regulation of Rest. Notch2-mutant nonneuroendocrine cells highly express innate immune signaling genes including stimulator of interferon genes (STING) and were sensitive to STING agonists. This work identifies a Notch-independent mechanism to promote nonneuroendocrine plasticity and suggests that therapeutic approaches to activate STING could be selectively beneficial for SCLCs with NOTCH2 mutations.
A genetically engineered mouse model of NOTCH-mutant SCLC reveals that nonneuroendocrine plasticity persists in the absence of NOTCH, driven by a RUNX2-REST–dependent pathway and innate immune signaling.
Introduction
Small cell lung cancers (SCLC) are high-grade neuroendocrine tumors that account for approximately 15% of lung cancers (1–3). SCLCs are characterized by near universal loss of the RB1 and TP53 tumor-suppressor genes (4–6), and approximately 70% of SCLCs highly express the neuroendocrine transcription factor ASCL1 (4, 7, 8), which is required for SCLC tumorigenesis (9) and a dependency in a subset of SCLC cell lines (10–12). NOTCH receptors and target genes are repressed in nearly all ASCL1-positive SCLCs (4), suggesting that NOTCH and ASCL1 oppose each other during SCLC tumorigenesis (13, 14). Approximately 25% of SCLCs harbor mutually exclusive loss-of-function (LOF) mutations in NOTCH receptors (NOTCH1, NOTCH2, NOTCH3, and NOTCH4; ref. 4), demonstrating that genetic loss of NOTCH is selected during SCLC tumorigenesis and nominating NOTCH as a tumor-suppressor gene in SCLC.
Functional studies in SCLC cell lines and SCLC genetically engineered mouse models (GEMM) support the role of NOTCH as an SCLC tumor suppressor (4, 13). Re-expression of transcriptionally active NOTCH1 (NOTCH1-ICD) or NOTCH2 (NOTCH2-ICD) in the Rb1−/−, Trp53−/−, Rbl2−/− (RPP, also known as RPR2) SCLC GEMM, which normally forms SCLC tumors of the ASCL1 + subtype, dramatically inhibits SCLC tumor formation, demonstrating that low NOTCH activity is required for SCLC tumor initiation (4). The SCLCs that escape tumor suppression by NOTCH and form with high NOTCH activity lose ASCL1 expression and neuroendocrine differentiation, demonstrating that NOTCH functionally opposes the ASCL1-driven neuroendocrine phenotype in SCLC. In support of these in vivo findings, re-expression of NOTCH in SCLC cell lines blocks cellular proliferation and neuroendocrine differentiation (4, 13, 15, 16). These data, together with selection for LOF NOTCH mutations in human SCLC tumors, support a tumor-suppressive role for NOTCH during SCLC tumorigenesis.
More recent data suggest that NOTCH has a dual role in SCLC and can function as both a suppressor and promoter of SCLC tumorigenesis (15). In the RPP SCLC GEMM, NOTCH-active nonneuroendocrine subpopulations of cells highly express the canonical NOTCH target gene HES1. These HES1-positive cells have decreased expression of ASCL1, and other neuroendocrine markers can be induced by canonical DLL4–NOTCH signaling, and the NOTCH target gene REST is necessary and sufficient for their formation (15). Functionally, their presence promotes the growth of the bulk ASCL1-positive neuroendocrine cells in the tumor. These HES1-positive nonneuroendocrine cells are more resistant to standard chemotherapies compared with their neuroendocrine counterparts, which is supported by human data showing increased NOTCH expression in CDX models from patients with SCLC after chemotherapy resistance (17). In addition, c-MYC can drive plasticity toward a nonneuroendocrine state during tumor evolution with high NOTCH and REST expression (18). Together, these data suggest that NOTCH also has a tumor-promoting function during SCLC tumor maintenance.
Several studies have now demonstrated that SCLCs are highly plastic and can adopt distinct neuroendocrine and nonneuroendocrine fates (15, 16, 18–23). Nonneuroendocrine adherent/mesenchymal subpopulations within SCLC tumors promote SCLC tumorigenesis and metastasis (19) and share many common features with HES1-positive nonneuroendocrine cells (15). These subpopulations are present in SCLC GEMMs, spontaneously form in culture from primary cells derived from SCLC GEMM tumors, display loss of the oncogenic neuroendocrine regulators ASCL1 and MYCL, and can promote metastasis of their neuroendocrine counterparts (19). Moreover, two recent articles found that nonneuroendocrine SCLCs have restored antigen presentation and are enriched for antitumor immune responses with PD-1/PD-L1 immunotherapy (21, 22).
Our laboratory recently developed a CRISPR-based SCLC GEMM using an adenovirus encoding Cre-recombinase and sgRNAs targeting Rb1, Trp53, and Rbl2 (RPP) to intratracheally inject into lox-stop-lox (LSL)-Cas9 mice. Somatic CRISPR/Cas9 editing of RPP in the lung results in SCLC tumors (referred to hereafter as CRISPR RPP GEMM; ref. 16). Our model has the capability to inactivate a candidate target gene of interest at tumor initiation and study its role during SCLC tumorigenesis. The selection for LOF NOTCH mutations in human SCLC is paradoxical given data demonstrating that NOTCH functions both as a tumor suppressor and tumor promoter. We therefore used our CRISPR RPP GEMM to generate Notch1-mutant or Notch2-mutant SCLCs and study the functional role and consequences of LOF Notch1 or Notch2 mutations during SCLC tumorigenesis.
Materials and Methods
Cell lines and cell culture
NCI-H1694 (obtained November 2018) were obtained from the ATCC. CORL47 were obtained from Sigma (November 2018). 293FT were from Dr. Myles Brown laboratory at DFCI. NCI-H1694 cells were maintained in DMEM/F12 media supplemented with HITES [10 nmol/L hydrocortisone (Sigma Aldrich; #H0135), Insulin-Transferrin-Selenium (Gemini; #400-145), and 10 nmol/L β-estradiol (Sigma Aldrich; #E2257)]. CORL47 cells were maintained in RPMI medium. 293FT cells were maintained in DMEM media. The mouse cell lines derived from genetically engineered SCLC mouse tumors (see below for description of cell line generation) were maintained in RPMI-1640 media supplemented with HITES. All media were supplemented with 10% FBS (Gemini), 100 U/mL penicillin, and 100 μg/mL streptomycin except for NCI-H1694 cells, where the media were supplemented with 5% FBS, 100 U/mL penicillin, and 100 μg/mL streptomycin. Early passage cells of all cell lines listed above were tested for Mycoplasma (Lonza; #LT07-218) and then were frozen using Bambanker's freezing media (Bulldog Bio). The cells were last tested for Mycoplasma in June 2021. Cells were then maintained in culture for <3 months, at which point, new early passage vials were thawed. Nearly all experiments were performed in cells within the first 6 weeks after thawing. Where indicated, the following chemicals (stored at −20°C) were also added to the media as indicated in the text: doxycycline (Sigma, stock 1 mg/mL in H2O), Compound E (EMD Millipore; # 65790, stock 2 mmol/L in DMSO), ADU-S100 (ChemieTek; #CT-ADU-S100, stock 50 mmol/L in H2O), and 2′3′-cGAMP (InvivoGen; #tlrl-nacga23, stock 10 mg/mL in H2O).
Adenovirus production and purification
Five microgram of the adenovirus vector (pAd/PL Invitrogen; #V494-20) containing the desired sgRNA sequences and Cre recombinase expression cassette (see Adenoviral Cloning section in Supplementary Methods) was digested with PacI (New England Biolabs) for 2 hours at 37°C according to the manufacturer's instructions and column purified using QIAGEN's gel extraction kit. One microgram of PacI-digested pAd/PL was transfected into 1.5 × 106 293AD cells plated on a 6-cm tissue culture dish using Lipofectamine 2000. The following day, the media were exchanged, and subsequently every 48 hours thereafter. Once 293AD cells showed evidence of adenovirus production (determined by comet formation with lysis), the cells and supernatant were harvested, which were then subjected to 4 freeze-thaw cycles by alternating between an ethanol dry ice bath and 37°C. Cell debris was removed by centrifugation, and the supernatant was collected, passed through a 0.45 μm filter, aliquoted, and frozen at −80°C until use.
To generate high titer adenovirus for in vivo experiments, adenovirus was generated as described above. Fifty microliter of the adenovirus stock was added to each 10-cm tissue culture dish of 293FT cells plated at 3 × 106 cells per dish (four 10 cm dishes in total for each purification). When 293FT cells showed evidence of adenovirus production, as determined by cell rounding and partial detachment (∼48–72 hours after addition of adenoviral stock), the cells were collected, and adenovirus was purified using Virabind Adenovirus Purification Kit (VPK-100). The purified adenovirus was then dialyzed into PBS at 4°C overnight. Adenovirus was titered using the QuickTiter Adenovirus Quantitation Kit (Cell Biolabs; ##VPK106) according to the manufacturer's instructions.
Mouse experiments
All mouse experiments complied with NIH guidelines and were approved by Dana-Farber Cancer Institute Animal Care and Use Committee (DFCI, protocol 19-009).
Intratracheal injections
Intratracheal (IT) injections were performed as described previously (24). Briefly, mice were anesthetized with ketamine and xylazine, and pedal reflexes were monitored to ensure adequate anesthesia. Mice were maintained on a heated stage at 37°C while anesthetized. Mice were hung on stage with their top incisors and intubated with a 22-gauge 1 inch catheter (Thermo Fisher Scientific; #1484120). Once intubated, adenovirus (4 × 108 VP/mouse) in a total volume of 75 μL (diluted in PBS) was added to the catheter and subsequently inhaled by the mice.
Generation of GEMMs of SCLC using CRISPR/Cas9
For all experiments, LSL-Cas9 BL6NJ mice (The Jackson Laboratory; No. 026556) were crossed with LSL FLUC mice (The Jackson Laboratory; No. 005125) yielding progeny that were heterozygous for LSL-Cas9/+, LSL-FLUC/+, whose genotypes were confirmed (Transnetyx). Of note, the intention was to use the firefly luciferase for bioluminescence imaging (BLI) to track tumor burden over time. However, at early time points comparing BLI with MRI imaging, we found that BLI imaging yielded several false negatives for unclear reasons. Therefore, we carried out the study with lung MRI imaging only. Three- to 4 month-old transgenic LSL-Cas9/+, LSL-FLUC/+ were intratracheally injected with adenovirus (4 × 108 VP/mouse) encoding effective sgRNAs targeting Rb1, Trp53, and Rbl2 (RPP, also known as RPR2) and CMV-Cre recombinase that also encoded either an effective sgRNA targeting Notch1#1 (sgNotch1 RPP), Notch2#1 (sgNotch2 RPP), Ascl1#4 (sgAscl1 RPP), or a nontargeting sgRNA as a control (sgControl RPP). MRI of the lungs was performed on mice beginning 8 months after IT injection and was performed monthly thereafter until the mice became symptomatic defined as weight loss >15%, dyspnea, or had lung tumor volumes >200 mm3, and were then euthanized. Tumor volumes were calculated by lung MRIs (see Mouse MRI Imaging in Supplementary Methods).
Upon euthanization, approximately one third of the lung tumor was immediately flash frozen in dry ice for subsequent DNA and RNA analysis, approximately one third was fixed in 10% formalin for 24 hours and then stored in 70% ethanol, and the remaining one third was used to establish cell lines (see Materials and Methods section). Liver was also harvested and fixed as described above to determine metastatic burden. The tissues were then embedded in paraffin. Slides were made for hematoxylin and eosin (H&E) and IHC staining, and H&E slides were analyzed by Dr. Rod Bronson for diagnosis.
For experiments in Supplementary Fig. S2F only, LSL-Cas9 BL6J homozygous mice were used (The Jackson Laboratory; No. 026175).
Generation of cell lines from mouse SCLC tumors
To generate cell lines from tumors, LSL-Cas9/+, LSL-FLUC/+ mice (as described above) that grew tumors and were at their endpoint as defined above were euthanized, and their tumors were quickly extracted, washed in ice-cold PBS, and minced several times using an ethanol-sterilized razor blade. Three milliliter of collagenase/hyaluronidase (Stem cell biology #07912) diluted 1:10 in complete RPMI (10% FBS, penicillin/streptomycin, and HITES) and 1 mL dispase (Corning; #354235) was added to the tumor and incubated at 37°C for 20 to 40 minutes (until most tumor cells were dissociated). The cells were then collected, centrifuged at 1,000 rpm for 5 minutes, resuspended in RPMI HITES media, filtered through a 70 μm cell strainer (BD #352350), centrifuged again at 1,000 rpm for 5 minutes, resuspended in fresh complete RPMI HITES media, and placed in ultralow adherence tissue culture dishes (Corning; #3471). Media were subsequently replaced every 3 days.
SCLC adherent cell derivatives of the parental cell line were established by plating cells on tissue culture plates at 200,000 cells/mL for 4 days, and the supernatant containing suspension cells was removed and replaced with fresh complete media approximately every 3 days until the adherent cell lines were established.
RNA sequencing
For the RNA sequencing (RNA-seq) experiment using the SCLC tumors derived from the CRISPR-based SCLC GEMM (Fig. 3), tumors were harvested at necropsy and were flash-frozen. RNA was extracted using an RNeasy mini kit including a DNase digestion step according to the manufacturer's instructions, and RNA-seq was performed as described below.
For the RNA-seq experiment in mouse SCLC adherent cells (Fig. 5), cells were counted using Vi-Cell XR Cell Counter and plated at 200,000 cells/mL in one well of 6-well plate. Cells were harvested 72 hours later, and RNA was extracted as above.
Total RNA samples in each experiment were submitted to Novogene Inc. The libraries for RNA-seq are prepared using NEBNext Ultra II nonstranded kit. Paired-end 150 bp sequencing was performed on Novaseq6000 sequencer using S4 flow cell. Sequencing reads were mapped to the mm10 genome by STAR. Statistics for differentially expressed genes were calculated by DESeq2.
RNA-seq analysis
For gene set enrichment analysis (GSEA), software was downloaded from the gene set enrichment analysis website (http://www.broad.mit.edu/gsea/downloads.jsp). GSEA was performed using the “Hallmark,” cell-type signature gene sets, and neuroendocrine gene sets. Gene sets with an FDR < 0.25 and a nominal P value of <0.05 were considered significant.
For principal component analysis (PCA) and cluster analysis, top 500 genes in terms of the largest SD were subjected to PCA using the prcomp function of R software. Clustering and heatmap were generated using the heatmap.2 function in the gplots package of R software.
Transcriptional regulator and motif enrichment analyses were performed using epigenetic landscape in silico deletion analysis (LISA), 200 top upregulated and 200 top downregulated genes from the Notch-mutant versus Notch-WT adherent cell RNA-seq experiment were used as the input for Fig. 6A, our previous 631RPP adherent versus 631RPP suspension RNA-seq dataset was used for Fig. 6B (22), and Notch-mutant adherent cell line versus Notch-WT SCLC tumors were used for Supplementary Fig. S11C.
Analysis of publicly available cohort data
For the expression of MYC, MYCL, and MYCN, the processed RNA-seq, copy number variation, and single-nucleotide variant data were previously published (4). The samples were divided into two groups of NOTCH-mutant when there was at least one short variation observed in any NOTCH family genes. Otherwise, the sample was called NOTCH-WT. For the analysis in Fig. 3I–K, only samples with copy-number information were included (n = 29 total), and any samples with copy-number amplifications of MYC family members were excluded (n = 3), and therefore, 26 samples were analyzed. The gene expression of MYC, MYCL, and MYCN between two groups was represented in the bar plot. The Mann–Whitney test was used to calculate the P value between two distributions. The analysis in Supplementary Fig. S4A–S4C comparing NOTCH-mutant versus NOTCH-WT included all 81 SCLC human samples with RNA-seq data (4).
For the coexpression of RUNX1, RUNX2, REST, AXL, TAP1, POU2F3, ASCL1, NEUROD1, and INSM1 in SCLC tumors, the processed RNA-seq data were previously published (4), and data were downloaded from cBioPortal. All 81 samples were included in the analysis.
For the expression of STAT1, the RNA-seq of SCLC cell lines was from CCLE (25). The samples were divided into two groups of NOTCH-mutant and NOTCH-WT. The gene expression of STAT1 between two groups was represented in bar plot. The Mann–Whitney test was used to calculate the P value between two distributions.
Cistrome Data Browser was used to identify potential regulators of REST and the transcription factors and chromatin regulators binding to REST. For potential Rest regulators, 1 kb to mouse Rest transcriptional start site (chr5:77265493:77283696:NM_011263) was analyzed. For the transcription factors and chromatin regulators binding to REST, 1 kb to human REST transcriptional start site (chr4:56906899-56907899) was analyzed.
Mouse SCLC attachment assays
To determine which primary mouse SCLC cells could form attached nonneuroendocrine subpopulations, mouse SCLC suspension cells that were previously only grown in ultralow attachment flasks were counted on day 0 using a Vi-Cell XR Cell Counter and plated in tissue culture–treated 6-well plates at 200,000 cells/mL in 2 mL of complete media. After 3 days, all cells were collected, centrifuged, and replated in fresh media into the same well. After 6 days, representative images were acquired using brightfield microscopy with a 10× objective and then stained with crystal violet for visualization of the entire well.
For the attachment experiments with Compound E, mouse SCLC suspension cells that were previously only grown in ultralow attachment flasks were counted on day 0 using a Vi-Cell XR Cell Counter and plated in tissue culture–treated 6-well plates at 100,000 cells/mL in 2 mL with 1 μmol/L Compound E or DMSO. Four days later, representative images were acquired using brightfield microscopy with a 10× objective and then stained with crystal violet for visualization of the entire well. Attached cells were counted using a Vi-Cell XR Cell Counter and harvested for immunoblot analysis.
Stimulator of interferon genes agonist experiments
For the CXCL10 ELISA experiments, mouse SCLC adherent cells were counted on day 0 using a Vi-Cell XR Cell Counter and plated in tissue culture–treated 12-well plates at 300,000 cells/mL in 1 mL of complete media. Once the cells were attached, they were then treated with stimulator of interferon genes (STING) Agonist ADU-S100 (10 or 50 μmol/L), 2′3′-cGAMP (10 μg/mL), or ddH2O. The concentrations of ADU-S100 of 10 or 50 μmol/L were chosen based on the original publication from Aduro BioTech Inc. describing these cyclic dinucleotide derivatives (26). Twenty-four hours later, conditioned media from the cells were collected. Mouse CXCL10 ELISA (#DY466, R&D Systems) was performed according to manufacturer's instructions. The cell counting assays were performed as described above except that the cells were plated at 50,000 cells/mL in 1 mL of complete media and counted 72 hours later using a Vi-Cell XR Cell Counter. The combination Compound E and ADU-S100 CXCL10 ELISAs were performed as above with the following modification: K47 SCLC adherent cells were pretreated with 1 μmol/L Compound E or DMSO for 24 hours before the ADU-S100 was added.
Runx2 Rest sgRNA attachment experiments
K93 Notch-WT and K60 Notch-mutant mouse SCLC suspension cells were first infected with lentiviruses encoding an sgRNA targeting Runx2 (sgRunx2 #1,#2,#3), Rest (sgRest #1,#2,#3) or a nontargeting sgRNAs (nontargeting sgRNA #1,#2). Puromycin-resistant cells, indicative of successful infection with the sgRNA lentiviruses, were counted using the Vi-Cell XR Cell Counter and plated in tissue culture–treated 6-well plates at 100,000 cells/mL in 2 mL in complete media. After 4 days, representative images were acquired using brightfield microscopy with a 10× objective. Both the attached population and suspension populations in the same well were counted independently, and an identical plate was used to count the total cells (both suspension and adherent) using a Vi-Cell XR Cell Counter. The fraction of attached cells relative to the total number of cells (adherent + suspension) is plotted to account for any differences in proliferation during the 4-day assay. The data were then normalized to the nontargeting sgRNA #1. Isolated attached cells (from a separate plating experiment with identical design) were harvested for immunoblot analysis.
Statistical analysis
For the RNA-seq experiments and GSEA in Figs. 3 and 5, and Supplementary Fig. S11, statistical significance was calculated using FDR corrected for multiple hypothesis testing where q value of <0.25 is considered statistically significant.
For the in vivo studies using sgNotch1 RPP, sgNotch2 RPP, sgAscl1 RPP, and sgControl RPP in Fig. 1, the χ2 test for trend was used to determine the P value of percentage of mice with lung lesions in Fig. 1E, and the Gehan–Breslow–Wilcoxon test was used to determine the P value of the Kaplan–Meier survival analysis in Fig. 1G.
For all other experiments, statistical significance was calculated using unpaired, two-tailed Student t test. P values were considered statistically significant if the P value was <0.05. For all figures, *, P < 0.05; **, P < 0.01; ***, P < 0.001; and ****, P < 0.0001. Error bars represent SEM unless otherwise indicated.
Data and materials availability
Data from RNA-seq experiments will be deposited in the GEO database prior to publication. All fragments per kilobase million (FPKM) RNA-seq data are included in the Supplementary Tables. All other data and materials can be requested from the corresponding author upon reasonable request.
Results
Genetic inactivation of Notch1 or Notch2 promotes SCLC tumorigenesis in vivo
Because LOF NOTCH mutations are found in approximately 25% of human SCLCs, we hypothesized that genetic inactivation of NOTCH1 or NOTCH2 would accelerate SCLC tumorigenesis. To test this, we used our SCLC CRISPR RPP (also known as RPR2) GEMM (16) to generate autochthonous SCLCs that were Notch1-mutant, Notch2-mutant, or Notch-WT (Fig. 1A). We also used the same approach to generate autochthonous SCLCs with Ascl1 deletion (Fig. 1A), as we hypothesized that Ascl1 loss would have the opposite effect of Notch inactivation, and also to compare negative selection in our CRISPR RPP GEMM with the traditional RPP GEMM where Ascl1 is required for SCLC tumorigenesis (9). We first validated that individual sgRNAs targeting Notch1, Notch2, or Ascl1 effectively knocked out their intended target genes. We then generated adenoviruses that encoded Cre recombinase and sgRNAs targeting Rb1, Trp53, and Rbl2 (RPP) and Notch1 (sgNotch1 RPP), Notch2 (sgNotch2 RPP), Ascl1 (sgAscl1 RPP), or a nontargeting control (sgControl RPP). We confirmed that adenoviruses transduced into mouse embryonic fibroblasts (MEF) could simultaneously knock out Notch1, Notch2, or Ascl1 along with RPP (Fig. 1B and C).
These sgNotch1 RPP, sgNotch2 RPP, sgAscl1 RPP, or sgControl RPP adenoviruses were then introduced into the lungs of LSL-Cas9 mice by IT injection (Fig. 1D). Beginning 8 months after injection, we performed monthly lung MRIs to monitor lung tumors. Lung tumors were detected earlier (Fig. 1E, F, and H; Supplementary Fig. S1A) in mice injected with sgNotch1 RPP or sgNotch2 RPP compared with mice that received sgControl RPP. Eventually, approximately 100% of mice injected with sgNotch1 RPP, sgNotch2 RPP, and sgControl RPP developed lung tumors. Although not statistically significant by Kaplan–Meier survival estimate, there was a trend toward decreased overall survival in sgNotch1 RPP and sgNotch2 RPP mice compared with sgControl RPP mice (Fig. 1G; Supplementary Fig. S1B). Consistent with a previous study in the traditional RPP GEMM (9), tumor onset was dramatically delayed in sgAscl1 RPP mice, which translated to an increased overall survival compared with sgControl RPP mice (Fig. 1E–H). Only 42.8% (3/7 mice) of the mice injected with sgAscl1 RPP adenovirus developed SCLC lung tumors (Fig. 1E and F; Supplementary Fig. S2D). Zero of 15 mice in an additional cohort injected with three different sgRNAs targeting ASCL1 formed SCLCs, demonstrating that Ascl1 is required for SCLC tumorigenesis and that escape after Ascl1 inactivation is a rare event (Fig. 1E; Supplementary Fig. S2E). Collectively, these data demonstrate that loss of Notch1 or Notch2 modestly promotes SCLC tumorigenesis, whereas loss of Ascl1 inhibits SCLC tumorigenesis.
SCLCs select for LOF Notch mutations and inframe deletions in Ascl1 that restore partial Ascl1 expression
We hypothesized that LOF Notch1 and Notch2 mutations were positively selected, and LOF Ascl1 mutations were negatively selected during SCLC tumorigenesis. To test this, we performed CRISPR amplicon sequencing on SCLC tumors at the genomic loci targeted by Notch1, Notch2, and Ascl1 sgRNAs. Consistent with a tumor-suppressive role for Notch2, all sgNotch2 RPP tumors evaluated had insertions or deletions (indels), leading to LOF frameshift mutations in Notch2 (Fig. 2A, E, and F; Supplementary Fig. S2C). Interestingly, all three tumors that formed in mice injected with the sgAscl1 RPP harbored large inframe indels (24, 48, and 57 nucleotides) in roughly half of the Ascl1 sequencing reads, whereas the remaining sequencing reads had an LOF indel (Fig. 2B, E, and F; Supplementary Fig. S2D). Large inframe indels are a rare event during CRISPR/Cas9 editing and did not occur in MEFs infected with the Ascl1 sgRNA (Supplementary Fig. S2F), demonstrating strong selective pressure to retain 1 functional copy of Ascl1 during SCLC tumorigenesis. Finally, 11 of 15 sgNotch1 RPP tumors examined had indels that led to complete LOF frameshift mutations in Notch1 (Fig. 2C, E, and F; Supplementary Fig. S2B). Twenty-seven percent (4 of 15) of the sgNotch1 RPP lung tumors examined had larger inframe deletions in roughly half of the sequencing reads, whereas the remaining sequencing reads harbored an LOF indel (Fig. 2D–F; Supplementary Fig. S2B). This was unexpected given that the survival of the mice with these tumors was the same as the tumors with complete LOF Notch1 frameshift mutations (Supplementary Fig. S1B and S1C), and shorter than the sgControl RPP tumors (Supplementary Fig. S1B), suggesting that in some mice, there was selective pressure for SCLCs to have the capability to restore Notch1 expression.
We next asked whether the tumors that formed in sgNotch1 RPP, sgNotch2 RPP, and sgAscl1 RPP mice had histologic and neuroendocrine features of SCLC. Similar to sgControl RPP tumors, sgNotch1 RPP and sgNotch2 RPP tumors had histologic features of SCLC and high Ascl1 expression (Fig. 2G; Supplementary Fig. S2A–S2C). IHC for Notch1 or Notch2 showed that sgControl RPP tumors had subpopulations of Notch1- or Notch2-positive cells, whereas sgNotch1 RPP or sgNotch2 RPP tumors did not (Fig. 2H) consistent with previous work, demonstrating that Notch is expressed in only a small fraction of cells within SCLC RPP GEMMs (15, 16). Consistent with the presence of inframe Ascl1 indels, the tumors that formed in the sgAscl1 RPP mice had histologic features of SCLC and retained some Ascl1 expression (Fig. 2B and G; Supplementary Fig. S2D).
Loss of NOTCH drives MYCL expression in SCLC
To understand the transcriptional changes in tumors inactivated for Notch or Ascl1, we performed bulk RNA-seq on sgNotch1 RPP, sgNotch2 RPP, sgAscl1 RPP, and sgControl RPP lung tumors. Based on our Notch1 CRISPR amplicon sequencing results, we segregated sgNotch1 RPP tumors into two groups: (i) sgNotch1 RPP LOF tumors, where both copies of Notch1 harbored indels that resulted in LOF, or (ii) sgNotch1 RPP heterozygous (Het) tumors, where one copy of Notch1 harbored an inframe indel. Hierarchical clustering showed that sgNotch1 RPP LOF tumors were transcriptionally distinct from sgNotch1 RPP Het tumors, suggesting that this selection had functional consequences on transcription (Fig. 3A; Supplementary Table S1). Consistent with the dramatic effect of Ascl1 deletion on SCLC tumorigenesis, tumors that formed in the sgAscl1 RPP mice were most distinct from all other tumor genotypes (Fig. 3A). Similar to sgControl RPP and previous studies with the RPP GEMM (9, 27–29), the sgNotch1 RPP and sgNotch2 RPP tumors had high expression of Ascl1 and Insm1, but not Neurod1 or Pou2f3 (Fig. 3B).
GSEA of the RNA-seq data revealed that MYC and E2F target gene signatures were highly enriched in both sgNotch2 RPP and sgNotch1 LOF RPP tumors compared with sgControl RPP tumors, with a greater enrichment in sgNotch2 RPP tumors (Fig. 3C and D; Supplementary Table S2), suggesting that MYC pathway activation is linked to Notch inactivation in SCLC. To study this and other mechanisms by which loss of Notch drives SCLC tumorigenesis, we generated SCLC cell lines from sgNotch1 RPP, sgNotch2 RPP, and sgControl RPP tumors. Similar to ASCL1+ human SCLC cell lines, cells derived from sgNotch1 RPP, sgNotch2 RPP, and sgControl RPP tumors grew as clusters of cells in suspension when plated on ultralow attachment dishes and highly expressed Ascl1 (Fig. 3E; Supplementary Fig. S3A). As expected, Notch1 or Notch2 expression was absent in sgNotch1 RPP and sgNotch2 RPP tumors, respectively. Expression of Hes1, a canonical Notch target gene, was nearly completely lost in all cell lines derived from sgNotch1 RPP or sgNotch2 RPP tumors, demonstrating that loss of a single Notch receptor paralog markedly decreases Notch target gene expression. Pan-Myc expression was increased in nearly all sgNotch1 RPP and sgNotch2 RPP cell lines compared with the sgControl RPP cell lines (Fig. 3E). RT-qPCR for specific Myc paralogs demonstrated that Mycl was highly upregulated in four of the six Notch-mutant primary cell lines compared with sgControl RPP cell lines (Fig. 3F–H; Supplementary Fig. S3B–S3E) consistent with previous studies showing SCLC RPP GEMM models highly express L-Myc and not the other Myc paralogs (9, 27–29).
To further explore the link between Notch mutations and L-Myc expression, we analyzed data of 29 SCLC human tumors where both RNA-seq data and copy-number alteration data were available (4). Three of these tumors had MYC paralog amplifications and were excluded. Consistent with our data in mouse tumors, we found that human NOTCH-mutant SCLCs had higher expression of MYCL1, but not MYC and MYCN paralogs, compared with NOTCH-WT tumors (Fig. 3I–K). Analysis of the RNA-seq data from all 81 human SCLC tumors (4) showed a similar trend (Supplementary Fig. S4A–S4C).
We then asked whether Notch activity regulates L-Myc expression. To test this, we isolated subpopulations of adherent Notch-WT mouse SCLC derived from sgControl RPP tumors that had high levels of active Notch (Fig. 3L). Blocking NOTCH activity with the gamma secretase inhibitor Compound E (Fig. 3L; Supplementary Fig. S4D) increased L-Myc expression (Fig. 3M). Conversely, expression of a DOX-ON NOTCH1-ICD (transcriptionally active NOTCH) to induce NOTCH activity (Supplementary Fig. S4E) in the CORL47 human SCLC cell line, which has repressed NOTCH1 expression and activity at baseline (Fig. 3N), decreased MYCL and ASCL1 expression (Fig. 3N and O; Supplementary Fig. S4F). Together, these data demonstrate that low NOTCH activity is required for high MYCL expression in SCLC and suggest that NOTCH inactivation is a potential mechanism (and likely indirect given NOTCH canonically functions as a transcriptional activator) to increase MYCL1 expression in SCLC in the absence of a MYCL1 copy-number amplification.
Nonneuroendocrine adherent SCLC subpopulations can form in the absence of NOTCH activity
NOTCH activity promotes nonneuroendocrine plasticity within SCLC tumors. These cells are identified by HES1 positivity, and their formation is dependent on the NOTCH target gene REST (15). We performed HES1 IHC to ask whether these HES1-positive cells were reduced or absent in the sgNotch1 RPP or sgNotch2 RPP tumors. Consistent with previous observations, sgControl RPP tumors contained HES1-positive subpopulations (Fig. 4A and B). In contrast, HES1-positive cells were either markedly reduced or completely absent in the majority of sgNotch1 RPP and almost all sgNotch2 RPP tumors (Fig. 4A and B), consistent with our findings in Notch-mutant primary cell lines (Fig. 3E). This was paradoxical given previous data demonstrating a role for these HES1-positive subpopulations in promoting SCLC tumor growth (15) and our data demonstrating that inactivation of Notch, which ablates these HES1-positive subpopulations, accelerates SCLC tumorigenesis (Fig. 1). One explanation to reconcile these observations was that Notch-mutant tumors selected for Notch-independent mechanisms to promote the formation of nonneuroendocrine adherent cells. Alternatively, it is possible that nonneuroendocrine subpopulations were dispensable for Notch-mutant SCLC tumorigenesis.
To investigate whether nonneuroendocrine subpopulations exist in Notch-mutant tumors, we plated sgNotch1 RPP, sgNotch2 RPP, and sgControl RPP primary cell lines on tissue culture plastic. Four days later, some primary cell lines spontaneously adhered to tissue culture plastic and exhibited a mesenchymal-like phenotype (Fig. 4C and D), which has been observed previously (19). Sixty percent of cell lines derived from sgControl RPP tumors and 50% of sgNotch2 RPP tumors formed these adherent subpopulations, whereas only one of seven of the sgNotch1 RPP cell lines formed adherent cells (Fig. 4C and D). We then established these adherent cells as distinct cell lines from their suspension counterparts. CRISPR-amplicon sequencing for Rb1, Notch1, and Notch2 of neuroendocrine suspension cells and nonneuroendocrine adherent cells pairs confirmed that the adherent cell subpopulations harbored identical CRISPR-mediated indels as their suspension neuroendocrine counterparts confirming that they were tumor cells derived from neuroendocrine suspension cells and not contaminating cells from the tumor microenvironment (Supplementary Figs. S5 and S6). Although these data suggest that Notch1 is important for the formation of these adherent cell subpopulations, it is surprising that the Notch-mutant tumors formed adherent subpopulations at all, particularly the sgNotch2 RPP tumors, which formed adherent subpopulations at the same frequency as the sgControl RPP tumors but had near-absence of HES1-positive subpopulations in tumors (Fig. 4A and B).
Adherent cell lines derived from sgControl RPP, sgNotch1 RPP, and sgNotch2 RPP tumors had reduced expression of Ascl1 and Mycl compared with their suspension counterparts (Fig. 4E; Supplementary Fig. S7A–S7C). In sgControl RPP adherent cell lines, Notch, Hes1, and Rest expression was increased relative to their neuroendocrine suspension counterparts (Fig. 4E; Supplementary Fig. S7D and S7E), which was observed previously (15). Consistent with our data in Notch-mutant SCLC tumors, the sgNotch1 RPP and sgNotch2 RPP adherent subpopulations had markedly reduced expression of Hes1, demonstrating that canonical Notch signaling was decreased in Notch-mutant adherent cells compared with Notch-WT adherent cells (Fig. 4E–H; Supplementary Fig. S7D). Surprisingly, Rest mRNA expression was increased in Notch-mutant adherent cells compared with their suspension counterparts, and expressed at similar levels when compared with sgControl RPP adherent cell lines despite Notch inactivation and their reduction in Hes1 expression (Fig. 4F–J; Supplementary Fig. S7E), suggesting selection for Notch-independent mechanisms to activate Rest in sgNotch1 RPP and sgNotch2 RPP tumors. Loss of Ascl1 and gain of Rest persisted after replating adherent subpopulations on ultralow attachment dishes, suggesting that gene expression changes in adherent cells were not simply a consequence of plating the neuroendocrine cultures on tissue culture plastic (Supplementary Fig. S7F and S7G).
One trivial explanation for this would be that Rest expression is regulated by residual Notch activity in the Notch-mutant adherent cells. To test this, we used Compound E to block any residual Notch activity in the Notch-mutant suspension cells (Fig. 4K). Compound E treatment did not block the formation of adherent subpopulations in all Notch-mutant cell lines and only partially blocked the formation of adherent subpopulations in one of the three Notch-WT cell lines (Fig. 4L; Supplementary Fig. S8A and S8B), demonstrating that Notch activity is not required for adherent cell formation in all Notch-mutant cell lines and some Notch-WT cell lines. Furthermore, Rest expression in newly formed adherent cells was not significantly altered by Compound E, which effectively blocked all Notch activity and decreased Hes1 levels (Fig. 4K), consistent with a recent publication demonstrating that Notch activity is not required for Rest expression (30). Consistent with prior work (15), re-expression of transcriptionally active NOTCH markedly induced REST expression and promoted an adherent cell phenotype in the CORL47 cells (Supplementary Fig. S4G and S4H), demonstrating that increasing NOTCH activity is sufficient to induce REST expression and nonneuroendocrine plasticity. However, our data clearly suggest that additional NOTCH-independent mechanisms are capable of inducing REST and promoting nonneuroendocrine plasticity in SCLC.
Notch2-mutant adherent subpopulations have enriched IFN and STING signaling compared with Notch-WT adherent subpopulations
To examine whether loss of NOTCH might alter the nonneuroendocrine phenotype, we performed RNA-seq of adherent cell lines derived from three Notch-WT tumors compared with 3 Notch-mutant tumors (Fig. 4F). PCA and unsupervised hierarchical clustering analysis revealed significant differences between the Notch-mutant adherent cells compared with the Notch-WT adherent cells (Fig. 5A and B; Supplementary Table S3). Consistent with their genotype, Notch-mutant adherent cells had lower expression of canonical Notch target genes, demonstrating that inactivation of a single Notch receptor paralog is sufficient to decrease canonical Notch target gene expression (Fig. 5C), with the exception being REST, which we found is regulated in a Notch-independent manner (Fig. 4K and L).
To better understand the differences between Notch-mutant and Notch-WT adherent cells, we performed GSEA analysis using cell type signature gene sets. Notch-WT adherent cells were more similar to neurons and enteroendocrine cells (Fig. 5D; Supplementary Fig. S9A; Supplementary Table S4), whereas Notch-mutant adherent cells were most similar to mesenchymal and fibroblast cells (Fig. 5E; Supplementary Fig. S9B), suggesting that Notch confines the plasticity of SCLC within the neuroendocrine state, whereas loss of Notch selects for fibroblastic nonneuroendocrine plasticity. Consistent with this, the gene set for top 100 neuroendocrine markers (31) was decreased in the Notch-mutant adherent cells compared with Notch-WT adherent cells (Fig. 5F; Supplementary Fig. S9C).
GSEA for hallmarks gene sets revealed that Notch-mutant adherent cells were significantly enriched in interferon signaling, the inflammatory response, and JAK-STAT signaling (Fig. 5G; Supplementary Fig. S9D–S9F). Immunoblot analysis revealed that phospho-STAT1, total STAT1, and STING protein levels were selectively increased in the Notch2-mutant adherent cells compared with the Notch-WT adherent cells (Fig. 5H). IHC for STING expression in the limited set of NOTCH2-mutant human SCLCs that were available revealed that all NOTCH2-mutant tumors examined had high STING expression, whereas none of the NOTCH-WT tumors expressed STING (Fig. 5I). Although a limited sample size, these findings were intriguing as STING protein expression is largely absent in human SCLCs (20, 22). Also consistent with these findings, total STAT1 mRNA expression was modestly increased in NOTCH-mutant SCLC cell lines compared with NOTCH-WT SCLC cell lines (Supplementary Fig. S9G). Treatment of adherent cells with the STING agonist ADU-S100 dramatically induced CXCL10 production (a cytokine used to measure output of the STING pathway) in Notch2-mutant adherent cells to similar levels induced by the IFNβ-positive control (Fig. 5J). Consistent with this, ADU-S100 inhibited proliferation in Notch2-mutant adherent cells, but not Notch-WT adherent cells (Fig. 5K). Similarly, 2′3′-cGAMP, the endogenous second messenger that binds and stimulates STING (32), also selectively inhibited proliferation in Notch2-mutant adherent cells (Supplementary Fig. S10A), which was likely due to its ability to induce apoptosis in these cells (Supplementary Fig. S10B–S10G). CXCL10 production induced by ADU-S100 was significantly potentiated by blocking Notch activity with Compound E in K47 Notch-WT adherent cells, demonstrating that Notch activity represses STING pathway activation (Fig. 5L). Together, these data show that Notch2-mutant adherent cells are unique from Notch-WT adherent cells with high IFN signaling and high expression of components of the STING pathway, and are selectively vulnerable to STING agonists.
Runx2 regulates Rest and is necessary for neuroendocrine to nonneuroendocrine plasticity
Given that nonneuroendocrine adherent cells form in the absence of Notch activity (Fig. 4L), we performed Epigenetic LISA (33) using SCLC RNA-seq datasets to identify candidate master transcription factors that drive nonneuroendocrine plasticity in the absence of Notch. Rest was identified as a top regulator of downregulated genes when comparing Notch-mutant adherent cells with Notch-WT adherent cells and comparing Notch-mutant adherent cells with SCLC mouse tumors (Fig. 6A; Supplementary Fig. S11A and S11C; Supplementary Table S5), supporting our findings that Rest remains at least as active, if not more active, in Notch-mutant adherent cells compared with Notch-WT adherent cells (Fig. 4F, I, and J; Supplementary Fig. S7E). Rest was also one of the top regulators of differentially expressed genes in Notch-WT nonneuroendocrine adherent cells compared with their neuroendocrine suspension counterpart (Fig. 6B). Together, these data nominate Rest as a master repressor that drives nonneuroendocrine plasticity in both Notch-mutant and Notch-WT SCLC, which is consistent with previous findings showing that Rest is necessary and sufficient to drive nonneuroendocrine plasticity in SCLC (15).
We then used both LISA analysis of RNA-seq data and public chromatin immunoprecipitation sequencing (ChIP-seq) datasets on Cistrome.Db (33–35) to identify candidate transcriptional activators that upregulate Rest expression in a Notch-independent manner. LISA analysis of Notch-mutant versus Notch-WT adherent cells and Notch-WT adherent versus Notch-WT suspension cells both showed that the Runx family of transcription factors are master transcriptional regulators of upregulated genes in adherent nonneuroendocrine SCLC cells with some selectivity in Notch-mutant adherent cells compared with Notch-WT adherent cells (Fig. 6A and B; Supplementary Fig. S11B). Regulatory potential analysis of ChIP-seq data from Cistrome.Db (34, 35) 1 kilobase (kb) upstream of the REST transcription start site showed that RUNX2 is one of the top regulators of REST, and RUNX1 is one of the top transcription factors that bind to REST (Supplementary Fig. S11F and S11G). Publicly available ChIP-seq data showed that RUNX1 directly binds to the REST promoter in bone marrow–derived mast cells and RUNX2 directly binds to the REST promoter in preosteoblast cells (Supplementary Fig. S11J). Among the top transcription factors that bind REST, only RUNX and MYC were highly and selectively expressed in Notch-WT adherent cells compared with Notch-WT suspension cells (Supplementary Fig. S11G and S11H). In addition, RUNX (both RUNX1 and RUNX2) and REST mRNA expression is significantly correlated in human SCLC samples (Fig. 6C; Supplementary Fig. S11I). Interestingly, RUNX2 mRNA expression highly positively correlates with expression of genes that are enriched in the nonneuroendocrine subtypes of SCLC including AXL and TAP1 (Supplementary Fig. S12A and S12B), which are markers of the “inflammatory” subtype of SCLC (21, 22) and POU2F3 (Supplementary Fig. S12C; ref. 36), and negatively correlates with the neuroendocrine subtype markers ASCL1, NEUROD1, and INSM1 (Supplementary Fig. S12D–S12F).
Together, these findings nominate Runx1 and Runx2 as candidate activators of Rest expression to drive nonneuroendocrine SCLC plasticity. Immunoblot analysis of paired neuroendocrine suspension cells and nonneuroendocrine adherent cells showed that Runx2 and Rest were highly upregulated in adherent nonneuroendocrine subpopulations compared with their neuroendocrine counterparts and that Runx1 was not expressed in either subpopulation (Fig. 6D). To test whether Runx2 and Rest were required for neuroendocrine to nonneuroendocrine plasticity, CRISPR/Cas9 was used to inactivate Runx2 and Rest using multiple independent sgRNAs in the K60 Notch-mutant and K93 Notch-WT suspension SCLC cell lines. Consistent with previous data, CRISPR inactivation of Rest decreased nonneuroendocrine adherent cell formation (Fig. 6E, F, H, and I; Supplementary Fig. S11K and S11L). Similarly, CRISPR inactivation of Runx2 decreased the formation of nonneuroendocrine adherent cells and also dramatically downregulated REST expression in both Notch-WT and Notch-mutant cell lines, demonstrating that Runx2 is necessary for the transition from neuroendocrine to nonneuroendocrine adherent cells in both Notch-WT and Notch-mutant SCLC (Fig. 6E–J). Collectively, our data demonstrate that Notch-mutant SCLCs can still adopt plasticity states in the absence of canonical Notch signaling and use Runx2 activity to drive Rest and nonneuroendocrine SCLC phenotypes.
Discussion
Twenty percent to 25% of SCLCs select for LOF mutations in the NOTCH receptors with the most frequent mutations found in NOTCH1 or NOTCH2 (4). Restoration of NOTCH has potent tumor-suppressive effects in SCLC tumors, demonstrating that NOTCH has tumor-suppressor activity in SCLC (4). Our data showing that genetic inactivation of Notch1 or Notch2 accelerates SCLC tumorigenesis suggest a dominant tumor-suppressive role for Notch in SCLC tumorigenesis, which is supported by recent studies demonstrating that inhibition of NOTCH can cooperate with RB1 and TP53 loss to sustain neuroendocrine differentiation of pulmonary neuroendocrine cells during early events of SCLC initiation (37, 38) and is consistent with studies in cell lines where reactivation of NOTCH blocks ASCL1 expression and cellular proliferation (13, 14, 16, 23).
Notch also possesses tumor-promoting functions in SCLC through its ability to drive nonneuroendocrine subpopulations in vitro and within SCLC tumors in vivo marked by HES1 positivity (15). Our data are consistent with these findings in SCLC tumors and cell lines that are Notch-WT. We find that Notch is sufficient to drive the formation of nonneuroendocrine adherent cells in culture. Notch1 CRISPR-mediated genetic inactivation harbored inframe insertions/deletions (indels) in 1 Notch1 allele in approximately 25% of tumors, leaving open the possibility that these tumors selected for the ability to partially restore Notch1 expression. Alternatively, these inframe indels could partially attenuate Notch1 function given the tumor onset was similar to Notch1-mutant tumors where both alleles were deleted and slightly accelerated compared with Notch-WT tumors. All Notch2-mutant tumors had complete loss of Notch2 without any tumors harboring inframe indels. Both Notch1-mutant and Notch2-mutant tumors had near-complete depletion of Hes1-positive subpopulations, demonstrating that loss of a single Notch receptor, which is analogous to what has been observed in human SCLCs (4), was sufficient to significantly reduce Notch activity. This suggests that HES1-positive nonneuroendocrine subpopulations were not supporting tumorigenesis in these Notch-mutant mouse models. Notch has recently been shown to have a tumor-promoting function in ASCL1-negative SCLC GEMM tumors that are driven by MYC (18). Collectively, these data are consistent with Notch serving a tumor-promoting function in established SCLC tumors that are genetically NOTCH-WT or are ASCL1-negative SCLCs driven by MYC, but suggest that NOTCH dominantly functions as a tumor suppressor at earlier stages of SCLC tumorigenesis in the ASCL1 + SCLC subtype (4, 23, 37, 38). It remains an open question whether the tumors in our study that selected for the ability to restore partial Notch1 expression had a selective functional advantage.
This led us to ask the question why would approximately 25% of human SCLCs select for LOF NOTCH mutations if NOTCH is required to drive nonneuroendocrine subpopulations that support SCLC tumorigenesis? Consistent with previous findings (15), primary cells isolated from Notch-WT tumors spontaneously formed adherent nonneuroendocrine cells that highly express Notch, Hes1, and Rest, and NOTCH was sufficient to drive the formation of HES1-positive, REST-positive nonneuroendocrine cells. Surprisingly, we observed that some primary cells from the Notch-mutant tumors, primarily Notch2-mutant tumors, were able to form nonneuroendocrine subpopulations with exceedingly low Hes1 levels and still formed when Notch activity was completely blocked with the gamma-secretase inhibitor Compound E, demonstrating that these adherent nonneuroendocrine cells formed by Notch-independent mechanisms. Interestingly, all nonneuroendocrine subpopulations from both Notch-mutant and Notch-WT highly expressed the neuronal transcriptional repressor Rest, which drives nonneuroendocrine differentiation in SCLC (15, 18, 30, 39). We found that Runx2 promotes Rest expression and is necessary for the formation of adherent nonneuroendocrine cells in both Notch-mutant and Notch-WT SCLC models uncovering a Notch-independent mechanism to drive nonneuroendocrine plasticity in SCLC that converges on Rest (Fig. 6K). Consistent with our study, Runx2 has also recently been identified as an escape mechanism driving osteogenic nonneuroendocrine differentiation in MYC-driven SCLC tumors that are deleted for ASCL1 (40). RUNX1 and NOTCH often bind DNA at nearby sites and coordinately regulate target genes in T-ALL (41), and Runx1 rescues defective hematopoiesis in Notch1-null cells (42), providing evidence for their overlapping functions in different cancer types. In neuroendocrine SCLCs, RUNX1 and RUNX2 are largely repressed, whereas RUNX1T1, a fusion protein that contains the RUNX DNA–binding domain at its N-terminus, is highly expressed and can function to promote SCLC proliferation (43). An “inflammatory” subtype of human SCLC has recently been identified, and these SCLCs are nonneuroendocrine, have restored antigen presentation, and have prolonged responses to immune checkpoint blockade (21, 22). Our data show that RUNX2 expression is enriched in this subset of tumors and also in the nonneuroendocrine POU2F3 subtype (36). Future studies will focus on more detailed mechanisms for how RUNX2 regulates REST and promotes nonneuroendocrine differentiation in SCLC and whether RUNX2 could be used as a biomarker to identify nonneuroendocrine SCLCs.
Another recent study found that REST is regulated by MYC in a NOTCH-independent manner (30), which is supported by our LISA analysis, which also identified MYC as a candidate regulator of the transcriptional program in Notch-mutant adherent nonneuroendocrine subpopulations. Together, these studies demonstrate that plasticity mechanisms to promote nonneuroendocrine differentiation converge on REST, which can be regulated by at least three mechanisms including NOTCH, RUNX2, and MYC (15, 18, 30). It will be interesting to determine whether there are additional positive upstream regulators of REST and to uncover the functional consequences of inactivating REST at SCLC tumor initiation in vivo to better understand the function of nonneuroendocrine subpopulations in promoting SCLC tumorigenesis and therapeutic resistance. Alternatively, this could uncover additional escape mechanisms that drive nonneuroendocrine plasticity.
We found that nonneuroendocrine subpopulations in Notch2-mutant tumors highly express genes that promote innate immunity including STING and that Notch2-mutant adherent cell lines are highly sensitive to STING agonists. This was surprising as we and others (20–22) have shown that nonneuroendocrine SCLCs, which overall have higher NOTCH expression and can be driven by NOTCH, have restored antigen presentation and increased sensitivity to STING agonists relative to ASCL1-positive neuroendocrine SCLCs. More recent data support these findings showing that increased NOTCH signaling is a significant predictor of clinical benefit to immune checkpoint blockade (44). These studies together with our findings suggest that the nonneuroendocrine state, which can be driven in a NOTCH-dependent or NOTCH-independent manner, is enriched in genes that control antigen presentation and innate immune signaling. Therefore, a conserved gene set that controls the nonneuroendocrine state in both NOTCH-mutant and NOTCH-WT SCLCs, including RUNX2 and REST, should also be explored as a signature to predict immunotherapy response.
Our data show that inhibition of NOTCH activity augments the sensitivity to STING agonists in nonneuroendocrine SCLC cells that are Notch-WT consistent with a recent study showing that NOTCH-ICD directly inhibits STING activity (45). Together, this suggests that NOTCH-independent REST-dependent causal mechanisms that drive the nonneuroendocrine cell state are primed for STING agonism, and that the presence of NOTCH dampens the ability to stimulate the STING pathway (Fig. 6K). NOTCH-mutant non–small cell lung cancers (NSCLC), selectively NOTCH2-mutant and NOTCH3-mutant, show improved responses to immune checkpoint blockade compared with NOTCH-WT NSCLCs (46). Whether this observation is caused by NOTCH attenuating STING pathway activation or due to other mechanisms by which NOTCH opposes immunity, such as its ability to oppose JAK/STAT signaling (47, 48), is not known. Together, this suggests the intriguing possibility that LOF NOTCH mutations in lung cancer could enrich for responses to STING agonists, or that NOTCH inhibition could potentiate STING agonism in the NOTCH-WT setting. Additional work is needed to determine the mechanisms for how NOTCH inhibits STING and whether the link between NOTCH and STING has therapeutic utility in NOTCH-mutant SCLC.
Authors' Disclosures
E.H. Knelson reports grants from NIH during the conduct of the study and grants from Takeda outside the submitted work. S. Signoretti reports grants and personal fees from Bristol Myers Squibb and AstraZeneca; grants from Exelixis; and personal fees from Merck, CRISPR Therapeutics, the AACR, and NCI outside the submitted work; in addition, S. Signoretti has a patent for Biogenex with royalties paid. D.A. Barbie reports other support from Xsphera Biosciences as cofounder, personal fees from QIAGEN/N of One, and grants from BMS, Eli Lilly, Novartis, and Gilead Sciences outside the submitted work. M.G. Oser reports grants from Damon Runyon Cancer Research Foundation, NCI, and DF/HCC Lung Cancer Program during the conduct of the study; and grants from Eli Lilly, Takeda, BMS, Novartis, and Circle Pharma outside the submitted work. No disclosures were reported by the other authors.
Authors' Contributions
D. Hong: Conceptualization, data curation, formal analysis, validation, investigation, writing–original draft, writing–review and editing, first-author. E.H. Knelson: Formal analysis, investigation, writing–review and editing. Y. Li: Software, formal analysis. Y.T. Durmaz: Investigation. W. Gao: Investigation. E. Walton: Investigation, methodology. A. Vajdi: Software, formal analysis. T. Thai: Investigation. M. Sticco-Ivins: Formal analysis, investigation. A.H. Sabet: Investigation. K.L. Jones: Investigation. A.C. Schinzel: Investigation. R.T. Bronson: Formal analysis. Q.-D. Nguyen: Software, formal analysis, investigation. M.Y. Tolstorukov: Software, supervision. M. Vivero: Formal analysis. S. Signoretti: Resources, formal analysis, supervision, investigation. D.A. Barbie: Conceptualization, resources, supervision. M.G. Oser: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, writing–review and editing, senior author.
Acknowledgments
M.G. Oser is supported as a William Raveis Charitable Fund Clinical Investigator of the Damon Runyon Cancer Research Foundation (CI-101-19), by an NCI/NIH K08 grant (no. K08CA222657), the Kaplan Family Fund, and a developmental research project award from the DF/HCC Lung Cancer Program. E.H. Knelson is supported by the Gross-Loh research fellowship. The authors especially thank Pingping Mao for help with GSEA analysis, Shengqing Stan Gu for help with LISA analysis, and members of the Oser, Barbie, and Janne laboratories for helpful discussions. They thank Dana-Farber/Harvard Cancer Center in Boston, MA, for the use of the Specialized Histopathology Core, which provided histology and IHC service. Dana-Farber/Harvard Cancer Center is supported in part by an NCI Cancer Center Support Grant # NIH 5 P30 CA06516.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.