|Home | About | Journals | Submit | Contact Us | Français|
Small cell lung carcinoma (SCLC) is a high-grade pulmonary neuroendocrine tumor. The transcription factors ASCL1 and NEUROD1 play crucial roles in promoting malignant behavior and survival of human SCLC cell lines. We find ASCL1 and NEUROD1 identify heterogeneity in SCLC, bind distinct genomic loci, and regulate mostly distinct genes. ASCL1 but not NEUROD1 is present in mouse pulmonary neuroendocrine cells and only ASCL1 is required in vivo for tumor formation in mouse models of SCLC. ASCL1 targets oncogenic genes including MYCL1, RET, SOX2, and NFIB, while NEUROD1 targets MYC. ASCL1 and NEUROD1 regulate different genes that commonly contribute to neuronal function. ASCL1 also regulates multiple genes in the NOTCH pathway including DLL3. Together, ASCL1 and NEUROD1 distinguish heterogeneity in SCLC with distinct genomic landscapes and distinct gene expression programs.
Small cell lung cancer (SCLC) accounts for 10%-15% of all lung cancers (Siegel et al., 2013). Clinically, SCLC presents as a highly aggressive malignant lung tumor that displays poorly differentiated neuroendocrine features (Stovold et al., 2012). At the genetic level, SCLC can harbor several different mutations, but the majority of SCLC tumors have undergone deletion or inactivation of the TP53 and RB1 genes (Peifer et al., 2012). Moreover, it is common to find transcription factors (TFs) to be genetically amplified, including at least one member of the MYC Family (MYC, MYCN, or MYCL1), and at times Nuclear Factor I B-type (NFIB) (Dooley et al., 2011; Kim et al., 2006; McFadden et al., 2014), and SOX2 (Rudin et al., 2012). SCLC classified histologically is a heterogeneous group of cancers based on gene expression and selective tropism for subsets of SCLC cell lines by the Seneca Valley virus (George et al., 2015; Poirier et al., 2013; Poirier et al., 2015). One criterion for delineating subgroups with different expression profiles is by differential expression of the basic helix-loop-helix transcription (bHLH) factors ASCL1 and NEUROD1.
Molecular studies using human SCLC (hSCLC) cell lines show that ASCL1 and NEUROD1 are necessary for the survival of the cells where they are present and are required for tumor-initiating capacity in xenograft assays (Jiang et al., 2009; Osada et al., 2005; Osborne et al., 2013). NEUROD1 is also important for migratory capabilities of NEUROD1High hSCLC cells (Osborne et al., 2013). Even though genetic alterations of ASCL1 and NEUROD1 have not been reported, epigenetic analysis of hSCLC cells shows that their loci are in active chromatin regions and are encompassed within super-enhancers in their respective cell lines (Christensen et al., 2014). In addition, ASCL1 is sufficient to activate neuroendocrine markers in an adenocarcinoma cell line (Osada et al., 2008), and ASCL1 with SV40 large T antigen in lung club cells generated aggressive adenocarcinoma with neuroendocrine phenotype (Linnoila et al., 2000). Similarly, over-expression of NEUROD1 in non-endocrine lung cancer cell lines is sufficient to increase cell proliferation and activate a neuroendocrine program (Neptune et al., 2008).
SCLC is thought to originate from pulmonary neuroendocrine cells (PNECs) (Park et al., 2011; Sutherland et al., 2011). During normal development, ASCL1 is critical for the proper development of several neuronal and neuroendocrine populations, including PNECs in the lung. In mice null for Ascl1 there is a complete loss of PNECs (Borges et al., 1997; Ito et al., 2000). Thus, ASCL1 plays a critical role in both the normal development of mouse PNECs, and the survival of hSCLC cell lines. In mice null for Neurod1, PNECs are still present but they are enriched in clusters (NEBs) while the number of solitary cells is decreased (Neptune et al., 2008). The relationship between ASCL1 and NEUROD1, if any, in this neuroendocrine lineage is not known, but these factors have a temporal relationship in some neuronal lineages such as in adult neurogenesis in the hippocampal dendate gyrus and development of the olfactory epithelium (Cau et al., 1997; Kim et al., 2011). Although ASCL1 and NEUROD1 are both capable of increasing cell proliferation while activating neuroendocrine genes, it is unclear whether these two bHLH factors are lineally related in the PNEC, where they bind in the genome, what genes they regulate, and if these related factors share a common function in SCLC.
To better understand the role of the bHLH factors in SCLC, we define ASCL1 and NEUROD1 gene targets in pulmonary neuroendocrine tumors by probing the chromatin landscape occupied by these TFs in SCLC. Although ASCL1 and NEUROD1 are related bHLH factors and bind a similar DNA core motif, the vast majority of ASCL1 and NEUROD1 bound sites do not overlap, contain a distinct recognition motif, and are enriched for motifs for distinct families of TFs. The genes defined as direct downstream targets of ASCL1 and NEUROD1 are largely non-overlapping and show how these two bHLH factors differentially regulate key oncogenes in SCLC. Transcriptional targets of ASCL1 found in both mouse and human cancer models highlight ASCL1’s role in regulating NOTCH signaling. Importantly, we demonstrate the current mouse models for SCLC reflect ASCL1High high-grade neuroendocrine tumors, and ASCL1 but not NEUROD1 is required for tumor formation. We discuss the implications of the substantial differences in gene expression and chromatin landscapes in the ASCL1 versus NEUROD1 defined hSCLC cell lines, and highlight the importance of these distinctions when assessing therapeutic outcomes in preclinical models.
For decades, human-derived SCLC cell lines have been used as model systems to uncover the genetic and molecular characteristics that drive malignant behavior of the tumor cells, yet there is variation in gene expression and genetic alterations in these lines (Dooley et al., 2011; Johnson et al., 1996; Kim et al., 2006) (Fig. 1A-D). There are two sub-types of SCLC cell lines, classical and variant, distinguished by histology, MYC amplification, and lack of expression of a subset of neuroendocrine markers such as Dopa decarboxylase (DDC) and Gastrin-releasing peptide (GRP) (Carney et al., 1985; Gazdar et al., 1985). Microarray analysis of 38 of these hSCLC cell lines distinguishes multiple subgroups and reveals differential expression of the TFs ASCL1 and NEUROD1 (Fig. 1A, Table S1). Notably, cells with a high level of ASCL1 and low level of NEUROD1 (ASCL1High) show stronger expression of classical neuroendocrine markers DDC and GRP, compared to cells that express high levels of NEUROD1 and lower levels of ASCL1 (NEUROD1High) (Fig. 1A,B). In contrast, the neural genes Brain Creatine Kinase (CKB) and Enolase 2 (ENO2) are expressed in all SCLC lines (Fig. 1A). RNA-Seq from NCI-H889, NCI-H2107, NCI-H82 and NCI-H524 confirmed that ASCL1 and NEUROD1 have contrasting expression levels in these cells (Fig. 1C, Table S2). Furthermore, whereas a majority of these cells are mutant for TP53 and/or RB1, ASCL1High SCLC cell lines have classic SCLC morphology and often have MYCL1 amplification, while NEUROD1High cell lines have variant features and are MYC amplified (Fig. 1D) (Carney et al., 1985; Gazdar et al., 1985). We used hSCLC cell line microarray data to generate ASCL1High and NEUROD1High gene signatures. Using these signatures, 81 human primary SCLC tumor samples profiled by RNASeq (George et al., 2015) were clustered into ASCL1High and NEUROD1High subgroups, as well as those expressing low or high levels of both (Fig. 1E, Table S3) confirming the cell line expression heterogeneity is reflected in primary tumor data, although the primary tumors have more complexity. ASCL1High levels are much higher than NEUROD1High in the primary tumor samples (ASCL1 FPKM range from 0.1 to 976 with a median of 198; NEUROD1 FPKM range from 0 to 179 with a median of 4). Taken together, ASCL1 and NEUROD1 are differentially expressed across hSCLC cell lines, and across primary tumor samples that represent heterogeneous cell populations, with a majority of the samples being ASCL1High.
ASCL1 and NEUROD1 activate transcription through similar motifs. Given that ASCL1 and NEUROD1 are required in the SCLC lines in which they are expressed (Jiang et al., 2009; Osborne et al., 2013), we hypothesized that they may share targets required for tumor cell maintenance. To gain insight into this possibility, we performed ChIP-Seq in hSCLC cell lines that express high levels of ASCL1 (NCI-H889, NCI-H2107, and NCI-H128) or NEUROD1 (NCIH82 and NCI-H524) (Fig. S1, Table S4). We identify 6,250 ASCL1 bound sites (shared in NCIH889, NCI-H2107, and NCI-H128), and 4,193 NEUROD1 bound sites (shared in NCI-H82 and NCI-H524) (Fig. 2A). Of these, only 304 sites were bound by both ASCL1 and NEUROD1. These common sites are significantly more frequent (p-value <5×10−300) than expected by random chance (<10 sites calculated for random chance) demonstrating ASCL1 and NEUROD1 associate with a small but significant shared set of genes (some expressed highly in SCLC such as INSM1, NCAM1, and HES6) in addition to the much larger set of distinct genes (Table S4).
The epigenetic landscape in different cell types has shown that large domains of H3K27Ac marked chromatin, called super-enhancers, are associated with genes important in regulating cell identity (Whyte et al., 2013), and in cancer cells at key oncogenic genes (Hnisz et al., 2013). It was recently reported that ASCL1 but not NEUROD1 is associated with a super-enhancer in multiple ASCL1High hSCLC cell lines, while in NEUROD1High cells, NEUROD1 is associated with a super-enhancer (Christensen et al., 2014). Using H3K27Ac ChIP-Seq data for NCI-H128 and NCI-H82 (GSE36354, GSE62412), we identify super-enhancers and find only 5-10% of these overlap (Fig. 2B,C) (Table S5). INSM1, a commonly expressed gene across multiple neuroendocrine cell-types and in SCLC (Fujino et al., 2015; Pedersen et al., 2006) is among the few genes associated with super-enhancers in both cell lines. Notably, in addition to ASCL1, NCI-H128 super-enhancers associated with oncogenic genes for SCLC including MYCL1, NFIB, and BCL2, and lineage-specific genes for lung including NKX2-1, FOXA1, and FOXA2 (Fig. 2B). In contrast, NCI-H82 had super-enhancers associated with NEUROD1 and the oncogenic gene MYC but not the other lineage-specific genes for lung (Fig. 2B).
As stated above, only 5-10% of the super-enhancers are overlapping between the hSCLC cell lines NCI-H128 and NCI-H82 (Fig. 2C). To determine how this level of overlap compares to the overlap of super-enhancers seen in lineage-related cells, we compared super-enhancers identified in multiple tissues by Hnisz et al., 2013 (Table S6). We found that a 5-10% overlap is in the range of that seen between cells with no related lineal history such as brain and hematopoietic cells (10-20%) or brain and mammillary epithelial cells (14-22%) (see Table S6 for more comparisons). In contrast, cells from a shared lineage such as hematopoietic stems cells, T-cells and B-cells have a much higher overlap of super-enhancers (40-86%) (Table S6). Thus, differential super-enhancer locations in ASCL1High versus NEUROD1High SCLC cell lines support a fundamental difference in the genomic landscape, and possibly a distinct lineal history of the cell-types comprising these hSCLC models.
Since ASCL1 and NEUROD1 are transcriptional activators and bind sites that reside in active regulatory regions, we asked if these two factors are enriched in super-enhancer regions. We found ASCL1 and NEUROD1 bound sites are enriched in super-enhancer regions over that seen in regular enhancers (as defined by H3K27Ac enrichment) (Fig. 2D). In NCI-H128 (ASCL1High), 81% of the super-enhancers contained at least one ASCL1 bound site; and in NCIH82 (NEUROD1High), 79% contained at least one NEUROD1 bound site (in contrast only 5% of regular enhancers contain ASCL1 or NEUROD1 bound sites). Thus, within the distinct genomic landscapes of the ASCL1High and the NEUROD1High hSCLC, ASCL1 and NEUROD1 occupy largely distinct sites that localize with cell-type specific super-enhancers.
Both ASCL1 and NEUROD1 bind DNA sequences containing the E-box motif (CANNTG) (Bertrand et al., 2002). We performed de novo motif analysis using Homer (Heinz et al., 2010) on the ASCL1 and NEUROD1 binding sites in each cell line to identify similarities or differences in their binding preferences (Fig. 2E). As expected, the primary motif returned was the E-box motif, which was found in 92-94% of the peak regions and enriched at the peak center (Fig. 2E,F). Notably however, ASCL1 and NEUROD1 preferentially bind distinct E-box motifs. The differences in genomic sites bound and E-box motif preference support distinct roles for ASCL1 and NEUROD1 in SCLC cell lines.
The differential binding of ASCL1 and NEUROD1, and the distinct molecular profiles found in each cell line, may be a result of cell line specific co-factors. We searched for additional motifs enriched in ASCL1-only and NEUROD1-only regions. In ASCL1-only regions, motifs for NFI half sites (Roulet et al., 2002) and Forkhead factors (Badis et al., 2009) are enriched (Fig. 2E,F). In contrast, NEUROD1-only regions are enriched for the motif recognized by the homeodomain factors CRX and OTX2. The enrichment of these motifs suggests that one or more members of these TF families are capable of binding the same genomic regions as ASCL1 and NEUROD1. FOXA1 and FOXA2 from the Forkhead family, and NFIB and NFIX from the NFI family are specifically expressed in the ASCL1High hSCLC cells (Table S2). Notably, NFIB, FOXA1, and FOXA2 are also associated with super-enhancers in ASCL1High cells (Fig. 2B). The NEUROD1High cells specifically express OTX2 (Table S2), which is associated with a super-enhancer in these cells (Fig. 2B). Similar TF motifs, including the specific E-box motifs for ASCL1 and NEUROD1, are enriched in H3K27Ac marked regions in the respective cell lines supporting models for the importance of a combination of these factors in regulating transcription in SCLC (Fig. 2G). These results highlight distinct candidate co-factors that are associated with ASCL1 or NEUROD1 in the hSCLC models.
To determine candidate downstream targets of ASCL1 or NEUROD1, we called associated genes from our ChIP-Seq data using GREAT (McLean et al., 2010) (Table S4). Among the genes uniquely bound by ASCL1 are genes with a known function as oncogenes for SCLC including MYCL1, RET, SOX2 and NFIB, and the anti-apoptotic gene, BCL2. In addition, ASCL1 targets include genes important for lung development such as FOXA2 and NKX2-1(Minoo et al., 1999; Zhou et al., 1997) as well as genes characteristic of the neuroendocrine phenotype including CALCA/B (encoding CGRP), NCAM1, DLL3, and INSM1 (Fig. 3A). In contrast, NEUROD1 unique targets include the oncogene MYC as well as genes typically associated with neuronal function such as the TF OTX2 (Fig. 3B).
In order to test whether predicted target genes require ASCL1 for expression, we knocked down ASCL1 in the hSCLC cell line NCI-H889 and assessed mRNA levels of several of the identified targets. We show that with an 80% knock down of ASCL1 there is a significant decrease in CALCA, CALCB, RET, BCL2, MYCL1, and FOX2A, supporting the requirement for ASCL1 for their expression in this assay (Fig. 3C-D). Thus, in multiple models of SCLC, we find ASCL1 is a key regulator of several genes known to be important for cancer growth properties as well as lineage-specific genes for pulmonary neuroendocrine cell identity. Furthermore, ASCL1 and NEUROD1 binding is associated with mostly distinct genes highlighting their independent requirement in hSCLC.
To focus the candidate downstream target list of ASCL1 or NEUROD1, we intersected the ChIP-Seq associated genes with a union set of genes correlated with ASCL1 or NEUROD1 expression from primary SCLC tumors and cell lines. This union set includes genes identified as differentially expressed between ASCL1High and NEUROD1High from the 81 primary hSCLC samples (George et al., 2015) and from the hSCLC cell lines (Table S2). In addition, we include differentially expressed genes from the hSCLC cell line microarray data for ASCL1High versus NEUROD1High, and ASCL1High or NEURODHigh versus the hSCLC not expressing these TFs (Table S1). This analysis identified 620 ASCL1 and 443 NEUROD1 candidate target genes (Supplementary Figs. S2 and Table S7). Notably, there are 49 genes regulated in common including INSM1. In addition, G-protein and synapse related genes such as SYT1, RGS7, and GRM2 are in this group (Fig. 4F). Examination of ASCL1 or NEUROD1 unique target genes by gene ontology reveals enrichment of genes in neuronal systems in general. Particular to ASCL1, however, are multiple potassium channel genes and genes associated with NEB (Fig. 4F; Table S7). The latter group includes CALCA, DDC, RET, FOXA2, and NKX2-1 as discussed above. In contrast, NEUROD1 target genes are enriched in axon guidance, cholinergic synapse, and cell adhesion related genes (Fig. 4F, Table S7). These results highlight the distinct genetic programs regulated by ASCL1 and NEUROD1. Strikingly, as noted in the previous section, ASCL1 appears to be directly regulating multiple oncogenes identified in SCLC including MYCL1, SOX2, RET, and BCL2, suggesting a possible mechanism for its requirement in SCLC survival.
Mouse models for SCLC (mSCLC) express high levels of ASCL1 (Meuwissen et al., 2003; Schaffer et al., 2010) (Fig. 6). We reasoned that common gene targets of ASCL1 in both species would provide high value candidates for understanding the dependence of these tumor cells on ASCL1. We used a mouse model for SCLC that deletes Trp53, Rb1, and Rbl2 (TCKO) in the lung epithelium upon administration of Adenoviral-CMV::Cre intratracheally (Schaffer et al., 2010). ChIP-Seq for ASCL1 in dissected tumors from mSCLC resulted in 3,499 DNA bound regions (Supplementary Figs. S1, 2, Table S8). Analysis of these sites identified the same motifs as in hSCLC cell lines including the E-box as the primary motif with Forkhead and NFI as secondary motifs (Supplementary Fig. S1). Genes associated with ASCL1 bound sites in mSCLC identified 4,033 associated genes (Table S8). Focusing on those expressed with ASCL1 in both mouse and human SCLC tumors identified 140 common targets (Fig. 4D, Supplementary Fig. S2, Table S9). This common list retains important regulators in NEB including ASCL1, DDC, CALCA/B, FOXA2, INSM1, and RET. In addition, NOTCH pathway genes including DLL3 that encodes the NOTCH ligand recently used in a strategy to deliver a cytotoxic agent to SCLC cells (Saunders et al., 2015), and LFNG, encoding an O-fucosylpeptide 3-β-N-acetylglucosaminyltransferase known to modify the NOTCH extracellular domain (Fig. 4). Thus, ASCL1 is using homologous DNA regions to regulate cancer related genes in both mouse and human models of SCLC.
ASCL1 and NEUROD1 are required for growth and survival of hSCLC cells in vitro, in colony formation assays, and in xenografts (Jiang et al., 2009; Osborne et al., 2013). To determine if these factors are required in vivo for formation of neuroendocrine tumors in the lung, we turned to the TCKO mouse model for SCLC (Fig. 5A,B). In this model, large neuroendocrine (NE) tumors – both SCLC and large cell neuroendocrine carcinoma (LCNEC) – arise within 5 months (Gazdar et al., 2015) (Figs. 5C, 6A-C). To test the role of ASCL1 and NEUROD1, we crossed in mutant alleles of Ascl1 (Ascl1neoFlox) (Pacary et al., 2011) or Ascl1GFP (Leung et al., 2007), or Neurod1 (Neurod1Flox) (Goebbels et al., 2005). Strikingly, if Ascl1 is removed (Ascl1neoFlox/neoFlox or Ascl1neoFlox/GFP collectively called Ascl1CKO) along with the TCKO, no NE tumors form (Fig. 5D). This was assessed histologically by determining tumor load or tumor count 22-27 weeks post infection with Adenoviral-CMV::Cre (Fig. 5F,G). Even animals heterozygous for Ascl1 (Ascl1CKO/+) had a significant decrease in tumor formation. It is likely that a requirement for ASCL1 in maintenance/survival of pulmonary NE cells, whether they are Rb/TP53 mutant or not, contributes to the lack of tumor formation detected in this paradigm since we found 60% fewer NEB and solitary PNEC cells in Ascl1f/f;RosaLSLtdTOM relative to controls RosaLSLtdTOM after intratracheally infection with Adeno-CMV::CRE (Supplementary Fig. S3). Nevertheless, these results highlight the requirement for ASCL1 for NE tumor formation in the lung in vivo.
In contrast to the dramatic absence of tumor formation in the SCLC mouse model in the absence of Ascl1, loss of Neurod1 had no effect. When Neurod1Flox/Flox (Neurod1CKO) was included in the TCKO model, tumor load and tumor count were indistinguishable from animals in the presence of Neurod1 (Fig. 5E-G). Efficient deletion of Neurod1 in the tumors was verified by PCR (Fig. 5H). In addition, tumors continued to express high levels of NE markers such as CGRP, SYP and ASCL1. Thus, ASCL1 but not NEUROD1 is required for tumor formation in this mouse model of high-grade neuroendocrine cancer.
The NE tumors formed in the TCKO co-express high levels of ASCL1 and the characteristic NE markers synaptophysin (SYP) and CGRP (Fig. 6A-E). No NEUROD1 was detected in these tumors either by protein or RNA assays. Immunocytochemistry on CGRP+ tumors with antibodies against NEUROD1 detected no protein (Fig. 6F-F”) even when NEUROD1 was detected in other tissues such as olfactory epithelium (Fig. 6G). In addition, Western blots do not detect NEUROD1 in the mSCLC although ASCL1 is present (Fig. 6H). And finally, RNA-Seq from mouse SCLC tumor tissue showed high expression of Ascl1, Syp, and Calca but no Neurod1 (Fig. 6I). These results demonstrate the current mouse models of SCLC reflect the ASCL1High-type but not the NEUROD1High-type of SCLC.
The lack of NEUROD1 in the mSCLC tumors prompted us to revisit expression of NEUROD1 in normal lung, particularly in PNECs in neonates and adult mice. We first performed immunocytochemistry for NEUROD1 in P0 and P40 mouse lung. Although we could detect NEUROD1 in neuronal tissues like olfactory epithelium, cerebellum, and hippocampus, we could not definitively detect NEUROD1+ cells in lung tissue (data not shown). Because these are negative data and we assessed only 2 postnatal stages, we turned to a Neurod1::Cre mouse (Li et al., 2014) to label cells that have expressed Neurod1 at anytime during development. Using a RosaLSLtdTOM Cre reporter line (Ai14), we examined Neurod1-lineage cells in P0 and P40 lung. We detected Neurod1-lineage cells (TOM+) in the lung but they always co-labeled with the neuronal marker TUJ1 not the PNEC marker CGRP (Fig. 6J-J’”) consistent with PNECs receiving neuronal input (Van Lommel, 2001) and NEUROD1 expression in developing sensory neurons (Ma et al., 1996). As a positive control, we also show Neurod1-lineage cells in the granule neurons in the cerebellum at P0 (Fig. 6K). Examination of 276 NEBs and 244 solitary PNEC in these lungs revealed no Neurod1-lineage cells in the neuroendocrine compartment (Fig. 6L). Use of a second Neurod1::Cre mouse line and a Neurod1::GFP mouse line from the GENSAT project (Gong et al., 2003) also detected no Neurod1-lineage cells in NE cells in lung (data not shown). From this analysis, we conclude that NEUROD1 is not normally present in mouse pulmonary NE lineages, although Neurod1-lineage neurons innervate NEBs.
ASCL1 and NEUROD1 are required for cell survival in multiple human derived SCLC cell lines in which they are expressed (Jiang et al., 2009; Osborne et al., 2013). In experiments to uncover transcriptional targets for ASCL1 and NEUROD1 and to test their requirement in vivo, we reveal that 1) these factors stratify the hSCLC cell lines into those likely derived from the typical ASCL1 defined lung neuroendocrine cell lineage versus those that have a distinctly different chromatin landscape and are of unknown origin, 2) these bHLH factors regulate largely distinct targets, 3) ASCL1 directly regulates known SCLC oncogenes, and 4) ASCL1 but not NEUROD1 is required in vivo for tumor formation in a genetically engineered mouse model of pulmonary neuroendocrine tumors.
For decades, biological and preclinical studies of SCLC have largely depended on the availability of human tumor cell lines (Gazdar et al., 2010). There is an acknowledged heterogeneity in these cells given that they can harbor different genetic alterations, display distinct epigenetic profiles, have variant morphologies, can vary in expression of neuroendocrine markers, and have distinct vulnerabilities to viral infection (Carney et al., 1985; Gazdar et al., 1985; Poirier et al., 2013; Poirier et al., 2015). ASCL1 and NEUROD1 levels in these cell lines stratify them into subtypes (Fig. 1) (Poirier et al., 2013; Poirier et al., 2015). Recent RNA-Seq data from primary SCLC tumors (George et al., 2015) supports the prevalence of high ASCL1 across samples, the presence of NEUROD1 in a subset of tumors, and a gene signature for the ASCL1High and NEUROD1High subtypes derived from the cell lines that stratify these primary SCLC samples (Fig. 1).
Multiple lines of evidence illustrate that the ASCL1High and NEUROD1High SCLC lines have distinct chromatin landscapes. First, super-enhancers identified by H3K27Ac have characteristic genomic locations that can be used to infer broad lineage relationships (Hnisz et al., 2013). Notably, the super-enhancers in the two subtypes of hSCLC are in largely distinct locations (Fig. 2, Christensen et al., 2014). And the low percentage of super-enhancers common between the subtypes is in the range seen between lineally unrelated cells (Hnisz et al., 2013) (Table S6). Furthermore, de novo motif analysis performed either from ASCL1 and NEUROD1 bound regions, or enhancers defined by H3K27Ac, identify motifs for distinct TFs. Differences in chromatin landscapes and TF motifs found in H3K27Ac enriched regions have been used to distinguish cell-lineage relationships across tissues (Heinz et al., 2010; Stergachis et al., 2013). Together, these findings highlight multiple differences at the chromatin level between the ASCL1High and NEUROD1High subgroups of hSCLC.
The striking differences seen between the ASCL1High and NEUROD1High subgroups of hSCLC, combined with the absence of Neurod1 in mouse lung NE cells and in the mouse SCLC suggest the possibility that the heterogeneity arises from different cell-types of origin. The ASCL1High subgroup almost certainly arises from mutations originating in the resident ASCL1+ NEBs in the lung. However, the origin of the NEUROD1High tumor cells is not clear. It is possible that NEUROD1High hSCLC originate from a rare NEUROD1+ neuroendocrine cell-type that is more prevalent in humans versus mouse lung that has not yet been identified. It is also possible that they originate from neuroendocrine cells from extrapulmonary sites and have metastasized to the lung. And finally, the NEUROD1High hSCLC may represent a cancer cell-type not present in normal tissue but represents a NEB lineage cell substantially altered at the chromatin level from the ASCL1 defined NEB. Uncovering the origin of the NEUROD1 expressing cells will be important in understanding the heterogeneity seen in hSCLC.
It is important to note that studies of SCLC using the hSCLC cell lines to model the biology of SCLC often use the ASCL1High and NEUROD1High cell lines interchangeably. Indeed, many studies preferentially use the NEUROD1High cells such as NCI-H82 that represent the minority of hSCLC. This practice may cause confusion in interpretation of results and future studies should consider the differences in the hSCLC cell line subtypes when designing and interpreting results from these experiments.
de novo identification of motifs in ASCL1 and NEUROD1 bound regions identified distinct co-factor motifs implicating specific families of TFs that may collaborate with the bHLH factors in tumor formation and/or maintenance. Forkhead and NFI motifs were found enriched within the ASCL1 bound regions (Fig. 2). Of the four NFI factors, NFIB is often over-expressed and genetically amplified in SCLC, and has been shown to regulate cell viability and proliferation during tumor formation (Dooley et al., 2011). In addition, NFIB is associated with a super-enhancer in the ASCL1High SCLC line (Fig. 2). There are dozens of Forkhead family members that could work through the Forkhead motif identified in ASCL1 bound sites, and several of these factors have been shown to be mis-regulated and important for several types of cancers (Myatt and Lam, 2007). FOXO3 has overlapping sites with ASCL1 in neural progenitor cells (Webb et al., 2013). However, FOXO3 and other FOXO factor mRNA are present only at low levels in the hSCLC and mSCLC models. In contrast, FOXA1 and FOXA2 are highly expressed in the SCLC tumors in human and mouse (Table S7,9), and are associated with super-enhancers in the ASCL1High SCLC cells, highlighting them as candidates for a role with ASCL1 in SCLC. In addition, ASCL1 is bound near the FOXA2 gene and FOXA2 is decreased when ASCL1 is knocked down in hSCLC (Fig. 3). In future studies, FOXA1 and FOXA2, along with ASCL1 and NFIB, should be assessed for a role in regulating genes important for the survival of SCLC.
In contrast, NEUROD1 bound regions are enriched for a motif recognized by the homeodomain factor OTX2. OTX2 is specifically expressed in some NEUROD1High SCLC cells, and NEUROD1 appears to directly regulate OTX2 (Fig. 3). OTX2 is required for cell proliferation in pre-clinical models of medulloblastoma (Bunt et al., 2012). Consistent with this function in SCLC, we find NEUROD1 sites containing the OTX2 motif associated with the oncogene MYC. And lastly, OTX2 is identified near a super-enhancer in NEUROD1High SCLC cells (Christensen et al., 2014; and Fig. 2). Thus, OTX2 may collaborate with NEUROD1 in maintaining cell proliferation in NEUROD1High tumors. Altogether, ChIP-Seq has implicated TFs that may work with ASCL1 and NEUROD1 for survival of the tumor cells in which they are expressed.
Neuroendocrine lung tumors in both mouse and humans have a tendency to harbor several somatic mutations such as genetic amplifications of MYCL1, NFIB, and SOX2 (Dooley et al., 2011; McFadden et al., 2014; Rudin et al., 2012). ASCL1 binds near these loci suggesting it directly regulates the expression of these key oncogenes. Other ASCL1 targets such as RET and DLL3 are important for SCLC. RET is an oncogene in SCLC (Dabir et al., 2014; Jiang et al., 2009), and recently, an antibody-drug conjugate targeted to DLL3 blocked survival of pulmonary neuroendocrine tumor-initiating cells in patient-derived xenograft models of SCLC (Saunders et al., 2015). It is notable that even in SCLC, ASCL1 regulates NOTCH signaling components such as DLL3 and LNFG originally identified in neural development (Fig. 4) (Castro et al., 2006; Henke et al., 2009). In addition, BCL2 appears to be strongly expressed and required for several ASCL1High human pulmonary neuroendocrine tumors and treatment of mice carrying ASCL1High SCLC or NE-NSCLC xenografts with a BCL2 inhibitor showed inhibition of tumor growth (Augustyn et al., 2014; Tse et al., 2008). Thus, although there are no reported genetic alterations in the ASCL1 loci in SCLC, ASCL1 is a transcriptional activator directly targeting known oncogenes for SCLC. A loss of ASCL1 as a driver of these oncogenic factors may partially explain the requirement for ASCL1 in tumor survival.
The ASCL1High subtype of SCLC represents a majority of SCLC based on multiple lines of evidence. First, most hSCLC cell lines and primary tumor samples are ASCL1High (Fig. 1; George et al., 2015). Second, ASCL1 is normally present in lung neuroendocrine cells during development and in the adult, and it is essential for this lineage to develop (Ito et al., 2000). In addition, ASCL1 is highly expressed in the pulmonary neuroendocrine tumors in the mouse model and it is essential for these tumors to form (Fig. 5). Thus, ASCL1 is a lineage-specific oncogene for SCLC. On the other hand, although NEUROD1 expression was reported in neonatal lung tissue in mice (Neptune et al., 2008), and loss of NEUROD1 disrupts the pattern of NEBs formed, NEUROD1 itself is not present in the mouse lung neuroendocrine cells. Rather, neurons that innervate the NEBs and whose cell bodies are expected to be in sensory ganglia outside of the lung are from the Neurod1-lineage. Furthermore, mSCLC tumor tissue does not have detectable NEUROD1 mRNA, and Neurod1 is not required for tumor formation or maintenance in the mouse model (Figs. 5,,6).6). Thus, unlike adult neurogenesis where ASCL1 and NEUROD1 are present in the same lineage at different stages as progenitors transition to neurons, in the lung, ASCL1 and NEUROD1 do not appear to be lineally related.
The Trp53, Rb1, Rbl2 triple knockout mouse model most closely relates to the ASCL1High hSCLC subtype and reveals a possible progression from LCNEC to SCLC over time (Gazdar et al., 2015; Schaffer et al., 2010). Both LCNEC and SCLC express high levels of ASCL1 and both tumor types require ASCL1 for formation since in the mSCLC model lacking Ascl1, no neuroendocrine tumors form. Unexpectedly, even the loss of one allele of Ascl1 dramatically reduces tumor load and incidence. This may have therapeutic implications as a partial inhibition of ASCL1 may substantially attenuate tumor growth in humans and indicates the need for focusing therapeutic targeting of the ASCL1 pathway.
(detailed Experimental Procedures can be found in Supplementary Information)
The SCLC mouse model with conditional deletions in Trp53 (p53), Rb1, and Rbl2 (p130) was previously described (Schaffer et al., 2010). The Ascl1neoFlox (Pacary et al., 2011), Ascl1GFP (Ascl1tm1Reed/J) (Leung et al., 2007), Neurod1Flox (Goebbels et al., 2005), and Neurod1::Cre (Li et al., 2014) mice were previously described. Adenovirus expressing CMV::Cre (University of Iowa Viral Vector Core) was administered intratracheally to 6 week old mice (males and females) at 2.5 ×107 PFU as described (DuPage et al., 2009). Lungs from all infected mice were harvested between 22-27 weeks post infection. All procedures on animals follow NIH Guidelines and were approved by the UT Southwestern Institutional Animal Care and Use Committee.
Lungs were inflated then submerged for 1 hour using 4% paraformaldehyde, rinsed, paraffin embedded, sectioned at 10 μm, random sections were processed for H&E, and imaged on a digital slide scanner, Nanozoomer XR-12000. Antibodies used include guinea pig anti-ASCL1 (1:5000, Johnson lab generated TX518) (Kim et al., 2008), rabbit anti-SYP (1:500, Peirce PA5-27286), rabbit anti-CGRP (1:500, Sigma C8198), mouse anti-TUJ1 (1:1000, Biolegend 801201), and goat anti-NEUROD1 (1:500, Santa Cruz N-19).
Cell lines used were from the Hamon Cancer Center Collection (UT Southwestern Medical Center), maintained in RPMI-1640 (Life Technologies) supplemented with 5-10% FBS. The lentiviral shRNA vector pTY-shRNA-EF1a-puroR-2a-GFP was used. The shRNA sequences are as follows: shASCL1(1), 5′-GCAGCACACGCGTTATAGTTT-3′; 5′-ACTATAACGCGTGTGCTGCTC-3′ and shASCL1(2): 5′-CAACTGCAATTTTCCCTATTT-3′; 5′-ATAGGGAAAATTGCAGTTGTA-3′. Primers for qPCR for specific genes are included in Supplementary Information. Antibodies for Western blotting were guinea pig anti-ASCL1 (1:1000, Johnson lab generated TX518) (Kim et al., 2008), and rabbit anti-NEUROD1 (1:1000, Abcam ab109224).
Tumors from lungs of Trp53;Rb1;Rbl2 triple knock out mice were pooled from an individual mouse and RNA or chromatin extracted. For hSCLC cell lines, 107 cells were used. The mSCLC sequencing library preparation used Illumina Tru-Seq and single-end 50-bp sequencing on the Illumina Hi-Seq 2000. hSCLC cell line libraries were made using the SOLiD Fragment Library Barcoding Kit (Applied Biosystems) and sequenced (50-bp reads) on an ABI SOLiD. RNA-Seq data are available (GSE69398). Mouse mm9 and human hg19 genome assemblies were used for sequence alignment. Microarray data was generated for gene expression in human tumor cell lines (GSE32036). RNA-Seq from primary SCLC tumors was from George et al., 2015.
For ChIP-Seq, 250 μg mouse tumor or 100 μg human cell line chromatin was immunoprecipitated with 5 μg mouse anti-ASCL1 antibody (BD Biosciences, cat# 556604) or with 5 μg goat anti-NEUROD1 (Santa Cruz, N-19). ChIP-Seq libraries were sequenced on an Illumina High-Seq 2000 or Illumina GAIIx (GSE69398). Sequence reads for each sample were mapped to the hg19 or mm9 genome assemblies as relevant with Bowtie (Trapnell et al., 2009). Duplicate reads were removed, and the remaining unique reads were normalized to 10 million reads. Peak calling was performed by HOMER (Heinz et al., 2010) using an FDR cutoff of 0.001, a cumulative Poisson p-value of <0.0001, and required a 4-fold enrichment of normalized sequenced reads in the treatment sample over the control/input sample. Normalized sequence reads around each peak were counted in 25 bp bins, and loaded into Matlab® to generate heatmaps.
Motif discovery was conducted with HOMER package v4.2. In ChIP-Seq data we used the following settings: -size 150 –S 10 –bits (Heinz et al., 2010). We limited the motif analysis to a 150 bp DNA region around each peak summit. In H3K27AC enriched regions we used –size given as the parameter as these regions are broad. Distance to gene and gene annotations for ChIP-Seq peaks were obtained using GREAT v1.82 (McLean et al., 2010).
Super-enhancers were determined using H3K27Ac ChIP-Seq data from (Lin et al., 2012)(GSE36354) and (Christensen et al., 2014)(GSE62412) calculated by HOMER using bound regions that were significant at Poisson p-value threshold of 1E-09 for the super-enhancer prediction. Rank ordering of super-enhancers was by ROSE (Loven et al., 2013; Whyte et al., 2013).
Pathway enrichment analysis was performed using ConsensusPathDB tool (Kamburov et al., 2011) to identify overrepresented genesets. Significantly enriched pathways (FDR ≤ 5%) were selected (Fig. 4, Table S7, S9).
We acknowledge the generous contribution by Dr. Julien Sage (Stanford University) for the Trp53;Rb1;Rbl2 triple floxed mice at the initiation of this project, the outstanding Next Generation Sequencing service from the Genomic and Microarray Cores at UT Southwestern, and Ms Erin Kibodeaux and Svetlana Earnest for excellent technical assistance. We thank Drs. Francois Guillemot for Ascl1neoFlox mice, Klaus Nave for Neurod1Flox mice, and Andrew Leiter for Neurod1::Cre mice. We appreciate lung tissues generously provided by Dr. Hatten’s laboratory from Neurod1::GFP and Neurod1::Cre/ROSA26LSL-GFP mice. Funding for this project was provided by a Technology Development Award to JEJ from the Cancer Center NCI Spore Grant in Lung Cancer P50CA70907, CPRIT RP110383, RP140143 to MHC and JEJ, and NIH F31 NS705592 to MDB.
Conceptualization, M.D.B., M.H.C., and J.E.J.; Investigation, M.D.B., T.K.S., R.K.K., M.H., A.A., J.K.O., A.F.G.; Resources, L.G., A.F.G. and J.D.M.; Writing, M.D.B., T.K.S., R.K.K. and J.E.J.; Funding Acquisition, M.H.C. and J.E.J.; Supervision, M.H.C., J.D.M., and J.E.J.
Accession codes for all data used in this study include GSE36354, GSE62412, GSE69398, GSE32036.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.