Metastasis development is the leading cause of cancer-related mortality in pancreatic ductal adenocarcinoma (PDAC) and yet, few preclinical systems to recapitulate its full spreading process are available. Thus, modeling of tumor progression to metastasis is urgently needed. In this work, we describe the generation of highly metastatic PDAC patient-derived xenograft (PDX) mouse models and subsequent single-cell RNA-sequencing (RNA-seq) of circulating tumor cells (CTC), isolated by human HLA sorting, to identify altered signaling and metabolic pathways, as well as potential therapeutic targets. The mouse models developed liver and lung metastasis with a high reproducibility rate. Isolated CTCs were highly tumorigenic, had metastatic potential, and single-cell RNA-seq showed that their expression profiles clustered separately from those of their matched primary and metastatic tumors and were characterized by low expression of cell-cycle and extracellular matrix–associated genes. CTC transcriptomics identified survivin (BIRC5), a key regulator of mitosis and apoptosis, as one of the highest upregulated genes during metastatic spread. Pharmacologic inhibition of survivin with YM155 or survivin knockdown promoted cell death in organoid models as well as anoikis, suggesting that survivin facilitates cancer cell survival in circulation. Treatment of metastatic PDX models with YM155 alone and in combination with chemotherapy hindered the metastatic development resulting in improved survival. Metastatic PDX mouse model development allowed the identification of survivin as a promising therapeutic target to prevent the metastatic dissemination in PDAC.
Pancreatic ductal adenocarcinoma (PDAC) is the third cause of cancer-related deaths in the United States and expected to become the second by 2030 (1). The majority of patients present with advanced metastatic disease for whom palliative chemotherapy is the only available option (2, 3). For patients diagnosed at earlier stages and treated with curative intent, metastatic relapse is common and the 5-year survival rate is still poor (4, 5). Understanding the mechanisms governing the metastatic process is critical to identify novel therapeutic strategies aiming at improving survival of PDAC. Preclinical and necropsy studies have established that metastasis occurs late in the genetic evolution of PDAC and that epithelial-to-mesenchymal transition (EMT) precedes dissemination (6, 7). One of the most studied mediators of EMT transition and metastatic development is the TGFβ/SMAD4 signaling pathway (8). In genetically engineered mouse models (GEMM) of PDAC, Smad4 functions as a tumor suppressor, blocking the progression of KRas-initiated neoplasms. However, in a subset of advanced tumors, intact SMAD4 facilitates EMT and TGFβ-dependent growth (9). Furthermore, studies using human samples have established that the number of mutations in driver genes such as KRAS, TP53, SMAD4, and CDKN2A are associated with worse outcomes because of higher metastatic potential (10). Other studies have found complex reprogramming of chromatin modifications during the evolution of distant metastasis, as well as dependence of specific metabolic pathways (11). While these studies provide insights into the timing and molecular events leading to metastatic spread in PDAC, thus far no therapeutic targets to avert this process have been identified.
It has been proposed that circulating tumor cells (CTC) have the clonal capacity to initiate tumor growth in distant organs, thus their analysis may uncover vulnerabilities to hamper the metastatic spread (12, 13). However, the limitations to study CTCs are numerous. Circulating cancer cells are exceedingly rare in the blood and traditional methods for their detection have relied on selection of cells positive for epithelial markers, such as EPCAM, which may not be optimal because poorly differentiated carcinomas or with EMT phenotype can have diminished expression of those proteins (6, 14–18). Furthermore, those methods typically incorporate cell fixation steps precluding subsequent functional studies (15).
To overcome these technical challenges, we developed a model with human tumor grafts in a murine background that was permissive for spreading and employed HLA sorting to ensure that it is the human-derived CTC that is captured and its transcripts that are analyzed, thereby preventing contamination by murine cells. This platform allowed us to identify aberrant signaling and metabolic pathways as well as novel therapeutic opportunities to hinder the metastatic process.
Materials and Methods
Establishment of metastatic patient-derived xenograft mouse models
All experiments using mice were approved by CNIO-ISCIII Ethics Committee for Research and Animal Welfare (CEIyBA). The patient-derived xenograft (PDX) models Panc265, Panc198, Panc020, Panc042, Panc047, Panc026, and Panc219 from our tumor bank collection were used for the study. These xenografts were generated from freshly biopsied tumor samples from patients with PDAC and propagated subcutaneously into successive mouse generations (19). Extended information about the PDX is provided in Supplementary Table S2. Freshly collected subcutaneously grown tumors were cut with a sterile scalpel into small pieces of 1–2 mm3 and embedded in Matrigel Matrix (Corning). Six- to 8-week-old NSG mice (purchased from Charles River Laboratories or bred into the CNIO animal facility) were anesthetized using isoflurane gas and treated with buprenorphine (0.2 mg/kg). The left abdominal flank was shaved, and the skin was sterilized with 70% ethanol. Using a sterile surgical microscissors, a small incision was made in the upper left abdomen, and the pancreas was exposed. Small tumor pieces from each xenograft model were implanted into the splenic lobe of the pancreas using 6-0 absorbable sutures (B. Braun). Mice were monitored during the whole study and they were humanely sacrificed after appearance of several or all of the following symptoms: large tumor size, loss of body weight > 20%, lethargy, dyspnea, and/or pain. The date of sacrifice was recorded and the organs: spleen, orthotopic primary graft, liver, and lung were collected and preserved in 10% formalin solution for subsequent histologic studies. Description of procedures for histologic preparation, tissue dissociation, and flow cytometry are provided in Supplementary Materials and Methods.
Establishment of PDAC organoids
Organoids were cultured from PDX tumor tissues following a protocol described previously (20). Tissues were first digested with collagenase/dispase (1 mg/mL) for 30–40 minutes and then with Accutase (Sigma) for 40 minutes. After centrifuging the mixture at 1,500 rpm for 5 minutes, cell pellets were resuspended in organoid culture medium supplemented with 5% Matrigel and 10 μmol/L Y-27632. In one well of 12-well plate (precoated with Matrigel), 50,000 tumor cells were seeded. Culture medium was replaced every 4 days. For drug treatments, 96-well plates were used. Each well coated with 15 μL of Matrigel was seeded with 5,000 cells for organoid growth. After 4 days in culture, organoids were treated with different concentrations of drugs diluted in fresh culture media, and then analyzed 4 days later. Cell death and total cell numbers were quantified with CytoTox Glo (Promega) and normalized to DMSO-treated organoids. For morphometric analysis of organoid, the program Organoseg was used to measure organoid size and number of organoids. For each organoid line, at least 200 organoids were measured.
For survivin knockdown, short hairpin RNA (shRNA; TRCN0000073720 and TRCN0000073721) plasmids were obtained from Sigma. shRNA lentiviruses were packaged in 293T cells and concentrated with ultracentrifugation (25,000 rpm for 2 hours). For tumor cell infection, tumor cells were seeded in 60 mm culture dish precoated with a thin layer of Matrigel and cultured with organoid growth medium overnight. The next day, tumor cells were incubated with lentivirus, organoid culture medium, and 4 μg/mL polybrene overnight. On day 3, tumor cells were digested with Accutase, then grown as 3D-cultured organoids, and selected with 2 μg/mL puromycin for 8 days. To quantify survivin knockdown, organoids with shRNA survivin expression or shRNA GFP expression were isolated with Cell Recovery Solution (BD Biosciences) and lysed with RIPA buffer. One antibody (ab76424) from Abcam was used to detect survivin expression in control and survivin-knockdown organoids.
ISH to ALU repetitive sequences
ALU-positive control probe II is a cocktail of dinitrophenol (DNP)-labeled oligonucleotide probes, which hybridized to ALU repetitive sequences present within the genome of primates. Tissue samples were fixed in 10% neutral buffered formalin (4% formaldehyde in solution), paraffin-embedded and cut at 3 μm, mounted in superfrost plus slides, and dried overnight. For the whole technique, an automated immunostaining platform was used (Ventana Discovery XT, Roche). Antigen retrieval was first performed with low pH buffer (RiboCC, Roche) and Protease III (Roche). Slides were then incubated with the probe ALU-positive control probe II (Ventana, Roche 05272041001). After the probe, stringency washes were necessary three times and the slides were incubated with an intermediate, Rabbit anti-DNP. Visualization systems were needed (OmniRabbit, Ventana, Roche) conjugated with horseradish peroxidase; and IHC reaction was developed using Silver as a chromogen (Silver Kit, Ventana, Roche) and nuclei were counterstained with Carazzi hematoxylin. Finally, the slides were dehydrated, cleared, and mounted with a permanent mounting medium for microscopic evaluation.
Quantitative PCR: targeting human DNA (ALU sequences) in blood of metastatic PDX models
Fresh EDTA blood was collected via cardiac puncture from NSG mice implanted orthotopically with the highly metastatic Panc265 model. The blood from each animal was first centrifugated at 400 × g for 10 minutes and the plasma was separated from the cellular pellet. Next, the cellular pellets were resuspended in 2 mL of Red Blood Cell Lysis Solution (Qiagen, catalog no. 79217) for 10 minutes at37°C. The samples were then centrifuged for 7 minutes at 300 × g, followed by two washing steps using 1 mL of sterile PBS (1×). After the washing steps, the cellular pellets were resuspended in 200 μL of sterile PBS (1×) and the DNA extraction from each sample was performed using DNeasy Blood and Tissue Kit (Qiagen, catalog no. 69504) following the manufacturer's instructions. The DNA was then eluted in 100 μL of nuclease-free water. Finally, the DNA concentrations were measured using a NanoDrop ND-1000 spectrophotometer. Real‐time PCR using TaqMan was performed on Real-Time PCR System 7500 (Applied Biosystems), according to the manufacturer's instructions. Oligonucleotide primers for ALU sequences were:
The TaqMan probe was:
The primers, probe, and TaqMan Universal PCR Master Mix were purchased from Applied Biosystems.
Isolation of human tumor cells from metastatic PDX models
Human tumor cells from the mouse tissues (primary tumor and liver metastases) were purified by autoMACS Pro Separator using MS Columns (Miltenyi Biotec) in accordance with the manufacturer's instructions. The cell suspensions were filtered through a 40-μm Cell Strainer (BD Biosciences). For blood collection, tumor-bearing mice were euthanized and bled using a 1 mL syringe via a cardiac puncture, and collected in EDTA-containing tubes. The blood was then diluted 1:1 with PBS 1×, and carefully layered upon 1.5 volume of Ficoll‐Paque PLUS (VWR International Eurolab) and centrifuged for 30–40 minutes at 400 × g, without brake, at room temperature. The mononuclear cell layer was then collected, washed in PBS 1×, counted, and prepared for further application.
Human single-cell suspensions were enriched by a two-step process: first incubating with a PE-conjugated primary antibody for the human antigen HLA-ABC (clone G46-2.6, BD Pharmingen), followed by incubation with Anti-PE MicroBeads UltraPure (Miltenyi Biotec). The enriched human tumor cell suspensions were checked using a Confocal Microscope (Leica SP5-MP).
Single-cell RNA-sequencing, processing, and statistical analysis
Cells enriched from primary tumor, blood, and liver of PDX metastatic model were individually separated on the C1 Single-Cell Auto Prep System (Fluidigm). In total, 137 cells were captured on three C1 array chips for mRNA sequencing. After single-cell sequencing, any putative murine cell was discarded by FastQ Screen software (v. 0.4.4,). The mRNA reads derived from human single cells were aligned on the human genome reference (hg19 assembly, UCSC) using the TopHat2 algorithm (v. 2.0.10). The gene-level expression was quantified using HTSeq-count (v. 0.5.4p5). Sequencing data are deposited at Gene Expression Omnibus (GSE114704).
Principal component analysis on the log2-transformed single-cell gene expression profiles (read counts per million) was performed. The differential gene expression test based on read counts was carried out using SCDE (Bioconductor R package, v. 1.99.4). After adjusting P values for multiple testing (Benjamini and Hochberg method), genes with FDR < 0.05 were considered statistically significant. A preranked gene list by the Z-score of the expression difference between CTCs and primary tumor cells was used to perform gene set enrichment analysis (GSEA).
Description of extended procedures for single-cell RNA-sequencing (RNA-seq) and processing, as well as statistical analysis on single-cell RNA-seq data are provided in Supplementary Materials and Methods.
In vivo treatment experiments
YM155 (purchased from Selleckchem; ref. 21) was dissolved and diluted in saline, danusertib (purchased from MedChemExpress; ref. 22) was dissolved in an in situ salt prepared as previously described and diluted in 5% dextrose, nab-paclitaxel was dissolved and diluted in saline, and PAC-1 (purchased from MedChemExpress; ref. 23) was dissolved in DMSO and diluted in saline prior to administration. For the in vivo efficacy studies, the following drug treatment regimens were established: YM155 (2 mg/kg/day, i.v.); nab-paclitaxel (50 mg/kg once a week, i.v.); PAC-1 (5 mg/kg/day, i.v.); and danusertib (PHA-739358; 15 mg/kg/day, i.p.) alone and in combination in 6–8 mice per group. The mice were treated during 4 weeks with all compounds and for 3 weeks with nab-paclitaxel. Animals were monitored daily for signs of toxicity and were weighed thrice weekly. Animals that developed adverse effects as described above were humanely euthanized. Once treatment was completed, a small number of mice on danusertib and/or YM155 treatment group were euthanized at the time of ethical endpoint sacrifice of control group and the remaining animals were subsequently monitored and sacrificed when they manifested the above mentioned distress signs. For evaluation of metastatic tumor burden in Panc265 metastatic model, mouse livers and lungs of control and treatment groups (danusertib and/or YM155) were subjected to ALU-ISH staining and the percentage of ALU-positive cells was determined by Axio Vision. Kaplan–Meier survival plots were prepared using GraphPad software and median survival times were determined for all experimental groups. Description of procedures for the computational drug prescription are provided in Supplementary Materials and Methods.
Suspension cell culture and anoikis assay
For suspension cell culture, low attachment 96-well plates (pretreated with poly- HEMA) were used. Tumor cells were obtained from organoid culture by the following procedures. Organoid cultures were treated with collagenase/dispase (1 mg/mL) for 1.5 hours, and then with TrypLE (Thermo Fisher Scientific) for 40 minutes to obtain single cells. Cells were resuspended in organoid culture medium supplemented 0.5% Methylcellulose (1500cps, R&D Systems) and seeded in 96-well plates (20,000 cells, 200 μL per well). After 24 hours in suspension culture, cell death was analyzed using an LDH Assay Kit (Abcam) following the manufacturer's protocol. For each well, 100 μL of suspension was transferred into a 1.5 mL tube and centrifuged at 5,000 rpm for 5 minutes to pellet cells, and 25 μL of supernatant was transferred and mixed with 50 μL lactate dehydrogenase (LDH) assay solution. After 30 minutes, LDH activity was measured by absorbance at 405 nm.
Establishment of metastatic PDAC PDX models
A series of metastatic PDX models was developed by orthotopically implanting tumor materials in NOD/SCID/IL2g mice. Similar experiments conducted in different mouse strains (Nude and SCID) did not result in metastatic spread (Supplementary Fig. S1; Supplementary Table S1). A total of seven metastatic models were established. Panc265 and Panc198 generated both lung and liver metastasis, while the others only spread to the lungs (Fig. 1A and B). The presence of human tumor cells in host liver and lung was confirmed by ISH to primate ALU sequences (ALU ISH; Fig. 1A). The median survival of the entire cohort of mice was 57.6 days (range 25–95; Fig. 1B). At later stages of the disease, mice developed signs of lethargy, weight loss, and/or dyspnea. In Panc265, metastasis was detected by 2 weeks after implantation and progressed exponentially with time leading to terminal cancer by week 4 (Fig. 1C). Metastatic progression was parallel by presence of human DNA in mouse blood (Fig. 1D) and an increase in the size of primary tumor (Supplementary Fig. S2). Increasing amounts of human cancer cells were detected in lung and liver of Panc265 tumor–bearing mice from week 1 to 4 (Supplementary Fig. S3). Brain micrometastasis also developed in this model (Supplementary Fig. S4).
Identification of human cancer cells involved in the metastatic spread
Expression of the MHC class I (HLA-ABC) antigen, which is present on the surface of all nucleated cells and is a commonly used marker to identify human cells in PDX models, was determined (24). To identify mouse cells in grafted tumors, H-2kD marker (mouse MHC class I), which is present on the surface of all mouse nucleated cells, was utilized. By flow cytometry, we found that HLA-ABC was present in 83%–86% of the tumor cells dissociated from Panc265 and Panc198 models, with negligible presence in control mouse peripheral tissues (Supplementary Materials and Methods; Supplementary Fig. S5). Viable human tumor cells from metastatic lesions were isolated by magnetic-activated cell sorting assays. This approach allowed capturing and purification of human primary tumor cells, CTCs, and metastatic cells from the PDX models (Fig. 2A).
To evaluate whether the CTCs isolated from the highly metastatic Panc265 model possessed tumorigenic capacity, 5,000 and 10,000 HLA-ABC–positive cells were transplanted subcutaneously in the flank of NSG mice. Seven of 8 mice developed detectable tumors after 1 month postimplantation (Fig. 2B). These human CTC-derived tumors also exhibited metastatic potential (Fig. 2C); but not the high incidence (100%) of liver metastases observed in the tumor derived model characterized previously (Fig. 1). Probably, the reason for this is the subcutaneous versus orthotopic implantation of the CTCs.
Identification of potential drivers of the metastatic process with actionable potential in PDAC
To identify potential drivers of the metastatic process in PDAC, we performed single-cell RNA-seq of individual PDAC cells obtained from the primary tumor (orthotopic graft), blood, and liver metastasis (as more common site of spreading in PDAC) of model Panc265 (Supplementary Materials and Methods). We successfully sequenced 37 human primary tumor cells, 23 liver metastatic cells, and 10 CTCs with high quality characteristics (Supplementary Table S3). Applying principal component analysis, the first and second principal components separated the cells into three distinctive main clusters: CTCs, primary tumor, and liver metastatic cells (Fig. 3A). The first principal component separated cells by their tissue environment (solid tissue and bloodstream). The second principal component separated primary tumor from liver metastasis.
We then used pathway and gene set overdispersion analysis (PAGODA) to further characterize the main functional features of this transcriptional heterogeneity and to analyze the transcriptional diversity on biological pathways and gene ontologies (25). This method identified extracellular matrix (ECM) interaction and focal adhesion in addition to cell cycle as the key functional patterns of heterogeneity among the single-cell populations (Fig. 3B). These functional features showed a distinct pattern characteristic of each cell population that differs from each other between primary tumor, liver metastasis, and CTCs. Genes such as FN1, ITGB5, and COL1A1, which are the known markers of ECM interaction, had a decreased expression in CTCs. GSEA of the CTC population showed a strong downregulation of genes implicated in cell-to-cell attachment including ECM receptor, apical surface and apical junctions, and cell adhesion molecules, and upregulation of oxidative phosphorylation metabolism (Fig. 3C and D). To identify deregulated genes, we tested for differential expression between the different cell populations. We obtained 279 genes differentially expressed (FDR < 0.05) when comparing each one of the three biological samples (Supplementary Tables S4–S6).
To identify potential antimetastatic targets in these models, we employed two complementary approaches:
(i) Gene expression profiling of human CTCs: single-cell RNA-seq analysis identified BIRC5 (survivin), a member of the inhibitors of apoptosis protein (IAP) family, as highly expressed in CTCs (a 3.2-log fold change in expression) and metastasis (a 1.9-log fold change) when compared with the expression level in primary tumor (Supplementary Tables S4 and S5; Supplementary Fig. S6A). These findings were confirmed by IHC analysis that revealed nuclear expression of survivin protein in primary and metastatic tumors (Supplementary Fig. S7A). Furthermore, we also identified survivin expression in isolated human CTC by immunofluorescence (Supplementary Fig. S7B).
(ii) Connectivity map: the analysis revealed an enrichment of antiapoptotic signals in CTCs including CASP9 and BIRC5 and proapoptotic drug response signatures (Supplementary Fig. S6D). The top opposite drug signatures were PAC-1 (promoter of caspase signaling) and YM155 (BIRC5 inhibitor). GSEA preranked analysis of the comparison of primary tumor versus CTCs gene expression data showed that AURKA-, AURKB-, and PLK1-knockdown signatures had opposing expression signatures in CTCs. In addition, their kinase substrates were significantly enriched in CTCs compared with primary tumors (Supplementary Materials and Methods; Supplementary Fig. S6B and S6C).
Efficacy of agents targeting the pathways involved in the metastatic process in PDX models
First, we tested inhibitors of survivin (YM155) and aurora kinase (danusertib), and caspase activator (PAC-1) in the Panc265 model. Tumors were orthotopically implanted and treatments were initiated 1 week after implantation (Supplementary Fig. S8A). As shown in Fig. 4A, PAC-1 was ineffective in this model. Danusertib and YM155 treatment resulted in a significant improvement in median survival, which was further extended when combined with nab-paclitaxel. Importantly, analysis of ALU-ISH–positive human cancer cells in metastatic lesions from lung and liver showed that YM155 was significantly more effective than danusertib in hampering the metastatic spread (Fig. 4B).
For these reasons, we focused our next in vivo studies on YM155 using a different experimental design, in which drugs were administered 2 days after implantation to understand whether early inhibition of survivin would further abrogate the metastatic spread (Supplementary Fig. S8B). As shown in Fig. 4C and D, YM155 as well as nab-paclitaxel significantly improved median survival compared with control, and the combination of these two agents was statistically superior to either one alone in both PDX models. In addition, it completely eliminated metastatic spread in Panc265 (Supplementary Fig. S9).
Understanding the mechanisms by which survivin promotes metastasis
Two independent shRNAs were used to knockdown the expression of survivin in organoid model generated from Panc265 PDX (Supplementary Fig. S10). Loss of survivin resulted in a significantly decreased efficiency of organoid formation, as determined by the number of organoids established in culture (Supplementary Fig. S11A). However, once established, shRNA-survivin organoids did not differ in growth potential compared with controls, as determined by the final size (Supplementary Fig. S11B), suggesting that survivin does not impact the proliferative processes that regulates growth of organoids, while playing a critical role during the early stages of culture. Similarly, pharmacologic inhibition of survivin with YM155 in two different organoid models confirmed that it was sufficient to inhibit viability of organoid cultures (Fig. 4E).
Survivin has been associated with anoikis, a form of programmed cell death induced by loss of cellular adhesion to ECM (26). To directly test whether survivin regulates anoikis in PDAC cells, we cultured control and shRNA-survivin organoids under anoikis conditions. As shown in Fig. 4F, shRNA-survivin organoids showed a significant increase in cell death compared with controls suggesting that survivin is a key regulator of anoikis.
Pancreatic cancer is highly lethal with rapid dissemination of tumor cells leading to widespread disease, yet its metastatic initiation process remains poorly understood. While numerous models for studying PDAC exist including cell lines, organoids, GEMMs, and PDXs, there are important limitations in systems to model the role of CTCs in metastatic initiation. The aim of this work was to develop highly metastatic PDAC PDX models for subsequent studies of human CTCs, to identify aberrant signaling and metabolic pathways as well as potential therapeutic targets.
We established a cohort of seven PDAC PDX models with substantial metastatic potential representing a unique resource for further studies. The observation that the models only metastasize in the NOD/SCID/IL2g mice, with a more immunosuppressed background, suggests a role for elements of the immune system in restraining metastatic dissemination. In addition, the development of brain metastasis in model Panc265 makes it particularly useful for the study of metastatic spread to the central nervous system. Of note, we observed that our model resulted in a proportionally higher number of metastasis to the lungs as compared with liver, which is not typically observed in clinical practice. This pattern of disease spread in orthotopic models, while uncommon, has been previously observed and is a matter of investigation by other groups (27).
Single-cell transcriptome analysis of orthotopic graft, considered as primary tumor, CTCs, and liver metastasis showed a distinct expression pattern. This suggests considerable clonal heterogeneity during metastatic progression, mostly noticeable in the CTC cluster. The functional features of this transcriptional heterogeneity were different between primary tumor, liver metastasis, and CTCs. Genes such as FN1, ITGB5, and COL1A1, which are known markers of ECM interaction, had a decreased expression in CTCs, which could explain their ability to detach from the primary tumor. GSEA of the CTC population showed a strong downregulation of genes implicated in cell-to-cell attachment and hypoxia, as well as upregulation of oxidative phosphorylation. This complex genetic and metabolic reprogramming supports the idea that CTCs need to adapt to transition from the primary tumor microenvironment, which is more hypoxic and depended on multiple mechanical cues from neighboring tissues, to survive in the bloodstream with significantly more oxidative stress and limited reliance on cell–cell and cell–ECM interactions (28, 29).
The findings of downregulation of certain EMT genes in CTCs differ from another study (17). This can be explained, at least in part, by distinct systems employed. Here, we established human tumor xenografts, with the genomic complexity known to PDAC, on an immunocompromised murine system to analyze human CTCs, while the other study utilized KPC models to analyze murine CTCs. In addition, we employed an unbiased method to isolate CTCs (HLA based), therefore having no contamination from mouse cells, while the other study utilized a microfluidic approach to separate mouse CTCs from blood cells. The expression of EMT genes in CTCs is not yet fully understood and is likely regulated by multiple factors, including interaction of CTCs with blood cell elements and oxygen content (30–32). For instance, while some studies reported that the interaction of neutrophils with CTCs did not affect the expression of EMT genes, others have shown that the interaction with platelets promoted a transition to a mesenchymal-like phenotype in CTCs, in a process governed by TGFβ (30, 31).
To identify potential antimetastatic targets, we employed two complementary approaches. First, we interrogated the gene expression signatures that were upregulated in CTCs as compared with primary tumors to identify dependencies that could be targeted. In particular, those that promote viability of CTCs in the bloodstream. Unexpectedly, single-cell RNA-seq analysis identified BIRC5 (survivin) as highly expressed in both CTCs and metastasis when compared with primary tumor. Next, we utilized connectivity map to discover potential drugs to revert those signatures associated with the transition from primary tumor to CTC (33). We observed an enrichment of antiapoptotic signals in CTCs including CASP9 and BIRC5 and proapoptotic drug response signatures. The compounds with highest opposite effect to those signatures were PAC-1 and the BIRC5 inhibitor, YM155. Interestingly, gene set enrichment preranked analysis of the comparison of primary tumor against CTCs showed that AURKA-, AURKB-, and PLK1-knockdown signatures also had opposing expression signatures in CTCs. Moreover, their kinase substrates were significantly enriched in CTCs compared with primary tumors, suggesting that the inhibition of those kinases might revert the CTCs transcriptional phenotype.
Targeting the IAP family of proteins, including survivin, is an emerging therapeutic strategy (34). We showed that while survivin does not impact the proliferative processes of organoids that are already established, it plays a critical role during the early stages of culture. We also showed that BIRC5 knockdown significantly increased cell death of organoids under anoikis conditions, supporting the concept that survivin regulates cell viability when cellular adhesion to ECM is lost (26). Newer inhibitors of survivin as well as other IAP protein members are currently in preclinical and clinical development (35). The models developed in this study could indeed be used to test whether these agents prevent or delay PDAC metastasis and should, therefore, be explored in this disease.
In conclusion, we generated a cohort of PDX models of PDAC that provided an efficient and reproducible system to model the metastatic process. This in vivo platform allowed us to isolate and characterize human CTCs and their corresponding primary tumor and metastatic cells, facilitating their transcriptional dissection at the single-cell level. We uncovered important functional reprogramming of this cellular component such as low proliferative state, downregulation of EMT-associated genes, and a metabolic switch to oxidative phosphorylation. Transcriptomic analysis revealed several potential targets to prevent or delay the metastatic process, including survivin. The finding that survivin inhibition alone and in combination with chemotherapy decreased the metastatic burden and increased survival in PDX models should spur the conduction of clinical trials with IAP drugs to thwart the metastatic process in pancreatic cancer.
Disclosure of Potential Conflicts of Interest
M. Hidalgo is a consultant (paid consultant) at InxMed, Agenus, Oncomatrix, Bioncotech, Pharmacy, and Takeda, reports receiving a commercial research grant from Bioline, has ownership interest (including patents) in Agenus, Pharmacy, Bioncotech, Champions Oncology, and Nelum. No potential conflicts of interest were disclosed by the other authors.
Conception and design: S. Dimitrov-Markov, J. Perales-Patón, P.P. Lopez-Casas, M. Hidalgo
Development of methodology: S. Dimitrov-Markov, J. Perales-Patón, M. Muñoz, N. Baños, F. Al-Shahrour, P.P. Lopez-Casas, M. Hidalgo
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Dimitrov-Markov, A. Dopazo, M. Muñoz, N. Baños, V. Bonilla, C. Menendez, L. Huang, S. Perea, M. Hidalgo
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Dimitrov-Markov, J. Perales-Patón, L. Huang, F. Al-Shahrour, P.P. Lopez-Casas, M. Hidalgo
Writing, review, and/or revision of the manuscript: S. Dimitrov-Markov, J. Perales-Patón, B. Bockorny, S. Perea, F. Al-Shahrour, P.P. Lopez-Casas, M. Hidalgo
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Dimitrov-Markov, J. Perales-Patón, M. Muñoz, C. Menendez, Y. Duran
Study supervision: S. Perea, F. Al-Shahrour, P.P. Lopez-Casas, M. Hidalgo
Other (conception, design, and supervision of a part of the study represented in Fig. 4E and 4F): S.K. Muthuswamy
S. Dimitrov-Markov was the recipient of an FPI fellowship from the Spanish Ministry of Economy and Competitiveness (BES-2012-055779). CNIO Bioinformatics Unit was supported by the Instituto de Salud Carlos III (ISCIII); Spanish National Bioinformatics Institute (ELIXIR-ES, INB) grant (PT17/0009/0011 - ISCIII-SGEFI/ERDF); Marie-Curie Career Integration grant (CIG334361); and J. Perales-Patón was supported by Severo Ochoa FPI grant doctoral fellowship by the Spanish Ministry of Science and Innovation.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.