Stress Granules Determine the Development of Obesity-Associated Pancreatic Cancer

Cancer cells share a set of properties that are collectively referred to as the stress phenotype of cancer, which reflects the prominent levels of various cellular stresses that are present in cancer tissues (1). As cellular stress is detrimental to survival, tumorigenesis is interlocked with the capacity of cancer cells to activate stress-adaptive mechanisms such as macropinocytosis, autophagy, and the unfolded protein response (UPR; refs. 2–4). Each of these stress adaptive mechanisms is specific to the type of stress and promotes tumorigenesis by enhancing the fitness of cancer cells as well as resistance to therapies. In contrast to stress-specific adaptive mechanisms, SGs are nonmembranous organelles that assemble in response to several tumor-associated stresses including oxidative stress, hypoxia, endoplasmic reticulum (ER) stress, nutrient deprivation, and osmotic stress (5–10). SGs are biomolecular condensates that form by liquid–liquid phase separation (LLPS) and compartmentalize hundreds of proteins and thousands of mRNA molecules (11–17). Although much remains to be discovered on the biological pathways affected by SGs, they are critical to cell survival under stress with particular relevance to pancreatic ductal adenocarcinoma (PDAC) cells (18–20). PDAC is the prototypical KRAS-driven cancer and an aggressive disease that is continually increasing in incidence; the 5-year survival rate for PDAC is only 10%, and therapeutic options are severely deficient (21). Mutant KRAS was shown to upregulate the capacity of PDAC cells to form SGs, leading to enhanced resistance to several stress stimuli and chemotherapeutic agents (18). Direct evidence linking SGs to tumorigenesis, however, is lacking.

In considering the relevance of SGs to tumorigenesis, it is important to note the ample epidemiologic evidence and studies in mouse models that show that preexisting stress and inflammatory conditions promote cancer occurrence and development (22–24). Among these conditions, obesity affects ∼2/3 of adults in the United States and ∼50% worldwide and is a growing global epidemic (25–28). Obesity doubles the incidence risk and mortality for pancreatic cancer, and even higher relative risks and mortality rates have been observed for prevalent cancers such as colon, breast, liver, kidney, stomach, and uterine cancers (29, 30). The precise mechanisms through which obesity promotes cancer occurrence and progression are unknown. However, obesity is a strong inducer of ER, oxidative, genotoxic, and biomechanical stress, and evidence points to obesity-associated cellular stress as a critical intermediary in obesity-associated cancer (31–34). Here we aimed to ascertain that SGs, as a pan-stress adaptive mechanism, would be required in PDAC development and that the preexisting stress of obesity would dictate a higher selective pressure for SG-mediated stress adaptation and a higher reliance of obesity-associated PDAC on SGs.

Recent work has shown that SG formation is determined by the collective interactions of ∼36 proteins and their associated mRNAs and that G3BP proteins are a central node of this network (14). To assess the role of SGs in PDAC growth, we utilized PDAC cell lines derived from the KPC (LSL-KrasG12D/+;LSL-Trp53R172H/+; Pdx-1-Cre) genetically engineered mouse model (GEMM) of KRAS-driven murine PDAC (mPDAC; refs. 35–38). KPC cells were engineered to stably express one of two independent doxycycline (Dox)-inducible small hairpin (sh) RNAs that target G3bp1 or a control nontargeting (NT) shRNA (Fig. 1A and B). Immunofluorescence microscopy for two SG markers, Pumilio and endogenous eukaryotic initiation factor 4G (Eif4g), revealed that the specific knockdown of G3bp1 attenuated SG formation in KPC cells exposed to oxidative stress via treatment with sodium arsenate (SA; Fig. 1A). As expected, SGs are not present in the absence of stress stimuli in cell culture conditions. Furthermore, oxidative stress induces the translocation of Eif4G and Pumilio to SGs but has no impact on their expression levels. Quantification of the SG index by computing the cell area occupied by SGs (labeled by Pumilio or Eif4g) as a fraction of total cell area showed that the knockdown of G3bp1 attenuated SG formation by 80% for G3bp1 sh #1 and by ∼60% for G3bp1 sh #2 (Fig. 1C; ref. 18). Attenuation of the SG index reflects a diminished percentage of SG-positive cells and average size of SGs (Supplementary Fig. S1A and S1B). As cancer cells within a tumor are exposed to diverse types of stress, and it has been suggested that the key nucleator molecules driving SG formation may change accordingly, we assessed whether G3bp1 knockdown affected SG formation under ER stress and hypoxia (Fig. 1D; refs. 15, 39, 40). Knockdown of G3bp1 attenuated SG formation in KPC cells exposed to ER stress and hypoxia by ∼60% and ∼95%, respectively (Fig. 1D). Consistent with a role for G3bp1 in stress adaptation, knockdown of G3bp1 had no impact on cell proliferation under normal growth conditions but impaired cell survival under oxidative stress (Fig. 1E; Supplementary Fig. S1C and S1D). Altogether, these results show that the knockdown of G3bp1 attenuates SG formation in PDAC cells under several prevalent tumor-associated stressors.

To assess the contribution of SGs to PDAC development, we initially implanted KPC cells harboring NT or G3bp1 shRNAs into the pancreata of syngeneic (C57BL/6) mice and induced shRNA expression via Dox administration through drinking water, starting 24 hours after implantation (Fig. 1F). As previously reported, tumors arising from the orthotopic implantation of KPC-4662 cells recapitulate the histopathologic complexity and dense, desmoplastic microenvironment of human disease (Supplementary Fig. S1E; refs. 36–38). G3bp1 knockdown led to a ∼50% reduction in mPDAC growth (60% and 40% reduction for G3bp1 sh #1 and sh #2, respectively) compared with control (shNT; Fig. 1G and H). Knockdown of G3BP1 also led to robust inhibition of SG formation in human PDAC (hPDAC) cell lines (by 80% and 90% in HPAC and MiaPaCa-2 cells, respectively; Fig. 1I; Supplementary Fig. S1F and S1G) and, as seen with KPC cells, had no impact on cell proliferation under normal growth conditions (Supplementary Fig. S1H). Consistent with our observations in mPDAC, G3BP1 knockdown in xenografts of hPDAC cells in athymic nu/nu mice reduced tumor growth by ∼50% (Fig. 1J).

To establish that the impact of G3BP1 knockdown on tumor growth is due to its function as an SG nucleator, we assessed whether targeting another SG nucleator, the T cell–restricted intracellular antigen like 1 (TIAL1) protein, would phenocopy the effect of G3BP1 knockdown (14, 41). Dox treatment of MiaPaCa-2 cells harboring a TIAL1 shRNA downregulated TIAL1 levels and inhibited SG formation under oxidative stress comparable with G3BP1 knockdown (Fig. 1I; Supplementary Fig. S1F and S1G). TIAL1 knockdown had no impact on cell proliferation under normal growth conditions (Supplementary Fig. S1I). TIAL1 suppression in vivo, however, impaired tumor growth by ∼50% (Fig. 1J). These data show that individual downregulation of two otherwise functionally distinct SG nucleator molecules has a similar impact on hPDAC growth. As such, these results support the idea that the impact of G3BP1 and TIAL1 knockdown on tumor growth is mediated by SGs.

Previous work has shown that the ability of G3BP1 to regulate SG formation is determined by its capacity to undergo LLPS with RNA, which is controlled by the dimerization of G3BP1 via its NTF2-like (NTF2L) domain and the interplay between its three intrinsically disordered regions (IDR; refs. 14). To causally link SG formation with PDAC growth, we initially took advantage of a G3bp1 mutant in which the NTF2L domain has been deleted (dNTF2L-G3bp1) and that has been shown to be deficient in SG formation (14, 42). As such, G3bp1-knockdown KPC cells were engineered to express shRNA-resistant GFP-G3bp1 wild-type (WT) or GFP-dNTF2L-G3bp1 to comparable levels with endogenous G3bp1 (Fig. 2A). Expression of G3bp1 WT or GFP-dNTF2L-G3bp1 in G3bp1-knockdown cells had no impact on proliferation under normal growth conditions (Supplementary Fig. S2A). As expected, the formation of SGs under oxidative stress conditions was fully rescued by the expression of shRNA-resistant GFP-G3bp1 WT but not GFP-dNTF2L-G3bp1 (Fig. 2B; Supplementary Fig. S2B–S2D). Furthermore, GFP-G3bp1 WT expression rescued the growth of G3bp1-knockdown mPDAC tumors, whereas expression of GFP-dNTF2L-G3bp1 failed to do so (Fig. 2C and D; Supplementary Fig. S2E). Given that the NTF2L domain is critical to SG formation, these results show that the rescue of tumor growth by the expression of GFP-G3bp1 WT is determined by its SG-nucleating capacity.

To conclusively determine that the impact of G3BP1 knockdown on PDAC growth is due to the inhibition of SG formation, we utilized a “synthetic” construct comprised of domains heterologous to the domains of G3BP1 that are involved in SG formation, which was previously shown to rescue SG formation in G3BP1/G3BP2-knockout cells (Fig. 2E; ref. 14). Expression of GFP-“synthetic” in hPDAC cells in which G3BP1 was knocked down had no impact on cell proliferation under normal growth conditions but led to a ∼3× increase in SGs when cells were under oxidative stress (Supplementary Fig. S2F and S2G). However, at the attainable expression levels, GFP-synthetic was weaker (by 40%) than shRNA-resistant GFP-G3BP1 WT, which rescued SG formation to levels comparable with shNT-expressing cells (Fig. 2F and G; Supplementary Fig. S2G). Nonetheless, in agreement with its capacity to rescue SG formation, GFP-synthetic rescued the growth deficiency of G3BP1-knockdown tumors by ∼2.5× (Fig. 2H). Altogether, these results causally link SGs to PDAC development.

Previous work reporting that mutant KRAS upregulated SG formation showed that SG inhibition led to higher cell death in mutant KRAS versus KRAS WT cells under oxidative stress, hence indicating that the levels of SGs can correlate with their requirement for survival (18). To determine the dependence of obesity-associated PDAC on SGs, we initially assessed SG levels in syngeneic orthotopic tumors in mice fed a high-fat chow to model diet-induced obesity (DIO; Fig. 3A). For comparison, we utilized standard weight mice that were fed a standard diet (ST; Fig. 3A). Consistent with previous studies, DIO promoted mPDAC growth (Fig. 3B; Supplementary Fig. S3A and S3B; refs. 43–45). As such, tumors in three orthotopic mPDAC models (KPC-4662, KPC-6560, and ES-149) in DIO mice were ∼3× larger than their counterparts in ST mice (Fig. 3B; Supplementary Fig. S3A and S3B; refs. 36, 37, 46). In agreement with previous studies, assessment of cell death and proliferation by quantification of the fraction of tumor area positive (+) for cleaved caspase-3 (CC3) and Ki-67, respectively, indicated a ∼2.5× decrease in cell death and a ∼5× increase in proliferation in DIO versus ST tumors (Supplementary Fig. S3C and S3D; ref. 47). Quantification of the SG index revealed a ∼5–6× increase in SG levels in DIO versus ST tumors for KPC-4662, KPC-6560, and ES-149 orthotopic models (Fig. 3C and D; Supplementary Fig. S3E). Analysis of tumors of comparable size in the DIO and ST cohorts (Supplementary Fig. S3F) also revealed a ∼5× difference in SG levels, indicating that enhanced SG levels in obesity-associated PDAC are linked to obesity and not tumor size. Of note, no SGs were detected in normal pancreata from DIO and ST mice (Supplementary Fig. S3G). In addition, quantification of tumor SGs reflects predominantly cancer cell SGs, as stroma cells form relatively significantly less (1/10th–1/20th) SGs (Supplementary Fig. S3H).

To determine whether the heightened levels of SGs in mPDAC tumors in DIO versus ST mice were truly a consequence of obesity and not solely a high-fat diet, we next utilized the leptin-deficient (ob/ob) genetic model of obesity (Fig. 3E; Supplementary Fig. S3I and S3J). Tumors arising from KPC cells implanted in the pancreata of ob/ob mice versus age-matched WT mice on standard chow were ∼3–4× larger and showed a ∼5× increase in SG levels (Fig. 3F and G; Supplementary Fig. S3J and S3K). The same difference in SG levels was observed when tumors of comparable size in the ob/ob and ST cohorts were analyzed (Supplementary Fig. S3L). These results are consistent with a model whereby the stress phenotype of obesity-associated PDAC dictates a heightened SG formation capacity.

To assess the dependence of obesity-associated PDAC on SGs, we transduced KPC-4662 cells harboring NT or G3bp1 shRNAs (Fig. 1A and B) with a firefly luciferase lentiviral construct to track the development of mPDAC orthotopic tumors in real time by bioluminescence imaging. Western blotting confirmed equal levels of luciferase expression in the established cell lines, and IVIS imaging upon implantation confirmed equal luciferase activity among all cohorts (Supplementary Fig. S4A). As expected, control (shNT) tumors in DIO mice showed an earlier onset, whereby tumor burden at the earliest measurement on day 7 in DIO mice was comparable with day 35 in ST mice (Fig. 4A–C; Supplementary Fig. S4B). In addition, the growth of control DIO tumors proceeded at an accelerated rate compared with control ST tumors, and control DIO tumors were significantly larger at endpoint (Fig. 4B and C). Inhibition of SGs in tumors in ST mice had no initial impact on tumor burden, which became significantly smaller from control tumors only from day 38 onward; from this point, SG-inhibited tumors in ST mice progressed with a slower growth rate compared with control ST tumors and showed a ∼50% decrease in bioluminescence at endpoint (60% and 40% for G3bp1 sh #1 and sh #2, respectively). The 50% decrease in bioluminescence is consistent with our findings with tumor weight measurements (Fig. 4B and C and Fig. 1H). Distinct from their ST counterparts, SG-inhibited DIO tumors showed a drastic difference in tumor burden compared with control DIO or ST tumors throughout the duration of the study (Fig. 4B and C; Supplementary Fig. S4B). In DIO mice, SG-inhibited tumors showed a significantly smaller tumor burden compared with control from day 7 (Fig. 4C; Supplementary Fig. S4B). This difference expanded as control tumors grew exponentially, whereas SG-inhibited tumors showed little growth. SG inhibition had no impact on the body weight of ST or DIO mice (Supplementary Fig. S4C). Confirming that the bioluminescence readings accurately reflect tumor size in the DIO model, tumor weight measurements and bioluminescence reads on day 35 showed a similar ∼5-fold difference between control and SG-inhibited tumors (Fig. 4B; Supplementary Fig. S4D). At endpoint measurement on day 42, SG-inhibited G3bp1 sh #1 and sh #2 tumors in DIO mice were ∼1/14th (93% smaller) and ∼1/7th (87% smaller), respectively, of control tumors (Fig. 4B and C). A higher dependence of obesity-associated PDAC on SGs was also observed in a second (KPC-6560) orthotopic model (Supplementary Fig. S4E–S4G). Knockdown of G3bp1 in KPC-6560 cells inhibited SG formation to a similar extent as in KPC-4662 cells and had no impact on cell proliferation under normal growth conditions (Supplementary Fig. S4E and S4F). The tumor weight of SG-inhibited KPC-6560 tumors in DIO mice was ∼70% smaller than control tumors on day 21; at this point, consistent with the KPC-4662 model, no difference in tumor weight was observed between SG-inhibited and control tumors in ST mice (Supplementary Fig. S4G and S4H). Inhibition of SGs also had a heightened impact on established KPC-4662 tumors in DIO versus ST mice (Supplementary Fig. S4I and S4J; Supplementary Methods). Dox administration in ST and DIO mice harboring similar sized tumors led to a ∼7× inhibition of the growth of shG3bp1- versus shNT-expressing tumors in DIO mice but only a ∼2× inhibition in ST mice. In addition, orthotopic mPDAC tumors in the ob/ob model of obesity showed a dependence on SGs comparable to DIO (Fig. 4D and E; Supplementary Fig. S4K). SG-inhibited tumors in ob/ob mice were smaller than control starting at the first measurement on day 6 and throughout the duration of the study, with a 91% reduction at endpoint on day 25 (Fig. 4D and E). Of note, tumor burden data past day 42 for DIO, day 45 for ST mice, and day 25 for ob/ob mice were excluded from the analysis, as mice in control conditions were euthanized due to disease progression. Altogether, these data show that orthotopic PDAC onset, progression, and maintenance in obese mice are significantly impaired by SG inhibition and that the growth and maintenance of obesity-associated PDAC have a higher dependence on SGs.

We next evaluated whether the differential impact of SG inhibition on the growth of obesity-associated versus ST mPDAC was due to cell death and/or proliferation. SG inhibition led to higher cell death in DIO mice compared with ST mice (Fig. 4F and G). As such, CC3⁺ fractions for G3bp1 sh #1 and sh #2 tumors were respectively ∼17× and 8× higher than control (shNT) tumors in DIO mice, but only ∼2× and ∼4× higher than control tumors in ST mice (Fig. 4F and G). No differences in CC3⁺ fractions were observed in control tumors in DIO versus ST. In contrast to cell death, the impact of SG inhibition on proliferation in DIO and ST tumors was similar; G3bp1 sh #1 and sh #2 reduced Ki-67 levels by ∼3× and ∼7×, respectively, in DIO mice, and by ∼4× in ST mice (Fig. 4F and G). These results show that SGs can contribute to both cancer cell proliferation and survival in vivo. However, the specific enhancement of tumor cell death that was observed when SGs were inhibited in the DIO setting indicates that obesity-associated PDAC has a higher dependence on SGs for survival.

Following the initial 45-day period, mice were continually monitored for a total of 300 days to determine their rate of survival (Fig. 4H and I). Despite accelerated tumor growth, the median survival between DIO and ST mice bearing control tumors was comparable—62 and 64 days, respectively (Fig. 4H). The median survival of ST mice bearing G3bp1-knockdown tumors relative to control was extended by ∼1.5× to 89 days. G3bp1 knockdown in tumors in DIO mice, on the other hand, led to significantly longer survival (Fig. 4H). Expression of a shRNA-resistant GFP-G3bp1 WT in G3bp1-knockdown cells rescued SG formation and countered the impact of SG inhibition on survival similarly in DIO and ST mice, leading to a median survival of 45 and 37 days, respectively (Fig. 4H; Supplementary Fig. S4L and S4M). Notably, ∼40% of DIO mice with SG-inhibited tumors showed no detectable tumors by bioluminescence or gross examination of pancreata and surrounding tissues at endpoint on day 300 (Fig. 4I). Altogether, these data demonstrate that the development and progression of obesity-associated PDAC are highly dependent on SGs.

The heightened dependence of obesity-associated PDAC on SGs and enhanced SG levels suggests that obesity may modulate cellular signaling to promote SG formation in cancer cells under stress. The pathobiology of obesity is intricate and multisystemic, including major signaling alterations due to secreted factors in the bloodstream from adipose tissue, and perturbed metabolic pathways and tissue homeostasis mechanisms (48). To determine the mechanisms through which obesity enhances SG levels in PDAC, we initially assessed how secreted factors with altered circulating levels in obese subjects—namely, IFNγ, TNFα, insulin growth factor 1 (IGF1), IL1β, IL6, IL8, IL10, IL4, IL13, plasminogen activator inhibitor (PAI), leptin, and adiponectin—affect the capacity of PDAC cells to form SGs (49–57). Evaluation of SGs in hPDAC cells treated with each of the obesity-associated secreted factors for 2 hours prior to oxidative stress revealed that IGF1 enhanced SG formation by ∼4.5× (Fig. 5A and B). Of note, oxidative stress had no impact on IGF1 levels (Supplementary Fig. S5A). A modest upregulation of SGs was observed with IL6, IFNγ, IL10, and IL13 (∼1.25× for IL6, IFNγ, and IL10 and ∼1.14× for IL13), whereas TNFα, adiponectin, and IL1β led to a slight reduction (∼0.75× for TNFα, adiponectin, and ∼0.9× for IL1β; Fig. 5A). IL8 and leptin had no effect. A previous study linked obesity-associated PDAC growth to a cross-talk between locally produced, obesity-driven cholecystokinin (CCK) and pancreatic epithelium (45). Although the concentration at which CCK can accumulate locally is unknown, the stimulation of hPDAC cells with the reported concentration of CCK in circulation in obesity-associated mPDAC or 3× above had no impact on SG formation (Supplementary Fig. S5B and Supplementary Methods). The capacity of IGF1 to potently promote SG formation was maintained throughout a panel of mouse and human PDAC cell lines under oxidative and ER stress (Fig. 5C and D; Supplementary Fig. S5C). As such, these data identify IGF1 as the top obesity-associated secreted factor that promotes SG formation in PDAC cells.

IGF1 binds specifically to the IGF1 receptor (IGF1R) leading to phosphorylation and activation of the tyrosine kinase activity of the receptor and its downstream effector pathways. The IGF1R inhibitor picropodophyllin (PPP), which specifically inhibits phosphorylation of tyrosine Y1136 in the activation loop of the IGF1R kinase domain, impaired the IGF1-stimulated phosphorylation of IGF1R and downstream activation of AKT in hPDAC cells and diminished SGs to levels equivalent to control vehicle-treated cells (Fig. 5E; Supplementary Fig. S5D; ref. 58). Moreover, SG inhibition by PPP is specific to the IGF1-treated condition, as in the absence of IGF1, PPP had no impact on SG levels. These results suggest that the upregulation of SGs in obesity-associated PDAC is mediated by a distinct IGF1/IGF1R-driven pathway. In principle, however, circulating insulin, which is also increased in obesity, can stimulate IGF1R and consequently upregulate SGs (59–61). Nonetheless, treatment of hPDAC cells with insulin, at concentrations reported in the pancreas of obese mice and in circulation in obese patients, had no impact on IGF1R phosphorylation, activation of downstream signaling, or SG levels (Fig. 5E; Supplementary Fig. S5D; refs. 59, 60). These findings suggest that the activation of IGF1R in obesity-associated PDAC is preferentially driven by elevated IGF1 versus insulin.

Consistent with reported heightened levels of IGF1 and insulin in obesity, quantification of phospho-Igf1r (pIgf1r)/phospho-Insulin receptor (pIr) in DIO and ST tumor tissues revealed a ∼2.5× increase in pIgf1r/pIr in DIO versus ST tumor tissues (Fig. 5F; Supplementary Fig. S5E). A similar stimulation was observed when mPDAC cells were treated with Igf1 (Supplementary Fig. S5F). To discern whether SG upregulation in obesity-associated PDAC is mediated by Igf1r activation, we next administered the specific Igf1r inhibitor PPP (every 12 hours for 48 hours) or vehicle to DIO mice with established orthotopic tumors and evaluated SG levels (Fig. 5G; Supplementary Fig. S5G). PPP treatment resulted in a ∼2.5× reduction in SG levels in DIO tumors, thus demonstrating that SG upregulation in obesity-associated PDAC is driven by Igf1r activation. To determine whether the Igf1r-mediated upregulation of SGs in obesity-associated PDAC is driven by Igf1, we administered an Igf1-neutralizing antibody or IgG control (once per day for 72 hours) to DIO and ST mice with orthotopic PDAC tumors of similar size (Fig. 5H; Supplementary Fig. S5H; Supplementary Methods; ref. 62). Confirming target engagement, tumors from DIO mice treated with the Igf1-neutralizing antibody showed impaired phosphorylation of the downstream effector molecule S6 kinase 1 (S6k1) compared with tumors from IgG control–treated DIO mice (Supplementary Fig. S5I). Moreover, Igf1-neutralizing antibody treatment reduced SGs (∼5× reduction) in DIO tumors to levels comparable with control-treated ST tumors (Fig. 5I). Treatment of ST mice with the Igf1-neutralizing antibody had no impact on tumor SG levels. These data demonstrate that neutralizing Igf1 inhibits SG formation in DIO tumors specifically and indicate that SG upregulation in obesity-associated PDAC is driven by elevated Igf1. Based on these observations we next tested whether the administration of Igf1 at levels found in obesity is sufficient to stimulate SG formation in PDAC tumors in ST mice (Fig. 5J; Supplementary Fig. S5J; Supplementary Methods). Confirming target engagement, Igf1 treatment enhanced S6k1 phosphorylation in ST tumors (Supplementary Fig. S5K). Notably, a single administration of Igf1 enhanced SG levels by ∼3× (Fig. 5J). Altogether, these results indicate that SG upregulation in obesity-associated PDAC is driven by IGF1/IGF1R activation.

Previous work reported that mutant KRAS regulation of SG formation in hPDAC cells is driven by 15-deoxy-12,14-prostaglandin J2 (15-d-PGJ2) via the KRAS/MAPK regulation of cyclooxygenase 2 and 15-hydroxyprostaglandin dehydrogenase (HPGD; ref. 18). Assessment of the interstitial fluid of DIO and ST tumors of comparable size revealed similar levels of 15-d-PGJ2 (Supplementary Fig. S5L and Supplementary Methods). In addition, no changes in the transcript levels of Cox1, Cox2, and Hpgd were observed between ST and DIO tumors (Supplementary Fig. S5L and Supplementary Methods). These observations support a model whereby Igf1 promotes SGs independent of prostaglandins (Supplementary Fig. S5M).

Given that PI3K/AKT and MAPK are the two major pathways downstream of IGF1/IGF1R activation, we assessed the impact of PI3K and MEK inhibition on SG levels in IGF1- and vehicle-treated cells (Fig. 6A and B; Supplementary Fig. S6A; ref. 63). Consistent with previous work that the mutant KRAS regulation of SG formation in hPDAC cells is MAPK dependent, the protein kinase MEK1 inhibitor PD98059 impaired SG formation in the absence of IGF1 by ∼60% and had a modest impact (25% inhibition) in SG levels in IGF1-treated cells (Fig. 6A and B; ref. 18). On the other hand, the PI3K inhibitor LY294002, as well as inhibition of the downstream PI3K/AKT effector mTOR with rapamycin, attenuated SG formation in IGF1-treated cells but had no impact in the absence of IGF1 (Fig. 6A and B). Of note, LY294002 and rapamycin diminished SGs in IGF1-treated cells to levels equivalent to vehicle-treated cells. Inhibitor treatments at the indicated doses and duration had no impact on cell death or cell cycle (Supplementary Fig. S6B and S6C). These results show that IGF1 promotes SG formation via a MEK-independent but PI3K/AKT/mTOR-dependent mechanism.

A previous study reported that the PI3K/AKT/mTOR effector molecules S6K1 and S6K2 mediated SG formation in Hela cells under oxidative stress by regulating the phosphorylation of eIF2α via an unknown mechanism and eIF2α-independent mechanisms (64). Given these findings, we sought to determine the contribution of S6K1/S6K2 signaling to IGF1-mediated SG formation in hPDAC cells. Treatment with the S6K1-specific inhibitor PF-4708671 blocked IGF1-mediated SG formation in hPDAC cells but had no impact on SG formation in the absence of IGF1 (Fig. 6A–C). In addition, IGF1 treatment had no impact on eIF2α phosphorylation in the presence or absence of oxidative stress, thus indicating that IGF1 promotes SG formation in hPDAC cells via an eIF2α-independent mechanism (Supplementary Fig. S6D). No changes were observed in the expression levels of the SG nucleators G3BP1 and TIAL or the KRAS mediators of SG formation, namely, COX2 and HPGD (Supplementary Fig. S6D; ref. 18). Altogether, these results show that IGF1 promotes SG formation through an S6K1-dependent but MEK- and eIF2α-independent mechanism.

Because mutant KRAS–mediated SG formation is MEK dependent, the observation that IGF1 promotes SG formation independent of MEK suggests that IGF1-driven SG formation is independent of mutant KRAS. To test this idea, hPDAC cells expressing a shRNA targeting KRAS or control shRNA were subjected to oxidative stress and assessed for SG formation (Supplementary Fig. S6E and Supplementary Methods). As expected, KRAS knockdown impaired SG formation in the absence of IGF1 (18). KRAS knockdown had no impact on SG formation in the presence of IGF1, however, demonstrating that IGF1 stimulates SGs independent of mutant KRAS. Given that SGs function to enhance cellular stress resistance, we tested whether IGF1 stimulation would also restore the fitness of KRAS-knockdown cells concomitant with SG upregulation (Supplementary Fig. S6F; Supplementary Methods). As expected, in the absence of IGF1, KRAS knockdown lowered stress resistance and led to a ∼3× increase in cell death under lethal oxidative stress conditions. In contrast, IGF1 treatment fully abrogated cell death in KRAS-knockdown cells to levels comparable with control IGF1–treated cells. Of note, IGF1 treatment enhanced the resistance of control cells by impairing cell death by ∼40%. Altogether, these results indicate that IGF1 drives SG formation independent of mutant KRAS and can counter the impact of KRAS inhibition on SG formation and cellular fitness.

To decipher the molecular mediators through which IGF1/S6K1 promotes SG formation, we initially assessed if the temporal dynamics of IGF1 stimulation affected SG levels (Fig. 6D). The quantification of SG levels under oxidative stress following a time course (10-minute, 30-minute, 1-hour, and 2-hour pretreatment with IGF1) revealed a ∼2× enhancement with as short as 10-minute pretreatment with IGF1; SG levels increased over time and plateaued at ∼4× over control at the 1-hour time point onward, mirroring the dynamics of S6K1 phosphorylation (Fig. 6E). The evidence that a short treatment with IGF1 and S6K1 activation promotes SG formation points to IGF1/S6K1-driven posttranslational modifications of an effector as a mediator of this process. S6K1 regulates several effector molecules with roles in cellular processes, including cytoskeletal organization, mRNA splicing, inflammation, apoptosis, and metabolism (65). In addition, S6K1 also stimulates protein translation via the elongation factor 2 kinase (eEF2K), eukaryotic initiation factor 3 (eIF3), and ribosomal protein 6 (S6), which as such would counteract SG formation (66). Focusing on the former group of effectors, therefore, we identified the SRSF2 protein kinase 2 (SRPK2) as distinct in that several of its binding partners are components of SGs (13, 67–70). For this reason, we next investigated the localization of SRPK2 in IGF1- and vehicle-treated hPDAC cells under oxidative stress (Fig. 6F; Supplementary Fig. S6G). Notably, SRPK2 was detected in SGs in both IGF1- and vehicle-treated cells. However, the partitioning of SRPK2 to SGs was significantly higher in IGF1-treated cells (Fig. 6F; Supplementary Fig. S6G). As such, the quantification of the relative fraction of SRPK2 in G3BP1-SGs showed a ∼4× enrichment in IGF1-treated cells (Fig. 6G). No changes were observed in SRPK2 levels with IGF1 stimulation, whereas the phosphorylation of SRPK2 followed the same temporal dynamics as IGF1-induced S6K1 phosphorylation (Fig. 6E). To investigate if the IGF1/S6K1-induced phosphorylation of SRPK2 affected the partitioning of SRPK2 to SGs, we assessed the effect of the S6K1 inhibitor PF-4708671 on the localization of SRPK2 (Fig. 6G; Supplementary Fig. S6H). PF-4708671 impaired SRPK2 phosphor­ylation and attenuated the partitioning of SRPK2 into SGs in IGF1-treated cells to levels comparable with those in the absence of IGF1. These results identify SRPK2 as a novel component of SGs and show that IGF1 promotes the partitioning of SRPK2 into SGs via S6K1.

The IGF1/S6K1-dependent partitioning of SRPK2 into SGs suggests that the capacity of IGF1 to promote SG formation may be mediated by SRPK2 phosphorylation. To evaluate this idea, we initially utilized Dox-inducible lentiviral shRNAs to target SRPK2 in hPDAC cells (Fig. 6H). Knockdown of SRPK2 attenuated SG formation in hPDAC cells stimulated with IGF1 to levels comparable with unstimulated control (Fig. 6H–J; Supplementary Fig. S6I). Expression of a shRNA-resistant SRPK2 WT in SRPK2-knockdown cells or of a phosphomimetic SRPK2 mutant in which the S6K1 phosphorylation amino acid Ser 494 was mutated to aspartate (SRPK2 S494D) fully rescued IGF1-driven SG formation, whereas expression of a phosphodeficient SRPK2 in which Ser 494 was mutated to an alanine (SRPK2 S494A) failed to do so (Fig. 6K–M; Supplementary Fig. S6J; ref. 69). In addition, PF-4708671 treatment attenuated SG formation in SRPK2-knockdown cells expressing shRNA-resistant SRPK2 WT but had no effect when shRNA-resistant, phosphomimetic SRPK2 S494D was expressed (Supplementary Fig. S6K). Altogether, these results identify S6K1-phosphory­lated SRPK2 as an essential mediator of IGF1-stimulated SG formation in hPDAC cells.

Our results indicate that SG upregulation in obesity-associated PDAC is driven by IGF1/IGF1R activation and that IGF1 promotes SG formation specifically through S6K1/SRPK2 (Supplementary Fig. S7A). Along with the heightened dependence of obesity-associated PDAC on SGs, these data suggest that the SG formation and growth of obesity-associated PDAC may be selectively dependent on S6K1 activity. To evaluate this idea, KPC-4662 and KPC-6560 cells expressing luciferase were orthotopically implanted in the pancreata of syngeneic DIO and ST mice; 24 hours after implantation (day 1), mice were treated with the S6K1 inhibitor PF-4708671 or vehicle control via a once-a-day intraperitoneal injection for the duration of the study (Fig. 7A). Bioluminescence measurements on day 1 demonstrated equal cell number implantations in all cohorts (Supplementary Fig. S7B). As expected, S6K1 was hyperactivated in DIO versus ST tumors (Supplementary Fig. S7C). Confirming target engagement, treatment of DIO mice with PF-4708671 led to the diminished phosphorylation of the S6 protein in orthotopic tumors (Fig. 7B). PF-4708671 treatment attenuated SG levels in DIO tumors by ∼80% but had no effect on SG levels in tumors in ST mice, establishing that SG formation in obesity-associated PDAC is specifically dependent on S6K1 activity (Fig. 7C; Supplementary Fig. S7D). Consistent with the role of S6K1 as a mediator of SG formation specific to IGF1 stimulation and the heightened dependence of obesity-associated PDAC on SGs, PF-4708671 treatment significantly impaired mPDAC growth in DIO mice but had no effect on mPDAC growth in ST mice (Fig. 7D and E; Supplementary Fig. S7E–S7H). Of note, PF-4708671 treatment had no impact on the body weight of DIO or ST animals (Supplementary Fig. S7E and S7F). KPC-4662 and KPC-6560 tumors in PF-4708671-treated DIO mice diverged in growth rate from their vehicle-treated counterparts starting on day 25, and at endpoint were respectively ∼1/5th and ∼1/8th (82% and 88% smaller) of tumors in vehicle-treated DIO mice (Fig. 7E; Supplementary Fig. S7G). In contrast, the growth rate of tumors in PF-4708671- and vehicle-treated ST mice was indistinguishable for the duration of the study.

SG abrogation by G3BP1 knockdown affected cell death and proliferation in both ST and DIO mPDAC but resulted in greater cell death in DIO tumors (Fig. 4). Inhibition of S6K1 does not affect the basal levels of SGs, as S6K1 specifically augments SG levels in obesity-associated PDAC. Therefore, S6K1-inhibited PDAC tumors would still retain the basal levels/function of SGs, and treatment with PF-4708671 would be expected to selectively affect cell death in DIO tumors. Consistent with this model, PF-4708671 treatment enhanced cell death specifically in DIO tumors but had no impact on proliferation in DIO or ST tumors (Fig. 7F and G). Altogether, these observations identify IGF1/S6K1 as a specific pathway mediating SG formation in obesity-associated PDAC and a potential target for the treatment of this disease.

To confirm the relevance of our findings to hPDAC, we obtained tumor samples of 65 patients and examined the SG index in the epithelial compartments as marked by cytokeratin 8 (CK8; Fig. 7H). SGs were consistently observed across all patient samples, and their prevalence was heterogeneous (Fig. 7H). Approximately 50% of the patients had a body mass index (BMI) >25 and as such were qualified as overweight/obese (Fig. 7I). Notably, SG levels differed significantly between the different BMI cohorts (Fig. 7J). As such, we found a ∼5-fold increase in SG levels in tumor samples from patients with BMI >25 compared with those with BMI <25 (Fig. 7J). Altogether, these results establish SG upregulation as a key feature and unique vulnerability of obesity-associated PDAC.

Biomolecular condensates that form through LLPS are increasingly recognized as important means of organizing and compartmentalizing cellular signaling (71). However, the understanding of their physiologic relevance is lacking. SGs are a prime example of such biomolecular condensates that are found in vivo in tissue samples from patients and mouse models of pancreatic cancer and sarcomas and have been implicated in the stress adaptation of cancer cells and resistance to therapy (18, 72, 73). Recent work has identified that the shift in saturation concentration that enables phase separation, and consequently SG formation, is determined by the collective interactions of ∼36 proteins and their associated mRNAs. Notably, G3BP1 and G3BP2 are a central node of this network (14). In agreement with this model, we show that G3BP1 knockdown significantly diminishes SG formation in PDAC cells. Importantly, we show that G3BP1 knockdown impairs PDAC growth. Three lines of evidence indicate that SGs specifically, and not an alternative function of G3BP1, are required for tumorigenesis. First, targeting another node in the core SG network, namely, TIAL1, phenocopies the effect of G3BP1 knockdown on PDAC growth. Second, deletion of the dimerization domain of G3BP1, which results in impaired SG formation, also impairs PDAC growth. Third, a “synthetic” protein consisting of dimerization, central IDR, and tandem RNA binding domains that are heterologous to the domains of G3BP1 necessary for SG formation can substitute for G3BP1 in SG formation and PDAC tumorigenesis to a similar extent despite harboring completely different primary sequences. Collectively, these results establish a pathophysiologic function of SGs as critical mediators of pancreatic tumorigenesis. Adding to these findings, our results provide insight into the specific functions of SGs under physiologic stress stimuli. In vitro evidence has linked SGs to stress adaptation primarily through the role SGs play in blocking cell death. Our results show that in the setting of tumorigenic stress, SGs contribute to both tumor proliferation and evasion of cell death. Whether proliferation is regulated directly by SGs, or indirectly as a corollary of stress adaptation, remains to be determined.

We predicted that stress acts as a selective pressure in tumorigenesis and that tumors that arise in preexisting stress conditions like obesity would face a greater selective pressure and have a higher dependence on SGs. In support of this idea, our results show that SG inhibition has a heightened impact on obesity-associated PDAC: The relative decrease in tumor growth is several-fold larger than standard weight, survival is prolonged, and 40% of the mice are tumor free at endpoint (300 days after implantation), whereas all standard weight mice succumb to disease. These findings provide important insights into the pathophysiologic role of SGs and inform potential chemopreventive strategies. Given the heightened dependence of obesity-associated PDAC, approaches targeting SG formation may succeed at blunting the occurrence risk imparted by obesity. In addition, these approaches may be relevant to the myriad of tumors associated with obesity, several of which are continually increasing in occurrence. Moreover, the findings that the heightened dependence of established obesity-associated PDAC on SGs for tumor maintenance indicates that targeting SGs may also prove a beneficial therapeutic option for the treatment of established tumors. On this note, the observation that the heightened dependence of obesity-associated PDAC on SGs is accompanied by augmented SG levels provides further evidence that the level of SG dependence correlates with the cellular capacity to form SGs and that SG levels may serve as biomarkers dictating the optimized outcome.

Our results establish a specific pathway, namely, IGF1/IGF1R/S6K1/SRPK2, through which obesity promotes SG formation in PDAC. Although IGF1 and insulin downstream signaling overlap significantly and circulating levels of both IGF1 and insulin are elevated in obesity, we show that IGF1 is the driver of IGF1R and downstream effector activation. This may be due to higher levels of IGF1 compared with circulating and pancreatic levels of insulin in obesity. In addition, ∼40% to 90% of IGF1R in tissues is present in complex with IR and these heterodimers have a higher affinity for IGF1 than for Insulin (74–76). SRPK2 is a recently identified substrate of S6K1 with roles in mRNA stability, splicing, and lipid metabolism (69). Our results add to these roles by identifying SRPK2 as an IGF1/S6K1-dependent constituent of SGs and a determinant of IGF1-driven SG formation. The mechanisms through which SRPK2 mediates SG formation are yet to be characterized. The nodes of the core SG network are known to contribute to SG formation to varying degrees, and it is possible that S6K1-stimulated SRPK2 may affect their contribution; this is consistent with the predicted model whereby SG assembly is subject to regulation by positive and negative cooperativity of extrinsic factors with the core network interactions (14). In support of this possibility, SRPK2 is known to interact with SG constituents such SRSF proteins and Tau in a phosphorylation-dependent manner (67, 70). SRSF3 is a component of the core SG network and therefore may represent a means through which SRPK2 alters the core interaction network to promote SG assembly. Another possibility is that as a constituent of SGs, SRPK2 itself may contribute to the core SG network in IGF-treated cells. Clearly, further investigation will be required to decipher the mechanisms through which SRPK2 can mediate SG assembly. Nonetheless, the IGF1/S6K1-dependent localization of SRPK2 to SGs highlights that the composition of SGs is context dependent. Previous studies have shown that the composition of SGs can vary with stress. The varying composition therefore extends not only to different stress stimuli but also to macromolecular stimuli.

In addition to SGs, tumors engage several stress-specific adaptive mechanisms. As such, it will be important to determine whether the heightened dependence of obesity-associated PDAC on SGs indicates a broad characteristic of these tumors and extends to other stress adaptive mechanisms. Furthermore, it will be critical to determine whether any of these dependencies extend to the many obesity-associated cancers. On this note, the findings that IGF1 stimulates SG formation even under conditions of mutant KRAS inhibition support a model whereby obesity-associated SG upregulation is independent of the tumor mutational status. Together with a better understanding of signaling alteration in obesity and other cancer-predisposing pathologic conditions, as they pertain to SG formation and stress adaptation, such studies can have important implications for the development of new and optimized therapeutic approaches.

All animal work was approved by the Thomas Jefferson University Institutional Animal Care and Use Committee. Standard C57BL/6 (ST; RRID:IMSR_JAX:000664), diet-induced obese C57BL/6 (DIO; RRID:IMSR_JAX:380050), and Lep^ob (ob/ob; RRID:IMSR_JAX:000632) mice were obtained from The Jackson Laboratory. CrTac:NCr-Foxn1^nu (nude; RRID:IMSR_TAC:ncrnu) mice were obtained from Taconic Farms. Eleven- to 13-week-old ST and DIO and 5- to 7-week-old ST and ob/ob mice were used for orthotopic tumor models and associated controls. Six- to 8-week-old nude mice were used for xenograft tumor models and associated controls.

Lentiviral transductions of all cancer cell lines were approved by the Thomas Jefferson University Institutional Biosafety Committee. The human pancreatic cancer cell lines MiaPaCa-2 (AddexBio; cat. #C0018002/59, RRID:CVCL_0428), HPAC (ATCC; cat. #CRL-2119, RRID:CVCL_3517), CFPAC (ATCC; cat. #CRL-1918, RRID:CVCL_1119), AsPC-1 (ATCC; cat. #CRL-1918, RRID:CVCL_1119), and Capan-2 (ATCC; cat. #HTB-80, RRID:CVCL_0026) were obtained and originally authenticated by short tandem repeat from ATCC. MiaPaCa-2, HPAC, and CFPAC cells were cultured in Dulbecco's modified Eagle medium (DMEM; Corning, 10-017-CV) supplemented with 10% FBS (Corning, 35-011-CV), 1% penicillin/streptomycin (Gibco, 15140122), and 1% HEPES Buffer 1M solution (Fisher, BP299-100). Capan-2 were cultured in McCoy's 5A medium (Gibco, 16600082), and AsPC-1 were cultured RPMI 1640 medium (Fisher Scientific, 11-875-093) supplemented as above. Mouse pancreatic cancer cell lines KPC-4662 and KPC-6560 were isolated from the LSL-KrasG12D/+;LSL-Trp53R172H/+; Pdx-1-Cre GEMM of KRAS-driven mPDAC and were a kind gift by Robert H. Vonderheide (University of Pennsylvania; ref. 36) and cultured in DMEM supplemented as above. All cells were cultured at 37°C with 5% CO₂ and routinely tested for Mycoplasma contamination every 4 months after the first thaw by the Mycoplasma PCR detection kit (ABM, cat. #G238). All cells were frozen at low passage and were no longer used after 20 passages after the first thaw.

The following constructs were purchased from Dharmacon: SMARTvector inducible mCMV/TurboRFP, G3BP1 shRNA (V3SH11252-226176431; V3IHSMCR_6114081), TIAL1 shRNA (V3SH11252-227445578; V3IHSMCR_7383228), SRPK2 shRNA #1 (V3SH11252-225843065), SRPK2 shRNA #2 (V3SH11252-226279721), pTRIPZ inducible/TurboRFP, G3BP1 shRNA #1 (RHS4696-200750396; V3THS_329105); SMARTvector inducible hCMV/TurboGFP, NT shRNA (VSC11707) used in human cell lines, NT shRNA (VSC11651) used in mouse cell lines, G3bp1 shRNA #1 (V3SM11253-231226612; V3IMMMCG_11164262), G3bp1 shRNA #2 (V3SM11253-231617629; V3IMMMCG_11555279), and G3bp1 shRNA (V3SM11253-234970000; V3IMMMCG_14907650). pTRIPZ inducible/TurboRFP, KRAS shRNA was purchased from Open Biosystems (V2THS-275818). The luciferase expression plasmid pLenti-PGK-Blast-V5-LUC (w528-1; RRID:Addgene_19166) was a gift from Eric Campeau and Paul Kaufman (Addgene plasmid #19166; https://www.addgene.org/19166/, RRID:Addgene_19166). pLenti-CMV-Blast-DEST (706-1; RRID:Addgene_17451) was a gift from Eric Campeau and Paul Kaufman (Addgene plasmid #17451; https://www.addgene.org/17451/, RRID:Addgene_17451). The Dox-inducible pCW57.1 plasmid was a gift from David Root (Addgene plasmid #41393; http://n2t.net/addgene:41393; RRID: Addgene_41393). Mouse G3bp1-turboGFP in pCMV-6-AC was obtained from Origene (MG207441). BP-LR cloning was used to insert G3bp1-tGFP into pcW57.1. The following primers were used to add attB1 sites to generate the full-length G3bp1-GFP and the N-terminus deletion dN-G3bp1-GFP constructs cloned into pcW57.1: G3bp1-GFP forward: 5′ GGG GAC AAG TTT GTA CAA AAA AGC AGG CTC GAG ATC TGC CGC GAT CGC C 3′; dN-G3bp1-GFP forward: 5′ GGG GAC AAG TTT GTA CAA AAA AGC AGG CTC GAT GTT TGT CAC AGA GCC TCA AGA GGA AT 3′; GFP-reverse: 5′ GGG GAC CAC TTT GTA CAA GAA AGC TGG GTC AGG CAC TGG GGA GGG GTC ACA GG 3′. The following primers were used to generate a full-length G3bp1-GFP resistant to G3bp1 shRNA #1 via QuickChange mutagenesis (QuickChange II Site-Directed Mutagenesis Kit, Agilent, cat. #200523): P1 forward: 5′ GAC ACC AGA GGT CGT CCC CGA TGA TTC TGG AAC TTT CTA TG 3′; P1 reverse: 5′ CAT AGA AAG TTC CAG AAT CAT CGG GGA CGA CCT CTG GTG TC 3′; P2 forward: 5′ GAC ACC AGA GGT CGT CCC CGA CTC CGG AAC TTT CTA TG 3′; P2 reverse: 5′ CAT AGA AAG TTC CGG AGT CGT CGG GGA CGA CCT CTG GTG TC 3′. Plasmids containing EGFP-G3BP1-WT and EGFP-GST-2xAsh1-IDR-hnRNPA1-tandem RRM (“synthetic”) in pEGFP-C3 (RRID:Addgene_53755) were kindly obtained from Paul Taylor (St. Jude Children's Research Hospital). BP-LR cloning was used to insert EGFP-G3BP1-WT and GFP-“synthetic” into pLenti-CMV-Blast-DEST (RRID:Addgene_17451). The following primers were used to add attB1 sites for BP-LR cloning: Synthetic forward: 5′ GGG GAC AAG TTT GTA CAA AAA AGC AGG CTC CAC CAT GGT GAG CAA G 3′; Synthetic reverse: 5′ GGG GAC CAC TTT GTA CAA GAA AGC TGG GTT CAG TTA TCT AGA TCC GGT GGA TCC 3′; EGFP-G3BP1-WT forward: 5′ GGG GAC AAG TTT GTA CAA AAA AGC AGG CTC GCT ACC GGT CGC CAC C 3′; EGFP-G3BP1-WT reverse: 5′ GGG GAC CAC TTT GTA CAA GAA AGC TGG GTT CAG TTA TCT AGA TCC GGT GGA TCC 3′.

Human pDONR223-SRPK2 was a gift from William Hahn and David Root (Addgene plasmid #23766; http://n2t.net/addgene:23766; RRID:Addgene_23766), and human SRPK2-MycD in pCMV-6-AC was obtained from Origene (cat. #RC205134). The following primers were used to generate point mutations in SRPK2-MycD-pCMV-6-AC via QuickChange mutagenesis: S494D forward: 5′ CAT GAC AGA AGC AGA ACG GTT GAT GCC TCC AGT ACT GGG GA 3′; S494D reverse: 5′ TCC CCA GTA CTG GAG GCA TCA ACC GTT CTG CTT CTG TCA TG 3′; S494A forward: 5′ GAC AGA AGC AGA ACG GTT GCC TCC AGT ACT GG 3′; S494A reverse: 5′ CCA GTA CTG GAG GCG GCA ACC GTT CTG CTT CTG TC 3′. The following primers were used to add attB1 sites to these SRPK2 mutants: SRPK2 forward: 5′ GGG GAC AAG TTT GTA CAA AAA AGC AGG CTC GCG AGG AGA TCT GCC GCG ATC G 3′ reverse: 5′ GGG GAC CAC TTT GTA CAA GAA AGC TGG GTC GGA AAC AGC TAT GAC CGC G 3′. BP-LR cloning was used to insert WT SRPK2, SRPK2 S494D, or SRPK2 S494A into pLenti-CMV-BLAST-DEST.

HEK293T (RRID:CVCL_HA71) cells were transduced using the calcium phosphate transfection method. Briefly, 500 μL of CaCl2 2.5 M, 7 μg pMD2G (RRID:Addgene_12259; gift from Didier Trono, Addgene plasmid #12259; http://n2t.net/addgene:12259; RRID:Addgene_12259), 13 μg psPAX2 (RRID:Addgene_12260; gift from Didier Trono, Addgene plasmid #12260; http://n2t.net/addgene:12260; RRID:Addgene_12260), and 20 μg of construct vector were added dropwise to 500 μL of HEPES buffered saline (HEBS) 2× while vortexing. After 25 to 40 minutes, the DNA CaPO4 coprecipitate was transferred dropwise onto HEK293T (RRID:CVCL_HA71) cells. Media were harvested 24 hours and 48 hours after transfection and passed through 0.45-μm filters. Finally, viruses were concentrated by spinning in 100 KD cutoff Amicon Filter tubes for 30 minutes at 4°C at 4,000 RPM.

A total of 40,000 cells of interest were seeded in serum-free media and transduced with the lentiviral particles containing the indicated constructs. Multiplicity of infection (MOI) for KPC cells was 5; all other cell lines, MOI = 10. Following transduction, KPC cells were selected with puromycin 5 μg/mL for 3 days and the top 20% cells based on their GFP expression after a 16-hour induction with 1 μg/mL Dox (Sigma-Aldrich, D9891) were sorted by flow cytometry with a BD FACSAria II Cell Sorter. For tumor growth luminescence tracking by IVIS, sorted KPC-4662 and KPC-6560 cells harboring the indicated constructs were additionally transduced with lentiviral particles containing the luciferase expression construct and selected with blasticidin (10 μg/mL) for 3 days. For KPC-4662 cells expressing G3bp1-GFP or dN-G3bp1-GFP resistant to shRNA, cells were transduced with lentivirus particles containing G3bp1-GFP or dN-G3bp1-GFP expression construct cells and selected with puromycin (5 μg/mL) for 2 days. Subsequently, cells were transduced with lentivirus particles containing G3bp1 shRNA #3, which targets the 3′ untranslated region (UTR) of G3bp1, or an NT shRNA construct, and selected with puromycin (5 μg/mL) for an additional 3 days.

MiaPaCa-2 and HPAC cells transduced with lentiviral particles containing the indicated shRNA constructs were selected with puromycin 5 μg/mL for 3 days. For MiaPaCa-2 cells expressing shRNA-resistant GFP-G3BP1 or GFP-synthetic, cells harboring G3BP1 shRNA #2, which targets the 3′ UTR of G3BP1, were transduced with lentiviral particles containing the GFP-G3BP1 or GFP-synthetic expression construct and selected for 3 days with blasticidin (10 μg/mL). Cells were subjected to sorting by flow cytometry with a BD FACSAria II Cell Sorter for the top 20% GFP-expressing cells. G3BP1 knockdown, expression of G3BP1-GFP, dN-G3bp1-GFP, GFP-G3BP1, GFP-synthetic, and luciferase were confirmed by Western blot after 3 to 5 days of Dox induction at 1 μg/mL.

Cells were plated on coverslips and stressed when at 70% confluence. Where indicated, cells were treated with Dox (1 μg/mL) for 3 to 5 days to induce shRNA or protein expression prior to stress. Oxidative stress was induced with SA for 1 hour at 200 μmol/L for KPC-4662 and KPC-6560 and 100 μmol/L for hPDAC cell lines; ER stress was induced by treatment with thapsigargin (Tocris, cat. #1138) for 6 hours at 25 μmol/L for KPCs 4662 and Capan-2, and 16 μmol/L for MiaPaCa-2. Hypoxia was induced using the Celldiscoverer 7 Automated Microscope incubation chamber system (Zeiss) set up at 0.5% O₂ and 5% CO₂ for 12 hours for MiaPaCa-2 or 24 hours for KPC-4662 cells.

For treatment with obesity-associated factors, cells were cultured as described above, starved for 16 hours in serum-free DMEM, and then treated for 2 hours with vehicle control (PBS), 100 ng/mL IGF-1 (Human IGF-I, PeproTech, cat. #100-11; murine IGF1, PeproTech, cat. #250-19), 50 ng/mL PAI (PeproTech, cat. #140-04), Leptin (PeproTech, cat. #300-27), 10 pg/mL recombinant IFNγ (PeproTech, cat. #300-02), 5 ng/mL or 15 ng/mL cholecystokinin (CCK; Thermo Fisher Scientific, J66669.EXD), 5 ng/mL insulin (EMD Millipore, cat. #407709), IL1β (PeproTech, cat. #200-01B), IL6 (PeproTech, cat. #200-06), IL8 (PeproTech, cat. #200-08), IL10 (PeproTech, cat. #200-10), IL4 (PeproTech, cat. #200-04), IL13 (PeproTech, cat. #200-13), adiponectin (PeproTech, cat. #450-24), and TNFα (PeproTech, cat. #300-01A) according to reported serum levels in obese subjects. Subsequently, stress was induced as described above.

For inhibitor treatments, cells were starved for 16 hours in serum-free DMEM or McCoy's 5A. Cells were then treated with recombinant IGF1 or vehicle control, in combination with each of the following inhibitors for 1 hour: the IGF1R inhibitor PPP (EMD Millipore, 407247) at 0.5 μmol/L, the PI3K inhibitor LY294002 (Cayman, cat. #70920) at 0.2 μmol/L, the mTOR inhibitor rapamycin (Selleck Chemicals, cat. #S1039) at 2 nmol/L, the MEK inhibitor PD98059 (Selleck Chemicals, cat. #S1177) at 0.7 μmol/L, and the S6K1 inhibitor (PF-4708671; Selleck Chemical, cat. #S2163, or MedChem, cat. #HY-15773) at 200 nmol/L for Capan-2 cells and 20 nmol/L for all other cells. Treatment concentrations were determined empirically and based on EC₅₀ values. Finally, stress was induced for 1 hour with 100 μmol/L SA for hPDAC cells or 200 μmol/L for KPC cells.

All cells were grown on coverslips, fixed in 4% paraformaldehyde, immunostained, and mounted in ProlongGold (Invitrogen, P36934). Immunostaining was performed using three well-described markers of stress granules: G3BP1 (1:2,000 Bethyl; cat. #A302-033A, RRID:AB_1576539), EIF4G (1:500 Cell Signaling Technology; cat. #2498S, RRID:AB_2096025 or 1:500 Santa Cruz Biotechnology; cat. #sc-133155, RRID:AB_2095748), and Pumilio (1:500 Abcam; cat. #ab10361, RRID:AB_297098). Additionally, cells were stained for SRPK2 (1:200 Proteintech; cat. #25417–1-AP, RRID:AB_11205747), phospho-IGF1R (1:200, Abcam; cat. #ab39398, RRID:AB_731544), and CC3 (1:500 Cell Signaling Technology; cat. #9661, RRID:AB_2341188).

All tissue samples were rinsed in PBS and fixed in 10% formalin for 24 hours. All tissue samples were then placed in cassettes and stored at 4°C in 70% ethanol until embedded in paraffin, sectioned (5 μm), and mounted on glass slides. Slides of tissue sections were deparaffinized in xylene and rehydrated through a reversed ethanol gradient. Then sections were subjected to heat-mediated antigen retrieval with citrate buffer (0.01M sodium citrate/0.05% Tween-20 pH 6.0). Tissue samples were blocked for 1 hour at room temperature (RT) in a 10 mmol/L Tris-HCl, 0.1M MgCl₂, 0.05% Tween-20, 1% BSA, and 10% FBS solution. Slides were stained with primary antibody overnight at 4°C in a humified chamber. Primary antibodies were used as follows: phospho-IGF1R (1:200, Abcam; cat. #ab39398, RRID:AB_731544), CK8 (1:200, DSHB; cat. #TROMA-I, RRID:AB_531826), G3BP1 (1:400, Bethyl; cat. #A302-033A, RRID:AB_1576539), phospho-S6K1 (1:100; Aviva Systems Biology; cat. #OAAF07416, RRID:AB_2630782), phospho-S6 (1:800; Cell Signaling Technology; cat. #5364S, RRID:AB_10694233), CC3 (1:200; Cell Signaling Technology; cat. #9664S, RRID:AB_2070042), and Ki-67 (1:200; Proteintech; cat. #27309–1-AP, RRID:AB_2756525). Slides were washed in TBST (0.1% Tween-20, TBS) and incubated with the appropriate secondary antibodies for 1 hour at RT. After washing in TBST, sections were stained with DAPI for 5 minutes at RT, and coverslips were mounted with ProlongGold (Invitrogen, P36934).

Imaging was performed with a Celldiscoverer 7 Automated Microscope system (Zeiss) using ZEISS Plan-APOCHROMAT 20×/0.95 Autocorr Objective and a ZEISS Plan-APOCHROMAT 20×/0.95 Autocorr Objective with a 2× tube lens. Images were acquired with an Axiocam 506 monocamera and the LSM 900 with Airyscan 2 imaging software. For SG index quantification in cells in culture, a z-stack of 12 serial optical sections every 0.49 μm at 20× or of 25 serial optical sections every 0.2 μm at 40× or 20× were captured in 15 to 30 nonoverlapping fields randomly positioned in grid format to span the coverslip. For SG quantification in tissues, a z-stack of 25 serial optical sections was captured every 0.2 μm at 40× in 30 random nonoverlapping fields with a similar percentage of epithelial composition. For tissue immunofluorescence (mean intensity and relative tumor area), a z-stack of six serial optical sections was captured every 1 μm for each field of view. Nonoverlapping fields of view were randomly positioned to span the tissue section in a grid-like fashion imaging ∼50% of each tissue section. For SRPK2, localization to SGs imaging was performed with an Axio Observer 7 microscope (ZEISS) with a Plan-Apochromat 63×/1.40 Oil M27 Oil objective and an Axiocam 702 camera.

Quantification of the SG index was performed as previously described (18). Quantification of the percentage of cells positive for SGs was based on a manual count of a minimum of 200 cells per experimental condition in a minimum of five random fields of view, each for at least three independent experiments. Quantification of protein levels in tissue sections was done by measuring the fluorescence intensity using ImageJ (https://imagej.nih.gov/ij, RRID:SCR_003070). Z-stacks from tissue sections were projected onto one maximum intensity projection image per field of view. Subsequently, the mean intensities of each field of view were averaged, and the background was subtracted to give the mean intensity of each tissue section. For the quantification of tumor area positive for Ki-67 and CC3, maximum projection images were thresholded to black and white, and subsequently used to determine the tumor area positive for the protein of interest or the total tumor area per field of view. Batch processing was utilized to apply set thresholds to all images from the same experiment. The relative Ki-67– or CC3-positive area per tumor was computed as a fraction of the total tumor area for each field of view and then averaged across all fields. The same method was applied to determine the cell area positive for CC3 in a minimum of eight random fields of views in each condition and then averaged across all fields.

KPC cells were induced with Dox (1 μg/mL) 3 days prior to seeding in a 24-well plate. Dox was maintained in the media for the duration of the experiments. Oxidative stress was induced by 300 μmol/L SA for 6 hours. Imaging was performed using a Celldiscoverer 7 Automated Microscope system (Zeiss) with a ZEISS Plan-APOCHROMAT 20×/0.95 Autocorr Objective. A z-stack of seven serial optical brightfield sections was captured every 1 μm, in a minimum of five random fields, hourly. The percentage of dead cells was derived by manual count based on morphology and blebbing and is the mean of five random fields of view for each condition.

Cells were lysed in 2× Laemli sample buffer. All protein samples were denaturized by heating at 95°C for 5 minutes. Protein samples were then separated by SDS-PAGE gel electrophoresis and transferred to nitrocellulose membranes (Odyssey Nitrocellulose Membranes; Licor), blocked with TBS-BSA 3% (w/v) at RT for 20 minutes, incubated with primary antibodies overnight at 4°C, and followed by 1 hour with secondary antibodies coupled with an Alexa Fluor dye 680 or 800. Fluorescence detection and quantification of immunoblot bands was achieved using the LI-COR imager and software (Li-COR Biosciences). The following antibodies were used: G3BP2 (1:1,000, Thermo Fisher Scientific; cat. #A302-033A, RRID:AB_1576539), G3BP1 (1:20,000, Bethyl Laboratories; cat. #A302-033A, RRID:AB_1576539), Pumilio (1:5,000, Abcam; cat. #ab10361, RRID:AB_297098), EIF4G (1:1,000, Cell Sig­naling Technology; cat. #2498S, RRID:AB_2096025), Alpha Tubulin (1:5,000, Sigma-Aldrich; cat. #T6199, RRID:AB_477583), eIF2α (1:1,000, Cell Signaling Technology; cat. #9722S, RRID:AB_2230924), phospho-eIF2α (1:1,000, Cell Signaling Technology; cat. #3398P, RRID:AB_2096481), Cox2 (1:500, Abcam; cat. #ab15191, RRID:AB_2085144), TIAL1 (1:1,000, Abcam; cat. #ab169547, RRID:AB_2910194), turbo GFP (1:2,000, Origene Technologies; cat. #TA150041, RRID:AB_2622256), GFP (1:1,000, Cell Signaling Technology; cat. #2555S, RRID:AB_10692764), HPGD (1:250, Abcam; cat. #ab187161, RRID:AB_2861359), IGF1 (1:500, R&D Systems; cat. #AF-291-NA, RRID:AB_2122119), and 1:1,000 for phospho-IGF1R (Abcam; cat. #ab39398, RRID:AB_731544), IGF1R (Cell Signaling Technology; cat. #3027S, RRID:AB_2122378), phospho-AKT (Cell Signaling Technology; cat. #4060S, RRID:AB_2315049), AKT (Abcam; cat. #ab8805, RRID:AB_306791), phospho-ERK1/2 (Cell Signaling Technology; cat. #4370S, RRID:AB_2315112), ERK1/2 (Cell Signaling Technology; cat. #9107S, RRID:AB_10695739), phospho-S6K1^thr389 (Aviva Systems Biology; cat. #OAAF07416, RRID:AB_2630782), SRPK2 (Proteintech; cat. #25417–1-AP, RRID:AB_2880068), and phospho-SRPK2^ser494 (EMD, Millipore; cat. #07–1817, RRID:AB_11205747).

Cells were induced with Dox (1 μg/mL), and Dox was replenished every 48 hours for the duration of the experiment. For growth curves, 5,000 cells were plated in a 24-well plate and fixed in 4% PFA at days 0, 2, 4, and 6. All samples were stained with SYTO 60 red fluorescent nucleic acid stain (5 mmol/L stock solution; Fisher Scientific, S11342) at 1/5,000 in 0.1% Triton/PBS for 1 hour. Results were imaged, and relative cell density was quantified with a Licor system.

DIO mice were fed with rodent high-fat (60% kcal from fat) chow (Research Diets, D12492i) during the entire experiment. Standard mice received standard chow. As female mice are not susceptible to DIO, ST and DIO mice for the comparative studies were male of 11 to 13 weeks of age. ob/ob mice were male and female of 5 to 7 weeks of age at the time of orthotopic implantation for IVIS experiments; ST mice for comparative studies were age-matched. Mice were weighed before cell implantation and at endpoint. For IVIS and survival experiments, mice were also weighed twice a week throughout the entire duration of experiments. For all orthotopic implantations, 50,000 cells resuspended in a 1:1 solution of DMEM and Matrigel at a final volume of 50 μL were injected in the tail of the pancreas. Prior to each surgery, mice were injected with Buprenex at 0.08 mg/kg (Buprenorphine Hydrochloride, injection 0.3 mg/mL; NDC 42023-179-05). For each surgery, mice were anesthetized with inhalation of ∼3% v/v vaporized isoflurane. Mice implanted with inducible cell lines received Dox in their water starting 24 hours after implantation at a final concentration of 200 μg/mL; Dox–water was changed every 3 days. At endpoint, tumors were weighed, measured, and pictured.

hPDAC cell lines (1M/flank; MiaPaCa-2 and HPAC) were resuspended in a 1:1 solution of PBS and Matrigel in a final volume of 100 μL and were implanted subcutaneously in both left and right flanks of 6- to 8-week-old male and female nude mice. Tumor volume was determined using electronic calipers to measure length (l), width (w), and the formula (w² × l)/2. Experiments were terminated when tumor volume in control mice reached ∼2,000 mm³.

For bioluminescence measurements, animals were anesthetized with inhalation of ∼3% v/v vaporized isoflurane. All mice were imaged with a positive control (orthotopically implanted with an NT shRNA cell line). To detect orthotropic pancreatic tumors, mice were injected with XenoLight D-Luciferin – K+ Salt Bioluminescent Substrate (15 mg/mL stock; D-luciferin potassium salt, Gold Biotech) by i.p. injection at a final concentration of 50 mg/kg. The bioluminescence images were acquired 8 minutes after IP substrate administration using the IVIS Lumina XR system. Measurements were set up as follows: exposure times ranged from 5 seconds to 1 minute (auto settings); binning on medium; f/stop at 1. The gray scale photograph and pseudocolor luminescent images were superimposed for the identification of the location of any bioluminescent signal of the labeled cells. Optical images were displayed and analyzed with IVIS Living Image software packages. Regions of interest (ROI) were drawn to assess the relative signal intensity emitted. Bioluminescent signal was expressed as photon intensity in units of photons/second (p/s) within the ROI.

After KPC cell implantation, mice were tracked every day after surgery until death. Survival studies were performed using humane endpoints including a 20% loss in body weight, loss of locomotor activity, or any sign of distress. Mice surviving over 300 days after KPC cell implantations were euthanized.

Experimental treatments with PPP were performed by i.p. injections (20 mg/kg/12 hours) of the compound in 10 μL volume of DMSO:vegetable oil, 10:1 (v/v). For experimental treatments with the S6K1 inhibitor, PF-4708671 (MedChem, HY-15773) was dissolved in 10% DMSO (Thermo Fisher Scientific, D1391) first and further diluted in 30% PEG400 (Selleck Chemicals, S6705), 0.5% Tween 80 (Selleck Chemicals, S6702), and 5% propylene glycol (Sigma-Aldrich, 528072) to achieve a final DMSO concentration of 1%. Vehicle was made of 10% DMSO diluted in 30% PEG400, 0.5% Tween 80, and 5% propylene glycol. Mice were then injected at a final concentration of 50 mg/kg daily for 35 (KPC-6560) or 40 (KPC-4662) days, and tumor development was assessed with the IVIS Lumina XR system.

To assess the target engagement of PF-4708671, DIO mice with established tumors (day 21 after implantation) or ST mice with established tumors (day 27 after implantation) were injected for 3 days with vehicle or PF-4708671 50 mg/kg. Subsequently, mice were euthanized, tumors were excised, and tissues were processed and stained for phosphorylated S6 as described above.

Mouse Igf1 antibody (R&D Systems; cat. #AF791, RRID:AB_2248752) and IgG control (R&D Systems; cat. #AB-108-C, RRID:AB_354267) were resuspended in PBS and administered to mice at a final concentration of 0.1 μg/g by i.p. injections. DIO and ST mice were injected daily for 3 days, starting at days 20 and 27 after KPC-4662 implantation, respectively. Recombinant mouse Igf-I/Igf1 protein (R&D Systems; cat. #791-MG) was resuspended in PBS and administered to ST mice at 100 ng/mouse by i.p. injection; PBS was used as a control. Mice were injected once on ∼day 29 after KPC-4662 implantation. Subsequently, mice were euthanized, tumors were excised, and tissues were processed and stained for phosphorylated S6 and the SG index as described above.

Human tissue samples of 65 patients with histologically confirmed PDAC were provided by the tissue bank of the University Medical Center Mainz in accordance with the regulations of the tissue biobank and the approval of the ethics committee of University Medical Center Mainz (2019-14390; Landesärztekammer RLP). All patients were therapy naïve and underwent primary surgical resection. Of this cohort, a tissue microarray was created. To overcome heterogeneity, four array spots of each tumor sample were generated (two center, two periphery; diameter: 1 mm). Clinical and pathologic data, including the BMI of the patients, were obtained. Tissue microarrays were immuonstained with anti-CK5, anti-G3BP1, and DAPI, and 50% of each tumor-center spot was imaged at 40×. Imaging and analysis were performed blinded. The SG index in CK8-positive cells was derived as described above.

Statistical analysis was performed with GraphPad 8 Prism Software (GraphPad Prism, RRID:SCR_002798). Statistical significance was determined with the Mann–Whitney two-tailed unpaired t test for nonparametric values and the Student two-tailed unpaired t test for parametric values. For tumor growth tracking with IVIS, data were analyzed by two-way ANOVA (time/total flux). P < 0.05 was considered statistically significant.

Requests for data and reagents may be directed to and will be fulfilled by the corresponding author, Dr. Elda Grabocka (elda.grabocka@jefferson.edu).

No disclosures were reported.

G. Fonteneau: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing–original draft. A. Redding: Formal analysis, validation, investigation, visualization, methodology, writing–original draft. H. Hoag-Lee: Formal analysis, validation, investigation, visualization, methodology, writing–original draft. E.S. Sim: Investigation, methodology. S. Heinrich: Resources, investigation, methodology. M.M. Gaida: Resources, data curation, funding acquisition, investigation, methodology, writing–review and editing. E. Grabocka: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration.

This study was supported by an NIH/NCI R37CA230645 grant and a V Scholar Award to E. Grabocka and grants from the German Research Foundation (Ga 1818/2-3, SFB1292 TPQ1) to M.M. Gaida. The Thomas Jefferson University Flow Cytometry Core, Translational Research/Pathology Core, and Animal Core are shared facilities partially supported by the National Cancer Center Support Grant (P30 CA056036) to the Sidney Kimmel Cancer Center at Jefferson.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.