Parallel Signaling through IRE1α and PERK Regulates Pancreatic Neuroendocrine Tumor Growth and Survival

When misfolded proteins in the endoplasmic reticulum (ER) accumulate above a critical threshold, the unfolded protein response (UPR) is initiated to restore homeostasis. The UPR is controlled by three ER transmembrane proteins, inositol-requiring enzyme 1α (IRE1α/ERN1), PKR-like ER kinase (PERK/EIF2AK3), and activating transcription factor 6 (ATF6), that detect misfolded proteins, expand ER protein-folding capacity, and decrease protein-folding demand (1–4).

The most ancient UPR sensor, IRE1α, contains an ER lumenal domain that recognizes unfolded proteins and undergoes dimerization or oligomerization depending on the degree of engagement (5–7). Remediable ER stress causes dimerization and subsequent trans-autophosphorylation of the cytosolic kinase domain, which allosterically activates the attached endoribonuclease (RNase) and initiates frameshift splicing of XBP1 mRNA. Resulting translation of the XBP1s (s, spliced) transcription factor upregulates genes encoding ER protein-folding and quality control components (8, 9). Analogously, recognition of misfolded proteins by the lumenal domain of PERK results in dimerization, trans-autophosphorylation, and activation of its cytosolic kinase domain (10, 11). PERK then phosphorylates eIF2α/EIF2S1, which downregulates Cap-dependent translation, and reduces ER protein load (12, 13). Concurrently, transcripts with upstream open reading frames (uORF), such as activating transcription factor 4 (ATF4), are preferentially translated and further promote ER protein folding and quality control (14). Finally, during ER stress, ATF6 translocates to the Golgi and is cleaved to release its N-terminal transcription factor domain, which upregulates ER quality control factors (15). Collectively, remediable UPR signaling induces an “Adaptive (A)-UPR,” which restores ER homeostasis (16).

However, under sustained, irremediable ER stress, the UPR regulators switch from prohomeostatic to proapoptotic outputs (17). High-level kinase autophosphorylation causes IRE1α oligomerization and relaxed RNase specificity, leading to regulated IRE1-dependent decay (RIDD) of ER-localized mRNAs that encode secretory proteins and essential components of the protein-folding machinery (18, 19). This activity not only results in active deterioration of ER function (18), but also degrades select miRNA precursors to upregulate key apoptotic signals, including thioredoxin-interacting protein (TXNIP; refs. 1, 20). Similarly, prolonged PERK activation results in upregulation of the transcription factor CHOP/GADD153, which attenuates expression of antiapoptotic BCL-2 and increases expression of proapoptotic BCL2 family proteins to promote cell death (21, 22). Finally, ATF6 has been reported to reduce expression of the antiapoptotic protein Mcl-1 and upregulate CHOP (23, 24). Unmitigated ER stress thus culminates in a “Terminal (T)-UPR,” whereby adaptive signaling through XBP1s, ATF4, and ATF6 are eclipsed by proapoptotic signals (Fig. 1A).

Elevated ER stress and UPR activity have been documented in various solid cancers, such as glioblastoma and carcinomas of the breast, stomach, esophagus, and liver (25–29). However, whether UPR signaling in these various neoplasms ultimately inhibits or promotes tumor growth remains an area of intense debate (30–34). We speculated that pancreatic neuroendocrine tumors (PanNET) might be particularly sensitive to protein-folding stress. Derived from endocrine cells, PanNETs universally hypersecrete one or more polypeptide hormones such as insulin or glucagon (35, 36). Moreover, the development and maintenance of pancreatic neuroendocrine cells is greatly impacted by genetic loss of the UPR in mice and humans (37–39).

Here, we show that the Adaptive-UPR, especially through the IRE1α and PERK arms, is strongly upregulated in human PanNET samples and in two distinct murine models. Using genetic tools, we discovered that disruption of the Adaptive-UPR or activation of the Terminal-UPR is detrimental to PanNET growth and survival in vivo. Likewise, administration of highly selective IRE1α and PERK kinase inhibitors in two murine preclinical PanNET models phenocopies the antitumor effects of genetic deletion. Specifically, inhibiting IRE1α or PERK exacerbates ER stress and leads to apoptotic signaling through the reciprocal UPR branch. Critically, pharmacologic targeting of IRE1α increased life expectancy without deleterious effects on animal health, highlighting its therapeutic potential in PanNETs and other ER stress–sensitive cancers.

We obtained 6 formalin-fixed paraffin-embedded samples of deidentified primary human PanNETs, 4 frozen human PanNET samples for RNA analysis, and matched normal pancreata from the UCSF Department of Pathology, IRB protocol number 13-11606.

Parental INS-1 cells were a gift from the laboratory of Gerhart Ryffel at Institut für Zellbiologie (Tumorforschung), Universitätsklinikum Essen. Expression of the pancreatic neuroendocrine markers insulin, synaptophysin and chromogranin A was confirmed frequently over the course of the project to authenticate the identity of the INS-1 cells. Generation of isogenic, stable INS-1 lines was performed as described previously (18). Mycoplasma testing was performed on parent stocks with the MycoAlert Mycoplasma Detection Kit (Lonza) and cells were verified to be Mycoplasma free. Cells were cultured in RPMI1640 media supplemented with 10% tetracycline-free FBS (Gemini Bioproducts 100-800), 10 mmol/L HEPES, 2 mmol/L glutamine, 110 μg/mL Na Pyr, 100 U/mL penicillin, 100 μg/mL streptomycin (UCSF Cell Culture Facility), and 100 μmol/L βME (Bio-Rad 161-0710). All cell stocks were used for ≤16 passages. Brefeldin A (9972S) was purchased from Cell Signaling Technology; tunicamycin (T7765), thapsigargin (T9033), and doxycycline (D9891) were purchased from Sigma-Aldrich.

KIRA8 was synthesized in-house and purified by reverse-phase chromatography (HPLC). The purity of KIRA8 was determined with two analytical RP-HPLC methods, using a Varian Microsorb-MV 100-5 C18 column (4.6 mm × 150 mm), and eluted with either H₂O/CH₃CN or H₂O/MeOH gradient solvent systems (+0.05% TFA) run over 30 minutes. Products were detected by UV at 254 nm. KIRA8 was found to be >95% pure in both solvent systems. GSK2656157 (GSK-PKI) was purchased at >98% purity from Advanced Chemblocks Inc. For use in tissue culture, KIRA8 and GSK-PKI stock solutions were prepared by dissolving in DMSO at a concentration of 20 mmol/L. For use in animal studies, KIRA8 was dissolved in a vehicle solution of 3% ethanol, 7% Tween-80, 1.2% ddH₂O, and 88.8% of 0.85% W/V saline at a working concentration of 10 mg/mL; GSK-PKI was dissolved in 5% 1-methyl-2-pyrrolidinone (NMP), 5% Kolliphor HS 15 (Solutol), and 90% of 20% W/V (2-Hydroxypropyl)-β-cyclodextrin (HP-β-CD) at a working concentration of 10 mg/mL.

Guide RNAs were designed using the Zhang Lab's Optimized Design Tool (crispr.mit.edu) and targeted to the 5′ end of each gene to create random insertions/deletions (indels) upstream of key structural and functional domains. For each gRNA, forward and reverse oligonucleotides were purchased from Integrated DNA Technologies (Supplementary Table S1), annealed, and ligated into the pSpCas9(BB)-2A-EGFP vector (pX458; a gift from Feng Zhang, Broad Institute of Massachusetts Institute of Technology and Harvard University, Cambridge, MA; Addgene Plasmid #48138) at the BbsI cloning site. The resulting plasmids were transfected into INS-1 cells using Lipofectamine 2000 (Thermo Fisher Scientific); a BD FACSAria II (BD Biosciences) was used to subsequently single-cell sort EGFP-positive cells and establish clonal lines. Clones were screened for knockout of target genes by Western blot analysis and/or by allelic sequencing with custom primers (Integrated DNA Technologies; Supplementary Table S2) after processing genomic DNA with the KAPA Mouse Genotyping Kit (KAPA Biosystems KK7352) and TOPO TA Cloning Kit (Thermo Fisher Scientific K457501).

For protein analysis, cells were lysed in T-PER buffer (Thermo Fisher Scientific 78510) plus phosphatase inhibitor cocktail (Cell Signaling Technology #5872S). Protein concentration was determined using Pierce BCA Protein Assay (Thermo Fisher Scientific 23225). Western blots were performed using 10% and 4%–12% gradient Bis-Tris precast gels (NuPage) on Invitrogen XCell SureLock Mini-Cell modules. Gels were run using 2-(N-morpholino)ethanesulfonic acid (MES) buffer (Invitrogen NP000202) and transferred onto nitrocellulose transfer membrane using an XCell II Blot Module or iBlot Dry Blotting System (Thermo Fisher Scientific). Antibody binding was visualized on CL-XPosure film using ECL SuperSignal West Extended Duration Substrate (both from Thermo Fisher Scientific) or using the Odyssey CLx Imaging System (LI-COR Biosciences). Antibodies used: ATF4 (Cell Signaling Technology #11815), BiP (CST #3177), CHOP (CST #2895), eIF2α (CST #9722), phospho-eIF2α (CST #3398), IRE1α (CST #3294), PERK (CST #3192), PERK p-T980 (CST #3179), Spliced XBP-1 (BioLegend #619502), and actin (Sigma A5441 1:3000).

RNA was isolated from whole cells using either Qiashredder and RNeasy Kits (Qiagen) or TRIzol (Invitrogen). TissueLyser LT (Qiagen) was used for RNA isolation from tumors. For cDNA synthesis, 500 to 1,000 ng total RNA was reverse transcribed using Superscript II Reverse Transcriptase and Oligo d(T)₁₆ primer (Invitrogen). For qPCR, we used Power SYBR Green and the StepOnePlus Real-Time PCR System (Applied Biosystems). qPCR primers are listed in Supplementary Table S3. Gene expression levels were normalized to actin.

RNA was isolated from whole cells or tissue and reverse transcribed as detailed above to obtain total cDNA. Sense (5′-AGGAAACTGAAAAACAGAGTAGCAGC-3′) and antisense (5′-TCCTTCTGGGTAGACCTCTGG-3′) primers were used in a standard GoTaq Green PCR reaction (Promega) to amplify a region spanning the 26-nucleotide intron that includes a single PstI restriction site, which is excised by active IRE1α. The resulting PCR fragments were then digested by PstI (New England Biolabs), resolved on 3% agarose gels, stained with ethidium bromide, and quantified by densitometry using ImageJ (NIH, Bethesda, MD).

To measure apoptosis by Annexin V staining, cells were plated in 12-well plates overnight. Cells were then treated as described for indicated times. On the day of analysis, cells were trypsinized, washed in PBS, and resuspended in Annexin V binding buffer (10 mmol/L HEPES 7.4, 140 mmol/L NaCl, 2.5 mmol/L CaCl₂) with Annexin-V FITC (BD Biosciences 556419). Flow cytometry was performed on a Becton Dickinson LSRFortessa or LSRII flow cytometer. To measure cell proliferation, cells were seeded at 5%–10% confluence in 96-well plates, treated as indicated, and assayed using the CellTiter-Glo Luminescent Cell Viability Assay (Promega) according to the manufacturer's protocol. Luminescence was quantified using a Cytation 5 Cell Imaging Multi-Mode Reader (BioTek).

All animal studies were reviewed and approved by the UCSF Institutional Animal Care and Use Committee. Animals were maintained in a specific pathogen-free animal facility on a 12-hour light/dark cycle at an ambient temperature of 21°C. They were given free access to water and food.

Five- to 8-week-old NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NSG, Stock #005557, The Jackson Laboratory) mice were injected subcutaneously with 5 × 10⁶ INS-1 cells, and tumor size was followed for up to 4 weeks. Where indicated, animals were provided doxycycline chow (Envigo TD.09761). For small-molecule studies, KIRA8, GSK-PKI, or the corresponding vehicle solutions were prepared as described above and delivered daily by intraperitoneal injection.

Tg(RIP1-Tag)2Dh mice (previously described in ref. 40) were initially obtained from the Bergers Laboratory at UCSF and maintained as heterozygotes by breeding wild-type C57BL/6 female mice with hemizygous RIP-Tag2 (RT2) male mice. RT2-positive mice were given supplemental diet with adjusted sucrose starting at 12 weeks (Envigo TD.86489). KIRA8 and GSK-PKI treatments, as detailed above, were initiated at 12 weeks and continued as described.

After completion of treatment (14 days), mice were starved for 5 hours and blood glucose levels were measured to mark time = 0. Mice were then injected with 1 mg/g D-glucose (Thermo Fisher Scientific A24940-01) and their blood glucose monitored over 180 minutes. To monitor blood glucose levels, a drop of blood was collected from the tail onto OneTouch Ultra Blue test strips and measured using the OneTouch Ultra 2 Meter (LifeScan). For plasma analysis, blood was collected by retro-orbital bleed into EDTA-coated tubes (BD #365974) and analyzed using Amylase Activity Assay Kit (Sigma MAK009-1KT) and Lipase Acitvity Assay Kit (MAK046-1KT) according to the manufacturer's protocols.

Samples were fixed in 4% buffered formaldehyde for 24 hours, washed in PBS, transferred into 70% EtOH in ddH₂O, and then embedded in paraffin and sectioned (5-mm thickness) using a Leica RM2255 rotary microtome, by the Brain Tumor Research Center (BTRC) Histology Core, or by the Biobank and Tissue Technology Lab at UCSF. Hematoxylin and eosin staining was performed using standard methods. Stained slides were imaged using an Aperio AT2 slide scanner and data were processed using QuPath software. Total cell counts, Ki67 stains in all tissues, and cleaved caspase-3 stains in RT2 tumors were automated in a blinded fashion; cleaved caspase-3 stains in INS-1 xenografts were quantified manually in a blinded fashion. Antibodies used for IHC: BiP (C50B12; Cell Signaling Technology #3177, 1:200), CD31 (Cell Signaling Technology #77699 1:100), chromogranin A (polyclonal; Cell Marque, 1:4), cleaved caspase-3 (Cell Signaling Technology #9661 1:200), insulin (DAKO A0564, 1:200), IRE1α (Cell Signaling Technology #3294, 1:100), Ki67 (Ventana #790-4286, Undiluted), Myc (Sigma M4439, 1:5,000), synaptophysin (LK2H10 clone; Cell Marque, 1:100).

Graphs were generated using Prism 6 software and represent the average of at least 3 independent experiments. Data are expressed as means ± SD for bar graphs and dot plots when n ≥ 5; for n < 5, error bars were excluded. Box plots show min., Q₁, median, Q₃, and max. Statistical significance was indicated when P < 0.05. Two-tailed Welch t test used for direct comparison between two groups (equal variance not assumed); one-way ANOVA with Tukey's multiple comparison test used for ≥3 groups; two-way ANOVA with Tukey or Sidak multiple comparison test used when two factors (independent variables) were analyzed; log-rank (Mantel–Cox) test used for survival analysis.

We obtained a panel of six human PanNETs (stage I) and performed IHC against the ER chaperone BiP/GRP78, which is upregulated by the Adaptive-UPR. We observed markedly higher BiP expression in 5 of the 6 human PanNETs compared with normal pancreas (Fig. 1B and C; Supplementary Fig. S1A). Moreover, XBP1 splicing (IRE1α signaling) and ATF4 mRNA expression (PERK signaling) were upregulated in human PanNETs compared with normal pancreas (Fig. 1D and E). We were unable to assess the activation state of ATF6 because all commercially available antibodies tested showed nonspecific staining by IHC.

To recapitulate UPR signaling in vivo, we employed a PanNET xenograft mouse model using rat insulinoma (INS-1) cells (41, 42). We injected INS-1 cells subcutaneously into the flank of immunodeficient NOD-scid IL2Rgamma^null (NSG) mice (Fig. 1F). Tumors became palpable at 1 to 2 weeks postinjection, grew locally without metastasizing (stage I), and closely resembled human PanNETs by histology and IHC for known markers (Fig. 1G–J). Moreover, CD31 staining demonstrated that the INS-1 xenografts showed similar vascular patterns to human PanNETs (Fig. 1K). Notably, compared with INS-1 cells grown in vitro, xenograft tumors displayed marked upregulation of IRE1α (XBP1s) and PERK (p-PERK) signaling (Fig. 1L; Supplementary Fig. S1B and S1C).

We previously engineered transgenic INS-1 cell lines stably integrated with doxycycline (Dox)-inducible constructs of Myc-tagged IRE1α (18, 26). In vitro, Dox induces supraphysiologic production of transgenic IRE1α, which spontaneously oligomerizes through mass action to induce Xbp1 splicing, ER-localized mRNA decay, and a Terminal-UPR (18). To test the effects of IRE1α hyperactivation in vivo, we injected INS-1::Vector and INS-1::Myc-IRE1α cells into the flanks of NSG mice and provided either regular or Dox chow (Fig. 1F). To study long- and short-term effects of IRE1α expression, Dox induction was provided either for the entire 4-week duration of tumor growth or for a 48-hour interval 2 weeks after tumor cell injection. Dox chow alone had no effect on the size of control INS-1 tumors over a 4-week time course; in contrast, Dox-induced overexpression of IRE1α markedly reduced tumor mass to <30% of control tumors (Fig. 2A–C; Supplementary Fig. S2A) and led to significant increases in both adaptive (Xbp1s; Fig. 2D; Supplementary Fig. S2B) and apoptotic/RIDD outputs (elevated Txnip; decreased Ins1 and Ins2; Fig. 2E–G; Supplementary Fig. S2C and S2D). Even short-term expression of IRE1α was sufficient to induce robust apoptosis as visualized by cleaved caspase-3 staining (Fig. 2H–I; Supplementary Fig. S2E and S2F). Interestingly, IRE1α overexpression resulted in some degree of PERK phosphorylation, perhaps through crowding effects at the ER membrane. However, this was not associated with activation of the downstream effector ATF4 or detectable levels of proapoptotic CHOP (Supplementary Fig. S2G and S2H).

Intriguingly, INS-1::IRE1α tumors were consistently larger than the INS-1 vector controls in the absence of Dox (Fig. 2A). The most likely explanation is that leaky expression of the Ire1α transgene (Fig. 2B and C; Supplementary Fig. S2A and S2E) promoted a modest increase in Xbp1 splicing without triggering IRE1α's apoptotic outputs, as seen in vitro (Fig. 2D–I; Supplementary Fig. S2B–S2D; ref. 18). These results support the notion that the balance of adaptive and apoptotic signals from IRE1α is optimized by PanNETs to favor growth and avert cell death.

Just as hyperactivation of IRE1α impairs tumor development, we reasoned that inhibition of UPR pathways would also unbalance UPR signaling and impede tumorigenesis. To initially test this concept, we used CRISPR/Cas9 gene editing to functionally knock out Ire1α, Xbp1, and Perk in INS-1 cells (Fig. 3A; Supplementary Fig. S3A). Interestingly, proliferation and survival rates of cultured INS-1 KO lines were not diminished in comparison to the parental INS-1 cells (Fig. 3B), suggesting that full UPR functionality is dispensable in vitro. We next injected the INS-1 KO cell lines into NSG mice and allowed tumors to develop for 4 weeks. Although scrambled (Scr) CRISPR/Cas9 controls achieved a tumor mass no different from that of the parental INS-1 cells (Supplementary Fig. S3B), all UPR KO clones showed markedly impaired tumor growth, in some cases attaining only approximately 10% of the mass of control tumors (Fig. 3C–H). This correlated with a >65% reduction of actively proliferating cells, as determined by Ki67 staining (Fig. 3I and J). Neither differences in apoptosis nor evidence of significant signaling compensation in IRE1α and PERK KOs were detected at the 4-week endpoint of the experiment (Supplementary Fig. S3C–S3E).

Our team developed ATP-competitive IRE1α Kinase Inhibiting RNase Attenuators, KIRAs, that bind IRE1α's kinase domain and allosterically inhibit its RNase (43, 44). We resynthesized a monoselective IRE1α inhibitor from a recent KIRA series (compound 18; ref. 45), which has recently been renamed KIRA8 (Supplementary Fig. S4A; ref. 46). Administration of KIRA8 to cultured INS-1 cells revealed an IC₅₀ of approximately 125 nmol/L; doses between 500 nmol/L and 1 μmol/L resulted in near-complete inhibition of Xbp1 splicing, reversal of RIDD (Ins1 mRNA decay), and prevention of apoptosis due to supraphysiologic IRE1α signaling (Supplementary Fig. S4B–S4D).

Correspondingly, we obtained a highly selective PERK inhibitor, GSK2656157, referred to here as GSK-PERK Kinase Inhibitor (GSK-PKI; Supplementary Fig. S4E; ref. 47). GSK-PKI displayed minimal off-target inhibition against a panel of 300 kinases in vitro, including other eIF2α kinases (47). Under conditions of elevated ER stress in cultured INS-1 cells, GSK-PKI also exhibited an IC₅₀ of approximately 125 nmol/L with near-optimal inhibition of PERK autophosphorylation and CHOP expression between 500 nmol/L and 1 μmol/L (Supplementary Fig. S4F).

Even at doses that essentially eliminate Xbp1 splicing, extensive exposure of cultured INS-1 cells to KIRA8 neither affected their viability (Fig. 4A) nor their proliferation (Fig. 4B). This is consistent with the original report (45) and our findings that IRE1α activity is relatively low and dispensable in cultured cells (Figs. 1L and 3B). Conversely, GSK-PKI treatment mildly increased apoptosis and dampened growth rate in cultured INS-1 cells (Fig. 4B and C). This discrepancy may stem from pathway compensation in PERK KO cells or off-target GSK-PKI effects (48), but the aggregate effect is modest under low ER stress.

To mildly induce ER stress and activate both IRE1α and PERK pathways, we used thapsigargin (ER Ca²⁺ pump inhibitor; Fig. 4G; Supplementary Fig. S4G) alone or in combination with 500 nmol/L of KIRA8 or GSK-PKI. Under these conditions, the addition of either inhibitor roughly doubled the number of apoptotic cells over 30 hours of treatment (Fig. 4D), raising the possibility of compensatory signaling through a parallel pathway. Although treatment with KIRA8 did not noticeably affect PERK autophosphorylation, high baseline levels may have masked any subtle changes (Fig. 4E). However, KIRA8 increased thapsigargin-induced CHOP expression (Fig. 4E) and sustained p-PERK and CHOP levels after thapsigargin washout (Fig. 4F; Supplementary Fig. S4H and S4I). Inversely, treatment with GSK-PKI boosted Xbp1 splicing, both with and without ER stress induction, indicating compensation through IRE1α (Fig. 4G).

We next tested the effectiveness of pharmacologic UPR inhibitors in the INS-1 xenograft model by administering doses of KIRA8 or GSK-PKI estimated to achieve serum concentrations above IC₅₀. Tumors were grown in NSG mice for two weeks, treated for 48 hours, harvested, and analyzed for target inhibition by their respective drugs. Dosing with KIRA8 lowered Xbp1 splicing from a baseline of approximately 15% to below 1.5% (Fig. 5A; Supplementary Fig. S5A); GSK-PKI treatment reduced PERK autophosphorylation below detectable levels (Fig. 5B).

To determine effects on tumor burden, we injected NSG mice with INS-1 cells and the following day began administering daily doses of KIRA8 or GSK-PKI for three weeks prior to harvest. Although less dramatic than IRE1α or PERK KO tumors, both KIRA8 and GSK-PKI significantly reduced tumor mass by ∼50% (Fig. 5C–F). In contrast to cultured INS-1 cells, INS-1 tumors exhibited markedly decreased Ki67 staining within 48 h of KIRA8 or GSK-PKI administration in vivo (Fig. 6A–D). More specifically, KIRA8 and GSK-PKI blocked G₁-S transition by impairing expression of Cyclin E1 and D1, respectively (Supplementary Fig. S5B and S5C).

Moreover, in vivo administration of either compound increased levels of cleaved caspase-3 roughly 1.5-fold (Fig. 6E–H). Because of high basal activation in vivo, we were unable to directly detect further increases in p-PERK under KIRA8 treatment (Supplementary Fig. S5D) but did note increased levels of PERK-associated markers for ER stress (Ire1α) and apoptosis (Txnip, Bim/Bcl2l11; Fig. 6I; refs. 1, 49–51). In regard to GSK-PKI, cytotoxicity was associated with increased Xbp1 splicing (Fig. 6J; Supplementary Fig. S5E) and decreases in Ins1, Ins2, and Blos1/Bloc1s1 mRNA (Fig. 6K), implicating upregulation of adaptive and proapoptotic (RIDD) IRE1α signaling, respectively.

We next tested the effects of KIRA8 and GSK-PKI in a second preclinical PanNET model. The RIP-Tag2 (RT2) mouse is a transgenic strain in which viral SV40 large T-antigen (Tag) expression is driven by the β-cell specific rat insulin promoter-1 (RIP; ref. 40), leading to predictable development of islet hyperplasia (5–10 wks), adenomas (10–12 weeks), and eventually invasive metastases/tumors (12–15 wks; Fig. 7A; ref. 52). Similar to INS-1 xenografts, RT2 tumors also display elevated ER stress compared with wild-type (WT) pancreata. By 10 to 14 weeks Xbp1 splicing, IRE1α expression, and PERK phosphorylation are all increased (Fig. 7B and C; Supplementary Fig. S6A and S6B); eIF2α phosphorylation, which integrates multiple stress pathways, is unchanged at the tested time points.

RT2 animals were subjected to a tumor regression trial: we began vehicle or drug (KIRA8 or GSK-PKI) administration at 12 weeks of age, when tumors have become invasive (Fig. 7A), and continued therapy for 14 days (d). After the 2-week treatment period, we sacrificed the animals and measured tumor burden histologically as a percentage of total pancreas area (Fig. 7D and E). In KIRA8-treated animals, Xbp1 splicing was completely ablated (Supplementary Fig. S6C) and tumor burden was decreased >75% compared with vehicle-treated control animals (Fig. 7D and F). There was no visible damage to the surrounding pancreas. As in INS-1 xenografts (Fig. 6; Supplementary Fig. S5), KIRA8 treatment resulted in moderate decrease of Ki67 staining and roughly 2-fold increase of cleaved caspase-3 (Supplementary Fig. S6D–S6G).

GSK-PKI treatment was associated with widespread degeneration of the exocrine pancreas (Fig. 7E), confounding our ability to measure tumor burden histologically; however, direct measurement of tumor masses revealed that GSK-PKI also reduces tumor burden (∼50%) in RT2 animals (Fig. 7G). Surprisingly, short-term treatment of RT2 tumors did not reveal changes in Ki67 or cleaved caspase-3 staining despite complete inhibition of PERK activation (Supplementary Fig. S6H–S6L).

The deleterious impact of GSK-PKI on pancreatic health spurred us to scrutinize the effects of both compounds on WT mice. Twelve-week-old C57BL/6 mice were treated for 14 days with vehicle, KIRA8, or GSK-PKI before sacrificing and removing the pancreas. Weekly blood draws and a terminal glucose tolerance test were performed on a parallel cohort of animals. As expected, KIRA8-treated animals had no discernible damage or loss of pancreas mass (Fig. 7H; Supplementary Fig. S7A). Moreover, they recovered normally in a glucose tolerance test and did not display elevated levels of serum amylase or lipase (Supplementary Fig. S7B–S7D). However, they did experience an approximately 10% decline in body mass (Supplementary Fig. S7E). In contrast, GSK-PKI treatment resulted in severe disruption of the exocrine pancreas, a corresponding ∼50% loss of pancreas mass, and a small loss of body mass (Fig. 7I; Supplementary Fig. S7F–S7G. Unexpectedly, they recovered well during a glucose tolerance test and had normal levels of serum amylase and lipase (Supplementary Fig. S7H–S7J).

On the basis of these data, we performed a survival study following initiation of KIRA8 treatment at 12 weeks of age. Although the vehicle-treated animals lived an average of 17 days after initiation of treatment, the KIRA8 animals survived over twice as long, with several animals surviving over 60 days and one animal up to 82 days (Fig. 7J). Because RT2 mice typically did not show overt symptoms prior to death and pancreatic autolysis occurs rapidly, necropsies were not performed. However, all indications are that KIRA8-treated animals eventually succumbed due to PanNET growth similar to the vehicle-treated animals.

Currently, PanNETs are only potentially curable through surgical resection. For the 20% to 30% of patients diagnosed with metastatic disease, treatment is limited to managing symptoms of hormonal hypersecretion and systemic chemotherapy, to which the tumor invariably develops resistance. The 5-year survival for these patients is as low as 4% to 25% (53). Treatments with targeted therapies, such as the FDA-approved drugs everolimus (an mTOR inhibitor) and sunitinib (a multi-kinase inhibitor) have an approximately 6 month increase in progression-free survival in patients with metastatic PanNETs (54, 55). Recently, peptide receptor radiotherapy (PRRT) has been FDA approved for some patients with metastatic somatostatin receptor (SSTR)-expressing PanNETs, but not all patients respond and the long-term outcomes remain unknown (56).

Notably, professional secretory cells of the endocrine pancreas that PanNETs originate from are critically dependent on the UPR for their development and survival. Generally, deletions of Perk, Ire1α, and Xbp1 in mice lead to endocrine cell apoptosis and pancreatic dysfunction (37–39). However, it is likely that PanNETs are even more dependent on the UPR as they constitutively hypersecrete one or more hormones (35, 36); some “insulinoma” PanNETs secrete over 10-fold higher amounts of insulin than normal pancreatic β-cells (57). Based on these observations, we predicted that PanNETs would be particularly dependent on the UPR to manage ER stress.

Experiments using inducible IRE1α expression provided the first evidence that the level of UPR activation has important consequences for tumor growth in vivo: hyperactivation of IRE1α resulted in massive apoptosis, while moderate expression enhanced adaptive signaling and promoted tumor growth. As such, we hypothesized that downregulation of adaptive UPR signaling would also impede tumor development. Interestingly, neither genetic deletion nor pharmacological blockade of IRE1α or PERK pathways had profound effects on INS-1 growth in vitro, consistent with the lack of KIRA8 cytotoxicity previously observed in cultured cancer cell lines (45). Although cancer cells face unique challenges to ER proteostasis in vivo (e.g., hypoxia, nutrient deprivation), these factors are not mimicked in normal cell culture conditions. Demonstrating this point, we found that INS-1 cells grown in culture have low levels of ER stress and UPR activation compared with the same cells grown in mice as xenografts. By artificially inducing ER stress in cultured INS-1 cells, we increased their dependency on the Adaptive-UPR and sensitized them to KIRA8- and GSK-PKI-induced apoptosis. In vivo, IRE1α or PERK inhibition markedly impaired tumor development by decreasing proliferative capacity and inducing apoptosis.

Together, these data provide mechanistic insights into the anti-tumor effects of KIRA8 and GSK-PKI. In vitro, inhibiting IRE1α leads to higher sustained activation of PERK and its pro-apoptotic target CHOP. Inversely, PERK inhibition results in IRE1α hyperactivation. In both cases, this leads to loss of an adaptive arm of the UPR, exacerbates ER stress, and shifts the burden to the remaining branch, triggering its apoptotic (Terminal) program. Likewise, in vivo inhibiting one arm of the UPR, led to compensatory hyperactivation of the other: IRE1α inhibition by KIRA8 led to transcriptional upregulation of PERK-associated ER stress and apoptotic markers; PERK blockade by GSK-PKI increased IRE1α's adaptive and apoptotic outputs (RIDD). Although GSK-PKI also inhibits the kinase RIPK1, this is unlikely to contribute to the phenomena observed here as RIPK1 is proapoptotic and the drug phenocopies on-target, genetic knockout of Perk. Whether ATF6 also experiences compensatory activation is unclear with currently available tools and will require further analysis.

To determine whether our results generalize to other PanNET models, we tested KIRA8 and GSK-PKI in the well-characterized transgenic RT2 model. In use for over 20 years, this model has predicted the clinical efficacy of several compounds that have gone on to FDA approval for PanNETs, including everolimus and sunitinib. In contrast to the INS-1 xenograft model, RT2 tumors arise directly from endogenous pancreatic β-cells and develop in their natural environment. Although both KIRA8 and GSK-PKI are effective at reducing tumor burden in this model, the stark difference in their pancreatic toxicities sets them apart. Within two weeks, daily GSK-PKI administration to WT mice severely damaged and reduced pancreatic mass, similar to results obtained using oral administration of this PERK inhibitor in CD-1 mice (47).

Because genetic deletion of either Perk or Ire1α can lead to β-cell dysfunction and apoptosis, it was assumed that KIRAs would have similar toxicities to PERK inhibitors. However, inhibiting IRE1α with KIRA8 (or related compound KIRA6) was well tolerated and preserved pancreatic β-cell health in multiple diabetes models (44, 46). Likewise, KIRA8 administration had no noticeable effect on pancreatic mass or histology in WT C57BL/6 or RT2 mice while reducing tumor growth and increasing survival in the RT2 model. This discrepancy in KIRA8 toxicity between tumor and normal pancreatic tissues likely hinges on the postmitotic nature and low proliferation rate of endogenous pancreatic β-cells. As such, IRE1α emerges as a more attractive therapeutic target than PERK, though more studies are needed to understand the long-term effects of IRE1α inhibition.

Other facets of our data hint at alternative strategies for targeting IRE1α in PanNETs. For one, KIRA8 had a more dramatic effect on tumor burden in the RT2 model than in the INS-1 xenograft model. The difference in immunocompetence between NSG (xenograft) and C57BL/6 mice (RT2 background) may be a decisive factor. For example, recent reports suggest that targeting the IRE1α–XBP1 axis may engage a two-pronged attack that restrains malignant cells while simultaneously eliciting concomitant antitumor immunity (58). The recent explosion in immunotherapy suggests that the immune system has a more complex and widespread role in cancer than initially appreciated, and manipulating the UPR may enhance this approach (59, 60).

Together, our findings reveal many crucial aspects of studying and targeting the UPR in PanNETs. Both the intrinsic secretory nature of PanNETs and their surrounding environments contribute to their level of ER stress, forcing them to lean on the Adaptive-UPR for survival and making them susceptible to pharmacologic intervention. Ultimately, optimization of current inhibitors, exploration of combination therapies, and enhancement of UPR activators will diversify our treatment strategies for targeting the UPR in secretory cancers. Moreover, because this approach is not dependent on the presence of a somatic mutation in a UPR component, it may have benefits in many other solid tumors where ER stress is documented.

D.J. Maly is a co-founder and consultant at and has ownership interest (including patents) in OptiKira. B.J. Backes is a co-founder at and has ownership interest (including patents) in OptiKira. F.R. Papa is a co-founder and consultant at and has ownership interest (including patents) in OptiKira. S.A. Oakes is a co-founder and consultant at and has ownership interest (including patents) in OptiKira. No potential conflicts of interest were disclosed by the other authors.

Conception and design: P.C. Moore, J.Y. Qi, M. Thamsen, M.J. Gliedt, F.R. Papa, S.A. Oakes

Development of methodology: P.C. Moore, J.Y. Qi, R. Ghosh, D.J. Maly, B.J. Backes, F.R. Papa

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P.C. Moore, J.Y. Qi, M. Thamsen, R. Ghosh, J. Peng, R. Meza-Acevedo, R.E. Warren, G.E. Kim

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P.C. Moore, J.Y. Qi, M. Thamsen, A. Hiniker, G.E. Kim, F.R. Papa, S.A. Oakes

Writing, review, and/or revision of the manuscript: P.C. Moore, J.Y. Qi, R. Ghosh, F.R. Papa, S.A. Oakes

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S.A. Oakes

Study supervision: F.R. Papa, S.A. Oakes

The work was supported by grants: American Cancer Society Research Scholar Award (to S. Oakes); 2015 Caring for Carcinoid Foundation-AACR Grant for Carcinoid Tumor and Pancreatic Neuroendocrine Tumor Research, Grant Number 15-60-33-OAKE (to S. Oakes); Neuroendocrine Tumor Research Foundation (to S. Oakes); Harrington Discovery Institute Scholar-Innovator Award (to S. Oakes and F. Papa); UCSF and Onyx Oncology Innovation Alliance (to S. Oakes and F. Papa); Alfred P. Sloan Foundation (to D. Maly); Camille and Henry Dreyfus Foundation (to D. Maly); National Science Foundation (to J. Qi); NIH/NCI Brain Tumor SPORE (UCSF BTRC), the P50 CA097257 NIH/NCI Brain Tumor SPORE Grant. R01CA219815 (to S. Oakes), R01EY027810 (to S. Oakes and F. Papa), U01DK108332 (to S. Oakes), and T32CA177555 (to S. Oakes). We thank the Tlsty lab for NSG mice, IHC protocols, and equipment; Dr. Rushika Perera for assistance with RT2 tumor analysis; the Cyster lab for their microscope; the Goga and Maltepe Labs for providing incubators; the Goga and Debnath Labs for antibodies; the Hebrok Lab for islet histology expertise; Dr. Ed Roberts for necropsy procedures; Vinh Nguyen for overseeing pancreas dissociation; Matthias Hebrok, Andrei Goga, and Oakes and Papa Lab members for discussions.

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.