Antigen-derived peptides directly engage the unfolded-protein sensor IRE1α to curb cross-presentation by dendritic cells

Dendritic cells (DCs) promote adaptive immunity by cross-presenting antigen-based epitopes to CD8+ T cells. DCs process internalized protein antigens into peptides that enter the endoplasmic reticulum (ER) and upload onto major histocompatibility type I (MHC-I) protein complexes for cell-surface transport and cross-presentation. Perplexingly, DCs often exhibit activation of the ER-stress sensor IRE1α in the absence of classical ER stress—leaving the underlying mechanism unexplained. Here we show that antigen-derived hydrophobic peptides directly engage ER-resident IRE1α by masquerading as unfolded proteins. Furthermore, IRE1α activation depletes MHC-I heavy-chain mRNAs through regulated IRE1α-dependent decay (RIDD), thereby curtailing antigen cross-presentation. In tumor-bearing mice, IRE1α disruption increased MHC-I expression on tumor-infiltrating DCs, and enhanced recruitment and activation of CD8+ T cells. Moreover, IRE1α inhibition synergized with anti-PD-L1 antibody treatment to cause tumor regression. Our findings elucidate the mechanism and consequence of antigen-driven IRE1α activation in DCs, yielding a promising combination strategy for cancer immunotherapy.

CD8 + T cells. Moreover, IRE1α inhibition synergized with anti-PD-L1 antibody treatment to cause tumor regression. Our findings elucidate the mechanism and consequence of antigen-driven IRE1α activation in DCs, yielding a promising combination strategy for cancer immunotherapy.

INTRODUCTION
Dendritic cells (DCs) comprise a unique myeloid cell subset that plays a crucial role in antigen presentation during the development and elaboration of adaptive immunity (Mellman & Steinman, 2001;Steinman, 2007). Certain DC lineages mediate the specialized process of antigen cross-presentation, which initiates cytotoxic CD8 + T cell responses (Gatti & Pierre, 2003;Mellman & Steinman, 2001;Palucka, Banchereau, & Mellman, 2010). DCs are highly proficient in acquiring antigens from tissue microenvironments, through endocytosis of soluble proteins or phagocytosis of cell fragments and corpses (Z. Chen et al., 2001;Sallusto, Cella, Danieli, & Lanzavecchia, 1995). During exposure to a pulse of protein antigen, DCs internalize the polypeptide, which reaches the cytoplasm and undergoes proteasomal processing into shorter peptides (Alloatti, Kotsias, Magalhaes, & Amigorena, 2016). Subsequently, the transporter associated with antigen processing (TAP), which resides in the ER membrane, enables importation of the peptides into the ER lumen. Within the ER, the peptides are uploaded through chaperone-aided events onto MHC-I protein complexes, which are composed of heavy and light polypeptide chains (Jhunjhunwala, Hammer, & Delamarre, 2021;Thomas & Tampe, 2017). The peptide-MHC complexes traffic to the DC surface, where the epitopes are cross-presented to engage cognate T cell receptors (TCRs) on juxtaposing T cells. In cancer, both intrinsic and therapeutic mechanisms require efficient DC-mediated cross-presentation of tumor antigens to CD8 + T cells to achieve effective anti-tumor immunity (Barber et al., 2006;Barry et al., 2018;D. S. Chen & Mellman, 2013;Garris et al., 2018;Jansen et al., 2019;Spranger et al., 2016).
However, few treatment strategies are currently available to directly modulate DC crosspresentation.
DCs exhibit IRE1α activation in the absence of canonical ER stress (Iwakoshi, Pypaert, & Glimcher, 2007;Osorio et al., 2014;Tavernier et al., 2017). Gene knockout (KO) studies of XBP1 have indicated divergent effects on antigen cross-presentation in different types of DCs. In CD8α + DCs, XBP1 KO led to a hyper-activated RIDD phenotype, which disrupted T cell activation by depleting mRNAs encoding specific components of the cross-presentation machinery, i.e., Lamp-1 and TAP binding protein (TAPBP) (Iwakoshi et al., 2007;Osorio et al., 2014;Tavernier et al., 2017). In contrast, in tumor-associated DCs, XBP1 KO led to changes in lipid metabolism, which improved cross-presentation of tumor antigens and consequent anti-tumor T cell activity (Cubillos-Ruiz et al., 2015). In conventional (c) DC1 subpopulations residing in the lung, XBP1 KO together with partial IRE1α gene disruption reduced DC viability (Tavernier et al., 2017). Pulsing of bone marrow-derived DCs (BMDCs) with melanoma cell lysates as a source of antigens upregulated XBP1s without affecting RIDD, while IRE1α inhibition attenuated cross-presentation to CD8 + T cells (Medel et al., 2018). While these studies implicate IRE1α in the biological regulation of DCs, the mechanism underpinning the activation of IRE1α in these cells in the absence of canonical ER stress remains a mystery.
In the present study, we reveal that antigen-derived peptides can directly engage IRE1α in antigen-pulsed DCs by mimicking the action of misfolded proteins. We further show that antigen-induced IRE1α stimulation curtails cross-presentation, through specific RIDD-mediated depletion of MHC-I heavy-chain mRNAs. Blocking this negative feedback in tumor-bearing mice by inhibiting IRE1α upregulated MHC-I levels on DCs and enhanced tumor recruitment and activation of CD8 + T cells. Moreover, IRE1α inhibition cooperated with anti-PD-L1 immune-checkpoint disruption to cause tumor regression. Our findings elucidate an unexpected mechanism underlying IRE1α activation in DCs, revealing a promising combinatorial strategy for cancer immunotherapy.

Antigen pulsing of BMDCs activates IRE1α
To seek insight into the mechanism of IRE1α activation in DCs, we pulsed mouse BMDCs, matured ex vivo with GM-CSF plus IL-4, with the classical protein antigen ovalbumin. In order to cross-present ovalbumin, DCs must internalize the pulsed protein and process it intracellularly; in contrast, DCs can directly present the ovalbumin-based octapeptide SIINFEKL, based on its ability to displace antigens already bound to MHC-I complexes at the cell surface (Alloatti et al., 2016). Ovalbumin pulsing of BMDCs induced concentration-and time-dependent activation of IRE1α, evident by increased protein levels of XBP1s and phospho-IRE1α ( Figure 1A and 1B). Comparable to the tolllike receptor (TLR) agonist polyinosinic:polycytidylic acid (poly-I:C)-previously shown to activate IRE1α in leukocytes (Martinon, Chen, Lee, & Glimcher, 2010)-ovalbumin pulsing induced much weaker IRE1α activation than did the potent pharmacological ERstressor tunicamycin ( Figure 1B). In contrast to ovalbumin, SIINFEKL had little effect on IRE1α activity ( Figure 1B), suggesting a requirement for intracellular events. Unlike IRE1α, other UPR sensors, i.e., PERK (assessed by induction of its downstream targets ATF4 and CHOP) and ATF6 (assessed by its proteolytic processing), which responded to ER-stress induction by the proteasome inhibitor MG132, did not show measurable activation in response to ovalbumin pulsing ( Figure S1A), indicating specific IRE1α engagement in the absence of classical ER stress. To verify bona fide IRE1α activation during ovalbumin pulsing, we added the highly selective kinase-based small molecule IRE1α inhibitor, G03089668 (G9668) (Figure S1B-D) (Harnoss et al., 2020;Harnoss et al., 2019;Harrington et al., 2015), which completely blocked IRE1α phosphorylation and XBP1s induction. During a 4-hour pulse of ovalbumin, IRE1α phosphorylation peaked at 2 hours while XBP1s protein peaked at 4 hours ( Figure 1C and S1A). Of note, a 1-hour pulsing with ovalbumin followed by washing and incubation in standard media for another 3 hours generated similar levels of XBP1s (data not shown). Taken together, these results indicate a rapid, yet transient, stimulation of IRE1α in BMDCs in response to ovalbumin pulsing. BMDC pulsing with lysates derived from CT26, 4T1 and EMT6 tumor cells (see below) also activated IRE1α ( Figure 1D). Moreover, pulsing with a different, highly purified recombinant protein, i.e., clinical-grade soluble CD4, also led to IRE1α activation ( Figure 1E), indicating a specific stimulation mechanism independent of potential contamination with bacterial endotoxin (lipopolysaccharide, LPS).
The micropinocytosis inhibitor amiloride (Koivusalo et al., 2010) blocked antigeninduced IRE1α stimulation in BMDCs ( Figure 1E and S1E), demonstrating a requirement for antigen internalization. Furthermore, CRISPR/Cas9-based disruption of the TAP1 gene prevented IRE1α stimulation in response to ovalbumin ( Figure 1F), indicating a requirement for antigenic peptide importation into the ER. As expected (Martinon et al., 2010), BMDCs lacking the TLR adapter MyD88 failed to activate IRE1α in response to LPS ( Figure S1F); importantly, these cells showed unimpeded IRE1α stimulation in response to ovalbumin, further validating an LPS-independent IRE1α activation. We obtained similar evidence for LPS-independent IRE1α activation upon ovalbumin pulsing of Flt3-ligand (Flt3L)-matured BMDCs from WT or MyD88 KO mice (data not shown). Moreover, unlike BMDCs, mouse embryonic fibroblasts (MEFs) did not display detectable IRE1α activation after exposure to ovalbumin (Figure S1G), suggesting a BMDC-specific mechanism. Taken together, these results suggest that IRE1α engagement in DCs in response to a pulse of a protein antigen occurs independently of LPS and requires antigen uptake as well as importation of antigenderived peptides into the ER.

Antigen-derived peptides can directly engage IRE1α
In the context of classical ER stress, otherwise buried hydrophobic segments of unfolded proteins can directly engage the ER-lumenal domain of IRE1α (Gardner & Walter, 2011). Although general UPR activation was absent during antigen pulsing of DCs ( Figure S1A and S1B), we reasoned that antigen-derived peptides may directly engage IRE1α by mimicking the action of unfolded proteins. To examine this possibility, we first compared the ability of heat-denatured and native forms of the ovalbumin antigen to bind to a recombinant protein comprising the IRE1α lumenal domain fused to an Fc tag (LD-Fc). Whereas heat-denatured ovalbumin displayed specific and saturable binding to immobilized IRE1α LD-Fc, native ovalbumin showed little binding over background ( Figure 2A). We estimated a Kd of ~ 300 ± 75 μM for denatured ovalbumin, in line with affinities previously reported for unfolded-protein binding to IRE1α LD (Gardner & Walter, 2011). Co-immunoprecipitation studies confirmed concentrationdependent association of heat-denatured ovalbumin to IRE1α LD-Fc, whereas native ovalbumin again did not show appreciable binding ( Figure 2B). We obtained similar results by acidic (pH=2) or basic (pH=10) ovalbumin denaturation (data not shown).
Next, to determine whether specific ovalbumin subsegments could interact with IRE1α, we generated a "tiled" peptide array spanning the polypeptide sequence-consisting of 18 amino-acid long synthetic peptides with a 3-residue overlap, as previously described (Gardner & Walter, 2011). We spotted the peptides onto a membrane and examined binding of IRE1α LD-Fc, using horseradish peroxidase-based colorimetric detection. To test an additional antigen, we generated a similar peptide array based on the GP70 protein, which is expressed by CT26 colorectal cancer cells (Takeda et al., 2000). The analyses revealed that 34/123 ovalbumin peptides (clustered in 10 regions) and 64/210 GP70 peptides (19 regions) displayed significant binding to IRE1α LD-Fc ( Figure 2C and S2A). In contrast, arrayed peptides derived from these antigens did not exhibit detectable binding to CD4-Fc under identical conditions (data not shown).
We next evaluated the importance of hydrophobic side-chains for peptide binding. We synthesized biotin-tagged peptides corresponding to two binding and one non-binding segments of the tiled ovalbumin array (B1, B2, and B3), and mutated variants of the binders with hydrophobic residues substituted by aspartic acids (B1', B2') ( Figure S2B). We incubated each peptide with FLAG-tagged IRE1α LD, stabilized formed complexes by chemical crosslinking, and visualized them by anti-biotin immunoblotting. Whereas peptide B3 showed no significant interaction, B1 and B2 exhibited specific binding, associating not only with monomers but also with apparent dimers or oligomers of IRE1α LD-FLAG ( Figure 2D). In contrast to B1 and B2, mutated peptides B1' and B2' failed to show significant binding, indicating a critical role for hydrophobic residues in B1 and B2 for interaction with IRE1α LD.
To test whether ovalbumin-based peptides can associate with IRE1α in a cellular setting, we transfected HEK293 cells with cDNA constructs encoding shorter Myc-tagged versions of B1, B2 and B3 (M1, M2 and M3)-fused to a signal sequence for direct ER targeting ( Figure S2B). Immunoprecipitation with anti-Myc antibody followed by immunoblotting with anti-IRE1α revealed specific co-immunoprecipitation of Myctagged peptides with IRE1α ( Figure 2E). Analyzing the signal ratios for IRE1α over Myc confirmed markedly greater IRE1α interaction for M1 and M2 as compared to M3. To assess functional IRE1α engagement, we transfected U20S cells with similar cDNA plasmids encoding a series of three peptides each representing M1, M2, or M3 ( Figure   S2B). We measured IRE1α activation by RT-QPCR analysis of mRNA transcripts for XBP1s and XBP1u, as well as the RIDD targets CD59 and DGAT2. The M1 and M2 peptides functionally engaged IRE1α, as evident by upregulation of XBP1s and depletion of the RIDD-targeted mRNAs encoding CD59 and DGAT2 in comparison to non-transfected controls and peptide M3, which showed no activation above vector controls ( Figure 2F). Thus, congruent with the results obtained with the corresponding synthetic peptides in a cell-free setting ( Figure 2D), ovalbumin-derived, ER-targeted peptides can specifically bind to the ER-resident IRE1α protein within cells and stimulate its endoribonuclease activity in a manner that corresponds to their direct ability to bind to IRE1α.

IRE1α inhibition in BMDCs augments cross-presentation
To examine whether antigen-induced IRE1α activation impacts cross-presentation, we pulsed BMDCs with ovalbumin in the absence or presence of G9668. We then tested BMDC capacity to activate mouse splenic OT-I CD8 + T cells, which express a transgenic TCR specific to the SIINFEKL epitope. During SIINFEKL pulsing, IRE1α inhibition with G9668 had little effect; however, upon ovalbumin pulsing, it significantly and reproducibly augmented subsequent induction of OT-I CD8 + T-cell proliferation by ~ 20% ( Figure 3A and S3A). To confirm LPS-independent augmentation, we pulsed BMDCs with endotoxin-free ovalbumin (EF-OVA); comparably, IRE1α inhibition in this setting augmented OT-I CD8 + T-cell proliferation by ~ 29% ( Figure 3A). Furthermore, G9668 treatment during pulsing of Flt3L-matured BMDCs with either ovalbumin or EF-OVA increased OT-I CD8 + T-cell proliferation by 23% or 24%, respectively ( Figure 3B and S3A). Importantly, G9668 did not directly affect proliferation or activation of naïve CD4 + or CD8 + splenic T cells upon TCR stimulation with anti-CD3 plus anti-CD28 antibodies ( Figure S3B and S3C). Moreover, in contrast to CD8 + T cell stimulation, IRE1α inhibition during ovalbumin pulsing did not alter MHC class II-restricted activation of splenic OT-II transgenic CD4 + T cells, which also harbor an ovalbumin-specific TCR transgene ( Figure S3D). Taken together, these results indicate that protein-antigen pulsing of BMDCs leads to LPS-independent IRE1α activation, which in turn specifically dampens MHC-I-restricted antigen cross-presentation to CD8 + T cells. Functional inhibition of IRE1α reverses this curbing mechanism, enhancing antigen crosspresentation.
To investigate the impact of IRE1α inhibition on cross-presentation of tumor antigens, we subcutaneously inoculated BALB/c mice with CT26 cells and allowed tumors to form. We then isolated splenic CD8 + T cells (likely possessing TCRs that can recognize CT26 antigens) from these mice and co-incubated them with BMDCs prepulsed with CT26 cell lysates. Addition of G9668 augmented cross-presentation of CT26 epitopes to cognate splenic CD8 + T cells by ~ 25% (Figure 3C), in keeping with ovalbumin cross-presentation. Thus, IRE1α inhibition in BMDCs enhances crosspresentation of tumor-derived antigens.
Next, we turned to interrogate which specific aspect of the cross-presentation process is modulated by IRE1α. IRE1α inhibition did not alter the uptake of fluorescently-labeled ovalbumin by BMDCs ( Figure S3E). On the other hand, IRE1α inhibition during exposure of BMDCs to ovalbumin or cell lysates significantly increased MHC-I protein expression at the cell surface ( Figure 3D), suggesting negative regulation of antigen-driven MHC-I dynamics by IRE1α.
We therefore considered the possibility that IRE1α activation in response to antigen pulsing attenuates cross-presentation by decreasing MHC-I heavy-chain To examine RIDD-mediated depletion in BMDCs, we prevented de novo transcription with actinomycin D. In control-pulsed BMDCs, the mRNA levels of H-2K as well as the canonical RIDD targets CD59 and DGAT2 remained stable over a 24-hour period; in contrast, in antigen-pulsed BMDCs, the levels of both H-2K and CD59 mRNAs substantially declined over time ( Figure 4C). Furthermore, both the kinasebased IRE1α inhibitor G9668 and the RNase-directed IRE1α inhibitor 4μ8c (Cross et al., 2012) substantially rescued mRNAs encoding H-2K, CD59, and DGAT2 ( Figure 4D), as well as several other known RIDD targets, i.e., BLOS1, RNF213 and IRF7 ( Figure S4B), demonstrating RIDD-based depletion.
Thus, antigen pulsing of DCs activates IRE1α, which curbs cross-presentation by depleting MHC-I heavy-chain mRNAs through RIDD.

IRE1α inhibition in tumor-bearing mice upregulates MHC-I on tumor DCs and augments CD8 + T cell engagement
To determine whether the dampening effect of IRE1α on antigen cross-presentation has functional consequences in vivo, we examined the impact of IRE1α disruption on syngeneic tumor growth in mice. Tumor-cell-autonomous knockout of IRE1α by CRISPR/Cas9 had little effect on subcutaneous growth of CT26 colon tumors ( Figure   S5A and S5B). In contrast, systemic pharmacological inhibition of IRE1α with G9668 substantially attenuated tumor progression as compared to vehicle treatment ( Figure   5A, S5C, and S5D), suggesting enhancement of host-mediated anti-tumor activity. To specifically elucidate potential immune effects, we analyzed by flow cytometry the tumor-associated leukocyte populations after 7 days of treatment. As compared to controls, tumors from G9668-treated mice showed significantly greater infiltration by CD11c + MHC class II high DCs ( Figure 5B), specifically belonging to the cDC1 (XCR1 + CD103 + ) subpopulation ( Figure 5C). Furthermore, tumor-infiltrating cDC1s showed significantly higher surface levels of MHC-I, as well as the RIDD marker CD59, in G9668-treated mice as compared to controls ( Figure 5D). Moreover, G9668 treatment led to significantly greater numbers of tumor-infiltrating cytotoxic CD8 + T cells ( Figure   5E), and to higher expression by these cells of the activation markers granzyme B, PD-1, and CD44 ( Figure 5F). Importantly, staining with recombinant MHC-I tetramer complexes presenting the CT26 tumor antigen GP70 revealed significantly higher levels of tumor-infiltrating GP70-specific CD8 + T cells in G9668-treated mice ( Figure 5G).
Thus, IRE1α inhibition attenuates CT26 tumor growth in conjunction with elevated tumor infiltration, MHC-I expression, and tumor-antigen cross-presentation by cDC1s, augmenting the recruitment and activation of tumor-reactive CD8 + T cells.

Single-cell RNAeq demonstrates IRE1α regulation of MHC-I mRNAs in tumor DCs
To examine immune modulation in a different tumor model, we used syngeneic 4T1 triple-negative breast cancer (TNBC) cells. In contrast to the CT26 model ( Figure S5A and S5B), cell-autonomous knockout of IRE1α in 4T1 cells caused notable tumorgrowth inhibition (TGI) of 53% ( Figure S6A and S6B). Nevertheless, systemic treatment of mice bearing parental IRE1α wildtype 4T1 tumors with G9668 led to a stronger TGI of 82% ( Figure 6A, S6C and S6D), suggesting both cell-autonomous and host-mediated anti-tumor effects. The myeloid compartment in tumors has been systematically studied by single-cell RNA sequencing (scRNAseq) (Cheng et al., 2021;Mariathasan et al., 2018). We performed scRNAseq after 6 days of treatment to analyze the tumor leukocytic populations. Tumors in G9668-treated mice showed enrichment in both DCs and CD8 + T effector cells, but not in naïve T cells ( Figure 6B and 6C). Importantly, tumor-infiltrating DCs in G9668-treated mice displayed significantly higher mRNA levels of H-2K and H-2D heavy-chains; though not of TAPBP transcripts ( Figure 6D). CD59 and DGAT2 mRNAs were insufficiently abundant to enable accurate quantification, but five other RIDD targets that were detected, i.e. BLOS1, FERMT3, IRF7, RNF213 and SPON1, showed significant increases ( Figure S6E), confirming RIDD inhibition. Tumors in G9668-treated mice had unaltered levels of M1 macrophages, but showed significantly fewer M2 macrophages and monocytes as compared to controls ( Figure   S6F). Flow cytometric analysis after 6 days of treatment showed that IRE1α inhibition significantly increased tumor infiltration by cytotoxic CD8 + T cells and their expression of the activation markers IFN-γ, PD-1, and CD69, independent of IRE1α status in the malignant cells ( Figure 6E). These results further support the conclusion that IRE1α inhibition augments anti-tumor immune responsiveness by increasing MHC-I expression on tumor-associated DCs and consequent engagement of tumor-infiltrating cytotoxic CD8 + T cells.

Systemic IRE1α inhibition cooperates with immune checkpoint blockade
Although immune-checkpoint disruption has transformed patient benefit in a number of cancer settings (Marinelli et al., 2020), further advances are needed to achieve wider effectiveness. To investigate whether IRE1α inhibition would complement immunecheckpoint blockade, we turned to the orthotopic EMT6 TNBC model, previously found to exhibit partial responsiveness to anti-programmed death ligand (PD-L)1 antibody therapy upon implantation in the mouse mammary fat pad (Mariathasan et al., 2018).
Similar to the CT26 model, cell-autonomous IRE1α KO in EMT6 cells had minimal impact on tumor growth ( Figure S7A and S7B), identifying an additional suitable model to specifically interrogate the impact of IRE1α inhibition on immune modulation.
Treatment of EMT6 tumor bearing mice with either the anti-mouse PD-L1 monoclonal antibody 6E11 or with G9668 partially impaired EMT6 tumor progression ( Figure 7A and S7C). Remarkably, combined administration of 6E11 and G9668 led to frank tumor regression with a mean TGI rate of 114%, resulting in significantly better efficacy than either monotherapy (p < 0.01). In keeping with the other models, treatment of EMT6 tumor-bearing mice with G9668 for 7 days significantly increased surface levels of MHC-I and CD59 in tumor-infiltrating cDC1s ( Figure 7B). Moreover, G9668 increased tumor invasion by cytotoxic CD8 + T cells and their expression of the activation markers IFN-γ and granzyme B ( Figure 7C). Thus, IRE1α inhibition effectively complements PD-L1-based immune-checkpoint disruption to reverse syngeneic orthotopic tumor progression.

DISCUSSION
The mechanism of IRE1α activation in the absence of classical ER stress in DCs has been mysterious (Iwakoshi et al., 2007;Medel et al., 2018;Osorio et al., 2014;Tavernier et al., 2017). Our present findings reveal that exposure of DCs to pulsed protein antigens drives IRE1α activation through an LPS-independent process analogous to the direct engagement of IRE1α by unfolded proteins under canonical ER stress. We further show that antigen-induced IRE1α activity curbs cross-presentation through RIDD-mediated depletion of MHC-I heavy-chain mRNAs. This functional consequence likely represents a negative feedback loop that fine-tunes crosspresentation, perhaps to prevent inappropriate or excessive T-cell activation during sterile tissue injury. Importantly, the disruption of this inherent negative feedback by inhibiting IRE1α cooperates with immune-checkpoint blockade to enhance anti-tumor immune responses, revealing exciting potential for therapeutic translation.
Our BMDC studies showed that antigen pulsing selectively activates IRE1α but not PERK or ATF6, excluding the standard UPR as a key mechanistic driver. This contrasts with plasmacytoid DCs (pDCs), which do not mediate cross-presentation, and interestingly display constitutive activation of PERK (Mendes et al., 2021). BMDC pulsing with different protein antigens, or with lysates of several cancer cell lines, induced significant levels of IRE1α activity. Although these levels were markedly weaker than those induced by strong pharmacological ER stressors, antigenic IRE1α stimulation was highly reproducible, reaching peak intensity within 2 -4 hours of exposure, and then declining. These rapid yet transient kinetics are consistent with the time frame of antigen uptake, processing, and ER entry. Indeed, our further mechanistic dissection indicated that antigen-driven IRE1α activation requires both pinocytosis and ER importation events, but not TLR signaling.
IRE1α resides in the ER membrane and responds through its lumenal domain to ER accumulation of unfolded or misfolded proteins during canonical ER stress (Hetz, 2012;Walter & Ron, 2011). Based on our observation that antigen-induced IRE1α activation required TAP1-a critical mediator for importation of antigen-derived peptides into the ER-we reasoned that antigen-based peptides entering the ER may directly engage IRE1α by masquerading as unfolded or misfolded proteins. Several lines of evidence supported this idea. First, whereas native ovalbumin failed to interact with the IRE1α LD, unfolded ovalbumin-generated by denaturation with heat or extreme pHwas capable of direct LD binding, with affinity comparable to that of unfolded proteins (Gardner & Walter, 2011). Second, specific ovalbumin-based peptides bound to the IRE1α LD in a manner that required their hydrophobic amino acids. Third, cellular expression of ER-directed ovalbumin-based peptides demonstrated congruent interaction with, and functional engagement of, cellular IRE1α. The lack of stimulation of the other UPR branches in response to antigen pulsing further indicates a direct IRE1α activation mechanism independent of BiP. Hence, although a nascent protein that is incorrectly folded by the ER and a peptide that lacks 3D structure due to proteolytic cleavage of its parent protein in the cytoplasm are distinct, both can be sensed by IRE1α. Although any protein-producing cell may harbor some constitutive level of IRE1α-peptide interactions, our data suggest that DCs are uniquely capable of IRE1α activation upon transient exposure to a high concentration of an extracellular protein.
Moreover, for DCs, which are highly specialized in antigen cross-presentation, such IRE1α activation can have unique functional consequences. The recent discovery of pervasive functional peptide translation in cells (J. Chen et al., 2020) raises an intriguing question of whether additional peptide modalities besides antigen processing may similarly engage IRE1α.
Earlier work interrogating the involvement of IRE1α in DC regulation relied primarily on XBP1 KO-a strategy that does not completely disrupt, and in some cases even augments, the kinase-endoribonuclease activity of IRE1α. In CD8 + DCs, XBP1 KO causes artificial RIDD hyper-activation, which leads to mRNA depletion of certain components of the cross-presentation machinery, i.e., LAMP-1, TAPBP, and β2M (Osorio et al., 2014). In contrast, BMDC pulsing with melanoma cell lysates activated XBP1 splicing but not RIDD, and XBP1s appeared to promote, rather than disrupt, efficient melanoma antigen cross-presentation (Medel et al., 2018). Direct infection of BMDCs by Toxoplasma gondii led to IRE1α activation via MyD88-dependent TLR signaling, with decreased cross-presentation upon XBP1 KO and partial disruption of IRE1α (Poncet et al., 2021). The partial disruption of IRE1α signaling by XBP1 KO left the biological consequence of IRE1α engagement during antigen cross-presentation incompletely understood.
To impede IRE1α's enzymatic activity more fully, we used the highly selective and potent kinase-based IRE1α inhibitor G9668, which blocks both the kinase and the kinase-controlled endoribonuclease activities of IRE1α. Indeed, G9668 fully disrupted antigen-induced IRE1α activation in DCs, preventing IRE1α auto-phosphorylation, as well as consequent RNase-dependent XBP1 splicing and RIDD. While G9668 did not alter DC-surface presentation of ovalbumin-derived SIINFEKL peptide to OT-I CD8 + T cells, it augmented cross-presentation of pulsed full-length ovalbumin, confirming the requirement of intracellular events for IRE1α activation. Of note, G9668 did not alter MHC-II-restricted antigen presentation to OT-II CD4 + T cells, nor did it affect costimulation of naïve CD8 + or CD4 + T cells. Enhancement by G9668 was not limited to the OT-I model, as it also applied to cross-presentation of CT26-derived antigens to splenic CD8 + T cells from CT26 tumor-bearing mice. Thus, the engagement of IRE1α during antigen processing within DCs selectively curtails MHC-I-restricted crosspresentation, providing negative feedback to modulate T cell activation.
Our investigation of how IRE1α curtails cross-presentation underscores RIDD as an important mechanism that depletes mRNAs encoding MHC-I heavy chains. This Importantly, combined treatment with anti-PD-L1 antibody and G9668 in the orthotopic EMT6 TNBC model, which is only partially responsive to anti-PD-L1, led to clear tumor regression, establishing non-redundant complementarity of these two modalities.
In conclusion, our studies reveal that pulsed-antigen-derived peptides can directly engage IRE1α in cross-presenting DCs, explaining the activation of this ERstress sensor in the absence of classical ER stress. Furthermore, by fully blocking IRE1α's enzymatic function, we have discovered that IRE1α controls a negative feedback loop, by depleting MHC-I heavy-chain mRNAs via RIDD, to dampen crosspresentation and curtail consequent CD8 + T cell activation. Excitingly, disruption of this feedback by small-molecule IRE1α inhibition holds promise for cancer immunotherapy, particularly in combination with anti-PD-L1 inhibition. These findings bring important conceptual advances to seminal previous work identifying a role for XBP1s in dysregulated lipid metabolism and in function of tumor-associated DCs (Cubillos-Ruiz et al., 2015), and to studies on IRE1α disruption in immunodeficient mice (Harnoss et al., 2020;Harnoss et al., 2019;Harrington et al., 2015).

Cell Cultures, BMDC differentiation and Experimental Reagents
CT26, 4T1, HEK293 and EMT6 cells were originally acquired from ATCC, authenticated by analysis of short tandem repeats and tested to ensure no presence of mycoplasma within 3 mo of use. U20S cells were kindly provided by the Walter Lab of the University of California, San Francisco (UCSF). Cells were grown in RPMI1640 media supplemented with 10% fetal bovine serum (FBS) (Sigma, St. Louis, MO), 2 mM glutaMAX (Gibco, Amarillo, TX), 100 U/ml penicillin (Gibco) and 100 μg/ml streptomycin (Gibco). MEFs were obtained as previously described (Holst et al., 2007).
The kinase-based IRE1α inhibitor G9668 (Harnoss et al., 2020;Harnoss et al., 2019;Harrington et al., 2015) was used as indicated. For pulsing with tumor cell lysates, indicated cell lines were grown to confluence, suspended in sterile PBS and subjected to five freeze-thaw cycles with liquid nitrogen and heating at 37°C. Cell lysates were then normalized by BCA protein concentration measurement (Thermo-Fisher, Waltham, MA).

In Vitro characterization of small-molecule IRE1α inhibitor G03089668
Potency of G9668 was analyzed in two assays of IRE1α activity, with dilutions covering a range of concentrations from 0.2 nM to 10 μM in order to determine IC50 values.

Inhibition of RNase activity was assessed by the incubation of G9668 with IRE1α
(Q470-L977) and 5 ' FAM-CAUGUCCGCAGCGCAUG-3 ' BHQ substrate. Substrate cleavage was monitored kinetically as an increase in fluorescence. Cellular activity was evaluated with the XBP1s-luciferase reporter assay in HEK293 cells stably transfected with the XBP1s-luciferse reporter construct. Briefly, cells were preincubated with G9668 for 2 hours and subsequently stimulated with Tg (100 nM) for 6 hours. IRE1α-mediated cleavage of the reporter led to luciferase expression which was detected with the addition of luciferin substrate. Kinase selectivity of G9668 against a panel of 220 kinases was measured at a concentration of 1 μM with KinomeScan TM (DiscoverX, Fremont, CA). Fold selectivity was determined by IC50 measurement of competition by G9668 for binding of ATP to each specific kinase that showed significant inhibition by G9668 via KinomeScan TM .

Generation of IRE1α KO syngeneic tumor cell lines
Individual IRE1α-specific sgRNAs were designed using a standard guide scaffold and

Generation of TAP1 KO BMDCs
Bone marrow cells were purified from Cas9-expressing C57BL/6 mice as described above, subjected to red blood cell lysis with ACK lysis buffer and electroporated with P3 Primary Cell 4D-Nucleaofactor TM X -kit (V4XP-3032, Lonza, Basel, Switzerland), as previously described (Freund et al., 2020). Once re-suspended in P3 buffer, cells were added a Cas9-ribonuceloprotein (RNP) complex (IDT) containing non-targeting or TAP1-targeting single guide RNAs (sgRNAs) (IDT). The sequences of TAP1-targeting sgRNA included: sgRNA A -GCGGCACCTCGGGAACCAAC, sgRNA B -TAACTGATAGCGAAGGCATC, sgRNA C -ACGGCCGTGCATGTGTCCCA. These sgRNA were used separately or all three combined. Bone marrow cells were then transfected with the appropriate program and grown for 9 days similarly to all other BMDC cultures.  For co-immunoprecipitation experiments, IRE1α LD-Fc (10 μg/ml) and ovalbumin (indicated concentrations) were co-incubated in binding buffer (as described above) for 2 hr and then immunoprecipitated with anti-IRE1α lumenal domain antibody (Genentech), conjugated to sepharose beads, overnight at 4°C. Beads were subsequently washed four times with lysis buffer and boiled in SDS sample buffer for 10 min. Samples were then analyzed by SDS-PAGE followed by immunoblot.

In vitro IRE1α LD binding assays
Peptide arrays were produced by the MIT Biopolymers Laboratory. The tiling arrays were composed of 18-mer peptides spanning the ovalbumin or GP70 sequences and overlapping by 3 amino acids. The arrays were incubated in methanol for 10 min and then in binding buffer (50 mM Tris pH 7, 250 mM NaCl, 10% glycerol, 2 mM DTT) for three 10 min wash cycles. The arrays were then incubated for 1 hr at room temperature with 500 nM IRE1α LD-Fc and washed again for three 10 min cycles in binding buffer to remove any unbound LD-Fc. Using a semi-dry transfer apparatus, bound IRE1α LD-Fc was transferred after washes to a PVDF membrane and detected with anti-human Fc antibody (ab977225, Abcam), ECL solution (Thermo-Fisher) and ChemiDoc ZRS+ imager (Biorad). To measure binding of IRE1α LD-Fc to each peptide, images containing developed membranes were quantified with ImageJ software (version 2.0.0). Pixel intensity was determined for all spots containing peptides, with background subtracted for spots containing no peptides. Peptides were considered to bind IRE1α LD-Fc if spot intensity was above the average of all array peptides.
For binding assays of biotin-tagged peptides, we generated a FLAG-tagged IRE1α lumenal domain (LD-FLAG) comprised of amino acids M1-D443 of IRE1α fused Cterminally to a linker (GNS) followed by a Flag tag (DYKDDDDK). LD-FLAG was incubated with synthetic N-terminal biotin-tagged peptides derived from ovalbumin in binding buffer (as described above) for 1.5 hr, cross-linked by 25 μM DSS (Thermo-Fisher) for 1 hr and subsequently incubated with 50 μM Tris (pH 7.5) for 15 min to quench cross-linking. Samples were then analyzed by SDS-PAGE followed by antibiotin immunoblot.

Real-time quantitative PCR assay of transcript abundance
For RT-qPCR analysis, RNA was purified from BMDCs, HEK293 or U20S cells with the

In vitro degradation of MHC-I heavy-chain transcripts
To search for IRE1α cleavage sites within MHC-I heavy chain mRNAs, sequences were loaded unto A Plasmid Editor (APE) software and subjected to a search function for consensus GCAG locations. The location most likely to provide a stable stem-loop structure within each transcript was then chosen.
T7 RNA (1 μg) was digested at room temperature by IRE1α recombinant KR protein (1 μg) for 15 min in RNA cleavage buffer (HEPES pH 7.5, 20mM, KOAc 50 mM, MGAc 1mM, Tritox X-100 0.05%). The digestion was terminated by addition of formamide (97%) and exposed to 70°C temperature to linearize the RNA. Immediately after linearization, samples were placed on ice for 5 min and then run on a 3% agarose gel.

Ex vivo T-cell activation and cross-presentation experiments
For ex vivo T-cell activation experiments, mice were euthanized and spleens were removed and mechanically disrupted with a GentleMacs tissue dissociator (Miltenyi Biotec Inc, Auburn, CA). Total spleen cells were washed with sterile PBS, counted and CD8 + or CD4 + T cells were magnetically separated with appropriate separation kits (Stemcell Technologies, Vancouver, Canada).
For antigen cross-presentation assays, 2 ⋅ 10 4 BMDCs were plated, activated with LPS (10 μg/ml) for 2 hours and pulsed with SIINFEKL (100 nM), ovalbumin, EF-OVA, or CT26 lysate (all at 250 μg/ml) overnight. BMDCs were then washed with media and 2 ⋅ 10 5 CD8 + or CD4 + T cells were added and co-incubated for 72 hr prior to flow cytometry analysis. T cells were stained with a Celltrace Violet Cell Proliferation reagent (Thermo-Fisher) prior to introduction to the co-culture. Proliferation was then determined by loss of Celltrace Violet signal in viable T cells after co-incubation. For in vivo studies, 7 days after tumor-cell inoculation animals were randomized into groups receiving vehicle control (50% PEG400, 40% water, 10% DMSO) or G9668 (250 mg/kg) compound bidaily (BID) by oral gavage (PO). For the EMT6 combination studies, mice were randomized into four groups, with groups similarly receiving vehicle or G9668 in combination with anti-GP120 control antibody or 6E11 anti-PD-L1 antibody (both with LALAPG Fc alterations, dissolved in PBS, at 10 mg/kg intravenously (IV) for the first dose and 5 mg/kg intraperitoneally (IP) biweekly (BIW) thereafter).

Statistical analysis
All values were represented as arithmetic mean ± standard deviation (SD). Statistical analysis was performed by unpaired, two-tailed t test or one-way ANOVA. A resulting P value < 0.05 was considered significant. All analyses were performed with the GraphPad Prism 7 software (GraphPad Software, Inc., San Diego, CA). Tavernier, S. J., Osorio, F., Vandersarren, L., Vetters, J., Vanlangenakker, N., Van Isterdael, G., . . . Janssens, S. (2017). Regulated IRE1-dependent mRNA decay sets the threshold for dendritic cell survival. Nat Cell Biol, 19 (6)    Polystyrene wells were coated with native or heat-denatured ovalbumin (10 μg/ml) and incubated with purified recombinant human IRE1α LD-Fc fusion protein (10 μg/ml), followed by colorimetric detection with an HRPconjugated anti-human Fc antibody. (B) Native or heat-denatured ovalbumin at indicated concentrations was incubated with IRE1α LD-Fc (10 μg/ml), immunoprecipitated via monoclonal anti-IRE1α LD antibody, and analyzed by IB. (C) A tiled 18 aa-long peptide array spanning ovalbumin was incubated with IRE1α LD-Fc (500 nM) followed by colorimetric detection with an HRP-conjugated anti-human Fc antibody. (D) Biotin-tagged ovalbumin-based peptides (100 μM) were incubated with FLAG-tagged IRE1α LD (50 μM), cross-linked with disuccinimidyl suberate (DSS) and analyzed by IB. B1, B2, B3 are WT peptides; B1', B2' are mutant peptides in which all hydrophobic residues were replaced by aspartic acid. (E) HEK293 cells were transfected with cDNA constructs encoding Myc-tagged peptides (M1, M2, M3) derived from corresponding ovalbumin regions (B1, B2, B3) and containing an ER-directed signal sequence for 48 hr, followed by immunoprecipitation with anti-Myc antibody and IB for IRE1α or Myc. Bar graph indicates signal ratio for IRE1α over Myc. (F) U2OS cells were transfected with cDNA constructs encoding Myc-tagged peptides (M1.1, M1.2, M1.3, M2.1, M2.2, M2.3, M3) derived from corresponding ovalbumin regions (B1, B2, B3) and containing an ER-targeting signal sequence for 48 hr, followed by real-time quantitative PCR (RT-qPCR) analysis for XBP1s, XBP1u, CD59, and DGAT2 mRNA levels. Bar graphs in panels A, C and F represent mean ± SD from three independent technical repeats; images in panels B, D and E represent at least two similar experiments.  Figure 1D; Myc-tagged signal peptides (labeled M) used in Fig. 2E; and Myc-tagged peptides (Labeled M) used in Figure 2F. Bar graphs in panel A represents mean ± SD from three independent technical repeats. BMDCs were pulsed with ovalbumin (500 μg/ml) with or without G9668 (3 μM) for 24 hr and subsequently co-cultured with magneticallyseparated CD8 + OT-I T cells for 72 hr, followed by flow cytometry analysis of T cell proliferation by Celltrace Violet. (C) BMDCs were pulsed with lysates derived from CT26 cells (500 μg/ml protein) for 24 hr and co-cultured with magnetically-separated CD8 + splenic T cells from CT26 tumorbearing mice for 72 hr, followed by analysis of T cell proliferation by flow cytometry of Celltrace Violet staining. (D) BMDCs were pulsed with ovalbumin or lysates derived from 4T1, EMT6 and CT26 cells (500 μg/ml protein) for 8 hr and surface levels of MHC-I were assayed by flow cytometry. Analysis was performed using unpaired, two-tailed t test, * P ≤ 0.05 **P ≤ 0.01, ***P ≤ 0.001. Bar graphs in all panels represent mean ± SD from three independent biological repeats; panels A and B represent data from at least independent experiments collated as fold from control for each experiment. Proliferation and activation were analyzed respectively by Cell Titer Glo and flow cytometry. (D) BMDCs were pulsed with ovalbumin (500 μg/ml) for 24 hr, with or without G9668 (3 μM) and subsequently co-cultured with magnetically-separated CD4 + OT-II T cells for 72 hr, followed by flow cytometry analysis of T cell proliferation by Celltrace Violet. (E) BMDCs were pulsed with APC-labelled ovalbumin (500 μg/ml) for indicated time and analyzed for internalization of ovalbumin by flow cytometry. (E) BMDCs were treated with G9668 (3 μM) for 8 hr and mRNA levels of indicated genes associated with cross-presentation were analyzed by RT-qPCR. Bar graphs in panels B-E represent mean ± SD from three independent biological repeats.    Figure S4. IRE1α RIDD activity specifically targets MHC-I heavy-chain transcripts. (A) BMDCs were pulsed with ovalbumin (500 μg/ml) for 8 hr in the absence or presence of G9668 (3 μM), followed by real time RTqPCR measurements of the indicated RIDD substrates. (B) BMDCs were treated with actinomycin D (2 μg/ml)) and pulsed with ovalbumin (500 μg/ml) for 8 hr combined with DMSO or G9668 (3 μM) or 4μ8C (1 μg/ml), followed by real time RTqPCR measurements of the indicated RIDD substrates. Analysis was performed using unpaired, two-tailed t test, * P ≤ 0.05. Bar graphs in all panels represent mean ± SD from three independent technical repeats. . Total DCs were characterized as F4/80 low , CD11c + and class II MHC high , while cDC1s were characterized as CD103 + XCR1 + CD11b -. (E-F) Measurement of tumorinfiltrating CD8 + T cell abundance (E), expression of indicated activation markers (F) (n = 9 for vehicle-treated group and 10 for G9668-treate group), and binding of GP70 tetramers (G) (n = 6 for vehicle and 8 for G9668 group). Analysis was performed using one-way ANOVA for panel A and unpaired, two-tailed t test for panels B-G, * P ≤ 0.05 *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001. Scatter plots in all panels represent mean ± SD.  Mice were inoculated s.c. with parental or IRE1α KO 4T1 cells, grouped out 7 days afterwards and treated with vehicle or G9668 (250 mg/kg) (n = 5). Abundance of CD8 + T cells in the tumor and expression of indicated activation markers were assayed by flow cytometry. Analysis was performed using one-way ANOVA for panel A and unpaired, two-tailed t test for panels C-E, *P ≤ 0.05, **P ≤ 0.01, ****P ≤ 0.0001. Scatter plots represent mean ± SD.   Total DCs were characterized as F4/80 low , CD11c + and class II MHC high , while cDC1s were characterized as CD103 + XCR1 + CD11b -. (C) Abundance and activation marker expression of tumor-infiltrating CD8 + T cells were assayed by flow cytometry. Analysis was performed using oneway ANOVA for panel A and unpaired, two-tailed t test for panels B and C, * P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001. Scatter plots in all panels represent mean ± SD.