Oncogene-Induced Senescence Limits the Progression of Pancreatic Neoplasia through Production of Activin A

Pancreatic ductal adenocarcinomas (PDAC) are associated with some of the worst survival outcomes among all cancers, with a median survival rate of approximately 6 months and a 5-year survival rate <8% (1). Although, several therapeutic options for PDAC have emerged over the last decade, their efficiencies remain relatively limited, and approaches designed to circumvent the commonly late diagnoses of those tumors remain lacking (2). Studies exploring the mechanisms and mutations associated with PDAC onset have confirmed the major role played by Kras-activating mutations during the formation of acinar-to-ductal metaplasia (ADM) and pancreatic intraepithelial neoplasia (PanIN), which precede PDAC development (3, 4). Although, the progression of these lesions to PDAC requires mutations in additional factors, such as the cyclin-dependent kinase inhibitors p16 and p53, the latency prior to the evolution of PanIN in aggressive and invasive tumors supports the existence of some mechanism that limits the transformation of neoplastic cells. Interestingly, although Kras-activating mutations are known to drive PDAC onset (5), Kras can also promote oncogenic-induced senescence (OIS), a safeguard program that limits cancer proliferation and initiation (6, 7). The regulating role played by OIS during pancreatic tumorigenesis has been demonstrated previously (8–12), and animal models expressing the gain-of-function Kras^G12D mutation support the existence of a senescent tumor-suppressive program that limits the expansion of Kras-mutant cells (13, 14). In addition to the OIS-induced arrest of growth in cancer-initiated cells, senescent cells are capable of producing and secreting a complex series of cytokines and molecules that are associated with the senescence-associated secretory phenotype (SASP; ref. 15). A large number of SASP-associated proteins have been identified, among which, several are involved in proinflammatory responses (16). Although, the exact roles of SASP-associated cytokines and proinflammatory components during PDAC initiation remain unclear (17, 18), recent evidence supports the importance of SASP-associated molecules, such as C-X-C motif chemokine ligand 1 (CXCL1) during pancreatic tumorigenesis (18).

Among the SASP candidates that are potentially involved in paracrine senescence, TGFβ superfamily ligands, including TGFβ1, inhibin subunit beta A (INHBA), bone morphogenetic protein 2 (BMP2), and growth differentiation factor 15 (GDF15), have been identified through secretome analyses of cells undergoing OIS (19). Interestingly, INHBA, which encodes activin A, is targeted by Kras during oncogenic activation in pancreatic duct cells (20), suggesting that activin A may be a candidate SASP-associated factor that is induced by Kras-OIS in pancreatic neoplastic lesions. Interestingly, although, the ablation of the activin cognate receptor, activin-like kinase 4 (ALK4), favors the development of intraductal papillary mucinous neoplasms (IPMN) and accelerates the formation of PDAC in mice harboring the Kras^G12D mutation (21, 22), little is known regarding the contributions of ALK4-mediated activin signaling and its downstream signaling factors, P-Smad2 and P-Smad3, to the earliest stages of Kras-driven pancreatic cell transformations or the involvement of this pathway to Kras-OIS in low-grade pancreatic neoplastic lesions. Here, we aimed to understand the mechanisms associated with ALK4-mediated signaling during PDAC initiation. We report the identification of activin A as the major ALK4 ligand expressed in pancreatic ductal neoplastic lesions. Our work demonstrates that activin A, which is absent in normal acinar pancreatic cells, is strongly induced in ADM lesions, promoted by Kras^G12D and pancreatitis, in the mouse and human pancreas. More importantly, we report that activin A–induced expression during ADM contributes to the senescence program induced in precursor lesions. By blocking activin A signaling, through the use of a soluble form of activin receptor IIB (sActRIIB-Fc) molecules or the genetic targeting of ALK4, we further demonstrate that activin A acts as a beneficial SASP factor that limits the proliferation and expansion of pancreatic neoplastic lesions.

All animal maintenance and experiments were performed in accordance with Animal Research: Reporting of In Vivo Experiments guidelines and French laws and were approved by the local Animal Ethics Evaluation Committee. The generation of Acvr1b-mutant mice has been described previously (23). Acvr1b^flox/flox;LSL-Kras^G12D/+;Ptf1a-Cre (termed 4KC) mice were created by breeding Acvr1b^flox/flox mice with the previously established LSL-Kras^G12D/+;Ptf1a-Cre (termed KC) model (24). LSL-Kras^G12D/+;Ink4a/Arf^flox/flox;Pdx1-Cre mice (termed KIC) have been described previously (25). Acute pancreatitis experiments were performed using a standardized procedure (26). Briefly, pancreatitis was induced in 1.5-month-old wild-type (WT) C57BL/6 mice by hourly intraperitoneal injection of caerulein (50 μg/kg body weight; Sigma-Aldrich) dissolved in PBS. Seven injections were done for 2 consecutive days, followed by pancreas collection 48 hours later. Control littermates received injections of PBS only.

Panc-1 (CRL-1469), Mia Paca-2 (CRL-1420), and Capan-1 (HTB-79) PDAC cell lines were purchased from the ATCC and grown in the recommended culture media. All cell lines were Mycoplasma tested and experiments performed between passages 12–15 (Panc-1), 6–9 (Capan-1), and 8–12 (Mia Paca-2) following their purchase. Activin A (25 μg/mL, PeproTech), TGFβ (5 μg/mL, PeproTech), SB431542 (10 μmol/L Sigma-Aldrich, 616464), and sActRIIB-Fc (0.5 μg/mL) were added to cells for 24 hours, in 2% FCS culture media.

Mouse pancreases were harvested and fixed overnight, in 4% buffered formalin, prior to paraffin embedding and 3 μm sectioning. All immunohistologic staining procedures, unless otherwise mentioned, were performed following heat-induced epitope retrieval (Antigen-Unmasking Solution, Vector Laboratories), and primary antibodies were incubated overnight at 4°C. IHC stains were revealed using 3,3′-diaminobenzidine (DAB Kit, Vector Laboratories) and sections were counterstained with hematoxylin. Immunofluorescence (IF) staining was performed using a standard protocol, and sections were counterstained with 4′,6-diamidino-2-phenylindole (DAPI, Vector Laboratories). All images were acquired on an Eclipse-NiE Nikon Microscope and analyzed using NIS-Elements Software. A complete list of the primary antibodies used is provided in Supplementary Table S1. BIC14011a and PA485 human tissue microarrays (TMA) were obtained from a commercial source (US Biomax, Inc.). BIC14011a is a pancreas array that contains 22 pancreatitis cases, 18 PanIN cases, and right pancreatic adenocarcinoma cases. PA485 is a pancreatitis and matching pancreatic adenocarcinoma array, containing 43 cases of pancreatitis and five matched pancreatic adenocarcinomas. Senescence-associated beta-galactosidase (SA-β-gal) whole-mount staining was performed overnight, according to the procedures described in the Senescence Cells Histochemical Staining Kit (Sigma-Aldrich, #CS0030). All incubations and washes were performed using fresh and filtered solutions. Stained tissues were subsequently washed three times in PBS and post-fixed with 4% paraformaldehyde, prior to paraffin embedding for sectioning. Combined SA-β-gal/IF and SA-β-gal/IHC staining were performed on 7 μm sections, obtained from whole-mount SA-β-gal–stained tissues. IF and bright field images were acquired on a Zeiss Axio Imager Microscope and merged using ImageJ. Activin A positivity in pancreatitis and PanIN lesions was visually quantified by the identification of at least one positive cell per lesion, within each individual TMA spot. The quantification of ADM/PanIN areas was performed on hematoxylin and eosin (H&E)-stained pancreatic scanned sections (Panoramic Scan 3D HISTECH Ltd.). The surface areas of individual regions containing ADM/PanIN lesions were determined using ImageJ and normalized against the surface areas occupied by normal pancreatic tissues.

Activin A signaling was inhibited through the use of recombinant sActRIIB-Fc protein, which was produced and used as reported previously (27, 28). For in vivo treatments, randomized 1.5-month-old KC mice were injected intraperitoneally with either 5 mg/kg body weight sActRIIB-Fc suspended in PBS or PBS alone, twice a week. Short-term and long-term treatments consisted of five and 12 injections, respectively, performed over 3 and 6 consecutive weeks, respectively. All pancreases were collected 24 hours following the last injection. For in vitro treatments, sActRIIB-Fc was used at a final concentration of 0.5 μg/mL.

Mouse acinar cells were isolated, as reported previously (29). Resected pancreases were transferred to ice-cold Hank's Balanced Salt Solution (HBSS 1 ×, Life Technologies) and subsequently minced/dissociated with 1 mg/mL Collagenase P (Sigma-Aldrich #11213865001) in 1 × HBSS, for 30 minutes at 37°C. Dissociated cells were washed with 1 × HBSS, containing 10% FBS and 10 mmol/L 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES), prior to filtration through a 100-μm Sterile Nylon Mesh (VWR). Following centrifugation, cells were suspended in three-dimensional (3D) culture medium (RPMI1640 Medium, Gibco, #72400-021), supplemented with 25 mmol/L HEPES, 1% FBS, 1% penicillin/streptomycin, 1 μg/mL dexamethasone, and 0.1 mg/mL soybean trypsin inhibitor (Sigma-Aldrich, #T6522; ref. 29). 3D cultures were performed in 24-well tissue culture plates, coated with 3 mg/mL Type I collagen from rat tail (Life Technologies, #A1048301). Acinar-isolated cells were embedded in a 1:1 mix of rat-tail collagen and 3D culture medium, prior to plating on the prepared collagen layers. The cell–collagen mixture was allowed to solidify for 1 hour at 37°C, before covering the collagen discs with 1 mL 3D culture media. Media were changed on days 1 and 3, and the discs and cells grown on collagen were harvested on day 6, for gene expression analysis, SA-β-gal whole-mount staining, or paraffin embedding, prior to immunohistologic staining.

RNA was extracted using the RNeasy Kit, according to the manufacturer's instruction (Qiagen). Real-time PCR analyses were performed on a Step-One RT-System (Applied Biosystem). The primer sequences used are detailed in Supplementary Table S2.

Gene expression analysis was performed using the Affymetrix gene expression array dataset, deposited in the Gene Expression Omnibus repository (accession no. GSE61412; ref. 30). Briefly, expression data from 9-week-old KIC (n = 3) or control [LSL-Kras^G12D/+;Ink4a/Arf^flox/flox (termed KI), n = 3] mice were normalized using the robust multichip average (RMA) method in Bioconductor package (R 3.4.4 software; refs. 30, 31). Heatmap illustrations showing the expression of genes coding ligands that trigger the ALK4 function were generated with Morpheus software (https://software.broadinstitute.org/morpheus). Each column represents the RMA data for a given gene, obtained from three independent control (C9) or PDAC (T9) pancreas samples, derived from 9-week-old KI and KIC mice, respectively. The significant differential expression of genes between KIC and control KI pancreases was determined from normalized RMA dataset, using a Student t test analysis. Genes with a P < 0.05 were the sole candidates considered as being significantly dysregulated in KIC samples and are represented with fold changes and P values transformed into log₂ and −log₁₀ scales, respectively.

All analyses were performed using Prism 8 Software (GraphPad). Statistical analyses were performed as described in the figure legends. Unpaired Student t tests were used for pairwise comparisons. A one-way ANOVA, with Tukey post hoc test, was used for multiple comparisons. *, P < 0.05; **, P < 0.01; and ***, P < 0.001. The significance of activin A expression in pancreatitis and PanIN lesions, assessed using TMA analysis, was determined using Fisher exact test.

To understand the mechanisms associated with the loss of ALK4 signaling during PDAC initiation, we first attempted to identify potential candidate ligands that trigger ALK4 activation and the downstream phosphorylation of its effectors, Smad2 and Smad3, in pancreatic lesions mediated by a Kras^G12D mutation. Taking advantage of a previous transcriptomic analysis, performed on the KIC PDAC mouse model (25, 30), we used data mining to screen the expression levels of eight putative ALK4 ligands. Ligand expression levels were analyzed in 9-week-old KIC mice after the development of PanIN and pancreatic adenocarcinomas (25). Among the tested genes (Fig. 1A), six were found to be either significantly up- (InhbA, InhbB, and Gdf11) or downregulated (Nodal, Gdf1, and Mstn) in KIC mouse pancreases (P < 0.05) compared with age-matched KI control pancreases (Fig. 1B). More interestingly, a fold change analysis indicated that InhbA and InhbB were the only genes that displayed a greater than 2-fold change, InhbA, which encodes activin A, showing the most significant change (Fig. 1B).

These results suggested that activin A may represent the major ALK4 signaling ligand during pancreatic tumorigenesis. We next used an immunohistologic approach to detect activin A expression within different grades of pancreatic lesions. Although, activin A was not expressed in acinar-, islet-, and Ck19-positive duct cells in the normal pancreas, activin A expression was significantly increased in all KIC-analyzed lesions, ranging from ADMs to adenocarcinomas (Fig. 1C). A reduced number of activin A–expressing cells was detected in the PanIN1/2 lining, and the diffuse expression of activin A was detected in PanIN3 and adenocarcinoma lesions (Fig. 1D), which led us to speculate that the ectopic induction of activin A expression in ADM may represent an acute response to an oncogenic insult, driven by the combined ablation of the Ink4a/Arf allele and the activation of Kras.

We next questioned whether activin A–induced expression occurred when ADM was induced by other pathologic contexts, such as the sole activation of Kras^G12D in acinar cells or the inflammatory cues that drive pancreatitis. Using KC and caerulein injections in WT animals, as respective models of low-grade neoplastic lesions (32) and pancreatitis (33), we found that activin A was strongly induced in acinar-to-duct converting cells, independent of the ADM-triggering signal (Fig. 2A and B). Interestingly, as demonstrated in KIC pancreases, we confirmed that following the induction of activin A expression in ADM, activin A was only detected in a subset of Ck19-negative cells lining the PanIN lesions in KC mice (see arrow in Fig. 2A). Finally, using human pancreatic TMA analysis, we further validated the lack of activin A expression in normal acinar cells and the existence of robust activin A expression in 60% of ADM lesions observed in patients (n = 64), whereas activin A immunoreactivity was reduced and limited to a subset of duct cells in low-grade PanIN (Fig. 2C).

Because activin A appears to be ectopically expressed in acinar cells during ADM, we next addressed the consequences of blocking activin A signaling. We first used sActRIIB-Fc, a soluble activin receptor, that was previously demonstrated to bind activin A with high affinity (K_D of 35.7 pmol/L) and to inhibit activin A signaling through ALK4 (27, 34). Short-term treatments, involving five intraperitoneal sActRIIB‐Fc injections, performed over 3 weeks, significantly increased the weights of 1.5-month-old, sActRIIB-Fc–treated KC mouse pancreases compared with those from age-matched WT and nontreated KC controls (Fig. 3A and B). Histologic analysis of KC-treated animals confirmed the inhibition of Smad2 phosphorylation (Fig. 3C) and revealed that sActRIIB-Fc administration accelerated the development of ADM/PanIN lesions, as assessed by surface area measurements (Fig. 3D). In contrast, no alterations were observed in the pancreases of WT sActRIIB-Fc–treated mice (Supplementary Fig. S1A–S1C). These observations were subsequently confirmed through the use of an ALK4 conditional–knockout (KO) mouse model, which was reported previously (23), to generate Acvr1b^flox/flox;LSL-Kras^G12D/+;Ptf1a-Cre compound–mutant mice (termed 4KC). Although, the analysis of 3-week-old pancreases from KC and 4KC animals did not reveal obvious ADM or PanIN lesions, we found that the pancreases from 1.5-month-old 4KC mice were enlarged, presenting an increased number of ADM lesions and reduced P-Smad2 expression (Fig. 3E; Supplementary Figs. S2A and S2B, and S3). The accelerated ADM formation observed in 4KC pancreases, compared with KC pancreases, was further confirmed by the identification of an enlarged area of Ck19 immunoreactivity, associated with more pronounced collagen deposition and the increased recruitment of Desmin⁺ pancreatic stellate cells (Supplementary Fig. S2C).

Gene expression screening performed on cells undergoing OIS has indicated that InhbA may represent a SASP factor candidate (19). Because InhbA has been reported to be a target of Kras oncogenic activation in pancreatic duct cells (20), we hypothesize that the induced expression of activin A during ADM may underline a Kras-OIS process occurring in those lesions. When driven by a Kras^G12D mutation or TGFα stimulation, primary cultures of pancreatic acinar cells grown in a 3D collagen matrix form cystic duct structures that resemble and recapitulate ADM (29). Using such a model, we found that InhbA expression was induced during the formation of duct structures generated from cultured KC acinar cells (Fig. 4A). Although, the loss of acinar markers and increased duct marker expression confirmed the formation of duct-like structures (Supplementary Fig. S4A–S4C), the use of a SA-β-gal assay indicated the existence of senescent cells lining the formed structures (Fig. 4B). Real-time PCR analysis further confirmed the induced expression of the putative senescence-associated cyclin-dependent kinase inhibitors Cdkn1a (p21) and Cdkn2a (p16) and several SASP-associated genes, such as Il6, Il1a, Ccl20, and Vegfc, in day 6 duct-like structures compared with day 1 acinar cells (Fig. 4C). Having demonstrated that acinar-derived duct-like structures were subjected to senescence and expressed InhbA, we next addressed whether activin A was produced by senescent cells in ADM lesions. Using pancreases from KC mice, we first confirmed the induction of a senescence phenotype in ADM and PanIN lesions (Fig. 4D), as described previously (12). Interestingly, we noted that the patterns of SA-β-gal⁺ cells detected in both ADM and PanIN lesions were reminiscent of the pattern of activin A⁺ cells observed in KIC, KC, and caerulein-induced lesions (Fig. 4D). The colocalization of activin A expression in SA-β-gal⁺ cells and the expression of the senescence markers p21 and γH2A histone family member X (H2AX) was confirmed in ADM lesions (Fig. 4E and F). The colocalization of ALK4 and activin A in KC lesions further suggests that these mechanisms are cell autonomous (Supplementary Fig. S5). Taken together, these results support the role of activin A as a SASP component in ADM lesions.

After demonstrating that activin A signaling increased ADM/PanIN formation, we next questioned whether this finding reflects an escape to the senescence program, promoted by Kras^G12D activation. The treatment of KC-derived acinar cells with sActRIIB-Fc confirmed that duct structures obtained from treated cells were less prone to senescence, demonstrated by reduced SA-β-gal staining and the decreased expression levels of the senescence-associated cyclin-dependent kinase inhibitors Cdkn1a and Cdkn2a (Fig. 5A and B; Supplementary Fig. S6A). Although, the expression levels of Ccl20 and Vegfc were reduced following sActRIIB-Fc treatment, the expression levels of Il6, Il1a, Il1b, and Ccl2 were not significantly impacted (Fig. 5B), suggesting that the inhibition of activin A signaling has moderate impacts on SASP regulation. Histologic analysis confirmed that senescence was reduced in the formed cysts during the differentiation of ADM structures, yet the expression levels of the duct markers Ck19 and Sox19 were not impacted following the inhibition of activin A signaling (Fig. 5C; Supplementary Fig. S6B). Finally, we found that the decreased senescence observed in SA-β-gal assays was further supported by significant reductions in p21 and γH2AX expression levels in Ck19⁺ cells within the duct structures formed from KC-sActRIIB-Fc–treated and 4KC acinar cells (Fig. 5D and E; Supplementary Fig. S6C). We confirmed a direct role of activin A on Cdkn1a and p21 expression, mediated by P-Smad2 in the Panc-1 adenocarcinoma cell line, which was responsive to activin A stimulation, unlike in Capan-2 cells that are not responsive to activin A stimulation (Supplementary Fig. S7A–S7C). Mia-Paca-2 cells were responsive to activin A stimulation but did not show p21 upregulation. On the basis of these observations, we speculated that blocking activin A signaling may not alter the formation of ADM but, instead, impact senescence mechanisms among preneoplastic ADM cells. We explored this hypothesis in 4KC and KC animals exposed to short-term sActRIIB-Fc treatments. As shown in Fig. 5F and Supplementary Fig. S8A, we found that blocking activin A signaling resulted in reduced ADM and PanIN SA-β-gal activity, for both models. Reduced senescence was further associated with significantly decreased expression levels of p21 and γH2AX (Fig. 5G and H; Supplementary Fig. S8B and S8C), indicating that in vivo activin A signaling likely contributes to the senescence cues observed in Kras-mediated ADM and PanIN.

Senescence bypass is a key feature of tumor progression, and OIS results in the transcriptional repression of proliferation through the activation of the p53–p21 and/or p16–Rb suppressor pathways (7). Because blocking activin A signaling resulted in reduced senescence and the decreased expression of the cell-cycle inhibitor p21, we next determined whether these observations are associated with the promotion of proliferation. The quantification of paraffin-embedded structures confirmed that the chemical and genetic inhibition of activin A signaling resulted in the formation of large diameter cysts (Fig. 6A; Supplementary Fig. S9A). Using Ki67 as a proliferative marker, large cysts were observed with a larger percentage of Ki67-positive cells per duct following activin A inhibition compared with controls (Fig. 6B; Supplementary Fig. S9B). Consistent with these observations, the analysis of 4KC and sActRIIB-Fc–treated KC pancreases further confirmed that ADM lesions demonstrated significantly increased proliferation compared with age-matched controls (Fig. 6C; Supplementary Fig. S9C). Interestingly, we found that proliferation was sustained in PanIN lesions and in Ck19⁺ cells expressing ALK4, following sActRIIB-Fc treatment (Fig. 6C; Supplementary Figs. S10A–S10D and S11). These results indicated that the inhibition of activin A signaling may favor the formation of lesions with increased proliferative potential in a cell autonomous manner. The administration of sActRIIB-Fc to 4KC animals did not impact the pancreas volume, activin A expression levels, or the numbers of Ck19⁺ pancreatic cells or Ki67⁺ proliferative cells in ADM and PanIN lesions (Supplementary Fig. S10).

We next examined the consequences of sustained activin A inhibition on Kras^G12D-expressing acinar cell transformation. The analysis of 14-week- and 30-week-old 4KC mice confirmed the accelerated progression of ADM to PanIN-graded lesions within the entire pancreas (Fig. 7A). 4KC-mutant animals developed Ck19⁺ cystic lesions lacking ALK4, which were reminiscent of the IPMNs reported in the Acvr1b^flox/flox;LSL-Kras^G12D/+;Pdx1-Cre model developed by Qiu and colleagues (Fig. 7A and B; Supplementary Fig. S12A–S12C; ref. 22). Subsequent analysis revealed that 4KC pancreases presented significantly increased numbers of proliferating Ki67⁺/Ck19⁺ adjacent cells, located within the lining of the cystic lesions (Fig. 7C). To further validate these observations, we explored the consequences of long-term sActRIIB-Fc treatment in KC mice, by administering sActRIIB-Fc via intraperitoneal injections to 1.5-month-old KC mice twice a week for 6 weeks, prior to the histologic analysis of pancreatic tissues. Consistent with our finding in aged 4KC mice, the analysis of long-term sActRIIB-Fc treatment in KC mice confirmed that the inhibition of activin A signaling and the subsequent senescence escape of ADM cells resulted in the accelerated formation of both PanIN and Ck19⁺ cystic lesions in the pancreases of long-term sActRIIB-Fc–treated mice (Fig. 7D and E). As observed within the 4KC cystic lesions, a significant increase in the number of proliferating Ck19⁺ cells was observed, which were frequently adjacent to each other and located in the lining of the cystic lesions (Fig. 7F).

The contribution of ALK4-mediated signaling to the onset of pancreatic cancer has been demonstrated by the identification of ACVR1B homozygous mutations associated with PDAC and the use of ALK4-KO mouse models (22, 35, 36). Although, the role played by ALK4-mediated signaling has previously been demonstrated in the promotion of aggressive pancreas cancers and the formation of IPMNs (22, 35, 36), little is known regarding the ligands that trigger ALK4 activation during pancreatic tumor initiation. Here, we report that activin A was identified as the most prominent ALK4 ligand expressed during the onset of mouse pancreatic tumors. We showed that activin A expression was strongly induced in ADM lesions and was subsequently limited to a subset of cells in lesions that progress to PanIN grades, in both mouse and human tissues. Our observations indicate that activin A may represent a major SASP factor in the OIS promoted by Kras within ADM. These results further suggest that activin A production in neoplastic, senescent cells inhibits the proliferation of ADM, in a cell autonomous manner, limiting progression to PanIN and more advanced lesions, through the modulation of p16 or p21 expression. In addition, our study indicates the importance of Kras-OIS for the limitation of ADM progression to more advanced lesions and further demonstrates that SASP factors could exert beneficial, antitumoral effects. These results emphasize that blocking activin A signaling through ALK4 reduced senescence, both in vitro and in vivo, leading to the increased proliferation of ADM and PanIN lesions, which is consistent with the reduced survival observed for patients carrying ALK4 mutations (36).

The involvement of activin A in PDAC is supported by clinical studies reporting the expression of activin A in a subset of human pancreatic cancers and the identification of elevated serum activin A levels in patients with PDAC (37, 38). Activin A has been reported to drive the self-renewal of pancreatic cancer stem cells (39, 40) and to promote pancreatic cancer–associated cachexia (38, 41). Our results indicated that, in addition to previously identified protumoral functions in advanced lesions, activin A signaling plays a suppressive role during the early stages of pancreatic tumor initiation, through SASP-associated functions. This observation supports the existence of antitumoral SASP factors, such as activin A, that exert beneficial in vivo effects during OIS.

Although, members of the TGFβ signaling family have been associated with a profibrotic senescence phenotype (42), in contrast with the proinflammatory SASP mediated by NF-κB, other ligands within the TGFβ superfamily have been proposed to be important factors in the paracrine senescence mediated by Kras (19). The blockade of TGFβ signaling has shown to enhance oncogenic Ras-induced tumorigenesis and metastasis in mammary epithelial cells, and to reduce senescence in pancreatic cancer models (19, 43). Proteomic analysis has revealed a dynamic pattern of Ras-related SASPs, indicating a role for TGFβ signaling during the early stages of the senescence process (42) and identifying INHBA, GDF15, BMP2, and BMP6 as major factors during paracrine SASP, whereas TGFβ1 was only moderately induced compared with the above-mentioned ligands (19). These observations indicate that activin A may represent a key element in TGFβ signaling–mediated paracrine senescence. This hypothesis is supported by the proposed use of activin A as a potential blood biomarker for senescent cell burden in vivo (44), and is consistent with the recently published role of activin A as a prominent autocrine regulator of hepatocyte growth arrest and cellular senescence through CDKN2B/p15^ink4a (45).

Although activin A downstream signaling, mediated by ALK4 and P-Smad2, appears to limit ADM/PanIN progression, cell and noncell autonomous molecular targets further downstream of this signaling pathway remain to be further defined. Previous studies have indicated that activin A induces p21 expression, through ALK4- and Smad-mediated signaling pathways, in various cells, including pancreatic tumor cell lines (36, 46–48). Interestingly, p21 has been reported to limit senescence and ADM formation during pancreatitis (49). Similarly, the suppression of p21 accelerates the progression of Kras-induced pancreatic lesions through increased proliferation (50). Taken together, these observations indicate that p21 may represent an activin A target downstream of ALK4/P-Smad2 activation, which is supported by the combined observations of reduced p21 expression and increased proliferation in ADM/PanIN lesions and the direct regulation of p21 transcript and protein levels observed in Panc-1 cells, following activin A inhibition.

Our results may have clinical relevance, as targeting activin A through the use of sActRIIB-Fc has been proposed to be an effective therapeutic intervention against muscle atrophic conditions, cancer cachexia, and, more recently, senescence-associated aging conditions (51). Promising therapeutic benefits have been shown for the reversion of cancer cachexia, promoted by activin A and myostatin production in ovarian tumor and xenografted lung and colon cancer cell lines (52–54). Our results indicate that the use of such therapy, which may be efficient for many pathologic conditions, should be considered with extreme caution for the treatment of pancreatic cancer. Indeed, given that low-grade pancreatic lesions are frequently found in the peritumoral regions surrounding PDAC, whether blocking activin A–induced cachexia using sActRIIB-Fc may favor the progression of neoplastic lesions into more advanced grades should be more thoroughly investigated.

In conclusion, our study provides the first demonstration that activin A signaling through ALK4 represents one of the first defense mechanisms induced by Kras-OIS to limit the development and progression of PanIN. Although, limited to the sole function of activin A as an antitumoral cell autonomous SASP factor, our study further emphasizes the necessity of better understanding the complex paracrine functions associated with activin A and the activities of SASP molecules produced by ADM lesions, particularly how they mediate anti- or protumorigenic effects in both preneoplastic cells and their cellular environments. Understanding the impacts of senescent cells and SASP factors within low-grade pancreatic neoplasm is likely to represent a key step toward developing innovative therapeutic strategies for pancreatic cancer.

No potential conflicts of interest were disclosed.

Y. Zhao: Conceptualization, formal analysis, validation, investigation, visualization. Z. Wu: Investigation. M. Chanal: Validation, investigation, visualization. F. Guillaumond: Data curation, formal analysis, visualization, writing-review and editing. D. Goehrig: Validation, investigation, visualization. S. Bachy: Validation, investigation, visualization. M. Principe: Validation, investigation, visualization. A. Ziverec: Validation, investigation, visualization. J.-M. Flaman: Investigation. G. Collin: Investigation. R. Tomasini: Resources, writing-review and editing, scientific discussion. A. Pasternack: Resources. O. Ritvos: Resources, scientific discussion. S. Vasseur: Resources, visualization, writing-review and editing, scientific discussion. D. Bernard: Writing-review and editing, scientific discussion. A. Hennino: Funding acquisition, investigation, visualization, writing-review and editing, scientific discussion. P. Bertolino: Conceptualization, supervision, funding acquisition, validation, investigation, visualization, writing-original draft, project administration, writing-review and editing.

The authors thank the animal care staff at ALECS and ANICAN for the maintenance of the transgenic mouse strains. We are grateful to Nicolas Gadot (Research Pathology Platform, CRCL), Christophe Vanbelle, and Christelle Carreira (IARC) for assistance with histologic staining and image acquisition. We thank Ivan Mikaelian, Ulrich Valcourt, and David Vincent for valuable scientific discussions. This study was supported by La Region-Rhone Alpes-Auvergne and La Ligue de la Loire et du Rhône et la Fondation ARC pour la Recherche sur le Cancer. Y. Zhao and Z. Wu were supported by a Chinese Scholarship Council Fellowship.

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.