Targeting Epithelial Mesenchymal Plasticity in Pancreatic Cancer: A Compendium of Preclinical Discovery in a Heterogeneous Disease

Pancreatic Ductal Adenocarcinoma (PDAC) is a particularly insidious and aggressive disease that causes significant mortality worldwide. The direct correlation between PDAC incidence, disease progression, and mortality highlights the critical need to understand the mechanisms by which PDAC cells rapidly progress to drive metastatic disease in order to identify actionable vulnerabilities. One such proposed vulnerability is epithelial mesenchymal plasticity (EMP), a process whereby neoplastic epithelial cells delaminate from their neighbours, either collectively or individually, allowing for their subsequent invasion into host tissue. This disruption of tissue homeostasis, particularly in PDAC, further promotes cellular transformation by inducing inflammatory interactions with the stromal compartment, which in turn contributes to intratumoural heterogeneity. This review describes the role of EMP in PDAC, and the preclinical target discovery that has been conducted to identify the molecular regulators and effectors of this EMP program. While inhibition of individual targets may provide therapeutic insights, a single ‘master-key’ remains elusive, making their collective interactions of greater importance in controlling the behaviours’ of heterogeneous tumour cell populations. Much work has been undertaken to understand key transcriptional programs that drive EMP in certain contexts, however, a collaborative appreciation for the subtle, context-dependent programs governing EMP regulation is needed in order to design therapeutic strategies to curb PDAC mortality.

Cancer is a genetic disease caused by the accumulation of somatic mutations, resulting in a functional imbalance between tumour suppressive and oncogenic signals [3]. While transformed cells retain characteristics of the host to efficiently avoid being detected as foreign by the immune system, many aberrant phenotypes caused by genetic mutations and dysregulated signaling potentially render these cells susceptible to selective therapeutic interventions. Extensive examinations of the molecular traits of PDAC aimed at identifying such vulnerabilities have been conducted to date. Indeed, genomic and transcriptional profiling of patient tumours as part of large-scale studies by the The Cancer Genome Atlas (TCGA) and International Cancer Genome Consortium (ICGC) have allowed for insights into the scale of inter-tumour heterogeneity in a breadth of patient cohorts [4][5][6].
These studies have identified four major genetic aberrations common to pancreatic tumours [7][8][9]. 90% of tumours carry gain-of-function mutations in KRAS2, activating proliferative and cell survival pathways, whilst 95% contain either partial or complete inactivating mutations in CDKN2A, contributing to loss of cell cycle regulation, furthering proliferation. TP53, responsible for responding to DNA damage and inducing apoptosis, is altered in 60% of cases. SMAD4 inactivation is also common in pancreatic cancer development, and is found in 50% of patient cancers, disrupting the tumour suppressive signals of TGFβ, aiding proliferation [10]. As well as these four common driver mutations, genomic sequencing of tumours has identified an additional panel of consistently mutated genes [6]. These genetic mutations implicate pathways often dysregulated in cancer, including KRAS, TGFβ, WNT, NOTCH, ROBO/SLT, G1/S, SWI-SNF, and chromatin/DNA/RNA modification and repair.
Transcriptional profiling of PDAC tumours has allowed researchers to define discrete regulatory mechanisms within these networks that are associated with particular prognostic indices in different molecular subtypes of PDAC, which include squamous, pancreatic progenitor, immunogenic and aberrantly differentiated endocrine/ exocrine tumours [6]. Such classification schemes may provide clinical value by aiding in patient treatment regimen selection and planning [11], however, to date they have provided limited clinical value due to lack of targetable phenomena. It is important to note that while these studies have aimed to characterise changes within carcinoma cells, the excessive presence of desmoplastic stroma may confound these results. Indeed, microdissection of the tumour from its associated stroma has allowed the retrospective re-evaluation of large-scale transcriptional profiling efforts, highlighting the overwhelming contribution of stromal contamination to many such studies. Deconvolution based on laser capture microdissection and RNASeq profiling of 60 matched tumour/stroma pairs suggested that ICGC and TCGA samples contained stromal fractions of 46% and 55%, respectively, highlighting difficulties in deriving definitive conclusions from whole tumour analyses [12].
Such studies are invaluable as a means of understanding the intertumoural heterogeneity that exists between patients, and they form a strong set of public data that have been analysed to better appreciate the diversity of tumour presentation [13]. An increasing focus on single cell analytic technologies has yielded exciting opportunities to understand the contributions that individual cells make towards intratumoural heterogeneity, tumour progression, and patient outcomes [14,15]. These studies highlight the need for efforts aimed at distinguishing the heterogeneous nature of a tumour's biology from that of the surrounding host tissue in which it propagates, so as to be better able to exploit cancer specific vulnerabilities [16].
As such, it is not surprising that the interactions between neoplastic epithelial cells and host myofibroblast and stellate populations, which can promote stromal inflammation, are increasingly being recognised. This desmoplastic reaction, which accounts for up to 90% of PDAC tumour volume, has pro-tumourigenic properties by leading to increased tissue stiffness and hypoxia as well as by providing physical barriers to both immune surveillance and chemotherapeutic penetrance [17][18][19]. The fibrillar collagen, hyaluronic acid and fibronectin rich extracellular matrix (ECM) deposited by stromal cells contains many soluble cytokines and growth factors secreted by both cancer and stromal compartments and contributes to both tumour initiation and progression [20][21][22][23]. Resident cells are forced to interact within this dynamic tumour microenvironment and are subject to stimuli that influence cell phenotypes in both stromal and carcinoma components. Such stimuli may propagate the invasion and dissemination of carcinoma cells by inducing epithelial mesenchymal plasticity (EMP), and thus this process is considered an important vulnerability that, when effectively targeted, may curb tumour progression [24,25].

EMP and PDAC Progression
EMP is often separated into two distinct but related processes-the forward process of epithelial-mesenchymal transition (EMT), and the reverse process of mesenchymal-epithelial transition (MET) [26]. These programs serve to describe the plasticity within epithelial cells that enables them to dedifferentiate into a more motile mesenchymal state, thereby allowing them to more effectively migrate. EMP is thought to play a significant role in several stages of tumour formation [27] and progression [28]. Initially, this plasticity allows tumour cells to detach and migrate from their site of origin (invasion), gaining access to lymphatic and blood vessels (intravasation), and then penetrating distant sites (extravasation), to form metastases.
Development of PDAC frequently progresses undetected, remaining asymptomatic until it becomes an advanced stage of disease. Non-invasive precursor lesions formed either by epithelial proliferations or mucinous cysts in the pancreatic ducts, termed pancreatic intraepithelial neoplasia (PanINs), or intraductal papillary mucinous neoplasms (IPMNs), respectively, mark the onset of a histologically definable neoplasm in PDAC [50]. Such neoplasms, namely PanINs, progress through stages of dysplasia within the ductal epithelium, giving rise to the most common form of PDAC, pancreatic ductal adenocarcinoma (PDAC). The full breadth of factors that contribute to the invasive and metastatic behaviour of PDAC are vast. In this form of PDAC, there is very little latency between primary tumour formation and local and distant metastasis, implying that PDAC carcinoma cells may be readily equipped to invade and disseminate from a very early stage of development [51,52].
Invasive regions of human carcinomas are typically characterised by the presence of tumour-derived, fibroblast-like cells expressing mesenchymal markers such as vimentin, fibronectin and N-cadherin, with decreased expression of epithelial adhesion molecule E-Cadherin and increased nuclear beta-catenin relative to surrounding cells [53][54][55][56][57]. Decreased expression of E-cadherin has been shown to correlate with invasive and undifferentiated PDAC [58]. Furthermore, PDAC patients with tumour cells that express decreased E-cadherin and higher amounts of vimentin, s100A4, fibronectin and SNAI1 are more likely to have distant metastases, lymph node invasion and lower overall survival [54,[59][60][61][62]. The EMP inducing transcription factor (TF) TWIST1 has been shown to be upregulated in PDAC compared to match normal tissues [63], and SNAI1 mRNA levels in PDAC fine needle aspirates are significantly correlated with lymph node and perineural invasion as well as with poorer survival [64]. A mediator of transforming growth factor beta (TGFβ) signaling, SMAD3, was also shown to accumulate in the nucleus of PDAC samples, and was correlated with higher grade tumours and lymph node metastasis, indicating a role for TGFβ in driving EMP in vivo [65]. Solitary infiltrating cancer cells displaying low E-cadherin and increased vimentin expression have proven to be significant prognostic indicators in resected clinical specimens from PDAC patients [66]. Tumour budding cells in PDAC have been observed with increased levels of ZEB1 and ZEB2, and reduced levels of E-cadherin and β-catenin, indicative of EMP mediated local invasion. ZEB2 overexpression in tumour-stroma associated cells also correlated with pathological assessment of tumour size, and lymph node metastasis [67]. Such striking pathology provides some of the clearest evidence for the role of EMP in PDAC progression.
While this clinical evidence strongly supports a role for EMP in mediating cancer invasion, the inability to accurately follow carcinoma epithelial dedifferentiation in vivo has led to some debate surrounding the extent of its role in tumour progression [68,69]. Such debate has necessitated the use of genetically engineered mouse models (GEMMs) to trace the role of EMP in cancer progression, specifically the pancreatic epithelium conditional Kras/P53 mutant (PKCY) mice Lineage labelling of epithelial cells in this spontaneous PDAC model has allowed researchers to track these cells as they adopt mesenchymal properties and migrate away from the primary tumour into the circulation, seeding liver metastases [70]. In one study, EMP was detected in 42% of labelled PDAC epithelial cells, as assessed by the expression of EMP markers Zeb1 or Fsp1 and/or lack of E-cadherin. These cells were mostly observed in regions of inflammation, supporting the idea that EMP is driven by inflammatory interactions within the tissue microenvironment. Interestingly, some labelled epithelial cells that had undergone EMP displayed evidence of delamination and fibroblast morphology prior to tumour formation, and were otherwise indistinguishable from host stromal cells [70]. This is supportive of the very early, integral role that EMP may play in PanIN formation prior to tumour development.
Further studies in this same PDAC mouse model have shown that suppression of EMP via the knock-out of Twist1 or Snai1 TFs does not reduce metastasis, despite the decreased expression of EMP markers and increased cell proliferation as evidence for EMP ablation [71]. Equivalent numbers of lineage labelled epithelial cells were found in circulation and in metastases regardless of Twist/ Snai1 knockout, suggesting that other mechanisms are involved in PDAC cellular invasion. PDAC cells do not possess a strong epithelial phenotype however, and may thus be insensitive to the loss of Snail TFs, which are potent repressors of epithelial programs but are less efficient in inducing mesenchymal properties. This possibly explains why Snail is dispensable for EMP and metastatic progression in this model [71,72], and points towards alternative mechanisms of EMP induction that may be driving factors in this PDAC system.
Indeed, there is evidence that the Zeb1 TF is largely responsible for driving EMP in this GEMM model of PDAC development [73]. Zeb1 ablation in PDAC cells was not found to affect Twist1 expression, however it was associated with decreased Zeb2, Slug and a slight reduction in Snai1 expression. Zeb1 depleted tumours were better differentiated, indicating less local invasion, and showed significantly reduced metastasis when compared to control PDAC mice [73]. This is in direct contrast to depletion of Twist1 or Snai1, which did not affect metastasis in this model system, highlighting the importance of recognising the context and tissue specific drivers of EMP.
Subsequent investigations aimed at overcoming the limitations of identifying single EMP regulatory TFs has shown that lineage labelled cancer cells are able to metastasize without expression of αSma or Fsp1, both of which are thought to be robust markers of EMP activation in this model [74]. Indeed, larger metastatic nodules were found containing exclusively cells that had never expressed αSma or Fsp1, while micrometastatic clusters of 3-5 cells were shown to have undergone EMP. Such evidence, combined with the fact that Zeb1 depletion in previous studies resulted in only a 50% reduction in metastasis underscores the pitfalls of seeking to identify individual master regulators and markers of such a complex process. Adding to this complexity, the emerging importance of hybrid EMP phenotypes, in which the expression of both epithelial and mesenchymal markers may occur at levels that are insufficient to drive the reporter constructs used in such lineage tracing models, adds a further technical challenge [75][76][77].
More recent attempts to understand EMP in individual PDAC cells has shown the activation of EMP transcriptional programs within certain subsets of tumour cell populations [14]. This study highlighted a clear role for cytokines from the stromal compartment in inducing EMP in certain PDAC cell lines, and indicated that EMP activation could be observed in discrete tumour gland subunits with prognostic utility. These models have provided considerable insights into the diverse mechanisms of PDAC development, and highlight that there are context-dependent EMP programs involved in both local invasion and metastatic dissemination that require further examination [72,78].

In Vitro EMP Models and Exogenous Stimuli
While GEMMS, in particular the PKCY model of spontaneous PDAC formation, are currently the gold standard for studies of the biology of EMP in tumourigenesis, in vitro studies form the basis for the majority of our current molecular understanding of intracellular events which occur in EMP. Many publicly available and in-house generated cell lines are used to study PDAC, but only a very limited number of these undergo well-characterised, stimulus-driven transitions that mimic the pathophysiological induction of EMP. This is perhaps consistent with the limited number of EMP events witnessed in in vivo models, highlighting the difficulties of studying such a dynamic process.
EMP is modulated by TGFβ, receptor tyrosine kinases (RTK) ligands, WNT ligands, interleukins, hypoxia via HIF1α signaling, as well as HIPPO, NOTCH signaling. Their mechanisms and specific impact on downstream EMP targets have been comprehensively reviewed elsewhere, however our understanding of their subtleties is on-going [79,80]. TGFβ acts as a tumour suppressor in normal tissue and early stage disease by regulating cell proliferation and inducing apoptosis through canonical signaling pathways, however this activity is lost as cellular transformation progresses [81][82][83][84][85]. Indeed, TGFβ is a potent activator of EMP in PDAC cells when its tumour suppressive signals are disrupted through SMAD4 mutations, found in 50% of PDAC tumours [81,86]. Similarly, activating KRAS mutations found almost ubiquitously in PDAC cooperate with TGFβ signaling to hyperactivate downstream RAS/RAF MAPK pathways to induce EMP [87]. While TGFB activates the greatest number of EMP signaling pathways, and may thus be considered a major driver in PDAC, the activation of additional pathways shown in Figure 1 by RTK, WNT and interleukin ligands may provide additional layers of crosstalk. Activation of SMAD, MAPK, PI3K, STAT, and NFκB pathways are commonly demonstrated in PDAC EMP research, however the relative extent to which each pathway governs EMP is unclear, as many studies evaluate these pathways independently [29,[88][89][90][91][92][93][94].
These complex pathways ultimately serve to influence transcriptional programs that co-operate directly and indirectly to control the plasticity that exists between epithelial and mesenchymal phenotypes of carcinoma cells ( Figure 1). Of note is the increasing recognition for the role of long non-coding RNAs (LncRNA) and micro-RNAs (miRNA) in EMP regulation. Among the cells that do undergo EMP-like transitions, there is a degree of selectivity for the ligands that are able to activate these EMP programs, and this is reflected in the limited number of commercial cell lines that are commonly manipulated within the field. This is consistent with the level of heterogeneity reported in PDAC, and suggests discrete differences in steady state signaling, which may predispose a given cell's response or resistance to exogenous stimuli. limited number of these undergo well-characterised, stimulus-driven transitions that mimic the pathophysiological induction of EMP. This is perhaps consistent with the limited number of EMP events witnessed in in vivo models, highlighting the difficulties of studying such a dynamic process. EMP is modulated by TGFβ, receptor tyrosine kinases (RTK) ligands, WNT ligands, interleukins, hypoxia via HIF1α signaling, as well as HIPPO, NOTCH signaling. Their mechanisms and specific impact on downstream EMP targets have been comprehensively reviewed elsewhere, however our understanding of their subtleties is on-going [79,80]. TGFβ acts as a tumour suppressor in normal tissue and early stage disease by regulating cell proliferation and inducing apoptosis through canonical signaling pathways, however this activity is lost as cellular transformation progresses [81][82][83][84][85]. Indeed, TGFβ is a potent activator of EMP in PDAC cells when its tumour suppressive signals are disrupted through SMAD4 mutations, found in 50% of PDAC tumours [81,86]. Similarly, activating KRAS mutations found almost ubiquitously in PDAC cooperate with TGFβ signaling to hyperactivate downstream RAS/RAF MAPK pathways to induce EMP [87]. While TGFB activates the greatest number of EMP signaling pathways, and may thus be considered a major driver in PDAC, the activation of additional pathways shown in Figure 1 by RTK, WNT and interleukin ligands may provide additional layers of crosstalk. Activation of SMAD, MAPK, PI3K, STAT, and NFκB pathways are commonly demonstrated in PDAC EMP research, however the relative extent to which each pathway governs EMP is unclear, as many studies evaluate these pathways independently [29,[88][89][90][91][92][93][94].
These complex pathways ultimately serve to influence transcriptional programs that co-operate directly and indirectly to control the plasticity that exists between epithelial and mesenchymal phenotypes of carcinoma cells ( Figure 1). Of note is the increasing recognition for the role of long non-coding RNAs (LncRNA) and micro-RNAs (miRNA) in EMP regulation. Among the cells that do undergo EMP-like transitions, there is a degree of selectivity for the ligands that are able to activate these EMP programs, and this is reflected in the limited number of commercial cell lines that are commonly manipulated within the field. This is consistent with the level of heterogeneity reported in PDAC, and suggests discrete differences in steady state signaling, which may predispose a given cell's response or resistance to exogenous stimuli. EMP is induced by stimuli shown within arrows on the left in order of potency. These signals activate signal transduction pathways that cooperate directly and indirectly to translocate signals to the nucleus (braced) to regulate EMP transcription factors, long non-coding RNAs (LncRNA), and micro RNAs (miRNAs).These factors then modulate EMP by discrete regulation of epithelial (Red Box) and mesenchymal (Green box) cellular properties, which in turn influence migration and invasion. Transforming growth factor (TGFB) activates the greatest number of these pathways, including direct cytoskeletal regulation by RhoA, aswell as canonical SMAD and non-canonical EMP is induced by stimuli shown within arrows on the left in order of potency. These signals activate signal transduction pathways that cooperate directly and indirectly to translocate signals to the nucleus (braced) to regulate EMP transcription factors, long non-coding RNAs (LncRNA), and micro RNAs (miRNAs).These factors then modulate EMP by discrete regulation of epithelial (Red Box) and mesenchymal (Green box) cellular properties, which in turn influence migration and invasion. Transforming growth factor (TGFB) activates the greatest number of these pathways, including direct cytoskeletal regulation by RhoA, aswell as canonical SMAD and non-canonical p38/JNK, MEK/ERK MAPK pathways and PI3K/AKT. Receptor tyrosine kinase (RTK) signaling is induced by binding of growth factor (GF) ligands such as EGF, IGF, FGF, HGF or VEGF and activates RAS/MEK/ERK, PI3K/AKT/NFκB and downstream SRC pathways. WNT signaling also modulates EMP by downstream stabilisation of B-catenin and subsequent nuclear translocation for EMP program activation by TCF/LEF transcription factors. Interleukins (ILs) can also induce EMP programs via STAT3 signaling. Additional mediators of EMP include Hypoxia, Hedgehog, Notch and Integrin signaling (not shown), and highlight the context dependent activation of EMP from micro-environmental cues.
While most studies rely upon knockdown and over-expression approaches to demonstrate the function of proteins in the context of cell migration, proliferation and EMP transitions, relatively few studies have investigated these targets in the context of the physiological induction of EMP in response to exogenous stimuli. Among PDAC cell lines, L3.6pl cells have been shown to respond to VEGF treatment [95], while the inflammatory cytokines TNF-α and IL-1β drive EMP in PaTu 8988T and AsPC-1 cells via Hedgehog signaling [96]. Collagen 1 also stimulated L3.6pl and BxPC-3 cells to become more invasive through interaction with DDR1 [97], and BMP2 was able to elicit a similar response in BxPC-3 cells [98]. PANC-1 cells are a well characterised model of inducible EMP, first shown by Ellenrieder et al to undergo a bidirectional change in response to TGFβ alongside CAPAN-1, COLO-357, IMIM-PC1 [99], HPAF-II, and CAPAN-2 cells [100]. PANC-1 cells have since been repeatedly modelled with regard to their EMP response, which has been shown to be inducible in response to TGFβ, TNF-α, HGF, or hypoxia through differing mechanisms [101][102][103][104]. SNAI1 appears to be a major driver in this model, being heavily regulated at the transcript and protein level, despite modest changes in E-cadherin and Vimentin proteins [105]. EMP is thus invariably the result of exogenous stimuli that activate discrete but conserved cellular pathways through novel intermediates that are an ongoing focus of basic cancer cell biology research.

Pre-Clinical Discovery of EMP Targets
As a result of the complexities of discerning cancer biology from native processes in vivo, the use of cell lines derived from primary tumours are a valuable means of modelling the molecular and phenotypic properties of cancers. Extensive investigation has been performed using gene silencing and overexpression approaches to evaluate the role that particular molecules have in regulating or effecting the EMP phenotypes of PDAC cells, however a concise summary of novel targets in the PDAC EMP field has to date been lacking. Thus, this review provides an exhaustive overview of such research as a platform for their integration, and progressive evaluation. The function of these candidate molecules can be broadly separated into secreted/soluble products (Table 1), receptors (Table 2), other membrane associated proteins (Table 3), cytoskeletal adaptors (Table 4), kinases (Table 5), intracellular mediators (Table 6), transcription factors (Table 7) and post transcriptional controllers ( Table 8). The candidates shown were selected by searching Pubmed for the terms 'pancreatic' and 'epithelial', and articles investigating a novel candidate's impact on EMP phenotypes were manually curated. These effectors have been characterised to varying extents for their influence on invasion, migration, xenograft tumour growth, prognostic associations, and impact on known EMP signaling pathways. The proposed mechanisms of candidates and assays used to assess such effects are shown within tables and may be used to gauge where further support may be warranted to confirm and extend such findings. Due to the inherent variation in models used, the statistical power granted by IHC for varying sized patient cohorts with accompanying clinical information, and the level of EMP as a primary context, it is difficult to draw direct conclusions regarding pivotal significance within the field and clinical importance from such singular studies. Candidate expression in primary patient material that correlated with lymph-node metastasis are shown in bold within tables, and provide the best surrogate for their role in EMP mediated invasion, and include membrane bound proteins IGFBP2, ITGB4, CEACAM6 [106][107][108]. The use of IHC to capture dynamic EMP processes may be limited however, as shown in the case of LIN28B, where its expression is both induced by TGFβ and high in PDAC tissue, despite its role to suppress the pro-EMP non-coding RNA LET7a [109,110]. Such studies highlight both the utility and limitations of the links between in vitro assays and clinical material, and emphasise the need for both wider cohorts of patient material for validation and the development of GEMM models to strengthen findings in a standardized manner. Figure 2 illustrates the proposed activity of some of these novel candidates, and how they may positively or negatively regulate discrete EMP signaling pathways. Of note are several candidates that converge to positively regulate EMP migratory phenotypes through FAK/Src and FAK/PI3K signaling, including the 5HT receptor and mucins, as well as EEF2K, USP22, and ZIP4. Their complete mechanisms of action and prevalence in PDAC tissue remain to be elucidated, however their inhibition may curb carcinoma invasion by blocking FAK activation and subsequent EMP modulation. Similarly, candidates participating in stability of EMP signaling and TF activity provide targets to modulate the EMP process specific for carcinoma cells. AURKA kinase has been shown to participate in a positive feedback loop with stabilization and activity of TWIST1, while PEAK1 and NES have been implicated in stabilization YAP/TAZ and SMAD TF activity. The discovery of discrete EMP regulation and development of combinatorial inhibitors may provide the opportunity for more personalized therapeutic approaches to curb metastatic disease.
Cancers 2019, 11, x 7 of 32 Figure 2 illustrates the proposed activity of some of these novel candidates, and how they may positively or negatively regulate discrete EMP signaling pathways. Of note are several candidates that converge to positively regulate EMP migratory phenotypes through FAK/Src and FAK/PI3K signaling, including the 5HT receptor and mucins, as well as EEF2K, USP22, and ZIP4. Their complete mechanisms of action and prevalence in PDAC tissue remain to be elucidated, however their inhibition may curb carcinoma invasion by blocking FAK activation and subsequent EMP modulation. Similarly, candidates participating in stability of EMP signaling and TF activity provide targets to modulate the EMP process specific for carcinoma cells. AURKA kinase has been shown to participate in a positive feedback loop with stabilization and activity of TWIST1, while PEAK1 and NES have been implicated in stabilization YAP/TAZ and SMAD TF activity. The discovery of discrete EMP regulation and development of combinatorial inhibitors may provide the opportunity for more personalized therapeutic approaches to curb metastatic disease.  EMP and cell migration (GREEN boxes) is induced through cell surface proteins (ITG, 5HTR, MUC, BLT2, SEMA3C, RTK, TGFβR) (RED) to activate signaling pathways (ORANGE boxes, blue arrows). These pathways are influenced by novel mediators (BLUE boxes) through activation (GREEN arrows) or inhibition (RED T) of known signaling members, however complete mechanisms of action remain to be elucidated. For full details, evidence of proposed mechanism and references of novel mediators, see tables below. Note signaling pathways shown have had intermediates removed for ease of visualisation.

Conclusions
Overall, investigation of the fundamental biology of EMP aims to combat local and metastatic invasion by providing a better understanding of the processes that allow cancer cells to dissociate from their epithelial adhesions to spread. EMP is a prominent driver of PDAC progression, thus highlighting the importance of our understanding of the subtleties of its regulation. The ability of EMP programs to direct cancer cells towards a drug resistant and migratory lineage capable of seeding local and distant recurrence presents a significant barrier to current treatment regimens. Therefore, the identification of new candidate molecules regulating these processes are crucial to inform targeted therapies and provide insights into the vulnerabilities of heterogeneous populations of tumour cells present in PDAC.
It is clear from this ever-growing list of EMP effectors in PDAC cells alone, that much work remains to delineate their collective interactions within and beyond our current understanding on EMP signaling pathways. While candidates have been shown to play roles in aspects of EMP signaling and associated phenotypes, significant support is required for their mechanisms of action to make concrete conclusions about their directive actions in cancer. Our understanding of receptor mediated canonical signaling through PI3K/AKT, MAPK, NFκB and other well studied cell cycle pathways has required decades to tease apart, and the subtleties of EMP programs provides a similar challenge. Open source integrative tools such as Reactome [200], WikiPathways [201], String [202], and Cytoscape [203] provide platforms for researchers to combine such analyses to build upon our current understanding and fill knowledge gaps in the field of cancer biology. In this way, progress may be made to better understand and discover properties that may be modulated in concert to control EMP in cancer.
In vitro and xenograft tumour modelling and manipulation of target molecules often demonstrates a role in cancer cell migration and tumour formation, however stronger evidence for their physiological role in regulating EMP, metastasis and therapy resistance may require GEMMs. The use of in vivo manipulation of PDAC GEMM models using targeted CRISPR approaches may be such a route towards a system that better recapitulates the spontaneity and heterogeneity of human tumours [204].

Conflicts of Interest:
The authors declare no conflict of interest.