Loss of PDPK1 abrogates resistance to gemcitabine in label-retaining pancreatic cancer cells

Pancreatic ductal adenocarcinoma (PDAC) is an especially lethal disease with 53,070 new cases diagnosed last year and 41,780 deaths due to disease [1]. Its 5-year survival rate of 5–8% has not substantially changed over the last three decades and the American Association for Cancer Research (AACR) estimates pancreas cancer to rank second in cancer-related mortality in the U. S by the year 2020 [2]. Despite recent significant advances in the knowledge of the underlying molecular mechanisms in PDAC, meaningful long term survival remains elusive [3]. More than 80% of patients present with locally advanced or distant metastatic disease at time of diagnosis, which precludes operative extirpation and, therefore the only modality associated with longer term survival. These patients are thus relegated to palliative systemic therapies with the best combination of conventional cytotoxic chemotherapy for advanced pancreas cancer conferring a median survival estimate of less than 1 year [4, 5]. Given the dismal long term survival for the vast majority of patients with this disease, new therapeutic approaches in treatment of this disease are needed.

The cancer stem cell (CSC) theory holds that: 1) cancer arises from cells with dysregulated self-renewal mechanisms; and, 2) cancer is comprised of a heterogeneous mass of cells, a small fraction of which consists of stem-like progenitor cells that drive tumor growth and metastasis [6, 7]. The theory itself is a progression of Knudson’s two-hit hypothesis of carcinogenesis (initiation and promotion), though the origin of the cell lineage involved with initiation and promotion of neoplastic growth is different. A detailed pancreas cancer-specific stem cell phenotype-genotype association remains elusive, which is, in part, due to the different standards of definition and isolation of such cells but also due to an increased recognition of the inherent heterogeneity of the CSC fraction [8‐12] While many groups have described cancer stem cells from multiple tissue sources using a variety of methods, these reported methods rely on cell surface moieties as a surrogate for the identification of these stem cells, but do not necessarily isolate CSCs in a manner reflective of their proposed function and hierarchy [12‐15].

Almost 40 years ago ‘mutational selection’ in cancer was described and followed 3 years later by the first description of label retaining cells (LRC) and the ‘immortal strand hypothesis’ [16, 17]. Label-retaining cells (LRC) are associated with populations of cells enriched with adult tissue stem cells [18‐21]. Many solid organ cancers develop in tissues found to harbor LRC and it is increasingly recognized that slowly cycling LRCC exhibit cancer stem cell and pluripotency traits representing a distinct subpopulation of the heterogeneous CSC pool [5, 22‐26]. The clinical importance of the LRCC subpopulation has recently been demonstrated in a sentinel report of repopulation of residual tumors post-chemotherapy treatment with new cancer cells from this pool of cells [27]. Other reports have linked slowly cycling LRCC to disseminated tumor cells (DTC), relapse, and metastasis in cancer patients [28, 29]. Recently, we demonstrated that label-retaining cancer cells (LRCC) undergo asymmetric cell division, and represent a unique subpopulation of tumor-initiating stem-like cells with pluripotency gene expression profiles [20]. While early reports described fixed cells, which precluded downstream analysis, we recently published on such a method for the isolation of live tissue-derived LRC allowing for future assays dependent on live functioning cells. Using these methods it was shown that LRC do, in fact, undergo asymmetric cell division with non-random chromosomal cosegregation (ACD-NRCC) [20, 30].

The identification of LRCs in PDAC [i.e. pancreatic cancer-derived label retaining cancer cells (LRCC)] would offer a unique opportunity to study features of cancer stemness, in particular with regard to identifying vulnerabilities of this cell population knowledge which has remained elusive for the design of more effective therapies in pancreas cancer and drug development in general. Despite the ability to potentially impact sentinel events in cancer recurrence and progression, there is significant paucity in the understanding of selective molecular mechanisms in LRCC.

In the following report, we compare the transcriptome of LRCC and non-LRRC in pancreas cancer cell lines and identify perturbations unique to LRCC. Targeting one of the genes selectively upregulated in LRCC, 3-phosphoinositide dependent protein kinase-1 (PDPK1), we demonstrate that the phenotype resistance to chemotherapy in pancreatic cancer LRCC can be abrogated as a potentially novel treatment avenue against this difficult to treat cell population possibly guiding novel combination therapies in this lethal disease.

Conventional treatment with systemic cytotoxic chemotherapeutics, which non-specifically targets rapidly dividing cells, is ideally replaced, or supplemented, by therapy that targets the cells driving recurrence and metastasis—the primary causes of most cancer-related death for patients afflicted by pancreas cancer [3]. Here we demonstrate that by inhibiting genes overexpressed in label-retaining cancer cells, an in vitro model of slowly cycling cells and subpopulation of CSC, can improve response to the standard cytotoxic chemotherapy with gemcitabine. The rationale of manipulating sub-populations of malignant cells within a tumor, in particular sub-populations involved in chemoresistance and metastasis, has recently been shown in elegant murine models of breast and pancreas cancer, and could have important implications in the treatment of patients with cancer [34‐36].

While the identification of cancer stem cells in the literature is largely based on cell surface phenotype [12‐14, 37‐39], previous work from our group and others has shown that identification of a population of cells based on proposed function and hierarchy within a tumor is possible [19, 40]. Our group has previously shown that LRCC undergo asymmetric cell division with non-random chromosomal cosegregation (ACD-NRCC), which is consistent with the carcinogenesis (initiation and promotion) portion of the cancer stem cell theory [18, 20, 41]. Further, LRCC have been shown to be more tumorigenic, which is consistent with the tumor progression (tumor growth and metastasis) portion of the cancer stem cell theory [18, 25]. More recently, there have been more reports on LRCC mediating chemoresistance, early tumor recurrence, and metastasis [29, 42, 43]. However, there has been only one report on slowly cycling cells in pancreas cancer using the label Dil, and a surprising paucity on intracellular signaling features unique to LRCC governing chemoresistance or tumor initiation [44]. In this report we aimed to assess the possible clinical significance of the LRCC population in terms of response to gemcitabine, one of the standard therapies for patients with pancreas cancer. Deriving LRCC from different pancreatic cancer cell lines we first showed that LRCC are, in fact, resistant to this’therapeutic’ agent and that decreased rates of apoptosis upon gemcitabine treatment contributed to gemcitabine resistance phenotype of LRCC. One of the initially vexing findings was the decreased reduction of NLRCC growth after gemcitabine administration without a concomitant increase in the LRCC cells. We attributed this to the relative short treatment course of two doubling times possibly too short to detect differences considering the small number of LRCC but also possibly to a to-date incompletely understood interplay between LRCC and non-cancer stem cells along the concept of the ‘stem cell niche’. We speculate that LRCC might feed resistance signals to the bulk population but as viability and growth of NLRCC decreases as a consequence of the cytotoxic treatment ultimately also the pool of stem cells or LRCC is afflicted. Tumor resistance has then developed when stem cells recreate the niche, a biological function our presented in vitro LRCC model with limited gemcitabine exposure times was not able to measure. Such dynamic interplay between non-cancer stem cells and cancer stem cells has now been observed in a number of in vitro models and early paracrine signaling cues involved in this crosstalk involve HIFα (hypoxia inducible factors’ α subunits), sonic hedgehog, TGFβ1 (transforming growth factor beta 1), and nodal/activin to name some described from pancreatic cancer stem cell niche models or, importantly, prostaglandin E2-induced repopulation of the tumor following chemotherapy treatment from the slowly cycling CSC as shown in patient-derived xenotransplanted bladder cancer [27, 45, 46].

In order to investigate differences between the two cell populations and possibly home in on mechanisms of action governing inherent chemoresistance features of the LRCC sub-population, we compared transcriptomic profiles of LRCC to the NLRCC population and identified a top-ranking network by IPA enriched for EGF ligand signaling. Of note, this network is different from the canonical erb receptor tyrosine kinase signal transduction cascades focusing on immediate EGFR (epidermal growth factor receptor) or HER2 (human epidermal growth factor receptor 2) downstream signaling studied in many cancers and more similar to finding from our prior work identifying EGF and MET-mediated signaling as top regulators of cell fate and lineage specific progenitor cell differentiation in the liver [41]. Studies by Takebe and Jeanes et al. in breast and other solid tumors on cell differentiation and stemness have also reported on an intimate role of a more extended, non-canonical erb signaling network and cancer stemness [47, 48]. We selected three kinases of the network for further loss-of-function studies and showed that NTRK2/TrkB, PDPK1/PDK1, and BMX/ETK were involved in regulation of gemcitabine sensitivity. Individual knockdown of PDPK1/PDK1 in both the LRCC and NLRCC subpopulations demonstrated that PDPK1/PDK1 inhibition lowered resistance to gemcitabine preferentially in the LRCC population and less in the NLRCC cells. Loss of PDPK1 had minimal impact on growth and apoptosis rates of untreated cells suggesting an essential role of this regulator in the LRCC population. Along these lines, recent work has PDPK1 also been implicated in the regulation of self-renewal, cellular transformation, and stemness in several diseases including cancer. Of importance, this work identified downstream signaling cascades regulated by PDPK1 outside the canonical receptor tyrosine kinase, PI3K and AKT-cascade PDKP1 was first described. These include phospholipase C, protein kinase C, or Hippo signaling governing stemness features through crosstalk with WNT [wingless-type MMTV (mouse mammary tumor virus) integration site] or β-catenin signaling [49‐52].

In pancreas cancer, Eser et al. have shown that PDK1 is an essential effector of KRAS, and that an intact PDK1/PI3K axis is an essential tumor initiating event in cooperation with KRAS for increased cell plasticity, acinar-to-ductal metaplasia (ADM), and pancreatic ductal adenocarcinoma (PDAC) formation [53]. The pro-tumor function of an intact PI3K/PDPK1 axis reported by Eser and colleagues appears to be at odds with our findings of decreased PDPK1 expression levels in tumor tissues, and the associations of membranous localization with well-differentiated tumors and a trend towards improved clinical outcome. PDPK1 signals in a PI3K-depedent manner activating AKT, S6K (ribosomal protein S6 kinase), and SGK (serum/glucocorticoid regulated kinase) but can also activate PKC and p90RSK independent of PI3K and PIP3-mediated activation. PKC (protein kinase C) and p90RSK (p90 ribosomal S6 kinase) were found activated by alternative RAS-mediated signaling pathways independent of PDPK1 in the study by Eser et al. suggesting a non-essential role of PDPK1 in these tumors [53]. Additionally, downstream PDPK1 canonical signaling measured by phospho-AKT and phospho-GSK3β (S9) levels was lower in the evolved pancreas cancers compared to pre-cursor ADM and PanIN lesions also raising the possibility of the more involved cancers having become independent of the PDPK1 signaling axis and more driven by signal transduction perturbations of additional pathways acquired during the later stages of pancreas cancer progression [53]. Such a hypothesis appears to be in line with the decreased PDPK1 expression levels in the tumor vs matched clinical specimens on tissue microarray staining. While correlative tissue studies with phospho-AKT measures have shown the more commonly found negative correlation between increased AKT activation and clinical outcome, it is possible that there is a subset of pancreas cancer, similar to studies in non-small cell lung cancer, where phospho-AKT levels as a measure of EGFR-PI3K-AKT signaling have been shown to be associated with improved outcome [54‐57]. It is intriguing to speculate that the hypothesis of a greater dependency on PDPK1 signaling in the early tumor-initiating events of ADM and PanIN formation, as seen in the transgenic animal studies, versus a later loss of addiction to canonical PDPK1 signaling and overtake by PDPK1 independent oncogenic events is commensurate with the cancer stem cell hallmark of tumor initiation by this cell population [12, 53]. Irrespective of possible differences in PDPK1 function in early vs late or primary vs metastatic tumors as a possible explanation of the observed PDPK1 expression pattern on our cancer tissue microarrays, the association of the heterogeneous PDPK1 expression pattern with important clinicopathological outcomes appears to be in line with the intratumoral heterogeneity of this regulator found to be overexpressed and activated in the LRCC vs NLRCC subpopulations where it is essential for chemoresistance.

On a preclinical level, PDPK1 has shown to confer oncogenic signaling and CSC renewal. The other targets derived from the differential gene expression screen in LRCC vs NLRCC have also been previously linked to stemness. The Tec kinase BMX non-receptor tyrosine kinase (BMX) has been shown a tumor promoting role in gliolastoma multiforme through mediation of self-renewal and growth, and the neurotrophic tyrosine receptor kinase 2 (NRTK2) in precursor growth and differentiation [14, 58, 59]. On the other hand, the gene expression differences identified in our in vitro model of cancer stemness, LRCC vs NRLCC, showed limited overlap with other in vitro models comparing gene expression and pathway alterations between stem cell and non-cancer stem cell fractions. For example, similar IPA network analysis comparing 3D spheroid Panc1 cells, originally described as the invasive Panc1 cell population, versus 2D monolayer cells resembling non-cancer stem cells identified as the top differentially regulated network genes involved in DNA damage repair [60]. When comparing side population versus non-side population bulk cells from patient-derived xenograft models an extensive network enriched with transcription factors of pluripotency, cell differentiation, and EMT (epithelial-mesenchymal transition) can be identified [61]. Whether these differences are due to the different platforms or due to the increasingly recognized heterogeneity of different CSC population is currently not known. The two currently available drug discovery studies performed in pancreas cancer stem cell in vitro models using either a reporter line for the 26S proteasome activity of pancreas cancer cells or a 3D pancreatic carcinoma spheroid model yielded drug activities in the 3D CSC model with inhibition of phosphoinositide 3-kinase signaling, however there was an underrepresentation of selective PDPK1 inhibitors [62, 63].

This study is not without its limitations. First, the isolation of LRCC requires the assumption that all cells within the Cy5⁺ labeled population are expanding at the same rate. Since long term cultured cell lines were used, this assumption is based on the clonality and homogeneity of these lines. While cells were synchronized via transfer to serum-free media 24 h prior to labeling we cannot rule out that present clonal subpopulation with different growth rates in the used pancreas cancer cell lines increased the NLRCC subpopulation. However, we believe this fraction to be small. Second, the apoptosis experiments upon administration of gemcitabine were based on the concentration of the drug for 50% of maximal inhibition of cell proliferation, GI50, after two doubling times (determined individually for each cell line). This includes experiments on induction of apoptosis upon silencing the genes found to be overexpressed in the LRCC population. We cannot rule out that the impact on induction of apoptosis, or silencing of PDPK1 on gemcitabine drug response, might have been higher or lower either in the LRCC or NLRCC populations when other concentrations were used. For the presented in vitro comparisons, it was necessary to establish a consistent cutoff point that allowed for comparisons including downstream evaluation (i.e. qRT-PCR, cell cycle analysis, FACS) without complete mortality of the cells after gemcitabine exposure.

In summary, we present in LRCC, an in vitro model of slowly cycling pancreatic cancer stem cells implicated in chemoresistance and tumor recurrence, the discovery of novel targets involved in response to gemcitabine treatment, one of the standard chemotherapy approaches used for pancreas cancer in the clinic. For decades, chemotherapy regimens have been based on the cell cycle differential that exists between normal and malignant tissue. Given the continued and unchanged dismal prognosis for patients with pancreas cancer this rationale has proven not specific enough and not sufficiently efficacious. Here, we have shown that LRCC harbor distinct transcriptomic profiles involved in mediation of chemoresistance and that targeting essential regulators of this program in LRCC can sensitize pancreas cancer cells. These findings offer novel hypotheses for the derivation of effective combination therapy approaches in a disease void of impactful interventions.