Molecular features and evolutionary trajectory of ASCL1⁺ and NEUROD1⁺ SCLC cells

Abstract

Background

Small cell lung cancer (SCLC) is the most aggressive subtype of lung cancer without recognised morphologic or genetic heterogeneity. Based on the expression of four transcription factors, ASCL1, NEUROD1, POU2F3, and YAP1, SCLCs are classified into four subtypes. However, biological functions of these different subtypes are largely uncharacterised.

Methods

We studied intratumoural heterogeneity of resected human primary SCLC tissues using single-cell RNA-Seq. In addition, we undertook a series of in vitro and in vivo functional studies to reveal the distinct features of SCLC subtypes.

Results

We identify the coexistence of ASCL1⁺ and NEUROD1⁺ SCLC cells within the same human primary SCLC tissue. Compared with ASCL1⁺ SCLC cells, NEUROD1⁺ SCLC cells show reduced epithelial features and lack EPCAM expression. Thus, EPCAM can be considered as a cell surface marker to distinguish ASCL1⁺ SCLC cells from NEUROD1⁺ SCLC cells. We further demonstrate that NEUROD1⁺ SCLC cells exhibit higher metastatic capability than ASCL1⁺ SCLC cells and can be derived from ASCL1⁺ SCLC cells.

Conclusions

Our studies unveil the biology and evolutionary trajectory of ASCL1⁺ and NEUROD1⁺ SCLC cells, shedding light on SCLC tumourigenesis and progression.

Background

Small cell lung cancer (SCLC) is a particularly aggressive, lethal and widely metastatic cancer with a 5-year survival rate of about 7% [1]. SCLC is characterised by rapid tumour growth, early metastasis and acquired therapeutic resistance [2]. Genetically, SCLCs usually exhibit almost universal inactivation of TP53 and RB1, recurrent alterations of CREBBP, EP300, PTEN, SLIT2, SOX2, MYC family, and frequent dysregulation of the Notch, TGF-β and Hippo pathways [1, 3–5]. Most patients with SCLC present with metastases; thus, surgical resection is not suitable for most of these patients. The lack of SCLC specimens hinders the understanding of the molecular mechanisms of SCLC. For approximately three decades, no major clinical advances in SCLC treatment have been achieved, and it is therefore designated a “recalcitrant cancer” [6].

SCLC was previously regarded as a “homogenous” disease, but recent studies have indicated that there is considerable heterogeneity among SCLCs [3, 7–11]. Complementary data from human tumours, cell lines and mouse models of SCLC have classified four SCLC subtypes based on differential expression of four key transcription regulators: ASCL1, NEUROD1, YAP1 and POU2F3 [2, 12–16]. Recently, Rudin and colleagues generated single-cell RNA-Seq data sets of treated or untreated human SCLCs and found the highly inter-patient heterogeneity of SCLC. In addition, they confirmed the ASCL1-expressing, NEUROD1-expressing, and POU2F3-expressing SCLC subtypes and demonstrated distinct immune features and metastatic capabilities in ASCL1-expressing SCLC cells and NEUROD1-expressing SCLC cells [12]. This study leads to a new question: whether there is a cell-surface marker that can be used to distinguish distinct SCLC subtypes. Identifying the cell surface marker will help to purify distinct subtype of SCLC cells for further functional study.

Here, we generated single-cell RNA-Seq data of two resected human primary SCLC tissues. SCLC1 exhibits highly heterogeneity. Using single-cell RNA-Seq data of SCLC1, we demonstrated the coexistence of ASCL1⁺ and NEUROD1⁺ SCLC cells and identified EPCAM as a marker to distinguish ASCL1⁺ and NEUROD1⁺ SCLC cells. Based on EPCAM expression, we purified ASCL1⁺ and NEUROD1⁺ cancer cells from SCLC and large cell neuroendocrine lung cancer (LCNEC) cell line and found that NEUROD1⁺ cancer cells can be derived from ASCL1⁺ cancer cells and exhibited higher metastatic capability than ASCL1⁺ cancer cells. These results identified a cell surface marker that can be used to distinguish ASCL1⁺ SCLC cells and NEUROD1⁺ SCLC cells and provided experimental evidence of SCLC plasticity.

Materials and methods

Human samples and cell lines

Human lung cancer tissues were collected at Tianjin Medical University Cancer Hospital and Tianjin Medical University General Hospital. The use of all human lung cancer tissues was approved by the Institutional Review Board of Tianjin Medical University. Informed consents were obtained from all patients, and samples were deidentified prior to analysis. H1155, H69, H526, H146, H446 and H82 cells were obtained from ATCC within the past 10 years and were maintained in ATCC-recommended media. All experiments were performed within 1 month after thawing early-passage cells. All the cells were authenticated within 3 years. DNA purified from above cell lines were tested by the short tandem repeat analysis method using Promega PowerPlex 1.2 analysis system (Genewiz Inc.). Data were analysed using GeneMapper4.0 software and then compared with the ATCC databases for reference matching.

Sample preparation and single-cell RNA-Seq

Two SCLC tissue specimens were collected at Tianjin Medical University Cancer Hospital and Tianjin Medical University General Hospital. Resected tumours were transported in MACS Tissue Storage Solution (Miltenyi Biotec) on ice immediately after surgical procurement. A small fragment was fixed with 4% paraformaldehyde (Solarbio) at 4 °C overnight and was then paraffin embedded for immunofluorescence analysis. The remaining tumour was minced into tiny cubes (<1 mm³) and enzymatically digested with a MACS tumour dissociation kit (Miltenyi Biotec) for 60 min on a rotor at 37 °C, according to the manufacturer’s instructions. After digestion, the cells were centrifuged at 400 × g at 4 °C for 5 min, resuspended in 100 U/ml DNase I (Sigma–Aldrich, D4527) and incubated at 37 °C for 15 min. After filtering and centrifugation, the pellet was suspended in red blood cell lysis buffer (Solarbio) and incubated on ice for 2 min to lyse red blood cells. Finally, the cells were spun down and resuspended in PBS with 2% FBS, and DAPI, CD45 and CD31 triple-negative cells were sorted for single-cell RNA-Seq analysis.

For droplet-based single-cell RNA-Seq, single cells were captured and barcoded in a 10× Chromium Controller (10× Genomics). Single-cell RNA-Seq libraries were prepared using the Chromium Single Cell 3’v3 Reagent Kit (10× Genomics). Sequencing libraries were loaded into the Illumina NocaSeq system with 2 × 150 paired-end kits. The FASTQ files were analysed with the Cell Ranger Software Suite (version 3.1; 10× Genomics) against the GRCh38 human reference genome.

To visualise the processed data from Cell Ranger, we utilised Seurat suite version 3.1 to perform cell clustering. First, the filtered gene-barcode matrix of the sample identified by Cell Ranger Count was input into Seurat. Low-quality cells/dying cells were removed, and we retained the cells with proper molecular read count and gene number information (nCount_RNA < 50000, nFeature_RNA < 9000, and nFeature_RNA > 200 for SCLC1, nCount_RNA < 100,000, nFeature_RNA < 10,000, and nFeature_RNA > 200 for SCLC2) and with low mitochondrial transcript proportion (percent.mt <25). Then, we normalised and scaled the data and performed linear dimensional reduction by principal components analysis (PCA) on the scaled data. Next, we clustered and visualised the cell distribution by UMAP.

Inferred large-scale copy number variations analysis and whole-exome sequencing

UMAP visualisation revealed 15 transcriptionally distinct clusters, including fibroblasts, normal epithelial cells and lung cancer cells, in SCLC1, and 7 lung cancer cell clusters in SCLC2. To distinguish malignant cells from nonmalignant epithelial cells, we first excluded fibroblasts, which show transcription profiles distinct from those of epithelial cells, and then performed inferred large-scale copy number variations (CNVs) based on the expression profiles of the remaining single cells by averaging the expression over stretches of 100 genes on their respective chromosomes. As a control, we included published single cell RNA-Seq data of epithelial cells from normal human lung tissues [17]. In the CNV hierarchical cluster analysis, the normal lung epithelial cells clustered with single cells in Clusters 3, 7, 8, 12, and 14 in SCLC1; cells in Clusters 0, 1, 4, 5, 6, 9, 10, and 11 in SCLC1 (Clusters 0–6 in SCLC2) were classified as malignant cells.

For the whole-exome sequencing (WES) copy number variation calling, we performed this analysis in tumours and tumour-adjacent normal tissues. WES data by CNVkit (v0.9.9) with cnvkit.py batch command. The calling regions in genome were limited to the gene loci filtered in inferred CNVs and finally aligned to the inferred CNVs on the plot.

Trajectory analysis

The trajectory and pseudotime ordering of single cells were estimated based on the continuity of expression changes during cell development using Monocle 3 [18]. In brief, the cell expression profiles filtered and processed in Seurat were input into a standard Monocle 3 workflow with annotated clusters. After running PCA and UMAP for dimensionality reduction in Monocle 3, we performed trajectory graph learning and pseudotime calculation, and the result was visualised by performing the functions learn_graph() order_cells() and plot_cells(). For the pseudotime calculation, AT2 cells were set as the root in the timescale.

RNA velocity analysis

The raw expression profile of single-cell RNA-Seq data was filtered to obtain spliced and unspliced matrices through the velocyto run10x (v0.17) workflow. Expressed repeat annotations were downloaded from the UCSC Genome Browser. After running the velocyto workflow, the output loom file was input into the scVelo package [19] (v0.2.3) in Scanpy to estimate RNA velocity and latent time. The spliced RNA profile was used to perform PCA and UMAP dimensional reduction, and the results were visualised on a 2-dimensional plot. The velocity of cells is shown by black arrows indicating the predicted direction of development.

Single-cell RT-PCR

H526 single cells were isolated and transferred into 0.2 ml thin-wall PCR tube containing 4.9 µl of freshly prepared cell lysis buffer (1× PCR buffer containing 1.35 mM MgCl₂, 0.45% NP40, 4.5 mM DTT, 0.36 U RNase inhibitor and 0.045 mM dNTP mix). To pick up single cells, mouth-controlled suction was used to control a micropipette under a dissection microscope, permitting swift and efficient control of individual cell collection and release. The single cell in lysis buffer was incubated at 70 °C for 90 s, then immediately transferred to ice to release all mRNAs from the single cell. In all, 15 µl of RT mix was added to each tube, so that the total volume of each RT reaction was 20 µl per tube. First strand synthesis was performed at 42 °C for 60 min followed by inactivation of reverse transcriptase at 70 °C for 15 min. In all, 10 µl of cDNA was added to a new PCR tube containing 10 µl of PCR mix to amplify GAPDH, NEUROD1, POU2F3 and ASCL1 using FAM-labelled primers. The PCR was carried out at 95 °C for 3 min followed by 60 cycles of 95 °C for 30 s, 60 °C for 30 s, and 72 °C for 1 min. Sequences of primers used are provided in Supplementary Table 2. The PCR products were analysed by capillary electrophoresis.

Immunoblotting and immunostaining

Western blot analysis was performed using standard protocols. Blots were developed using ECL (Millipore). The following antibodies were used: anti-ASCL1 (Abcam, ab211327), anti-NEUROD1 (Abcam, ab60704), and anti-ACTB (Millipore, MAB1501).

SCLC tissues were obtained from the patients undergoing surgical resection with a histological diagnosis of SCLC at Tianjin Medical University Cancer Institute and Hospital in China. Samples were fixed with 4% paraformaldehyde at 4 °C overnight and paraffin embedded. Tumour sections were immunostained with antibodies against ASCL1 (Abcam, ab211327), NEUROD1 (Abcam, ab60704), EPCAM (Cell Signaling Technology, 14452s), and HLA-I (Abcam, ab185706), and DAB staining was performed following the standard protocol according to the manufacturer’s instructions (ZSGB-BIO). For immunofluorescence staining, the secondary antibodies used were Alexa Fluor 555, goat anti-mouse IgG (H + L) (Invitrogen, A-21422) and Alexa Fluor 488, goat anti-rabbit IgG (H + L) (Invitrogen, A-11008).

Flow cytometry

Cell staining was performed according to standard protocols. Antibodies used were EPCAM (Biolegend, 324212), CD45-PE (Sungene Biotech, H20451-09H), CD31-PE (Biolegend, 303106) and DAPI (BD Pharmingen™, 564907) according to manufacturer recommendations. Cells were washed with PBS with 2% FBS, and were sorted on a BD FACS Aria II, or analysed on a BD FACSVerse (BD Biosciences) and data were analysed by FlowJo 10.6.1.

Transwell assay of cell migration

Transwell inserts containing membranes with 8-μm pores (BD Biosciences) precoated with a mix of polylysine (Sigma–Aldrich, P4707) and laminin (Sigma–Aldrich, L2020) to improve cell adhesion were prepared. Tumour cells in serum-free medium (1 × 10⁵ cells per well) were seeded into the upper chamber and complete medium was placed in the lower chambers as a chemoattractant. Cells were incubated for 72 h at 37 °C in 5% CO₂. Experiments were performed in triplicate. Migrated cells on the underside of filter membranes were fixed with 4% formalin and stained with crystal violet. Cells in the top well were removed by wiping the top of the membrane with cotton swabs. The remaining cells were counted using light microscopy.

Xenograft assay

All animal procedures were approved by the Animal Care and Use Committee at Tianjin Medical University and confirmed to the legal mandates and national guidelines for the care and maintenance of laboratory animals. EPCAM^hi or EPCAM^lo H1155 cells (1 × 10⁶) were mixed with 5 × 10⁵ human lung cancer-associated fibroblasts (CAFs) in 100 µl of PBS containing 50% Matrigel and injected into the flanks of 8-week-old male NOD-SCID mice (10 mice for EPCAM^hi H1155 cells, 10 mice for EPCAM^lo H1155 cells). All mice were sacrificed when the largest tumour was 500 mm³, and the tumours in situ and lungs were fixed and embedded in paraffin. Paraffin blocks were cut into 4 μm sections. Tumour sections were stained with haematoxylin–eosin and with antibodies against ASCL1 and NEUROD1 for immunofluorescence analysis. Lung sections were stained with an antibody against HLA-I for immunochemical analysis.

Cisplatin and etoposide (C/E) administration

EPCAM^hi or EPCAM^lo H1155 cells (5 × 10⁵) were mixed with 2.5 × 10⁵ human lung CAFs in 100 µl of PBS containing 50% Matrigel and injected into the flanks of 8-week-old male NOD-SCID mice (5 mice for EPCAM^hi H1155 cells, 5 mice for EPCAM^lo H1155 cells). When the subcutaneous tumours had grown to 50–60 mm³, etoposide (8 mg/kg) and cisplatin (5 mg/kg) or vehicle were intraperitoneally administrated into the NOD/SCID mice. Tumour dimensions were measured by calipers every 2 days for 2 weeks and tumour volumes were calculated with the formula V = 1/2 (Length × Width²).

Statistical analysis

All the results are reported as the means ± SEM values unless otherwise noted. The correlation of the NEUROD1 (or ASCL1) expression level and the metastasis rate was determined with Kaplan–Meier analysis using the Mantel–Cox log-rank test (GraphPad Prism). Correlations between the ASCL1 and NEUROD1 expression levels were determined by Pearson correlation analysis and Spearman rank correlation analysis. A P value of <0.05 was considered statistically significant for all tests.

Results

Single-cell RNA sequencing revealed the intratumoural heterogeneity of SCLCs

To study intratumoural heterogeneity in SCLC, we performed droplet-based single-cell RNA-Seq using the 10× Genomics Chromium platform on fluorescence-activated cell sorting (FACS)-enriched single cells (CD45^-/CD31^-) dissociated from two freshly resected human primary SCLC tissues (SCLC1 and SCLC2) (Supplementary Table 1). SCLC1 contained 20–30% LCNEC component (Fig. 1a upper plot). After quality control (Supplementary Fig. 1A), 7,425 cells (2,001 in SCLC1 and 5,424 cells in SCLC2) were retained and grouped individually based on the expression profiles using uniform manifold approximation and projection (UMAP) (Fig. 1a lower plot, Supplementary Fig. 1B). To distinguish malignant cells from nonmalignant cells, we attempted to infer large-scale copy number variations (CNVs) for each cell by averaging relative expression levels over large genomic regions [20, 21]. In this analysis, we used reported single-cell RNA-Seq data on normal human epithelial cells as control [17]. We also generated bulk whole-exome sequencing (WES) data of SCLC1 and its tumour-adjacent normal lung to obtain CNVs of the predominant population in the tumour tissue to confirm the inferred-CNVs. Cells that exhibited similar CNV patterns with the reference cells were classified as nonmalignant cells in SCLC1 (Fig. 1b). For SCLC2, we identified 5 cells that exhibited similar CNV patterns with the reference cells in hierarchical clustering analysis and classified them as nonmalignant cells (Supplementary Fig. 1C). The nonmalignant cells were further denoted based on their expression characteristics (Supplementary Fig. 1D–F).

Fig. 1. Single-cell RNA-Seq revealed the intratumoural heterogeneity of SCLC1.

a Representative image of H&E staining of SCLC1 (upper plot). UMAP visualisation of 2,001 cells from SCLC1 showed transcriptionally distinct clusters (lower plot). Scale bars, 40 μm. b Inferred CNV profiles (top) and DNA whole-exome sequencing (WES) (bottom) were performed to identify malignant and nonmalignant cells from single-cell RNA-Seq of SCLC1. Reference normal cells and the corresponding clusters were labelled by colour codes. c UMAP visualisation of malignant cells in SCLC1 after re-clustering. d UMAP visualisation of NE-cell signature scores (SYP, INSM1, BEX1, CHGA, HES6, NCAM1) in malignant SCLC1 cells. e UMAP visualisation of the relative expression of SOX1, ELAVL3 and ELAVL4. f Blended visualisation of the expression of ASCL1 and NEUROD1 in malignant SCLC1 cells. g Representative immunostaining of ASCL1 and NEUROD1 in primary human SCLC samples. Scale bars, 40 μm. h UMAP visualisation of the relative expression of POU2F3 and YAP1. i Heatmap showed the expression levels of the top 50 differentially expressed genes in Cluster 3 and Clusters 0, 1, 2, 4, 5, and 6. j GSEA showed the top ten enriched pathways positively correlated with cells in Cluster 3.

In SCLC1, we identified 8 subclusters of malignant cells. Clusters 0, 1, 2, 4, 5, and 6 exhibited high neuroendocrine (NE) scores (Fig. 1c, d), expressed high level of SOX1, ELAVL3, ELAVL4, markers that are highly expressed in SCLCs but repressed in LCNECs [22] (Fig. 1e), and were classified as NE SCLC cells. Consistent with the previously reported expression patterns of ASCL1 and NEUROD1 in SCLC cell lines [11, 13, 15, 23], ASCL1 and NEUROD1 were mutually exclusively expressed in NE SCLC cells. Cells in Clusters 0, 1, 2, 4, and 5 expressed ASCL1 but not NEUROD1, whereas those in Cluster 6 expressed NEUROD1 but not ASCL1. Very few malignant cells expressed both ASCL1 and NEUROD1 (Fig. 1f). The mutually exclusive expression of ASCL1 and NEUROD1 was confirmed by immunostaining of the corresponding SCLC tissue sections (Fig. 1g). In SCLC2, the majority of cancer cells expressed ASCL1. NEUROD1⁺ cells were not clustered together; however, the expression of NEUROD1 and ASCL1 still exhibited a mutually exclusive pattern (Supplementary Fig. 1G, H).

Interestingly, we observed that a small subset of malignant cells (Cluster 3) lacked ASCL1, NEUROD1, YAP1 and POU2F3 (Fig. 1h). Moreover, cluster 3 did not express SOX1, ELAVL3, or ELAVL4, suggesting that cluster 3 might be LCNEC cells (Fig. 1e). Compared to NE SCLC cells, cluster 3 expressed high levels of CHCHD2, NDUFB11, LSM4, HMGN2 and a large number of ribosomal proteins, suggesting that these malignant cells were undergoing very active protein synthesis (Fig. 1i). Consistent with these findings, gene set enrichment analysis (GSEA) showed that these cells exhibited increased activation of mitochondrial ATP synthesis coupled electron transport and oxidative phosphorylation pathways (Fig. 1j).

Identification of EPCAM as a cell surface marker to distinguish ASCL1⁺ SCLC cells from NEUROD1⁺ SCLC cells

We next sought to understand the cellular state of these ASCL1⁺ and NEUROD1⁺ SCLC cells. We ranked the differentially expressed genes in NEUROD1⁺ cells versus ASCL1⁺ cells by log2-transformed fold change values and applied GSEA to identify the differential functional gene sets for each cluster. Consistent with the recent report showing that NEUROD1⁺ SCLC cells exhibit enriched upregulated neuron differentiation and neuropeptide signalling pathway compared to ASCL1⁺ SCLC cells [12], we found that the neuron migration and forebrain cell migration pathways were significantly upregulated in NEUROD1⁺ SCLC cells compared with ASCL1⁺ SCLC cells in our single-cell RNA-Seq data set. We also observed that phasic smooth muscle contraction pathway was upregulated in NEUROD1⁺ SCLC cells. The enriched downregulated pathways in NEUROD1⁺ cells included the protein targeting to ER, ATP synthesis coupled electron transport, keratinisation, and epidermis development pathways (Fig. 2a). Epithelial marker genes, including EPCAM, CDH1, KRT8, KRT18, and KRT19; the cellular tight junction genes CLDN3, F11R, and CLDN7; and the cell-cell adhesion-related genes ARG2, EGR3 and DMTN, were reduced in NEUROD1⁺ cells compared with ASCL1⁺ cells (Fig. 2b). In contrast, genes related to cell–matrix adhesion, such as EFNA5, CASK, and ITGA7, as well as cell–neuron migration signature genes, including NRP2 and RELN, were upregulated in NEUROD1⁺ cells compared to ASCL1⁺ cells (Fig. 2b, c). Specifically, a cell surface protein EPCAM was differentially expressed in ASCL1⁺ and NEUROD1⁺ cells, which was confirmed by the immunostaining for EPCAM, NEUROD1, and ASCL1 in the corresponding SCLC tissue sections (Fig. 2d, e) and other SCLC tissues (Supplementary Fig. 2A). Consistent with these findings, analysis of single-cell RNA-Seq data of 4 primary human SCLC samples generated by Rudin and colleagues [12] also revealed a positive correlation between EPCAM and ASCL1 and a negative correlation between EPCAM and NEUROD1 (Fig. 2f, g and Supplementary Fig. 2B). These transcriptional differences suggest that NEUROD1⁺ SCLC cells exhibit a decreased epithelial phenotype but an enhanced neural phenotype compared with ASCL1⁺ SCLC cells and that EPCAM may be used as a cell-surface marker to distinguish ASCL1⁺ and NEUROD1⁺ SCLC cells.

Fig. 2. NEUROD1⁺ SCLC cells showed reduced epithelial features and lack EPCAM expression.

a Selected enriched pathways identified by GSEA in NEUROD1⁺ versus ASCL1⁺ SCLC cells. b Heatmap showed differentially expressed genes involved in epithelial markers, tight junctions, cell–cell adhesion and cell–matrix adhesion. c UMAP visualisation of neuron migration-related genes in malignant SCLC1 cells. d UMAP visualisation of EPCAM, ASCL1, NEUROD1 (left plots) and the quantitation of EPCAM expressing cells in ASCL1⁺ or NEUROD1⁻ cancer cells, and ASCL1 or NEUROD1 expressing cells in EPCAM^hi cancer cells in SCLC1 analysed by single-cell RNA-Seq (right plot). e Costaining of EPCAM, ASCL1 and NEUROD1 on serial sections of SCLC1 tissue (left plots). Quantitation of the NEUROD1⁺ or NEUROD1^- in EPCAM^hi, and ASCL1⁺ or ASCL1⁻ in NEUROD1⁺ cancer cells in SCLC1 tissue sections (right plot). The data were presented as the means ± SEMs (n = 3 per group). Scale bars, 40 μm. f The relative expression of EPCAM, ASCL1 and NEUROD1 in malignant cells of the published single-cell RNA-Seq data of lung biopsy from SCLC patient (RU426). g Correlation of ASCL1, EPCAM, and NEUROD1 expression in malignant SCLC cells (RU426, RU1066, RU1145 and RU1229A). Fitted curves are shown with 95% confidence intervals.

To further confirm the correlation between EPCAM, ASCL1 and NEUROD1 in SCLC cells, we examined their expression in multiple human SCLC cell lines. FACS analysis showed that EPCAM expression was extremely high in H69 and H146 cells but was low in H82 and H446 cells (Fig. 3a). A previous report defined H69 and H146 cells as ASCL1-expressing subtypes and H82 and H446 cells as NEUROD1-expressing subtypes [24], supporting that EPCAM is highly expressed in ASCL1⁺ but repressed in NEUROD1⁺ SCLC cells. H526 cells were heterogeneous for EPCAM expression with 3-4% cells lacking EPCAM (Fig. 3b). H526 cells were classified as a POU2F3-expressing subtype [25]. Consistently, analysis of a reported RNA-Seq data set of H526 cells generated by Gay et al. showed that POU2F3 was highly expressed [23]. However, transcription of ASCL1 and NEUROD1 were also detected, albeit at much lower levels compared to POU2F3 (Supplementary Fig. 3A). To further understand the molecular features of the H526 cell line, we sorted single H526 cells and tested the expression of POU2F3, ASCL1, NEUROD1, and GAPDH in a single PCR reaction. Among 60 GAPDH-expressing H526 cells, 35 cells expressed POU2F3, consistent with the previous finding that NCI-H526 is a POU2F3-expressing SCLC line. However, we also observed the expression of ASCL1 in 1 H526 cell and NEUROD1 in 2 H526 cells. These three proteins were expressed in an exclusive pattern. Another 22 H526 cells did not express any of these three proteins (Supplementary Fig. 3B, C). We next purified EPCAM^hi and EPCAM^lo H526 cells and evaluated the expression of ASCL1 and NEUROD1. Consistent with the above single-cell RNA-Seq data, EPCAM^hi H526 cells expressed a high level of ASCL1, whereas EPCAM^lo H526 cells expressed a high level of NEUROD1 (Fig. 3c). Likewise, a LCNEC cell line H1155, which exhibits similar transcription profiling as SCLC cells [26, 27] and is also heterogeneous for EPCAM expression (with 15% cells lacking EPCAM), exhibited similar EPCAM expression pattern: EPCAM^hi H1155 cells expressed a high level of ASCL1, whereas EPCAM^lo H1155 cells expressed a high level of NEUROD1 (Fig. 3d, e).

Fig. 3. Identification of EPCAM as a cell surface marker to distinguish ASCL1⁺ SCLC cells from NEUROD1⁺ SCLC cells.

a Flow cytometric analysis of cell surface expression of EPCAM in H69, H82, H146 and H446 cell lines. b Gating strategy for flow cytometric sorting of H526 cells with high and low EPCAM expression. c Immunoblot analysis of ASCL1 and NEUROD1 expression in EPCAM^hi and EPCAM^lo H526 cells on Day 0. d Gating strategy for flow cytometric sorting of H1155 cells with high and low EPCAM expression. e Immunoblot analysis of ASCL1 and NEUROD1 expression in EPCAM^hi and EPCAM^lo H1155 cells on Day 0. f Phase contrast micrographs of EPCAM^hi and EPCAM^lo H526 and H1155 cell morphologies after culture for 1, 3 and 30 days. Micrograph of EPCAM^lo H526 cells on day 30 was not shown as the cells died after several days in culture. Scale bars, 20 μm.

ASCL1⁺ cancer cells can transform into NEUROD1⁺ cancer cells

Interestingly, purified EPCAM^hi and EPCAM^lo H526 and H1155 cells exhibited different growth patterns when they were cultured. On Day 1 after seeding, EPCAM^hi H526 and H1155 cells formed aggregates whereas EPCAM^lo H526 and H1155 cells grew as single cells, consistent with previous reports [12, 28]. Purified EPCAM^hi H526 and H1155 cells grew well in culture. On Day 30 after seeding, the vast majority of EPCAM^hi H526 and H1155 cells still grew as nonadherent floating aggregates, although a small subset of H1155 cells was attached to the plate. EPCAM^lo H1155 cells grew as loose aggregates or as single cells, with slight substrate adherence (Fig. 3f). These differences in growth patterns were consistent with the lower expression of cell–cell adhesion genes and higher expression of cell–matrix adhesion genes in NEUROD1⁺ SCLC cells than in ASCL1⁺ SCLC cells. EPCAM^lo H526 cells died after several days in culture.

To investigate whether a transition occurs between EPCAM^hi and EPCAM^lo cells, we cultured purified EPCAM^hi H526 and H1155 cells and EPCAM^lo H1155 cells and evaluated the change in populations using flow cytometry. EPCAM^lo H1155 cells retained their EPCAM^lo population during culture. However, in EPCAM^hi H526 and H1155 cell cultures, EPCAM^lo cells started to appear after 3 days of culture, and the percentage of EPCAM^lo cells continued to increase during culture (Fig. 4a, c). Consistent with the appearance of EPCAM^lo populations, NEUROD1 expression increased (Fig. 4b, d). To further confirm this result, we isolated single clones of EPCAM^hi H1155 and tested the frequency of transformation from EPCAM^hi to EPCAM^lo cells. After 30 days of culture, eight EPCAM^hi (ASCL1⁺) clones grew. While one EPCAM^hi clone retained EPCAM^hi state, 7 out of 8 EPCAM^hi clones transformed and stayed at different intermediate transformation stages (Fig. 4e), suggesting that the transcriptional programme controlling the transformation are heritable during cell cycle and that H1155 cells are highly heterogenous. These results indicate that ASCL1⁺ SCLC cells can transform into NEUROD1⁺ SCLC cells; however, NEUROD1⁺ SCLC cells are not able to transform into ASCL1⁺ SCLC cells. These results are consistent with the previous study showing that temporal shift in SCLC from ASCL1⁺ to NEUROD1⁺ using mouse models [15]. Knockdown of EPCAM in ASCL1⁺ H526 and H1155 cells have no effect in expression of ASCL1 or NEUROD1 (Fig. 4f), indicating that EPCAM is not involved in this subtype transition.

Fig. 4. ASCL1⁺ cancer cells can transform into NEUROD1⁺ cancer cells.

a Flow cytometric analysis of cell surface expression of EPCAM on unsorted H526 cells and sorted EPCAM^hi and EPCAM^lo H526 cells after culture for 0, 3 and 30 days. Unstained H526 cells were used as controls. Three independent experiments were performed. b Immunoblot analysis of ASCL1 and NEUROD1 expression in EPCAM^hi H526 cells after 30 days of culture. c Flow cytometric analysis of cell surface expression of EPCAM on unsorted H1155 cells and sorted EPCAM^hi and EPCAM^lo H1155 cells after culture for 0, 3 and 30 days. Unstained H1155 cells were used as controls. Three independent experiments were performed. d Immunoblot analysis of ASCL1 and NEUROD1 expression in EPCAM^hi and EPCAM^lo H1155 cells after 30 days of culture. e Flow cytometric analysis of cell surface expression of EPCAM on unsorted H1155 cells and isolated single clones of EPCAM^hi H1155 cells after culture for 30 days. Unstained and sorted EPCAM^hi H1155 cells were used as controls. Three independent experiments were performed. f Immunoblot analysis of ASCL1, NEUROD1 and EPCAM expression in EPCAM^hi and EPCAM^lo H526 (left plots) and H1155 cells (right plots) after EPCAM knockdown. g Pseudotime analysis of malignant and AT2 cells by Monocle 3, indicating a developmental trajectory from ASCL1⁺ to NEUROD1⁺ SCLC cells. h RNA velocities were visualised on the diffusion map projection of cells in SCLC1.

To confirm the developmental trajectory from ASCL1⁺ to NEUROD1⁺ SCLC cells in tumour tissues, we performed pseudotime analysis on the single-cell RNA-Seq data using Monocle 3 and observed a developmental trajectory from a subset of ASCL1⁺ SCLC cells (Clusters 4, 6) to NEUROD1⁺ SCLC cells (Cluster 10) when normal epithelial cells (AT2 cells) were annotated as the trajectory’s root (Fig. 4g). Consistent with these findings, using RNA velocity, a method inferring precursor cell dynamics, we also observed a clear directional flow from a subset of ASCL1⁺ (Clusters 4 and 6) to NEUROD1⁺ SCLC cells (Cluster 10). However, we did not observe directional flow from other ASCL1⁺ (Clusters 0, 1, 9) to NEUROD1⁺ SCLC cells (Fig. 4h). These collective data suggest that heterogeneity exists among ASCL1⁺ SCLC cells and that not all ASCL1⁺ SCLC cells can transform into NEUROD1⁺ SCLC cells, which might be the reason that H69 and H146 cells maintain an ASCL1⁺ phenotype. This finding is consistent with the physiological expression pattern of ASCL1, which is expressed earlier than NEUROD1 during neural system development [29, 30].

NEUROD1⁺ SCLC cells exhibited higher metastatic capability than ASCL1⁺ SCLC cells

We next sought to explore the metastatic capability of ASCL1⁺ and NEUROD1⁺ cancer cells. Given that it is difficult to purify enough NEUROD1⁺ H526 cells for xenograft assay, we used H1155 cells to address this question. Purified EPCAM^lo H1155 cells exhibited an enhanced migratory capability compared with EPCAM^hi H1155 cells in the Transwell migration assay (Fig. 5a). Then, we subcutaneously inoculated EPCAM^hi and EPCAM^lo H1155 cells in combination with CAFs into NOD/SCID mice. Four weeks after inoculation, EPCAM^hi and EPCAM^lo H1155 cells generated subcutaneous tumours of comparable sizes, indicating that these two subtypes have similar proliferative indices in vivo (Fig. 5b). However, EPCAM^hi and EPCAM^lo H1155 subcutaneous tumours exhibited different histological morphologies. EPCAM^lo H1155 tumour cells contained one or more prominent nucleoli with paranucleolar chromatin clearing. However, EPCAM^hi H1155 tumours showed a mixed histological pattern: while the majority of tumour cells contained dense nuclei, a small subset of tumour cells exhibited a morphology similar to that of EPCAM^lo H1155 tumour cells. Immunostaining of the subcutaneous tumour sections detected both ASCL1⁺ and NEUROD1⁺ tumour cells in EPCAM^hi H1155 tumours but only NEUROD1⁺ tumour cells in EPCAM^lo H1155 tumours (Fig. 5c), further confirming the transition from ASCL1⁺ to NEUROD1⁺ tumour cells.

Fig. 5. NEUROD1⁺ cells exhibited higher metastatic capability than ASCL1⁺ cells both in vivo and in vitro analyses.

a EPCAM^hi and EPCAM^lo H1155 cells were subjected to a migration assay (left plots), and the numbers of migrated cells were quantified (right plot). Scale bars, 20 μm. Means ± SEM represent 5 visualised areas in one experiment. Three independent experiments were performed. ****P < 0.0001 by unpaired two-tailed Student’s t test. b EPCAM^hi and EPCAM^lo H1155 cells were subcutaneously injected into NOD/SCID mice. Tumour volumes were shown. The data were presented as the means ± SEM (n = 10 per group). ns non-significant by two-way ANOVA. c Representative images of H&E staining and immunostaining for ASCL1 and NEUROD1 of subcutaneous xenografts formed by EPCAM^hi and EPCAM^lo H1155 cells and quantitation of the ASCL1⁺ and NEUROD1⁺ cancer cells in xenograft tumour tissues. The data were presented as the means ± SEMs (n = 5 per group). Scale bars, 40 μm. d Representative H&E stained sections of EPCAM^hi or EPCAM^lo H1155 tumours. The invasion area was indicated by the black arrows. Scale bars, 20 μm. e Representative images of immunostaining for HLA-1 in lung sections from mice subcutaneously injected with EPCAM^hi or EPCAM^lo H1155 cells (left plots). Quantitation of lung metastatic nodules (right plot). Scale bars, 40 μm. The data were presented as the means ± SEMs (n = 12 per group). ****P < 0.0001 by unpaired two-tailed Student’s t test. f Table indicating the information of the human sample donors (top left plot: Patients #1 to #10, PT primary tumour, MET metastatic tumours). Representative immunofluorescence staining of (primary or metastatic) human SCLC sections for ASCL1 and NEUROD1 expression (bottom left plots). Scale bars, 20 μm. The percentages of ASCL1 and NEUROD1 positive cancer cells in SCLC primary and metastatic samples were shown as the means ± SEM (right plot). *P < 0.05; **P < 0.01; ****P < 0.0001 by unpaired two-tailed Student’s t test. g Sensitivity of EPCAM^hi and EPCAM^lo H1155 cells to standard SCLC chemotherapy. Tumour volumes were shown. The data were presented as the means ± SEM (n = 5 per group). ns non-significant; **P < 0.01; ***P < 0.001 by two-way ANOVA.

In addition, we observed that EPCAM^hi H1155 tumours were well circumscribed with no invasion, whereas EPCAM^lo H1155 cells invaded the surrounding tissues, suggesting a higher invasive capability of EPCAM^lo than EPCAM^hi H1155 cells (Fig. 5d). Consistent with these findings, when we stained the lung tissue sections for HLA-I, which can distinguish human tumour cells from mouse cells, we observed many more human cancer cells seeded in the lung tissue in mice bearing EPCAM^lo H1155 cells than in those bearing EPCAM^hi H1155 cells (Fig. 5e). These results suggest that NEUROD1⁺ cancer cells exhibit higher metastatic capability than ASCL1⁺ cancer cells, consistent with previous studies [12, 31].

To further confirm this result, we tested the expression of ASCL1 and NEUROD1 in 6 resected human primary SCLC tumour tissues and 8 lymph node (patients 8, 9, and 10) or brain metastases (patients 3, 4, 5, 6, and 7) from 10 individual patients using immunostaining. Consistent with the in vitro and in vivo analyses, ASCL1⁺ tumour cells were enriched in primary tumour tissues, whereas NEUROD1⁺ tumour cells were enriched in metastases (Fig. 5f).

ASCL1⁺ and NEUROD1⁺ SCLCs exhibit similar responses to etoposide and cisplatin

We next evaluated the response of the ASCL1⁺ and NEUROD1⁺ populations to SCLC standard systemic chemotherapeutic drugs for SCLC (etoposide and cisplatin) using a xenograft animal model. When the subcutaneous tumours had grown to 50–60 mm³, etoposide and cisplatin or vehicle were intraperitoneally administrated into NOD/SCID mice (5 mg/kg cisplatin on Day 1 and 8 mg/kg etoposide on Days 1, 2 and 3, or corresponding vehicle only) [32]. Both EPCAM^hi and EPCAM^lo H1155 subcutaneous tumours exhibited responses to etoposide and cisplatin (Fig. 5g). Thus, although EPCAM^hi and EPCAM^lo tumour cells exhibit multiple different features, they have similar sensitivity to standard SCLC chemotherapeutic drugs.

Discussion

In this study, we generated single-cell RNA-Seq data of two human primary SCLCs. SCLC1 is more heterogeneous than SCLC2. SCLC1 contains ASCL1⁺ and NEUROD1⁺ subtypes and each subtype clustered together. Compared with ASCL1⁺ cancer cells, NEUROD1⁺ cancer cells exhibit more neuron specification and neuronal migration characteristics, consistent with previous report [12]. Moreover, NEUROD1⁺ cells exhibit reduced epithelial features and lack EPCAM expression. Thus, our studies identify a cell surface marker to distinguish ASCL1⁺ SCLC cells and NEUROD1⁺ SCLC cells, which makes sorting distinct subtype of SCLC cells for deep molecular characterising possible.

It has been reported that some circulating tumour cells (CTCs) in peripheral blood of SCLC patients express EPCAM but some CTCs do not [33–35]. Our data support these observations and further show that EPCAM is highly expressed in ASCL1⁺ but repressed in NEUROD1⁺ cancer cells. Based on EPCAM expression, we purified ASCL1⁺ and NEUROD1⁺ cells from H1155 cells and revealed different migration and metastatic potential of these two subtypes. NEUROD1⁺ H1155 cells exhibit higher motility in vitro and higher metastatic capability in xenograft mouse model than ASCL1⁺ H1155 cells. These findings are consistent with mouse models [31] and patient samples showing that ASCL1-expressing SCLC cells are overrepresented in primary tumours, whereas NEUROD1-expressing SCLC cells are enriched in nodal and distant metastases [12].

Oliver and colleagues have reported that Myc can drive transition from ASCL1⁺ to NEUROD1⁺ state in SCLC mouse models [15]. Using human SCLC cell lines, we show that some ASCL1⁺ cancer cells can transform into NEUROD1⁺ cancer cells. ASCL1⁺ H526 cells can transform into NEUROD1⁺ H526 cells in vitro and ASCL1⁺ H1155 cells can transform into NEUROD1⁺ H1155 cells both in vitro and in xenograft mouse model. In addition, RNA velocity analysis of single-cell RNA-Seq confirms the flow from ASCL1⁺ SCLC cells to NEUROD1⁺ SCLC cells. These data suggest that ASCL1⁺ SCLC cells act as progenitors of NEUROD1⁺ SCLC cells, consistent with their physiological role in neural system development. Physiologically, ASCL1 is expressed in multipotent cells at very early neurogenesis and functions as a proneural protein to endow progenitors with a neuronal fate [36, 37]. NEUROD1 is expressed in cells already committed to a neuronal fate and promote their differentiation [29]. However, our studies show that not all ASCL1⁺ SCLC cells can transform into NEUROD1⁺ SCLC cells, i.e., SCLC lines H69 and H146. What determines the transformation from ASCL1⁺ SCLC cells to NEUROD1⁺ SCLC cells needs further experimentation. Given that NEUROD1⁺ SCLC cells have higher metastatic capability, blocking the transition might be an effective treatment strategy for SCLCs to slow down metastasis.

The limitation of this study is that the single-cell RNA-Seq data are generated from only 2 human SCLC tissues and these two SCLC tissues are very different. SCLC1 contains 25% LCNEC component and is highly heterogenous. ASCL1⁺ and NEUROD1⁺ cells in SCLC1 formed different clusters and exhibit similar CNV patterns, suggesting the highly intratumoural plasticity. However, cancer cells in SCLC2 predominantly express ASCL1. NERUOD1⁺ cells are rare and do not cluster together. In addition, the heterogeneity and plasticity are found in SCLC cell line H526 and LCNEC cell line H1155. Other four SCLC lines H69, H146, H82, and H446 are either positive for ASCL1 only or for NEUROD1 only. Thus, the frequency of this intratumoural heterogeneity and plasticity in SCLCs is not known. More SCLC samples are required to address this question.

Author contributions

ZL, XZ, and ZM designed the study and wrote the paper. XZ, HW, and WL performed experiments and analysed the results. XZ and WL performed bioinformatics. ZX, JC, WG and ZZ provided lung cancer specimens and resources. GW analysed pathological features of SCLCs. All authors reviewed the manuscript.

Funding

This work was supported by the National Natural Science Foundation of China (grants 81825017, 81773034 to ZL, 8217113342, 81872350 to ZM, 82273119 to ZZ), the Ministry of Science and Technology of China (grant 2018YFC1313002 to ZL), the Tianjin Municipal Science and Technology Commission (20JCZDJC00110 to ZL), and the Haihe Laboratory of Cell Ecosystem Innovation Fund (HH22KYZX0025 to ZL).

Data availability

The data generated in this study are publicly available in the Gene Expression Omnibus (GEO) database at GSE164145 and GSE164404. The data that supported the findings of this study are available in the Human Tumour Atlas Network (HTAN) database (RU426, RU1066, RU1145 and RU1229A) [12].

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

The use of all human lung cancer tissues was approved by the Institutional Review Board of Tianjin Medical University. Informed consents were obtained from all patients, and samples were deidentified prior to analysis. All animal procedures were approved by the Animal Care and Use Committee at Tianjin Medical University and confirmed to the legal mandates and national guidelines for the care and maintenance of laboratory animals.

Supplementary information

The online version contains supplementary material available at 10.1038/s41416-022-02103-y.