Highlights
- •PARCB temporal model profiling informs key aspects of SCNPC trans-differentiation
- •Trans-differentiation trajectory resembles a common developmental arc-like pattern
- •ASCL1 and ASCL2 mark distinct bifurcating SCNPC trans-differentiation trajectories
- •TFAP4 as a common regulator of ASCL1/2 implicated in SCNPC trans-differentiation
Summary
Trans-differentiation from an adenocarcinoma to a small cell neuroendocrine state is associated with therapy resistance in multiple cancer types. To gain insight into the underlying molecular events of the trans-differentiation, we perform a multi-omics time course analysis of a pan-small cell neuroendocrine cancer model (termed PARCB), a forward genetic transformation using human prostate basal cells and identify a shared developmental, arc-like, and entropy-high trajectory among all transformation model replicates. Further mapping with single cell resolution reveals two distinct lineages defined by mutually exclusive expression of ASCL1 or ASCL2. Temporal regulation by groups of transcription factors across developmental stages reveals that cellular reprogramming precedes the induction of neuronal programs. TFAP4 and ASCL1/2 feedback are identified as potential regulators of ASCL1 and ASCL2 expression. Our study provides temporal transcriptional patterns and uncovers pan-tissue parallels between prostate and lung cancers, as well as connections to normal neuroendocrine cell states.
Introduction
Small cell neuroendocrine (SCN) cancer is an aggressive variant that arises from multiple tissues such as the lung and prostate.1,2 SCN is characterized by its histologically defined small cell morphology of densely packed cells with scant cytoplasm, poor differentiation, and aggressive tumor growth, as well as expression of canonical neuroendocrine markers including SYP, CHGA, and NCAM1.3 In addition to their phenotypic resemblance, SCN cancers across multiple tissues show a striking transcriptional and epigenetic convergence in clinically annotated tumors.4,5 This molecular signature convergence is recapitulated by our established SCN transformation model that utilizes either normal lung epithelial cells, patient-derived benign prostate epithelial or bladder urothelial cells as the cells of origin.6,7
Small cell neuroendocrine prostate cancer (SCNPC) occurs either de novo (<1% of untreated prostate cancer cases), or through therapy-mediated transversion of castration resistant prostate cancer (CRPC) (∼20% of the resistance cases). The SCN terminology has been adopted to reflect the shared pan-tissue aspects of multiple SCN tumors, such as small cell lung cancer (SCLC). CRPC is a resistant variant of prostate adenocarcinoma (PRAD), which often responds to androgen deprivation therapy.8,9 Trans-differentiation from PRAD to the SCNPC state entails complicated epigenetic reprogramming at the chromatin level, resulting in transcriptional changes driven by a number of key master regulators.10,11 For example, methylation modulated by EZH2 and activation of transcriptional programs by SOX2 are required in TP53 and RB1 loss-mediated neuroendocrine differentiation in mouse transgenic models of SCNPC.12,13 Oncogenic mutation of FOXA1 potentiate pioneering activity and differentiation status of prostate cancer.14,15 Lastly, knockdown of transcription factors such as ONECUT2 has been shown to inhibit SCN differentiation.16,17 While the importance of these factors has been demonstrated, the chronological sequence of the associated epigenetic and transcriptional changes remains uncharacterized during the progression to SCNPC. Examination of the temporal evolution of lung cancer revealed a connection between transcription factor defined subtypes and cell plasticity.18,19 In our study, we sought to answer the following questions: (1) when do SCN-associated transcription factors emerge during SCNPC progression, (2) how do they coordinate SCN differentiation, and (3) can we identify a transition state defined by transcription factors that can be targeted?
Leveraging our previously developed human pan-small cell neuroendocrine cancer model, the PARCB forward genetics transformation model (driven by knockdown of RB1, alongside exogenous expression of dominant negative TP53, cMYC, BCL2, and myristoylated AKT1 via three lentiviral vectors),6,7 tumor samples were harvested at different time points for multi-omics analyses. The transcriptional and epigenetic status of each time point was determined using integrative bulk RNA sequencing, ATAC sequencing, and single cell RNA sequencing. This longitudinal study provides insight into the temporal evolution of the epigenetic and transcriptional landscape during trans-differentiation and small cell cancer progression. We found consistent transcriptional patterns and differentiation trajectories across samples generated from independent patient tissues, as well as a bifurcation of end-stage neuroendocrine lineages, defined by ASCL1 and ASCL2 and their associated programs.
Achates-scute complex (AS-C) proteins are basic-helix-loop-helix (bHLH) transcription factors, first identified in Drosophila melanogaster.20 They are important in the development of peripheral nervous systems and sensory organs.21 Mammalian ASCL1 is a well-known neuroendocrine transcription factor in small cell cancers.22,23,24 Independently, ASCL2 is involved in embryonic development, colorectal stem cell biology and cancer.25,26,27,28,29,30 ASCL2 is largely understudied in SCNPC, mainly shown to be co-expressed with POU2F3 in non-neuroendocrine cell populations.5,31 Here, our study reveals temporal transcriptional patterns during SCN differentiation in prostate cancer and associated lineage programs governed by general mutually exclusivity between ASCL1 and ASCL2. Follow-up analysis elucidated a transcriptional network circuity between ASCL1, ASCL2, and the transcription factor TFAP4 which was implicated by the trajectory data.
Results
Temporal gene expression programs of the PARCB transformation model reveal trans-differentiation pathways
To determine the timing of SCN differentiation events during prostate cancer development, we utilized the PARCB model system.6 Independent transformations were performed on basal cells extracted from benign regions of epithelial tissue from 10 PRAD patients. Basal cells were transformed by the oncogenic lentiviral PARCB cocktail and subsequently cultured in an organoid system in vitro.6 Transformed organoid-expanded cells from each patient tissue sample were subcutaneously implanted into multiple immunocompromised mice to allow for time-course collection of tumors from the matched starting material (Figure 1A). The tumors were collected at approximately two-week intervals until reaching 1 cm3 in size or occurrence of ulceration, whichever came first. The transformed tumor cells were triply fluorescent due to the lentiviral integration,6 which allowed for cancer cell purification by fluorescence-activated cell sorting (FACS) followed by multi-omics sequencing and analysis (Figure 1A). Each patient series (P1-P10) contains five to six time point samples ranging from basal cells (TP1) to organoids (TP2) to tumors (TP3-TP5/TP6) (Figure 1A). Upon histological examination of the tumor tissues by pathologists, we found that the time course tumors transitioned from squamous, to adenocarcinoma, then to mixed and eventually SCN phenotypes (Figures 1A and S1A–S1C). Furthermore, clinically defined neuroendocrine markers, including SYP and NCAM1, emerged during the transition to late stages of the tumor progression (Figure 1A). The basal cell marker p63 was only positive in early-stage tumors by immunohistochemistry (IHC) staining (Figure S1D).
We first performed a temporal analysis of gene expression using bulk RNA sequencing to understand the changes in the transcriptional landscape during SCNPC trans-differentiation. By projection of PARCB samples onto principal component analysis (PCA) of clinical lung and prostate cancer tumor samples,4,10,32,33,34,35,36 we validated that PARCB time course samples follow the transcriptionally defined convergence trajectory from adenocarcinoma to SCN states (Figures 1B and S1E). Additional SCNPC associated factors including ASCL1 and NEUROD1 were also elevated during the progression (Figure 1C). The mRNA of androgen receptor (AR) was expressed in tumors at the early stage (Figure 1C), but the protein level was not detectable by immunostaining (Figure S1D). Taken together, the histological and omics data indicate that PARCB time course tumors recapitulate both the phenotypic and transcriptional changes observed in the clinic and provide a model system for studying the temporal evolution of SCNPC development.
To determine the transformation trajectories among the time course series generated from 10 independent patient samples (P1–P10), we performed clustering and PCA of the transcriptomic data. To account for potential asynchronous development among each patient series and each individual tumor, we defined hierarchical clusters (HCs) of samples by their corresponding differential gene modules and found the resulting 6 clusters (HC1–6) to generally correspond with the time of collection (Figure 1D; Table S1A). This provides a clustering-based trans-differentiation reference frame and informs our subsequent multi-omics analyses. Unsupervised PCA demonstrates that the individual transformation paths of each series follow a generally consistent “arc-like” trajectory with a discernable bifurcation in late-stage samples (Figures 1E, S1E, and S1F; Table S1B). The late tumors were hence further defined as “Class I” and “Class II” tumors with correspondent HC5 and HC6 gene modules, respectively. HC2 to HC6 had elevated SCNPC signature scores compared to adenocarcinoma signature score (Figure S1G). This finding supports the existence of two transcriptional programs or endpoints defining the terminal SCNPC tumor phenotypes.
Gene ontology enrichment analysis of the corresponding 6 differential gene modules identified biological processes enriched uniquely or shared among HCs, including Inflammatory response (HC1 and HC3, patient derived basal cells and early tumors, respectively), cell proliferation (HC2, in vitro organoids), epidermis development (HC3, early tumors), cell activation (HC4, transitional tumors), stem cell differentiation (HC5, Class I late tumors) and neuro-/chemical synapse (HC5 and HC6, both classes of late tumors) (Figure 1E; Table S1C). The transcriptome evolution supports the idea that trans-differentiation from adenocarcinoma to the SCN state is a systematically coordinated process that involves a transitional stage followed by bifurcated pathways enriched in neuronal/neuroendocrine gene signatures.
Sequential transcription regulators modulate reprogramming and neuroendocrine programs through a highly entropic and accessible chromatin state
Temporal analyses on single transcription factor defined subtypes of SCLC models have delineated lineage plasticity in the development of lung neuroendocrine tumors.18 We sought to define the transcriptional evolution in SCNPC through an extensive survey of over 1,600 transcription factors37 by chromatin accessibility analyses using ATAC sequencing.38 A significant increase in overall accessible chromatin peaks across chromosomes is observed starting at the tumors at transitional stage (HC4) to late stages (HC5 and HC6) (Figure 2A). Unsupervised PCA using ATAC-sequencing data showed an arc-like and bifurcated trajectory consistent with the pattern observed using the RNA-sequencing data (Figures 1D and 2B). The Shannon entropy has been used to estimate the plasticity potential of a biological sample to change cellular state.39,40 We found that transitional samples (HC4) have the highest entropy (Figure 2B), suggesting there exists a high potential and less well-defined transcriptional state during the trans-differentiation process…
