Abstract

Simple Summary

   The shift from hormone-sensitive prostate cancer to
   castration-resistant prostate cancer (CRPC) has been hypothesized to be
   driven by prostatic luminal cells exhibiting castration tolerance,
   progenitor and tumor-initiating capacity. LSC^med cells that we
   recently isolated in a relevant mouse model of CRPC fulfil these three
   criteria. Using various bioinformatic pipelines, we here demonstrate
   that LSC^med cells match Club/Hillock cells recently identified in
   human prostate and prostate cancer. We identified EGFR/ERBB4, IGF-1 and
   MET pathways as key regulators of LSC^med cell progenitor and growth
   properties. We also demonstrate, for the first time in primary cultures
   of castration-tolerant prostatic progenitor cells, that the functional
   redundancy of these growth factor pathways confers to these cells the
   ability to bypass receptor-targeted pharmacological inhibition. Given
   the failure of EGFR- and MET-targeted monotherapies in CRPC patients,
   our data further support LSC^med cells as a relevant preclinical model
   to study the cellular and molecular mechanisms driving CRPC.

Abstract

   Background: The molecular and cellular mechanisms that drive
   castration-resistant prostate cancer (CRPC) remain poorly understood.
   LSC^med cells defines an FACS-enriched population of
   castration-tolerant luminal progenitor cells that has been proposed to
   promote tumorigenesis and CRPC in Pten-deficient mice. The goals of
   this study were to assess the relevance of LSC^med cells through the
   analysis of their molecular proximity with luminal progenitor-like cell
   clusters identified by single-cell (sc)RNA-seq analyses of mouse and
   human prostates, and to investigate their regulation by in
   silico-predicted growth factors present in the prostatic
   microenvironment. Methods: Several bioinformatic pipelines were used
   for pan-transcriptomic analyses. LSC^med cells isolated by cell sorting
   from healthy and malignant mouse prostates were characterized using
   RT-qPCR, immunofluorescence and organoid assays. Results: LSC^med cells
   match (i) mouse luminal progenitor cell clusters identified in
   scRNA-seq analyses for which we provide a common 15-gene signature
   including the previously identified LSC^med marker Krt4, and (ii)
   Club/Hillock cells of the human prostate. This transcriptional overlap
   was maintained in cancer contexts. EGFR/ERBB4, IGF-1R and MET pathways
   were identified as autocrine/paracrine regulators of progenitor,
   proliferation and differentiation properties of LSC^med cells. The
   functional redundancy of these signaling pathways allows them to bypass
   the effect of receptor-targeted pharmacological inhibitors.
   Conclusions: Based on transcriptomic profile and pharmacological
   resistance to monotherapies that failed in CRPC patients, this study
   supports LSC^med cells as a relevant model to investigate the role of
   castration-tolerant progenitor cells in human prostate cancer
   progression.

   Keywords: LSC^med, luminal progenitor, Club/Hillock cells, CRPC,
   signature, organoid, EGFR, IGF1-R, MET, drug resistance

1. Introduction

   Localized prostate cancer is successfully treated in more than 85% of
   cases by surgery (radical prostatectomy), external beam radiation
   therapy or brachytherapy [[42]1]. In patients with metastatic disease,
   androgen-deprivation therapy (ADT) is the gold standard treatment. ADT
   will induce cancer regression, relieve symptoms and prolong survival.
   However, after an initial response to ADT, all patients with prostate
   cancer will ultimately progress to castration-resistant prostate cancer
   (CRPC). Despite recent progress in the clinical management of those
   patients [[43]2], metastatic CRPC remains a lethal disease, and
   paracrine/autocrine androgen synthesis is thought to be one of the
   mechanisms of resistance to castration [[44]3,[45]4]. The
   identification of the cell(s) that drive cancer relapse and of the
   molecular pathways they use to promote tumor regrowth is needed. These
   cells should combine at least three properties: (1) castration
   tolerance, as they survive ADT; (2) luminal features, as the vast
   majority of hormone-sensitive prostate cancers (HSPC), from which CRPC
   arises, exhibit a luminal phenotype; (3) stemness, as they are assumed
   to rapidly regrow to form a tumor displaying phenotypic heterogeneity.

   We recently discovered, isolated and profiled in mouse prostate an
   unprecedentedly defined population of non-secretory luminal cells
   matching these three criteria [[46]5,[47]6,[48]7,[49]8]. We named these
   cells LSC^med according to their FACS profile (Lin^−/Sca-1^+/CD49f^med)
   using stem cell antigen-1 (SCA-1) and CD49f (integrin α6) as cell
   surface markers [[50]6]. These cells were identified by others as
   SCA-1^+ luminal cells [[51]9]. Cytokeratin 4 (CK4) was validated as a
   specific protein biomarker of LSC^med cells on prostate sections from
   various mouse models [[52]5,[53]10]. The intrinsic castration tolerance
   of these cells was shown by their increased prevalence in prostates
   from castrated versus intact mice [[54]5,[55]9], and by their
   insensitivity to enzalutamide [[56]9], a second-generation antiandrogen
   drug efficient at the HSPC and CRPC stages of the disease
   [[57]11,[58]12,[59]13]. The stem/progenitor properties of LSC^med cells
   have also been widely assessed by us and others through their enriched
   capacity versus mature luminal cells to form spheres and organoids in
   vitro, to self-renew, and to generate glandular structures when
   engrafted into host mice [[60]5,[61]6,[62]9]. Exhaustive description of
   LSC^med cell properties is provided in a recent review article [[63]8].

   The involvement of LSC^med cells in prostate cancer is supported by
   several observations in preclinical models for a review, Ref. [[64]8].
   While they are rare in healthy prostates (~5% of epithelial cells),
   LSC^med cells represent up to 80% of epithelial cells in prostate
   tumors driven by prostate-specific deficiency of the tumor suppressor
   gene Pten (mice are hereafter called Pten-null) [[65]5,[66]7]. In
   prostates of castrated Pten-null mice, LSC^med cells remain highly
   prevalent, and the detection of large clusters of CK4^+/Ki67^+ cells
   revealed that some LSC^med cells not only survive castration, but also
   proliferate [[67]5]. Finally, FACS-enriched Pten-null LSC^med cells
   generate invasive tumors when engrafted into host mice [[68]5].
   Together, these data suggest that, in mice, LSC^med luminal progenitor
   cells actively contribute to prostate cancer progression towards CRPC.
   Based on these observations, the goal of this study was to address the
   relevance of LSC^med luminal progenitor cells to model human prostate
   cancer progression, including the identification of actionable targets
   able to interfere with this process.

   Within the past few years, several groups published single cell (sc)
   atlases of the adult prostate based on RNA sequencing (scRNA-seq) data.
   These studies involved WT mice [[69]10,[70]14,[71]15,[72]16,[73]17],
   Pten-deficient mice [[74]18], and human specimens of healthy prostate
   [[75]10,[76]15,[77]16,[78]17,[79]19] and prostate cancer
   [[80]20,[81]21]. All mouse studies identified one computerized cluster
   of non-secretory luminal cells referred to as ‘luminal progenitors’
   based on their enrichment in stemness-related transcripts. The human
   studies identified one or two populations of non-secretory luminal-like
   cells called Club and Hillock based on their molecular similarity with
   eponymous epithelial progenitor cell types described in the lung
   [[82]19]. Noteworthy, cancer-associated Club cells has been recently
   identified in prostate cancer specimens [[83]20]. The correspondence of
   all these progenitor-like luminal cells has not been investigated
   beyond the qualitative overlap of a few markers. Using bona fide
   bioinformatic approaches, we here demonstrate that, in both healthy and
   cancer contexts, FACS-enriched LSC^med cells largely overlap with in
   silico-defined clusters of mouse luminal progenitor cells and human
   prostatic Club/Hillock cells. We provide a common 15-gene signature of
   castration-tolerant mouse luminal progenitor cells that should help to
   track these cells in preclinical models of prostate cancer in order to
   delineate their fate during cancer progression.

   The second aim of this study was to investigate the regulation of
   LSC^med cell proliferation by growth factors present in the prostatic
   microenvironment. Bioinformatic search for autocrine/paracrine
   ligand–receptor pairs identified receptors of the epidermal growth
   factor receptor (EGFR/ERBB4), insulin-like growth factor-1 receptor
   (IGF-1R) and MET pathways as top candidates. Strikingly, monotherapies
   targeting these receptors were disappointing in metastatic CRPC
   patients
   [[84]22,[85]23,[86]24,[87]25,[88]26,[89]27,[90]28,[91]29,[92]30,[93]31]
   . The interactive crosstalk between MET and EGFR family members is well
   documented and has been raised as a mechanism of resistance to targeted
   monotherapies in other cancers [[94]32,[95]33,[96]34]. Using the
   acknowledged organoid assay to monitor the regulation of LSC^med cell
   progenitor and growth properties by these growth factor pathways, we
   here document for the first time that castration-tolerant prostatic
   progenitor cells are able to evade EGFR, MET and IGF-1R pharmacological
   blockade. Together, our observations provide a strong rationale for the
   involvement of luminal progenitor cells in tumor progression observed
   in CRPC patients.

2. Materials and Methods

   Animals. Pten-null mice were generated by breeding Pten^flox/flox
   female mice with Pb-Cre4 transgenic males on the C57BL/6J background as
   previously described [[97]5]. Experiments were performed using 8 to
   11-month-old mice, i.e., when aggressive malignant phenotypes were well
   established. Non-transgenic C57BL/6J littermates were used as controls
   and are referred to as WT animals. Colonies were housed in conventional
   health status, on a 12/12 h light/dark cycle with normal chow diet and
   water provided ad libitum. Prostate samples were obtained by
   microdissection immediately after sacrifice by cervical dislocation.
   Animal experiments were approved by the local ethical committee for
   animal experimentation (APAFIS#1427-2017121915584941).

   Prostate cell subpopulation sorting by FACS. The procedures for cell
   sorting were performed as previously described [[98]5,[99]35]. Isolated
   cells (basal, luminal, LSC^med and stromal) were stained for FACS on
   ice for 30 min. Antibodies (eBioscience) used for FACS were fluorescein
   isothiocyanate-coupled lineage (Lin) antibodies (anti-CD31, CD45 and
   TER-119), phosphatidylethanolamine-Cyanine7-coupled anti-EpCAM,
   phosphatidylethanolamine-coupled anti-CD49f (integrin alpha-6) and
   allophycocyanin-coupled anti-Sca1 (lymphocyte antigen 6A-2/6E-1). Dead
   cells were colored with SYTOX blue. Cell sorting was performed on a BD
   FACS Aria III. Lin antibodies were used to deplete hematopoietic,
   endothelial and immune cells and EpCAM antibody was used to separate
   epithelial versus stromal cells. CD49f and SCA-1 markers were used to
   select basal cells, luminal cells and LSC^med cells in the EpCAM^+
   gate, and the stromal cells in the EpCAM^neg gate. Sorted cells were
   collected in DMEM medium, supplemented with 50% FBS, glutamine, and
   penicillin-streptomycin, or in RA1 Lysis Buffer (Macherey-Nagel, Düren,
   Germany) to perform RNA extraction as described in the manufacturer’s
   protocol.

   Reverse Transcription-Quantitative PCR (RT-qPCR). RNA extraction was
   performed with the Nucleospin RNA XS (Macherey-Nagel, Düren, Germany),
   as described in the manufacturer’s protocol. The reverse transcription
   was performed using the SuperScript™ VILO™ cDNA Synthesis Kit
   (Invitrogen) described protocol.

   For qPCR, iTaq Universal SYBR Green Supermix (Promega, Madison, WI,
   USA) was used, and reactions were run on a qTower 2.0 real-time thermal
   cycler (Analytik Jena). Primers are listed in [100]Table S1. Expression
   data obtained using Pten-null mouse prostates are presented as 2^−ΔΔCt
   normalized to WT mouse values (Ct represents the cycle threshold at
   which amplified cDNA is detected; the higher the Ct, the lower the gene
   expression).

   3D organoid culture. We used the reference protocol described by
   Clevers’ lab [[101]36], in which EGF is used as the growth factor. In
   our study, various growth factors were substituted for EGF. Culture
   media, additives, growth factors and drugs are listed in [102]Table S2.

   LSC^med cells sorted from Pten-null mouse prostates were plated in
   triplicate on a Low Growth Factor-containing Matrigel (Corning) layer
   in a 96-well plate (Falcon) in the presence or not of growth factors,
   according to the culture condition. After 1 day of incubation, the
   medium was removed and cells were covered by a new layer of Matrigel in
   order to perform 3D culture. The organoid-forming capacity did not
   differ from the efficacy obtained using 3D droplet culture. Drugs were
   added after 1 day of culture. Medium was changed every 2 days. After 10
   days of Matrigel embedding, organoids were fixed in 4% PFA, and photos
   were taken with a 4× objective under a M5000 EVOS inverted microscope
   in order to cover the entire surface of the well. Counting and
   surfacing were performed on Fiji Software by manually surrounding the
   organoid surface. The number of organoids obtained in the various
   experimental conditions was normalized to the mean value obtained in
   the EGF-containing medium.

   Organoid agarose embedding. After PFA fixation, all wells were
   combined, according to the culture conditions. Matrigel pellets were
   collected in BD Cell Recovery solution (Corning) and placed into ice
   during 30 min to 1 h, according to the number of pooled wells, in order
   to depolymerize the Matrigel. When the organoids started to settle at
   the bottom of the collecting tube, the supernatant was removed and
   pellets were resuspended in 2% low melting agarose. After
   solidification, the agarose pellets were transferred to 70% EtOH in
   order to perform paraffin wax protocol.

   Immunofluorescence (IF). All samples were fixed in 4% PFA, paraffin
   wax-embedded, and sections underwent heat-induced antigen retrieval in
   citrate buffer at pH 6 (95 °C, 30 min). IF was performed as described
   previously [[103]5] using antibodies directed against CK4 (1/150,
   BSM-52062R, ThermoFisher Scientific), CK5 (1/150, 905901, Biolegend)
   and CK8 (1/100, AB_531826, DSHB), KI-67 (1/150, RBK027-05, Diagomics)
   and E-cadherin (1/100, 610182, BD Transduction). Nuclei were stained
   with Hoechst dye. Samples were analyzed with a 40× objective under an
   Apotome 2 (Zeiss) microscope. Cell size was measured on
   E-cadherin-stained slides (350 cells per condition) by using the
   Cellpose 1.0 plugin associated with QuPath software.

   In silico identification of candidate pathways. Cellular interaction
   prediction was performed using CellPhoneDB [[104]37] with default
   parameters applied to three transcriptomic datasets for each sorted
   cell population of WT mouse prostates [[105]5]. Human gene orthologs
   were used to take advantage of the CellPhoneDB ligand–receptor database
   (human). The p-value threshold to consider an interaction as
   significant was set to 0.05 (p-value ≤ 0.05). Significant predicted
   interactions involved in the positive regulation of cell proliferation
   (GO:0008284) having the receptor expressed by LSC^med cells and a
   soluble mitogenic ligand (irrespective of the cell compartment of
   origin) were selected.

   scRNA-seq data retrieval. Data re-analyzed as part of this study were
   retrieved from the Gene Expression Omnibus (GEO) and The National Omics
   Data Encyclopedia (NODE) public databases ([106]Table 1). For mouse
   scRNA-seq data, we only retained samples taken from intact (i.e.,
   non-castrated) mice. Data retrieved from GEO repository [107]GSE151944
   were available as MULTI-seq sample barcodes; these were demultiplexed
   using the MULTIseqDemux function implemented in Seurat [[108]38]. For
   datasets [109]GSE145861 and [110]GSE145865 [[111]17], the data from the
   prostate and urethral regions were aggregated during the analysis step.
   For dataset [112]GSE164858 [[113]18] we retained the vehicle-treated
   sample only. Finally, data retrieved from the OEP000825 repository were
   raw fastq files which were processed (read alignment, generation of
   feature-barcode matrices) with Cell Ranger (10× Genomics) prior to any
   data analysis step.

Table 1.

   References of scRNA-seq datasets re-analyzed in this study.