Identification and targeting of long-term tumor-propagating cells in small cell lung cancer

Summary

Small cell lung cancer (SCLC) is a neuroendocrine lung cancer characterized by fast growth, early dissemination, and rapid resistance to chemotherapy. We identified a population of long-term tumor-propagating cells (TPCs) in a mouse model of SCLC. This population, marked by high levels of EpCAM and CD24, is also prevalent in human primary SCLC tumors. Murine SCLC TPCs are numerous and highly proliferative but not intrinsically chemoresistant, indicating that not all the clinical features of SCLC are linked to TPCs. SCLC TPCs possess a distinct transcriptional profile compared to non-TPCs, including elevated MYC activity. Genetic and pharmacological inhibition of MYC in SCLC cells to non-TPC levels inhibits long-term propagation but not short-term growth. These studies identify a highly tumorigenic population of SCLC cells in mouse models, cell lines, and patient tumors, and a means to target them in this most fatal form of lung cancer.

Graphical Abstract

Introduction

Small cell lung cancer (SCLC), which represents ~15% of lung cancers, is characterized by small cells with neuroendocrine features (Wistuba and Gazdar, 2006). Close to 200,000 people die from SCLC every year worldwide and the 5-year survival is a dismal 5–10%. SCLC disseminates early and is usually detected late when patients present with extensive metastases. Patients often respond well initially to chemotherapy (usually a combination of etoposide and a platinum-based agent), but they almost invariably relapse with disease that is resistant to their primary therapy and other agents. Despite numerous clinical trials, no new treatment has been approved in two decades and SCLC remains the most lethal form of lung cancer (Pietanza et al., 2015).

The cancer stem cell model assumes a hierarchical organization in which a subset of tumor cells is responsible for sustaining tumorigenesis and establishing the cellular heterogeneity of a primary tumor (Beck and Blanpain, 2013; Clarke et al., 2006; Magee et al., 2012; Visvader and Lindeman, 2012). Not all tumors may be organized in such a hierarchical manner (Meacham and Morrison, 2013; Quintana et al., 2010). The aggressive and highly metastatic nature of SCLC tumors suggests that SCLC tumors may harbor highly tumorigenic cells. However, the study of SCLC is challenging in patients because of the inherent complex genetic and environmental diversity of these patients. SCLC patients rarely undergo surgery and primary human material is scarce. Moreover, the establishment of SCLC cell lines and patient-derived xenografts can select for the growth of specific populations of tumor cells (Daniel et al., 2009; Leong et al., 2014), which may bias the analysis of cancer cell subpopulations. In contrast, relevant mouse models allow for the analysis of large number of independent primary tumors. The first mouse model for SCLC was developed based on the observation that human SCLCs are mutant for both the p53 and RB tumor suppressors (Meuwissen et al., 2003). The additional deletion of the Rb-related gene p130 enhances SCLC development (Schaffer et al., 2010). Rb/p130/p53 triple knockout (TKO) tumors have histopathological features of human SCLC, including an initial relative chemosensitivity followed by the acquisition of chemoresistance (Gazdar et al., 2015; Jahchan et al., 2013; Park et al., 2011).

Here we used mouse models and human SCLC cells to investigate tumor heterogeneity in SCLC. Because cancer stem cells may not possess the exact and full repertoire of normal tissue stem cell properties, we will instead use herein the term tumor-propagating cells (TPCs). We define TPCs as cells that are highly tumorigenic in transplantation assays and that can self-renew and differentiate into the bulk tumor population. We found that SCLC TPCs are highly abundant, proliferative, and not inherently chemoresistant in a mouse model. We also identified similar populations marked by high levels of the cell surface markers EpCAM and CD24 and low levels of CD44 in primary human explant models. Finally, we identified elevated MYC activity, in particular L-MYC, as a key determinant of the ability of SCLC TPCs to maintain the long-term growth of SCLC tumors.

Results

SCLC tumors contain a high fraction of tumor-propagating cells

To investigate the presence of TPCs in primary Rb/p130/p53 TKO tumors, we injected serial dilutions of tumor cell suspensions subcutaneously into NSG mice (Figure 1A–1B). In these assays, the calculated frequency of tumor initiation was ~1/128 (Figure 1C). This number is more than 10 times higher than what has been observed with mouse models of lung adenocarcinoma (Zheng et al., 2013) and similar to highly aggressive breast cancer models (Vaillant et al., 2008), suggesting that TPCs may be abundant in murine SCLC tumors.

Figure 1. Mouse SCLC tumors contain a high fraction of cells capable of tumor-propagating cells in transplantation assays.

(A) Workflow to identify tumor-propagating cells (TPCs) in a pre-clinical mouse model of SCLC (TKO, Rb/p53/p130 mutant).

(B) Representative flow cytometry analysis of TKO SCLC cells with markers of cell death (7AAD), Lineage (CD45, CD31, and Ter119), CD24, CD44, and EpCAM (n>20).

(C) Extreme limiting dilution analysis (ELDA) of Lineage-negative (Lin⁻, bulk tumor cells), CD24^High CD44^Low EpCAM^High/Low cells sorted from TKO tumors and injected subcutaneously in NSG mice.

Refer to Figure S1 for related information.

We next examined cell surface markers previously associated with TPCs in a few SCLC cell lines or in other solid tumor types, including CD133 (Jiang et al., 2009; Sarvi et al., 2014), CD90 (Salcido et al., 2010; Wang et al., 2013), c-KIT (Micke et al., 2003; Rygaard et al., 1993) and HGFR (Rygaard et al., 1993). Transplantation trials with 500–5,000 mouse tumor cells sorted for each of these markers independently showed no trend in enrichment (or loss) of tumorigenic potential (data not shown). In contrast, high levels of CD24 enriched for TPCs, and the transplantation ability of CD24^high cells was further increased by selecting for CD44^low and EpCAM^high cells (Figure 1B–1C and Supplemental Figure 1A–1B). Using a tumor-specific GFP reporter, we found that both TPC and non-TPC populations identified by these three markers are tumor cells (Supplemental Figure 1C–1D). These populations were also found in individually dissected tumors (Supplemental Figure 1E) and in an Rb/p53 mutant primary tumor (Supplemental Figure 1F). Overall, CD24^High CD44^Low EpCAM^High SCLC cells represent ~50% of live tumor cells (Supplemental Figure 1G). Thus, TKO tumors harbor a high number of SCLC cells with the ability to transplant tumors, and the combination of CD24^high, CD44^low, and EpCAM^high identifies cells that transplant 100 times more efficiently than other SCLC cell populations. We found that SCLC cells seed tumors in the liver and not the lungs following intravenous injection and do not reliably form lung tumors following intratracheal instillation, preventing us from investigating TPCs in an orthotopic model (data not shown).

SCLC patients rarely undergo surgery and we were not able to perform transplantation assays with fresh human specimens. Nevertheless, EPCAM and CD24 mRNA levels are high and CD44 expression is low in human SCLC cell lines and primary bulk tumors (Supplemental Figure 2A). Circulating tumor cells (CTCs) from SCLC patients form CDX (CTC-derived eXplant) tumors with a take rate of ~50% under the skin of NSG mice (Hodgkinson et al., 2014; Krebs et al., 2014) (Figure 2A). Although human SCLC CTCs are not enriched for EpCAM expression before assaying their transplantation ability and EpCAM-negative CTCs can be found in the blood of SCLC patients (Chudziak et al., 2016), CDX tumors express high mRNA levels of CD24 and EpCAM, and low levels of CD44 (Figure 2B). FACS analysis on three CDX models showed that an average of 41% of human SCLC cells was CD24^High CD44^Low EpCAM^High (42%, 27%, and 55% for CDX2, CDX3, and CDX4, respectively – Supplemental Figure 2C). These observations were confirmed by immunostaining for EpCAM and CD44 on tumor sections (Figure 2C). The frequency of TPCs using single cells dissociated from CDX models was similar to that seen with mouse TKO TPCs (Figure 2D, compare to Figure 1C). Cells in patient-derived xenografts (e.g. PDX model LU95) were nearly exclusively CD24^High CD44^Low EpCAM^High, and EpCAM-positive cells were able to confer tumorigenicity (Supplemental Figure 2D) (Saunders et al., 2015). Notably, high levels of EpCAM expression (but not CD24, data not shown) in early-stage tumors (George et al., 2015) correlated with decreased survival in SCLC patients (Supplemental Figure 2E), suggesting that a high level of EpCAM is associated with more aggressive tumors.

Figure 2. Xenografts derived from human SCLC circulating tumor cells harbor a high frequency of TPCs and high numbers of CD24^High CD44^Low EpCAM^High cells.

(A) Workflow to generate xenografts (CDXs) derived from human SCLC circulating tumor cells (CTCs).

(B) Expression (RPKM, from RNA-seq) of two genes expressed at high levels in SCLC (ASCL1, SYP) compared to the TPC markers EPCAM, CD24, and CD44 (n=3 CDXs). Negative controls: FOXN1, thymic and skin epithelium; MYBPC3, heart. Levels of the three MYC genes are also shown.

(C) Immunostaining for EpCAM and CD44 (brown signal) on CDX models. Lymphoma SUDHL8 cells were used as a negative control for EpCAM (inset) and H196 lung cancer cells were used as a positive control for CD44 (inset). Counterstain, hematoxylin. Scale bars, 100μm.

(D) Extreme limiting dilution analysis (ELDA) of cells from the CDX2, CDX3, and CDX4 models. Note that the data for CDX2 and CDX4 are not statistically different but the lower frequency for CDX3 is (p-values of 1.12×10⁻⁰⁷ for the CDX2 comparison and 3.51×10⁻⁰⁵ for CDX4).

Refer to Figure S2 for related information.

CD24^High CD44^Low EpCAM^High mouse SCLC cells give rise to TPCs and non-TPCs and can be serially transplanted

We next tested the long-term propagation ability of CD24^High CD44^Low EpCAM^High SCLC cells, focusing on TKO tumors because of the availability of numerous primary tumors (Figure 3A). TKO CD24^High CD44^Low EpCAM^High SCLC cells could generate new tumors at a similar frequency each time they were passaged (Figure 3A and Supplemental Figure 3A). The passaged tumors retained histopathological features of SCLC tumors, including expression of the neuroendocrine markers ubiquitin-C-terminal hydrolase 1 (Uchl1, also known as PGP9.5) and Ascl1 (Figure 3B and Supplemental Figure 3B). In addition, tumors initiated by CD24^High CD44^Low EpCAM^High cells gave rise to all the subpopulations of cells found in primary tumors (Figure 3C). Notably, CD24^High CD44^Low EpCAM^High cells were not quiescent and had similar proliferative rates compared to CD24^High CD44^Low EpCAM^Low cells or to all the non-TPCs populations combined (Figure 3D–3E and Supplemental Figure 3C). Thus, actively cycling CD24^High CD44^Low EpCAM^High SCLC cells self-renew and are enriched for the ability to propagate all tumor populations in this mouse model.

Figure 3. Serial transplantation reveals a stable TPC phenotype in the murine CD24^High CD44^Low EpCAM^High SCLC cell population.

(A) Serial transplantation assays (passages P1, P2, P3) of sorted Lineage-negative cells (Lin⁻: CD45⁻, CD31⁻, and Ter119⁻) and TPCs (CD24^High CD44^Low EpCAM^High) from TKO tumors (passage 0, P0). The estimated frequency of tumor formation is shown in red. Tumors from two TKO mice were passaged to P3 and analysis was done on 2 allografts per passage for each.

(B) Representative sections from the Lin⁻ P1, TPC P1, TPC P2, and TPC P3 allografts counterstained with hematoxylin and eosin (H&E) or immunostained for the neuroendocrine marker Uchl1. Scale bars, 50μm.

(C) Representative FACS plots of TPCs (red boxes) and non-TPCs (black boxes) from Lin⁻ P1, TPC P1, TPC P2, and TPC P3 allografts (n>2).

(D) Representative FACS histograms of TPCs and non-TPCs (CD24^High CD44^Low EpCAM^Low) tumor subpopulations labeled with the DNA replication marker EdU.

(E) Bar chart showing the frequency of EdU⁺ cycling cells in TPCs and non-TPCs (CD24^High CD44^Low EpCAM^Low) tumor subpopulations from 3 different Rb/p53/p130 mutant mice. CD24^High CD44^high and CD24^Low populations also showed non-significant differences in EdU incorporation (data not shown). Error bars indicate mean+/−SEM (n=3 mice); p-values are from paired t-test (p=0.1151); ns, not significant.

Refer to Figure S3 for information related to the frequency of the TPC population.

CD24^High CD44^Low EpCAM^High TPCs are not inherently chemoresistant

Cancer stem cells have been associated with chemoresistance and tumor relapse (Shafee et al., 2008; Visvader and Lindeman, 2012; Zheng et al., 2013). To investigate how TPCs responded to chemotherapy compared to non-TPCs, TKO mice carrying an inducible luciferase reporter allele (Rosa26^{LSL-Luciferase}) were monitored for tumor development in response to chemotherapy. At the end of each experiment, the relative frequency of TPCs was assessed by flow cytometry. Acute treatment with high doses of cisplatin and etoposide (Figure 4A) led to a significant induction of apoptosis in tumors, as monitored by cleaved caspase 3 (CC3 immunostaining) (Figure 4B and Supplemental Figure 4A) but there was no difference in the frequency of cells with TPC markers (CD24^High CD44^Low EpCAM^High) versus non-TPCs (CD24^High CD44^Low EpCAM^Low; CD24^low or CD24^High CD44^High) in this assay (Figure 4C and Supplemental Figure 4B). Analysis of TPCs and non-TPCs following 6 days of recovery after 3 days of acute treatment in a similar setting showed no differences in the relative percentages for these two populations; in addition, we observed no differences in the proliferative rate of TPCs from saline-treated mice and chemo-treated mice, suggesting that TPC populations do not rebound after treatment under these conditions (data not shown).

Figure 4. The murine CD24^High CD44^Low EpCAM^High TPC population is not inherently chemoresistant.

(A) Strategy used for the acute treatment of Rb/p53/p130 TKO mutant mice with saline and high doses of cisplatin (7.5mg/kg) + etoposide (15mg/kg) for 4 days.

(B) Analysis of cleaved caspase 3 (CC3) apoptotic cells on sections from TKO tumors treated acutely with saline or cisplatin and etoposide. Error bars indicate mean+/−SEM (n=3 mice); p-value is from a two-tailed unpaired Student’s t-test (p=0.0442).

(C) Relative frequency of CD24^High CD44^Low EpCAM^High cells from treated TKO mice. Error bars indicate mean+/−SEM (n=3 mice); ns, not significant, in a two-tailed unpaired Student’s t-test (p=0.8490).

(D) Strategy used for the treatment of Rb/p53/p130;Rosa26^{lox-Stop-lox-Luciferase} mice with saline and cisplatin (5mg/kg) once a week for 3 weeks.

(E) Fold-change of the tumor volume measured by luciferase activity in saline- and cisplatin-treated mice. Error bars indicate mean+/−SEM (n=3 mice); p-value is from a two-tailed unpaired Student’s t-test (p=0.0455).

(F) Relative frequency of TPCs (CD24^High CD44^Low EpCAM^High) in treated mice. Error bars indicate mean+/−SEM (n=3 mice); ns, not significant, in a two-tailed unpaired Student’s t-test (p=0.2674).

(G) Strategy used for the treatment of P2 allografts transplanted into NSG mice from single TKO tumors; pairs were treated with saline or cisplatin (5mg/kg) once a week for 3 weeks.

(H) Tumor mass from saline- (n=8 tumors) and cisplatin- (n=5 tumors) treated allografts. Error bars indicate mean+/−SEM; p-value is from a two-tailed unpaired Student’s t-test (p=0.0431).

(I) Relative frequency of TPCs (CD24^High CD44^Low EpCAM^High) in treated allografts. Error bars indicate mean+/−SEM (n=3 mice); ns, not significant, in a two-tailed unpaired Student’s t-test (p=0.1191).

(J) Strategy used for the treatment of Rb/p53/p130;Rosa26^{lox-Stop-lox-Luciferase} mice developing endogenous SCLC tumors and treated with saline or cisplatin (3mg/kg) weekly to generate chemonaïve and chemoresistant tumors (from Jahchan et al., 2013).

(K) Relative frequency of TPCs (CD24^High CD44^Low EpCAM^High) in treated mice. Error bars indicate mean+/−SEM (n=3 saline- and 3 cisplatin-treated mice); ns, not significant, in a two-tailed unpaired Student’s t-test (p=0.2172).

Refer to Figure S4 for related information.

The cisplatin/etoposide combination therapy was toxic to these tumor-bearing mice when applied for longer periods, so we used cisplatin only (5mg/kg) or control saline on a weekly basis for longer experiments (Figure 4D). Luciferase imaging confirmed that cisplatin inhibited overall tumor growth (Figure 4E). However, there was again no significant relative enrichment in CD24^High CD44^Low EpCAM^High populations versus non-TPCs in this assay (Figure 4F and Supplemental Figure 4C). To examine the response of single tumors to treatment, we grew isolated TKO tumors subcutaneously as allografts into NSG mice (Figure 4G). Cisplatin-treated allografts were significantly smaller than saline-treated tumors after three weeks of treatment, as expected (Figure 4H), but the relative frequency of CD24^High CD44^Low EpCAM^High cells remained the same as in the control group (Figure 4I and Supplemental Figure 4D). We recently described TKO mice carrying tumors treated weekly with cisplatin or saline until chemoresistant tumors were selected (Figure 4J) (Jahchan et al., 2013). The analysis of these chemoresistant tumors analyzed showed again no significant increase in the relative number of TPC cells versus non-TPCs (Figure 4K and Supplemental Figure 4E).

These experiments do not address how chemotherapy may affect the long-term propagation ability of CD24^High CD44^Low EpCAM^High cells but show that these cells are neither intrinsically resistant to chemotherapeutic agents used in SCLC patients nor enriched after chemoresistant tumors arise. Of note, the CDX3 and CDX4 models represent two extremes of chemotherapy response ((Hodgkinson et al., 2014) and unpublished data), yet in the limiting dilution studies, CDX4 tumors (least chemosensitive) had a lower frequency of tumor-initiating cells and fewer CD24^High CD44^Low EpCAM^High cells compared to CDX3 tumors (most chemosensitive) (Figure 2E–2F), uncoupling drug response from tumorigenicity. Thus, markers associated with TPCs may not be relevant biomarkers to predict relapse and the mechanisms of chemoresistance remains poorly understood in this cancer type.

Mouse SCLC TPCs retain their neuroendocrine differentiation

To identify candidate regulators of mouse SCLC TPCs and their long-term growth properties, we compared the transcriptional profiles of CD24^High CD44^Low EpCAM^High cells and non-TPCs (Figure 5A). Unsupervised clustering separated the TPCs and non-TPCs groups (Supplemental Figure 5A). Supervised clustering of genes differentially expressed at least 1.5-fold between the two populations identified 698 genes up-regulated in TPCs and 1919 up-regulated in non-TPCs (Figure 5B and Supplemental Table 1). Even at this low stringency for statistical cutoff, gene set enrichment analysis (GSEA) for Gene Ontology terms yielded no significantly enriched terms for TPCs, but the non-TPCs were enriched for terms related to extracellular matrix, wound remodeling, and chemokine/cytokine activity (Supplemental Table 2). The analysis of Oncogenic Signatures revealed that the TPCs genes were enriched in a neuronal/neuroendocrine signature (CAHOY_NEURONAL) (Supplemental Table 3).

Figure 5. SCLC TPCs retain their neuroendocrine differentiation and have elevated Mycl1 levels.

(A) Representative FACS plots to isolate TPCs (red gates, CD24^High CD44^Low EpCAM^High) and non-TPCs (black gates, CD24^Low; CD24^High CD44^High; CD24^High CD44^Low EpCAM^Low) populations in TKO tumors (n>3).

(B) Heatmap of the microarray analysis comparing TPCs and non-TPCs sorted from 3 different TKO mice. Yellow and blue indicate high and low expression, respectively. The values of the scale bar represent the median-centered log2 fold-change of each gene. Hierarchical clustering was performed on genes whose expression is at least 1.5-fold different between the two groups.

(C) Top 10 overrepresented transcription factors in the TPC population (based on total count of significantly enriched genesets).

(D) Quantitative RT-PCR analysis of Mycl1 mRNA levels in TPCs relative to non-TPCs (set to 1) from 3 TKO mice and one TKO P1 allograft. Arpp0 was used as an internal control to compare the different tumors. Error bars indicate mean+/−SEM (n=4 samples); p-values are from two-tailed paired Student’s t-test (p=0.0323).

(E) Quantitative RT-PCR analysis of Mycl1 mRNA levels in mouse Kp3 SCLC cells with shRNA knockdown using three different hairpins (sh6, sh7, and sh8). Arpp0 and B-actin were used as internal controls and data were plotted relative to the sh-scrambled vector control (shscr) (n=2 independent experiments; mean+/−SEM). p-values are from a two-tailed paired Student’s t-test (p=0.0142 for sh7).

(F) Representative L-Myc immunoblotting from Kp3 cells infected with a control vector (shGFP), and three sh-Mycl1 vectors. Hsp90 was used as a loading control. Quantification of the bands for L-Myc is shown below the blot.

(G) Cell growth assay (MTT) in Kp3 cells at different days upon Mycl1 knockdown. Fold growth is normalized to day 0. n=4 independent biological replicas with 3 technical replicas each; mean+/−SEM. Statistical significance was determined by two-tailed paired Student’s t-test (p= 0.016 at day 4 and p= 0.0416 at day 6 for sh8, all the other comparisons are not significantly different).

(H) Average of the number of colonies formed after single-cell sorting of Kp3 cells into 96-well plates. n=3 (for sh6) and n=4 (for sh7 and sh8) independent experiments with 3 technical replicas each; mean+/−SEM. Statistical significance was determined by two-tailed paired Student’s t-test (p= 0.0402 for sh6, p= 0.0017 for sh7, and p=0.0079 for sh8).

(I) Extreme limiting dilution analysis (ELDA) of shscr, sh6, sh7, and sh8 (Mycl1 knockdown) in mouse Kp1 SCLC cells injected at different dilutions into NSG recipient mice to assess tumor formation in vivo.

Refer to Figure S5 for related information.

Using a low stringency approach (see Experimental Procedures), 154 genes were identified as specifically up-regulated in TPCs (Supplemental Table 4). RT-qPCR analysis on an independent set of tumors confirmed the general up-regulation of a few of these genes in TPCs (Supplemental Figure 5B). Notably, Ascl1 expression was increased in TPCs compared to non-TPCs (Supplemental Table 4 and Supplemental Figure 5B). ASCL1 is a driver of neuroendocrine differentiation (Ito et al., 2000) and an oncogene in SCLC whose activity may promote the tumor-initiating capacity of SCLC cells (Augustyn et al., 2014; Jiang et al., 2009). We also found expression of the neuroendocrine marker CD56 (NCAM) on the surface of TPCs (Supplemental Figure 5C). Thus, the long-term self-renewal of mouse SCLC TPC populations is compatible with differentiated epithelial features.

Elevated MYC activity in SCLC TPCs drives the clonogenic ability of SCLC cells

We performed Enrichr analysis to identify transcriptional regulators of the 154 genes with elevated expression in TPCs. MYC was one of the top 10 overrepresented terms when counting the significantly enriched gene sets from all the databases in this analysis (Figure 5C and Supplemental Figure 5D). The MYC family of transcription factors (c-MYC, L-MYC, and N-MYC) is thought to be oncogenic in SCLC (Huijbers et al., 2014; Teicher, 2014; Wistuba et al., 2001), and a majority of human and mouse SCLC tumors express high levels of MYC factors (Supplemental Figure 2A–2B and Figure 2B); Mycl1 is the most frequently amplified Myc family gene in mouse SCLC tumors (Calbo et al., 2005; McFadden et al., 2014; Peifer et al., 2012). Expression of Mycl1 was specifically elevated in TPCs compared to non-TPCs (Supplemental Table 1), including in an independent set of mouse tumor samples (Figure 5D). MYCL1 levels were also higher in CD24^High CD44^Low EpCAM^high cells compared to CD24^High CD44^Low EpCAM^low cells in two CDX models (Supplemental Figure 5E).

To test for a possible functional role of Mycl1 elevated levels in mouse TPCs, we knocked down Mycl1 in murine Rb/p53 Kp3 cells, which are nearly entirely composed of CD24^High CD44^Low EpCAM^High cells (possibly by a selection process of tumorigenic cells upon passaging or simply illustrating the heterogeneity of SCLC tumors) (Supplemental Figure 5F). Three independent hairpins reduced L-MYC levels 2–5 fold (Figure 5E–5F) but had little to no effect on the growth of the tumor cells (Figure 5G). In contrast, the three hairpins strongly inhibited colony formation, an in vitro assay measuring the tumorigenic ability of single cells (Figure 5H). In murine Rb/p53 mutant Kp1 cells, in which the CD24^High CD44^Low EpCAM^High population represent less than half of the cells (data not shown), the sh6 hairpin knocked-down Mycl1 ~3-fold, had only modest inhibitory effects on growth, but also led to a significant decrease in the ability to form colonies (Supplemental Figure 5G–I). CD24^High CD44^Low EpCAM^High Kp1 cells formed more colonies than their CD24^High CD44^Low EpCAM^Low counterparts, and Mycl1 knockdown decreased the number of cells with TPC markers (Supplemental Figure 5J–5K). Decreased levels of Mycl1 significantly reduced the ability of Kp1 cells to form new tumors in the flanks of NSG mice (Figure 5I). Together, these experiments show that decreasing L-Myc levels inhibits the potential for long-term growth in mouse SCLC cell lines.

JQ1 treatment decreases the frequency of mouse SCLC TPCs and inhibits the tumorigenic potential of these cells

Inhibitors of bromodomain and extra-terminal (BET) proteins can target genes highly expressed in tumors, including MYC transcription factors in SCLC and other tumors (Bauer et al., 2012; Christensen et al., 2014; Delmore et al., 2011; Filippakopoulos et al., 2010; Loven et al., 2013; Puissant et al., 2013; Zuber et al., 2011). Treatment of Kp3 cells with escalating doses of the JQ1 inhibitor (Delmore et al., 2011) did not significantly affect the growth of these cell cultures at the lower concentrations tested (Figure 6A and Supplemental Figure 6A). In contrast, continuous JQ1 treatment strongly blocked the ability of single cells to form colonies at these lower concentrations (Figure 6B). Similar results were obtained for murine Kp1 cells and for human H69 cells (Supplemental Figures 6B–D and 6F–6G, respectively). The ability of various dilutions of mouse Kp1 cells to form tumors in NSG mice was significantly inhibited by JQ1 treatment at a relatively low dose (25mg/kg daily from the day of injection – compared to 50–100mg/kg in other studies (Bolden et al., 2014) (Delmore et al., 2011; Shimamura et al., 2013)) (Figure 6D). The tumors that ended up growing were smaller than control tumors, suggesting a dual effect of JQ1 on cancer re-initiation and long-term growth (Supplemental Figure 6J–6K). Pre-treatment with JQ1 for five days was sufficient to significantly decrease the frequency of CD24^High CD44^Low EpCAM^High cells (Figure 6E), which correlated with a decreased ability to form colonies (Figure 6F) and tumors (Figure 6G). These experiments suggest that a relatively low dose of JQ1 perturbs transcriptional networks in SCLC cells that are critical for the self-renewal and the long-term growth of these tumor cells.

Figure 6. Lowering Mycl1 levels in murine SCLC cells with JQ1 decreases the frequency of TPCs and inhibits the tumorigenic potential of these cells.

(A) MTT viability assay for mouse Kp1 SCLC cells after 48 hours of treatment with increasing doses of JQ1. Values from three independent experiments are shown as mean+/−SEM. The two-tailed paired Student’s t-test was used to calculate the p-values of the drug-treated cells versus control cells (p=0.0222 for JQ1 100nM; ns, not significant).

(B) Average of the number of colonies formed after single-cell sorting of Kp1 cells into 96-well plates. n=9 for JQ1 100nM and n=7 for JQ1 250nM; independent experiments with 3 technical replicas each; mean+/−SEM. Statistical significance was determined by two-tailed paired Student’s t-test (p=0.0003 for JQ1 100nM and p=0.0019 for JQ1 250nM).

(C) Quantitative RT-PCR analysis of Mycl1, Myc, and Mycn mRNA levels in mouse Kp1 SCLC cells treated with increasing doses of JQ1 for 24 hours. Arpp0 was used as an internal control and the numbers were plotted relative to the DMSO-treated Kp1 cells (n=6 independent experiments for JQ1 100nM and n=4 independent experiments for JQ1 250nM; mean+/−SEM). p-values are from a two-tailed paired Student’s t-test (p=0.0010 for Mycl1 and p=0.0464 for Mycn for JQ1 100nM, and p=0.0061 for Mycl1 and p=0.0450 for Mycn for JQ1 250nM; ns, not significant).

(D) Extreme limiting dilution analysis (ELDA) of Kp1 cells injected at different dilutions into NSG recipient mice to assess the frequency of tumor formation in vivo when treated with vehicle alone (Ctrl) or with JQ1 at 25mg/kg starting at day 0 of transplantation for 2 weeks. NA, not applicable.

(E) Relative frequency of TPCs (CD24^High CD44^Low EpCAM^High) in Kp1 cells treated with DMSO, or with 100nM and 250nM of JQ1 for 5 days. Error bars indicate mean+/−SEM (n=5 independent experiments); p-values are from two-tailed paired Student’s t-test (p=0.0230 for 100nM and p<0.0001 for 250nM).

(F) Average of the number of colonies formed after single-cell sorting into 96 well plates of pre-treated Kp1 cells with DMSO, JQ1 at 100nM (pre-JQ100), and JQ1 at 250nM (pre-JQ250) for 5 days. n=2 independent experiments with 3 technical replicas each; mean+/−SEM. Statistical significance was determined by two-tailed paired Student’s t-test (p= 0.0043 for pre-JQ250); ns, not significant.

(G) ELDA of pre-treated Kp1 cells with DMSO, and with JQ1 at 100nM and 250nM (JQ1 pre-treated) for 5 days that are injected at different dilutions into NSG recipient mice to assess the frequency of tumor formation in vivo.

(H) Quantitative RT-PCR analysis of Mycl1 mRNA levels in Kp1 cells stably infected with the pCDH-puro-L-Myc vector relative to uninfected Kp1 cells (Ctrl) (set to 1). Gapdh and ARPP0 were averaged and used as internal controls. Error bars indicate mean+/−SEM. p-values are from a two-tailed paired Student’s t-test (n=2 repeats; p=0.0063).

(I) Number of colonies from a single-cell sort of Kp1 cells stably infected with pCDH-puro-L-Myc and control uninfected Kp1 cells in media containing DMSO, 100nM, and 250nM of JQ1. Error bars indicate mean+/−SEM (n=3 independent experiments). Statistical significance was determined by two-tailed paired Student’s t-test between all the groups (p=0.046 and p=0.013 for the effects of JQ1 at 100mM and 250mM, respectively; p=0.021 for the effects of JQ1 in L-Myc-expressing cells both at 100mM and 250mM; p=0.0015 for the rescue effects of L-Myc on cells treated with JQ1 at 100mM; the effects of L-Myc on control cells are not significant).

Refer to Figure S6 for related information.

In all these experiments, JQ1 treatment reduced the expression of Myc family genes, especially Mycl1 (Figure 6C, Supplemental Figures 6E, 6H, and 6I). However, JQ1 has many more targets in cells than this gene family (Christensen et al., 2014; Lockwood et al., 2012; Tang et al., 2014). To determine if the inhibition of the clonogenicity of SCLC cells by JQ1 was at least in part mediated by its inhibitory effect on Myc genes, we overexpressed mouse L-Myc from retroviral constructs whose promoter may be less impacted by JQ1 than the genomic regulatory regions controlling the expression of endogenous genes. A 1.5–3-fold induction of L-Myc could in part rescue the effects of JQ1 treatment in a colony-formation assay in Kp3 cells (Figure 6H–6I); a similar partial rescue was observed in Kp1 cells using a different viral vector (Supplemental Figure 6L–6M). The mechanisms by which JQ1 treatment affects the biology of SCLC TPCs are likely complex but may in part be explained by the down-regulation of MYC activity.

While JQ1 treatment is clearly not equivalent to MYC inhibition, treatment with this small molecule provides a way to determine the consequences of a loss of long-term growth potential in SCLC TPCs in vivo. To test this possibility, we first repeated limited dilution transplantation assays of TPCs using mouse primary cells directly isolated from TKO tumors (Figure 7A). These assays showed a significant reduction in the number and the size of tumors growing in recipient mice after treatment with JQ1 compared to controls (Figure 7B–7C), indicating that JQ1 treatment inhibits the tumor re-initiation ability of SCLC TPCs as well as their long-term growth. Treatment of TKO;Rosa26^{LSL-Luciferase} mice after 5.5 months of Adeno-Cre infection with daily injections of JQ1 at 25mg/kg (Figure 7D) decreased the frequency of the CD24^High CD44^Low EpCAM^High population (Figure 7E–7F). At that time point, treatment significantly reduced the levels of Myc family genes, especially Mycl1 (Supplemental Figures 7A) but did not affect the general histology of the tumors (Supplemental Figure 7B). Importantly, this treatment led to a significant increase in survival compared to the control group (Figure 7G), indicating that a relatively low dose of JQ1 as a single agent can have therapeutic effects in SCLC.

Figure 7. Lowering Mycl1 levels in murine SCLC cells with JQ1 decreases the frequency of TPCs and inhibits the tumorigenic potential of these cells.

(A) Strategy used for sorting TPCs from TKO tumors and injecting them into recipient NSG mice treated with 25mg/kg of JQ1 for the first 2 weeks.

(B) Extreme limiting dilution analysis (ELDA) of the sorted TPCs in the vehicle (Ctrl) - and JQ1-treated groups was performed at the end of the experiment to assess the frequency of tumor formation in vivo.

(C) Representative images of the P1 allografts from the sorted TPCs injected in recipient NSG mice with the number of cells indicated and treated with vehicle (ctrl) or 25mg/kg of JQ1 for the first 2 weeks.

(D) Strategy used for the treatment of TKO mice 5.5 months after Adeno-Cre instillation with 25mg/kg of JQ1.

(E) Representative FACS plots of CD24^High CD44^Low EpCAM^High and CD24^High CD44^Low EpCAM^Low from tumors isolated from the vehicle (ctrl)- and JQ1- treated Rb/p53/p130 mutant mice ~1 month following treatment (n=4).

(F) Frequency of TPCs (CD24^High CD44^Low EpCAM^High) in Ctrl and JQ1-treated mice. Error bars indicate mean+/−SEM (n=4 pairs); p-values are from two-tailed paired Student’s t-test (p=0.0151).

(G) Survival curve generated from the Rb/p53/p130 mutant mice treated daily with IP injections of vehicle (Ctrl) and 25mg/kg of JQ1 starting at 5.5 months after Adeno-Cre instillation (Day 0 of treatment); median survival is 20.50 days for the vehicle- and 37 days for the JQ1-treated mutant mice; p=0.0043 by the Mantel-Cox test (n=12 vehicle-treated mice and n=12 JQ1-treated mice).

Refer to Figure S7 for related information.

Discussion

SCLC is the most fatal form of lung cancer. Here we found that mouse primary SCLC tumors include a significant fraction of long-term SCLC tumor-propagating cells and uncovered molecular features of these TPCs that may explain the aggressive nature of these tumors and may provide new therapeutic options in SCLC patients.

Intra-tumoral heterogeneity in SCLC and significant numbers of TPCs

Little is known about the subpopulations of tumor cells that may exist within SCLC tumors, and if these subpopulations are related to genetic or epigenetic changes (Calbo et al., 2005; Semenova et al., 2015). We found that that CD24^High CD44^Low EpCAM^High cells constitute around half of cells in primary murine SCLC tumors. The high number of TPCs is concordant with the high frequency of tumor formation from CTCs directly obtained from patients (Hodgkinson et al., 2014) and from patient-derived xenografts (PDX models) (Saunders et al., 2015). While we cannot exclude that a subpopulation of cells exist within CD24^High CD44^Low EpCAM^High TPCs that may be even more potent in its transplantation ability (and possibly more quiescent and/or more chemoresistant), our current studies point to a model in which cancer stem cells comprise a large fraction of SCLC tumors and are actively cycling.

A small subset of mouse CD44^High SCLC tumor cells express lower levels of neuroendocrine markers and can promote the metastasis of neuroendocrine tumor cells (Calbo et al., 2011; Kwon et al., 2015). These cells are present in the non-TPC subpopulations but our FACS analyses indicate that other non-neuroendocrine tumor cells (defined by low or no CD56 staining) are also present in this non-TPC subgroup (data not shown). The functional interactions between TPCs and various subtypes of non-TPCs may be critical in how SCLC tumors grow, spread, and respond to treatment. However, the identity and the exact roles that the various subpopulations within the non-TPC group may have in SCLC initiation, progression, and maintenance remain largely unknown. A better knowledge of these subpopulations may be critical to better understand tumor evolution and response to treatment.

Cell surface markers for SCLC TPCs

We found that the number of CellSearch-enumerated, EpCAM⁺ CTCs was an independent prognostic biomarker for patient overall survival using a cutoff of 50 CTCs per 7.5ml of blood (Hou et al., 2012). Moreover, in the 47 attempts where we failed to generate a CDX model from a SCLC patient’s blood sample, 34/47 samples (72%) had an EpCAM⁺ positive CTC count below this cutoff, whereas all 15 blood samples from patients where a CDX model was successfully derived had a EpCAM⁺ CTC count >50 (range 160 to >7000). Direct comparison of CellSearch, EpCAM⁺ CTC enumeration with a marker independent CTC enrichment based on cell size and deformability suggests that EpCAM⁻ SCLC CTCs also exist and may be abundant (data not shown), but overall, these human CTC data strongly suggest that it is the EpCAM⁺ CTC subpopulation that has tumor-initiating capacity. These observations do not exclude that additional markers exist that would further enrich for cells with even higher transplantation ability. Thus far, our experiments show no significant differences in the transplantation ability of CD24^High CD44^Low EpCAM^High cells that are CD90⁺ compared to CD90⁻ (frequencies of 1/320 versus 526, respectively, in these specific experiments). CD24^High CD44^Low EpCAM^High c-Kit⁺ and CD24^High CD44^Low EpCAM^High c-Kit⁻ also have similar calculated transplantation frequencies (1/102 versus 1/130, respectively, in these specific experiments). Other markers may be tested in the future in mouse and human models.

The rapid emergence of chemoresistant SCLC is not explained by the TPC model

The highly proliferative nature of TPCs in SCLC and their abundance may explain at least in part, why these tumors are initially responsive to chemotherapy (Pietanza et al., 2015; Semenova et al., 2015). However, the mechanisms underlying the rapid relapse of chemoresistant tumors remains unexplained and do not seem to be connected to the intrinsic biology of TPCs, which indicates that the frequency of TPCs would not be a good biomarker for the growth of chemoresistant disease in patients (also seen with a PDX model in (Saunders et al., 2015)). It is noteworthy that the least chemosensitive CDX model (CDX4) has a lower frequency of TPCs; this suggests that specific subpopulations of non-TPCs may be more chemoresistant and might serve as a protective niche for TPCs (Hartmann et al., 2005; Pardo et al., 2006). We did not observe a relative increase in CD44^high cells (Calbo et al., 2011) after chemotherapy treatment in our mouse models (data not shown). Similarly, while CD133 expression has been correlated with chemoresistance in SCLC cell lines (Kubo et al., 2013; Sarvi et al., 2014), we did not observe an induction of this marker in mouse chemoresistant tumors (data not shown). Thus, the existence of a possible chemoresistant reservoir in SCLC remains to be ascertained.

Molecular mechanisms underlying the differences between TPCs and non-TPCs

Our gene expression studies indicate that non-TPCs are less neuroendocrine than TPCs, but we do not fully understand yet why non-TPCs are less tumorigenic than TPCs. TPCs express high levels of neuroendocrine differentiation compared to non-TPCs, including Ascl1, a key oncogenic factor in SCLC (Augustyn et al., 2014; Jiang et al., 2009). Accordingly, treatment with a drug-conjugated antibody against DLL3, a transcriptional, target of ASCL1 (Nelson et al., 2009), was shown to inhibit the transplantation ability in human SCLC PDX models (Saunders et al., 2015). We focused on MYC based on its known oncogenic role in SCLC (Alves Rde et al., 2014; Calbo et al., 2005; George et al., 2015; Huijbers et al., 2014; Romero et al., 2014). Emerging evidence from in vitro studies in glioblastoma (Wang et al., 2008) and pancreatic cancer (Sancho et al., 2015) suggests that elevated MYC levels are required for the self-renewal and the long-term expansion of cells with features of cancer stem cells. Importantly, reducing MYC activity may be sufficient to achieve cancer inhibition without triggering the side effects that may be observed with more aggressive therapeutic interventions. However, the mechanisms by which MYC may control the self-renewal of adult stem cells and cancer stem cells are still largely unclear; in particular, whether these mechanisms overlap with mechanism by which MYC promotes cell growth and proliferation is not known (Wilson et al., 2004). We also do not know the mechanism by which Mycl1 transcription becomes elevated in TPCs compared to non-TPCs. In mouse neural cells, Ascl1 can directly bind to the Mycl1 promoter region (analysis not shown of data from (Webb et al., 2013)) and may control its expression. It is likely that a network of transcription factors controls the fate of SCLC TPCs, including MYC and ASCL1, and this network could be affected by JQ1 treatment (Augustyn et al., 2014; Lenhart et al., 2015).

Our ability to successfully identify highly tumorigenic cancer cells in genetic and patient-derived mouse models and the subsequent identification and targeting of transcription factors that regulate their growth, self-renewal and survival, could lead to more effective therapies based on the ability to eliminate TPCs rather than the bulk population of non-tumorigenic cancer cells.

Experimental Procedures

Human material and mouse tumors

PDX and CDX models were generated from extensive stage SCLC patients after ethical approval and patient consent as previously described (Anderson et al., 2015; Hodgkinson et al., 2014). Rb/p53/p130 triple knockout (TKO) mice that model human SCLC have been described extensively before (Gazdar et al., 2015; Jahchan et al., 2013; Park et al., 2011; Schaffer et al., 2010). For transplantation assays and analyses, SCLC tumors from TKO mice around 6–7 months of age were pooled, chopped, and then digested in 6ml of PBS containing 120μl of 100mg/ml of Collagenase/Dispase (Roche). Tumors were allowed to digest in a 37°C shaker for 45 minutes, followed by cooling down on ice before addition of 15μl of 1mg/ml DNAse (Sigma) for 5 minutes. Digested tissue was then passed through a 100μm filter then 40μm filter and red blood cells were lysed in 1ml of lysing solution (150mM NH4Cl, 10mM KHCO3, 0.1mM EDTA), to obtain a single cell suspension, which was counted and then stained with FACS antibodies. Nod.Cg-Prkdc^scidIL2rg^tm1Wjl/SzJ (NSG) immunocompromised mice were used for transplantation studies of allografts, and tumor cells were mixed with Matrigel (1:1) (BD Biosciences) before subcutaneous injection. For more details on the use of these models, including isolation and transplantation of single cell suspensions as well as drug treatments, see the Supplemental Experimental Procedures.

Cell culture

Growth conditions for mouse and human cell lines were previously described (Jahchan et al., 2013; Park et al., 2011; Schaffer et al., 2010). For more details on the various assays used in this study, see the Supplemental Experimental Procedures.

RNA and protein analyses

Protocols for immunoblotting and immunostaining, as well as the analysis of RNA expression were described before (George et al., 2015; Jahchan et al., 2010). For more details, see the Supplemental Experimental Procedures.

Image analysis and statistics

Statistical significance was assayed by Student’s t test with the Prism GraphPad software (two-tailed unpaired and paired t-test depending on the experiment). *: p-value<0.05; **: p-value<0.01; ***: p-value<0.005; ns: not significant. Data are represented as mean+/−SEM. For the survival curve analysis and comparison, the Mantel-Cox test was used. For limiting dilution analyses, ELDA software (Hu and Smyth, 2009) which uses the frequency of tumor-positive and negative injections at each transplant dose, was used to determine the stem cell frequency or tumor formation frequency of different groups by entering the numbers of successful outgrowths and numbers of total injections for each dilution (http://bioinf.wehi.edu.au/software/elda/). Expected frequencies are reported, as well as the 95% confidence intervals (lower and upper values are indicated). P-values comparing groups were calculated by the ELDA software (see http://www.statsci.org/smyth/pubs/ELDAPreprint.pdf for more information on how p-values were calculated).