Initially, Gata6 was presented as an important regulator of early pancreas specification and cell type differentiation, showing a partially overlapping expression with Gata4 (Ketola et al., 2004; Decker et al., 2006). Recently, Gata6 was demonstrated to be required for terminal differentiation and homeostasis of acinar cells and establishment of polarity (Martinelli et al., 2013). Evidently, Gata6 inactivation induces massive loss of acinar cells and fosters ADM in the pancreas (Martinelli et al., 2013). In addition, Gata6 ablation accelerates Kras^G12D driven tumorigenesis, demonstrating that Gata6 maintains acinar differentiation by driving expression of acinar master TFs and suppressing ectopic programs in the pancreas. Hence, in this context, Gata6 functions as a tumor suppressor (Martinelli et al., 2016). In fact, GATA6, among other genes encoding endodermal cell-fate determination TFs, is silenced via promoter hypermethylation in the squamous subtype of PDAC (Bailey et al., 2016; Seino et al., 2018). In line with that, GATA6 expression was preferentially detected in well-differentiated low-grade tumors upon transcription profiling (Collisson et al., 2011; Moffitt et al., 2015; Diaferia et al., 2016). Interestingly, silencing of Gata6 and the subsequent loss of acinar differentiation was observed during nicotine administration in mice, providing a possible link to cigarette smoking, which is a major risk factor contributing to pancreatitis and PDAC initiation (Hermann et al., 2014; Weissman et al., 2020). These findings altogether emphasize the importance of Gata6 maintenance to prevent tumor initiation and progression towards PDAC.

Mist1 is another acinar specification TF that is crucial for acinar cell maturation, function, stability, and identity and is involved in establishing granule organization and exocytosis pathways (Pin et al., 2001; Johnson et al., 2004; Direnzo et al., 2012). In the absence of Mist1 in pancreata with a Kras^G12D mutation, destabilization of the acinar phenotype leads to acceleration of PanIN formation (Shi et al., 2009). Furthermore, in cell culture models Mist1 was shown to reduce acinar cell proliferation rates by activating p21 (CIP1/WAF1) (Jia et al., 2008). Data from a 3D ADM culture model revealed that forced expression of Mist1 attenuates Kras^G12D-induced ADM and PanIN formation (Shi et al., 2013). Activation of Mist1 upon orthotopic transplantation of murine PDAC cells rescues the acinar gene expression program (Jakubison et al., 2018). Overall, the maintenance of a differentiated acinar identity by Mist1 protects acinar cells from early tumorigenesis.

Ptf1α maintains acinar cell identity and restrains Kras-mediated tumorigenesis (Rose et al., 2001; Thompson et al., 2012; Krah et al., 2015; Hoang et al., 2016). Nevertheless, Ptf1a is downregulated during inflammation-induced ADM and in acinar cells transformed by Kras^G12D and Notch co-activation (Molero et al., 2007; De La et al., 2008). Specifically, downregulation of Ptf1a is a necessary and rate-limiting step in ADM and neoplastic progression to PanINs and PDAC to overcome the Ptf1α-mediated maintenance of acinar gene signatures and quiescence in mice (Krah et al., 2015). Additionally, Ptf1a was shown to be epigenetically silenced in murine ADM and PDAC cells harboring an oncogenic Kras allele (Benitz et al., 2016). Moreover, the sustained expression of Ptf1a prevents and reverts Kras-driven pancreas tumorigenesis, rescues the acinar gene program in PDAC cells, and can inhibit tumor growth (Jakubison et al., 2018; Krah et al., 2019). These examples highlight the role of Ptf1α as a key transcriptional regulator of acinar cell identity rendering differentiated acinar cells less sensitive for cancer initiation.

In the adult pancreas, the primary function of Pdx1 is the specification and maintenance of mature β-cells (Ahlgren et al., 1998; Gannon et al., 2008; Gao et al., 2014). During tumor formation, Pdx1 is upregulated in ADM and PanINs upon overexpression of TGFα or expression of oncogenic Kras (Song et al., 1999; Hingorani et al., 2003; Park et al., 2011). In addition, gain- and loss-of-function analyses in human PDAC cell lines resulted in increased proliferation and invasion potential in the presence of PDX1. In contrast, its loss decreases cell survival and tumor growth in vivo, suggesting that PDX1 acts as an oncogene (Liu et al., 2008). In line with that, persistent Pdx1 expression in the normal pancreas promotes ADM induction via Stat3 activation. Simultaneous depletion of Stat3 blocks ADM formation (Miyatsuka et al., 2006). Despite its oncogenic function, Pdx1 often becomes downregulated by hypermethylation during progression towards the squamous and more aggressive subtype of PDAC. Conversely, PDX1 is part of a transcriptional network determining pancreatic endoderm cell fate and its presence results in a better prognosis in the pancreatic progenitor subtype (Bailey et al., 2016). In line with this conflicting data, Pdx1 was demonstrated to act as context-dependent TF during PDAC initiation and progression. Pdx1 switches from a safeguard of acinar cell identity during early tumorigenesis to an oncogene after the establishment of ADM (Roy et al., 2016). In summary, Pdx1 has two opposing functions that are activated in a context- and progression-dependent manner, emphasizing the necessity for more detailed analyses to better understand its bipartite function and prognostic value.

Expression of Sox9 in the adult pancreas is restricted to cytokeratin-positive duct cells including centroacinar cells (Miyamoto et al., 2003; Seymour et al., 2007; Furuyama et al., 2011; Shroff et al., 2014). During tumor formation, it was shown that Sox9 is induced in ADM and PanINs and is maintained in the pancreatic progenitor PDAC subtype (Morris et al., 2010; Kopp et al., 2012; Prevot et al., 2012; Meng et al., 2014; Grimont et al., 2015). Importantly, ADM and PanINs originating from the acinar compartment require ectopic induction of Sox9. Specific depletion of Sox9 from acinar cells efficiently blocks Kras-mediated PanIN formation in mouse models (Kopp et al., 2012). Furthermore, co-expression of oncogenic Kras and wild-type Sox9 promotes induction of precursor lesions from the acinar compartment (Kopp et al., 2012). Mechanistically, efficient repression of acinar genes and activation of ductal/progenitor genes in cells that undergo ADM is dependent on the combined expression of Sox9 and Hnf6, as Hnf6 overexpression also triggers ADM in mouse acinar cell lines and upon adenoviral gene delivery in vivo (Prevot et al., 2012). During pancreatitis, inflammation-induced EGFR signaling was shown to induce Nfatc1 and Nfatc4 expression, leading to ADM and PDAC progression due to upregulation of Sox9 (Chen et al., 2015; Hessmann et al., 2016). In addition, SOX9 may play a role in the IPMN-PDAC route, however, conflicting evidence have been observed. Some studies identified a gradual decrease in SOX9-positive cells in IPMNs during progression, while others report constant or even elevated SOX9 expression in both low-grade and high-grade IPMNs compared to the normal pancreas (Tanaka et al., 2013; Shroff et al., 2014; Gnerlich et al., 2019). In mice, Arid1a deficiency in the Kras^G12D pancreas results in reduced Sox9 expression and less PanINs, but increased IPMN and PDAC formation. Simultaneous SOX9 overexpression does not affect IPMN incidence, but reduces PDAC formation, demonstrating that Sox9 is a major downstream target of Arid1a and prevents tumor progression by promoting ductal differentiation (Kimura et al., 2018). Conclusively, Sox9 is a crucial mediator of ductal- or progenitor-like identity. Due to its embedding in multiple signaling pathways and feedback loops in cell-type specification, its deregulated expression is ultimately linked to early tumorigenesis.

In the adult pancreas, the expression of the Notch target Hes1 is limited to centroacinar and ductal cells associated with progenitor cell function (Miyamoto et al., 2003; Kopinke et al., 2011). Upregulation of Hes1 by active Notch signaling was observed during ADM and PanIN formation (Hingorani et al., 2003; Miyamoto et al., 2003; Jensen et al., 2005; De La et al., 2008; Plentz et al., 2009). Moreover, Notch-induced Hes1 was suggested to control the expansion of an undifferentiated precursor cell population, thereby promoting Kras-mediated tumor initiation and progression (Miyamoto et al., 2003; Jensen et al., 2005; De La et al., 2008). In fact, acinar-specific expression of mutant Kras induces Hes1 expression along with ADM and PanIN formation. In this context, Notch activation was shown to sensitize acinar cells to mutant Kras-induced ADM/PanIN initiation and progression (De La et al., 2008; Nishikawa et al., 2019). Interestingly, Elastase-mediated Hes1 depletion blocks the progression from ADM to PanINs, combined with a re-differentiation to acinar cells (Nishikawa et al., 2019). However, the role of Hes1 is likely more complex as in another mouse model using Ptf1α-mediated Hes1 ablation and oncogenic Kras induction, loss of Hes1 displayed increased ADM formation and accelerated PDAC tumorigenesis. Reduced numbers of high-grade PanINs were detected in this model, hinting towards tumor formation from a direct ADM-to-PDAC route that skips precancerous PanIN lesions (Hidalgo-Sastre et al., 2016). These findings convey that context-specificity and maintained activity of Notch and Hes1 during homeostasis are essential regulators of tumor initiation.

In summary, PDAC formation depends on early pre-neoplastic events like ADM, which relies on the downregulation of TFs that control acinar cell identity, including Gata6, Mist1, and Ptf1α, and a gain of TFs that promote duct or MPC-like specification, including Pdx1, Sox9, and Hes1 (Figure 2). However, some controversies and the incomplete understanding of the cellular origin of PDAC warrant further analyses to decipher the TF networks that are active during early tumorigenesis.

The oncogenic KRAS mutation is found in over 90% of PDAC patients and results in the persistent stimulation of downstream signaling leading to sustained cell proliferation, transformation, migration, and survival (Biankin et al., 2012; Buscail et al., 2020). Although Ras signaling is enriched in the pancreatic progenitor subtype, the activation of certain KRAS downstream mediators is able to foster the transition towards the squamous subtype. Elevation of Etv1, a downstream target of Kras, promotes stromal expansion and metastases through Sparc and Has2 activation in tumors generated by orthotopic transplantation of KPC cells (Kar and Gutierrez-Hartmann, 2013; Heeg et al., 2016). Etv1 overexpression induces all core EMT-TFs (Box 1) and molecular markers associated with the mesenchymal phenotype (e.g. Vim, Mmp3, and Mmp9), whereas knockdown of Etv1 reduces Zeb1 levels. Using human PaCa cell lines in vitro it was shown that elevation of HAS2 is able to fuel a self-enforcing feedback loop of CD44 and ZEB1 that involves differential splicing of CD44 by ESRP1, further promoting EMT (Preca et al., 2015, 2017). In addition, EMT and enhanced invasion can be activated by increased MAZ expression in human PaCa cell lines. MAZ acts downstream of KRAS and facilitates CRAF-MAPK signaling involving PAK and suppression of AKT/PKB (Maity et al., 2018). Moreover, the upregulation of MAPK or inactivation of TP53 leads to the overexpression of KLF7, promoting tumor growth and metastasis in mice (Gupta et al., 2020). Expression of KLF7 activates IFN-stimulated genes and stabilizes Golgi integrity and thus protein glycosylation to enhance the secretion of cancer-promoting growth factors. In cooperation with Myc Yap1 maintains the expression of metabolic genes required for proliferation and survival (Murakami et al., 2019). Ablation of Yap1 in a PDAC mouse model leads to the downregulation of Myc, inducing growth arrest and apoptosis (Murakami et al., 2019). Interestingly, a subset of tumor cells was able to restore Myc levels allowing cell survival through the induction of genes encoding EMT-TFs Snail, Zeb2, Twist2, and the stemness factor Sox2, thus compensating for Yap1 loss.

Multiple studies show that the pancreatic progenitor subtype is KRAS-dependent, whereas the squamous subtype is less dependent on KRAS (Singh et al., 2009; Collisson et al., 2011; Ischenko et al., 2021). Moreover, CRISPR/Cas9-mediated Kras knockout in tumor cells derived from the KPC mouse model showed pathway enrichment for EMT and TGFβ signaling, hinting that ablation of Kras drives a subtype switch towards the squamous subtype (Ischenko et al., 2021). Secondary ablation of Kras ^G12D in established tumors of a GEMM with doxycycline-inducible Kras ^G12D and conditional Tp53 inactivation leads to complete regression (Kapoor et al., 2014). Although these initial results are promising, the majority of mice show relapse and exhibit poorly-differentiated pancreatic tumors. The survival of tumor cells in this model in the absence of Kras is mediated by the upregulation of the transcriptional coactivator Yap1, a downstream mediator of the Hippo signaling cascade, and Tead2, forming Yap1/Tead2 complexes coordinating downstream gene expression. Other compensatory mechanisms have been identified, including the induction of the transcriptional repressor Gli2, a downstream mediator of the Shh pathway, upon in vitro Kras^G12D ablation (Adams et al., 2019; Ischenko et al., 2021). Gli2 induction rescued viability and induced upregulation of squamous-specific gene signatures (e.g. Vim and Zeb1). Moreover, GLI2 induction in human PaCa cell lines promotes a gene signature switch from the pancreatic progenitor towards the squamous subtype, accompanied by a decrease in epithelial identity markers (E-cadherin, ESRP1, GATA6, and SHH) and enrichment in expression of EMT/stemness markers (ZEB1, VIM, CK14, SOX2, and CD44). Primary tumor growth and metastatic outgrowth can be suppressed by ablation of SPP1, a downstream target of GLI2, emphasizing its role in promoting tumor aggressiveness. These findings demonstrate that aberrant activation of several TFs exacerbate PDAC progression (Table 1) with various degrees of KRAS-dependency. Interestingly, these deregulations frequently mediate the indirect upregulation of EMT-TFs (ZEB1/2, Snail, Slug, and Twist) and stemness factors (SOX2 and CD44) (Figure 3, Box 1). These findings corroborate that the induction of the reversible EMT program promotes an aggressive PDAC phenotype by enabling cellular plasticity, metastasis formation, chemoresistance, and the acquisition of CSC properties in PDAC (Satoh et al., 2015; Wang et al., 2015; Zheng et al., 2015; Krebs et al., 2017; Aiello et al., 2018; Recouvreux et al., 2020).

Apart from the activation of signaling cascades downstream of KRAS driving the subtype transition, the squamous subtype shows enrichment for TGFβ signaling (Bailey et al., 2016; Lomberk et al., 2018). In pre-malignant cells TGFβ acts as a tumor suppressor by inducing cell-cycle arrest, differentiation, and apoptosis (Massagué, 2008). Disruption of TGFβ signaling in PDAC prevents the tumor-suppressive effects, while activation of TGFβ signaling in a progressed state is a potent inducer of EMT (Dardare et al., 2020). A central player in canonical TGFβ signaling is SMAD4, whose inactivation or loss occurs as a late event during PDAC progression (Wilentz et al., 2000). The role of Smad4 in progression and metastasis remains controversial, as Smad4-deficiency attenuates EMT, leads to upregulated E-cadherin protein levels, and promotes a well-differentiated PDAC phenotype in the Kras ^G12D ;Ink4a/Arf ^∆ mouse model (Bardeesy et al., 2006; Ahmed et al., 2017; Shichi et al., 2019). Moreover, simultaneous knockout of Smad4 and Kras reduces EMT-related genes and promotes a Ras signaling signature (Ischenko et al., 2021). In contrast, other studies show that the loss of SMAD4 is associated with shorter overall survival and a squamous phenotype (Blackford et al., 2009; Yamada et al., 2015). Interestingly, the TF Runx3 promotes metastatic colonization but is dependent on the Smad4 state (Whittle et al., 2015). Heterozygous Smad4 inactivation in the KPC mouse model promotes progression and growth of the primary tumor, while loss of the remaining wild-type allele leads to a highly metastatic disease. Furthermore, the expression of TGFβ target genes can be repressed by elevated TGIF1, potentially decreasing PDAC progression as demonstrated by HEK-293 cell transfection experiments and in PDAC mice (Seo et al., 2004; Weng et al., 2019). In summary, the intricate balance among several TFs involved in TGFβ signaling and their mutational status determine the impact on progression.

On the other hand, several endodermal lineage specifiers promote the pancreatic progenitor molecular subtype in PDAC, such as GATA6, FOXA1/A2, and HNF4α, identified in silico (Bailey et al., 2016; Roe et al., 2017; Brunton et al., 2020). Although PDX1 expression is often increased during ADM and tumor onset, its downregulation or loss is mainly observed in poorly-differentiated tumors correlated with EMT and metastasis (Roy et al., 2016). Hence, high expression of PDX1 is observed in the pancreatic progenitor subtype and well-differentiated tumors (Ischenko et al., 2014; Roy et al., 2016; Lomberk et al., 2018). Dysregulation of these TFs by inactivating mutations or repression can affect the EMT-equilibrium towards a more squamous identity. As an example, loss of Gata6 in a mouse PDAC model decreases the Cdh1 inducers Foxa1/a2 and de-represses EGFR signaling in favor of dedifferentiation (Martinelli et al., 2016, 2017). In contrast, FOXA1 elevation was identified in patients’ primary lesions and well-differentiated low-grade tumors, in part by activating HNF4A and other endodermal lineage specifiers (Duncan et al., 1998; Diaferia et al., 2016; Roe et al., 2017). Together with Gata6, Foxa1/a2 block EMT and promote epithelial differentiation in GEMM PDAC models (Song et al., 2010; Martinelli et al., 2017). Hence, direct transcriptional repression of FOXA1 by BACH1 is required for metastatic colonization of AsPC-1 PaCa cells in an orthotopic implantation model (Sato et al., 2020). Interestingly, loss of FOXA1/A2 is frequently detected in the squamous subtype and is sufficient to induce EMT in human PaCa cell lines (Song et al., 2010; Roe et al., 2017). Apart from repressing FOXA1, BACH1 activates SNAI2, which further promotes EMT, assessed by gene inactivation in human cell lines (Sato et al., 2020). Transcriptomic analysis on primary tumors and patient-derived cell lines revealed that during tumorigenesis HNF4α directly activates HNF1A, and loss of the former enables a transition towards a more squamous phenotype (Brunton et al., 2020; Camolotto et al., 2021). Moreover, HNF4α directly represses the mesodermal and neural differentiation TFs SIX1/4, whose elevated expression was linked to the squamous subtype (Camolotto et al., 2021). Downregulation of HNF1A is observed in the tumor vs. normal pancreas, suggesting that decreased HNF1α levels are important for PDAC tumor progression (Hoskins et al., 2014; Luo et al., 2015). Overexpression of HNF1A decreases cell-doubling times, while its knockdown significantly increases proliferation in vitro. HNF1α downregulates apoptosis inhibitors and modulates the expression of cell cycle genes. However, whether HNF1α acts through the AKT/mTOR pathway requires additional investigation, since silencing of HNF1A activates AKT/mTOR signaling, but may also result in reduced expression of PI3K, AKT and mTOR (Hoskins et al., 2014; Luo et al., 2015; Subramani et al., 2020). Other studies indicate that HNF1A is an oncogene necessary for the regulation of cancer stem cell (CSC) populations in PDAC, promotes anchorage-independent growth, proliferation, as well as invasive and migratory capacities (Abel et al., 2018; Subramani et al., 2020). These contradictory findings could be explained by the hypothesis that cellular plasticity and thus the ability to induce partial-EMT is indeed necessary to acquire stemness, whereas reversal to an epithelial phenotype is crucial for metastatic outgrowth at secondary sites. Conclusively, the expression of several TFs involved in specifying pancreatic cell-fate maintain the pancreatic progenitor subtype by (in)directly promoting epithelial-identity markers and inhibiting EMT/dedifferentiation (Table 1). Their downregulation abrogates these effects, allowing a switch towards the more aggressive squamous subtype (Figure 3).

Comparisons between primary PDAC tumors and matched metastasis revealed no specific metastasis-inducing genetic mutations, hinting towards gene regulatory mechanisms affecting late PDAC progression and metastasis (Campbell et al., 2010; Yachida et al., 2010; Makohon-Moore et al., 2017). The involvement of EMT-TFs in PDAC invasion and metastasis was initially questioned due to experimental challenges to observe EMT and the metastatic cascade in vivo. In particular, depletion of either Twist or Snail in the KPC mouse model of PDAC is not affecting metastasis formation, indicating that they are dispensable for this process (Zheng et al., 2015). However, depletion of Zeb1 in the same mouse model suppresses metastasis formation as well as experimental lung colonization capacity, stemness, and cell and metabolic plasticity (Krebs et al., 2017). Moreover, glutamine depletion promotes metastasis of orthotopically and intravenously injected KPC cells through induction of EMT by upregulation of Snai2 via ERK signaling and ATF4 activation (Recouvreux et al., 2020). Collectively, these findings and research on core EMT-TFs in other cancers show that their individual contribution to invasion and metastasis is highly dependent on the cellular context (Elloul et al., 2005; Imamichi et al., 2007; Caramel et al., 2013; Denecker et al., 2014; Fischer et al., 2015; Zheng et al., 2015; Krebs et al., 2017; Stemmler et al., 2019; Recouvreux et al., 2020).

Liver metastases in KPC mouse models show elevated expression of Foxa1 and Prrx1a, whereas the expression of these factors decreases when the primary tumor acquires more squamous-associated features during progression (Takano et al., 2016; Roe et al., 2017). The decrease during progression and elevated expression in liver metastases suggests that re-expression of TFs associated with the pancreatic progenitor subtype is essential for successful liver colonization. Moreover, it was shown that Prrx1a enhances self-renewal, decreases invasiveness, and promotes metastatic outgrowth (Takano et al., 2016). Isoform b, on the other hand, fosters invasion, EMT, and dedifferentiation by promoting Hgf expression, suggesting that both isoforms distinctively regulate EMT and MET to form overt metastases. In addition, these two isoforms can form homo- and heterodimers, affecting transcriptional activity in human PaCa cell lines (Marchand et al., 2019). Simultaneous inactivation of Snai1 and Twist induces a shift of the EMT-equilibrium to a more epithelial-like state in the primary tumor of KPC mice while enhancing liver metastases (Carstens et al., 2021). Sca1 and Pdx1 levels are also regulating metastatic capacities: Sca1^- cell lines derived from the KPC mouse model express elevated levels of Pdx1 and successfully metastasize in lungs and lymph nodes upon tail vein injections, whereas Sca1⁺ cells with lower levels of Pdx1 fail to metastasize (Ischenko et al., 2014). These findings support the idea that the reinforcement of epithelial features and thus cellular plasticity are required for metastatic competence (Figure 3).