IRE1α Disruption in Triple-Negative Breast Cancer Cooperates with Antiangiogenic Therapy by Reversing ER Stress Adaptation and Remodeling the Tumor Microenvironment

Among the main breast cancer subtypes, triple-negative breast cancer (TNBC) accounts for 15%–20% of total incidence, and holds the most urgent need for effective therapy. It is defined immunohistochemically by absent expression of three key markers: estrogen receptor α (ERα), progesterone receptor (PR), and HER2 (or ERBB2/NEU). TNBC is an early-onset, highly aggressive malignancy, with dismal prognosis post standard-of-care chemotherapy (1, 2).

The unfolded protein response (UPR) is an intracellular sensing-signaling network that helps cells mitigate stress-driven perturbations to protein biosynthetic three-dimensional (3D) folding within the endoplasmic reticulum (ER; refs. 3–5). The mammalian UPR comprises a triad of ER membrane–resident sensors: IRE1α (inositol-requiring enzyme 1α), PERK (protein kinase-like ER kinase), and ATF6 (activating transcription factor-6). Upon direct or indirect detection of unfolded proteins through an ER-lumenal domain, each UPR sensor engages its own cytoplasmic moiety to adjust the ER's capacity to fold proteins and synthesize membranes, thereby helping to alleviate ER stress. If mitigation fails, the UPR induces apoptosis (6). IRE1α contains a cytosolic serine/threonine kinase domain, which controls activation of a tandem endoribonuclease (RNase) moiety (7, 8). Under ER stress, IRE1α dimerizes and transautophosphorylates, thereby activating its RNase module (8–11). The RNase excises 26 nucleotides from the mRNA encoding unspliced X-box protein 1 (XBP1u), causing a frame shift after RtcB-mediated ligation of the excised exons, to produce an mRNA encoding spliced XBP1 (XBP1s; refs. 3, 4, 12, 13). XBP1s is a transcription factor that stimulates multiple genes with adaptive and cytoprotective functions to facilitate ER-stress mitigation (14–16). In addition to promoting XBP1 mRNA splicing, the IRE1α RNase degrades ER-targeted mRNAs—a process termed regulated IRE1α-dependent decay, or RIDD—to abate translation (17, 18), suppress apoptosis (19, 20), and augment protective autophagy (21).

During cancer initiation, progression, and metastasis, tumor cells face various types of intrinsic and extrinsic stress, caused by activation of oncogenes and by metabolically restrictive tumor microenvironments (TME). Such stress conditions can overburden or perturb the ER's protein-folding functions, driving certain types of cancer cells to activate the UPR as a means of sustaining malignant growth while retaining viability (22–24). In ERα⁺ breast cancer, the UPR in general, and XBP1s in particular, contribute to acquired resistance against antiendocrine therapy (25). TNBC also coopts the UPR, as shown by seminal published work implicating XBP1s in conjunction with HIF1α in driving TNBC tumor angiogenesis and progression under hypoxia (26). More recent studies also have linked the IRE1α–XBP1s axis to the oncogenic transcription factor MYC—a potent driver of proliferation and protein synthesis (27)—in TNBC (28), prostate cancer (29), and B-cell lymphoma (30). Pharmacologic inhibition of IRE1α via its RNase domain markedly augmented the effectiveness of taxane chemotherapy in cell line–based or patient-derived xenograft TNBC models (28, 31). Analysis by transmission electron microscopy revealed a highly dilated ER in TNBC cell lines grown in vitro, suggesting volumetric ER expansion and adaptation to stress (26). Similar changes were observed in MYC expressing B-cell lymphomas (30).

The breast cancer TME has a complex cellular character, comprising both malignant and nonmalignant cells. Of the latter category, endothelial cells, cancer-associated fibroblasts (CAF), and myeloid-derived suppressor cells (MDSC) play key roles in supporting the development and progression of breast cancer (22, 32, 33). Higher microvascular density generally correlates with greater tumor burden, more advanced grade, more frequent lymph-node metastasis, and poorer prognosis (34, 35). Tumor vascularization is more extensive in TNBC relative to other breast cancers (34, 36), suggesting greater dependency on angiogenesis. Cancer cells actively promote tumor vascularization by secreting several proangiogenic growth factors, among which, VEGFA plays a prominent role (35). Breast cancer clinical trials have been performed with bevacizumab—a humanized mAb targeting VEGFA, which is FDA approved for certain other cancer indications—in combination with different chemotherapy regimens. Disappointingly, these trials failed to demonstrate an overall survival benefit. Nevertheless, subgroup analysis revealed a considerable improvement of overall response rates and progression-free survival, particularly in TNBC (34). Therefore, the treatment of TNBC may ultimately benefit from complementation of antiangiogenic therapy with mechanistically distinct novel strategies. Previous work established a biological link between XBP1s, hypoxia, and tumor angiogenesis (37, 38). Furthermore, XBP1s expression was shown to enhance VEGFA production and vascularization in TNBC tumor models and to correlate with worse prognosis in patients with TNBC (26, 28, 39).

CAFs are the most abundant stromal cell type in the TME and play a central role in promoting breast cancer tumor vascularization, growth, invasiveness, and treatment resistance (33). Indeed, CAFs correlate with a more aggressive breast cancer phenotype and poorer patient survival (40). Similarly, MDSCs exert not only immunosuppressive functions but also directly stimulate tumor growth, tumor vascularization, and metastasis (32). Accordingly, novel therapeutic strategies to diminish the abundance or activity of CAFs and MDSCs in the TME are potentially important.

Despite significant advances in understanding of the role of the IRE1α pathway in TNBC, several key mechanistic and therapeutic questions remain: (1) How important is IRE1α for ER adaptation in malignant TNBC cells? (2) Given its function in facilitating protein folding and secretion, does IRE1α modulate the cellular composition of the TME? (3) Does pharmacologic inhibition of IRE1α via its kinase domain, rather than RNase moiety, inhibit TNBC tumor growth? (4) Can IRE1α inhibition complement the insufficient efficacy of antiangiogenic therapy in TNBC?

Our results have mechanistic as well as therapeutic potential implications for the IRE1α pathway in TNBC. Mechanistically, we show that IRE1α knockout (KO) fully reverses the marked ultrastructural adaptation of the ER in TNBC cells growing in vivo. Furthermore, IRE1α disruption substantially remodels the TME in TNBC by increasing pericyte levels and promoting vascular normalization, while decreasing CAFs and MDSCs. Therapeutically, pharmacologic inhibition of IRE1α through its kinase module substantially attenuates tumor growth in models of both cell line–based and patient-derived TNBC xenografts in mice. Importantly, IRE1α inhibition cooperates with anti-VEGFA antibody treatment to achieve frequent TNBC tumor regression. Our results open promising new avenues to investigate these two evidently complementary therapeutic approaches to develop more effective biological treatments for TNBC.

Detailed methods are provided in Supplementary data.

All cell lines were obtained or generated from an internal repository maintained at Genentech. Short tandem repeat (STR) profiles were determined using the Promega PowerPlex 16 System, and compared with external STR profiles of cell lines to determine cell-line ancestry. SNP profiles were performed each time new stocks were expanded for cryopreservation. Cell-line identity was verified by high-throughput SNP profiling using Fluidigm multiplexed assays. SNPs were selected on the basis of minor allele frequency and presence on commercial genotyping platforms. SNP profiles were compared with SNP calls from available internal and external data (when available) to confirm ancestry.

Cell lines were tested to ensure they were mycoplasma free prior to and after cells were cryopreserved. Two methods were used to avoid false positive/negative results: Lonza Mycoalert and Stratagene Mycosensor. In addition, cell growth rates and morphology were monitored for any batch-to-batch changes. All cell lines were cultured in RPMI1640 media supplemented with 10% (vol/vol) FBS (Sigma), 2 mmol/L glutaMAX (Gibco), and 100 U/mL penicillin plus 100 μg/mL streptomycin (Gibco) over 2 to 3 passages before use. Thapsigargin (Sigma) was used at a concentration of 100 nmol/L.

A recombinant protein encoding the kinase and RNase domains of human IRE1α (amino acids 547–977) was generated via a baculovirus expression system in SF9 cells and purified to homogeneity using a tobacco etch virus-protease cleavable His6 tag. Mice were immunized using standard protocols and mAbs were screened by Western blot analysis against recombinant purified IRE1α lumenal and cytoplasmic domain proteins or lysates from MDA-MB-231 cells expressing wild-type IRE1 or harboring CRISPR/Cas9 KO of IRE1α. A mouse IgG2a mAb that specifically and selectively detected the human IRE1α (GN35-18) was thus isolated and cloned.

RNA-sequencing (seq) counts for genes of the ER16 signature were obtained from the Cancer Genome Atlas (TCGA), and normalized by multiplying the counts in each library by the size factor computed from DESeq2 package in R (R Foundation for Statistical Computing). For each gene, a Z-score was computed by subtracting its mean and dividing by its SD over all samples in the dataset. A signature score for ER16 was computed by taking the mean of these Z-scores over all 16 genes in the signature. Analysis of cell lines was performed in a similar fashion, except using RNA-seq data from Genentech centralized cell bank (gCell) dataset (https://www.gene.com). Correlations of each gene against the ER16 score were computed as the Pearson correlation coefficient using the cor function in R. All plots of ER16 signature scores utilize an exponential scale to show values from 0 to infinity, with the mean score over all samples indicated by a value of 1.

Annotation of breast cancer samples in TCGA was obtained from the Genomic Data Commons portal at https://portal.gdc.cancer.gov/legacy-archive/files/735bc5ff-86d1-421a-8693-6e6f92055563. Breast cancer samples were classified as HER2-positive if the field her2_status_by_ihc was “Positive”; as ER/PR-positive if the field her2_status_by_ihc was “Negative” and either er_status_by_ihc or pr_status_by_ihc was “Positive”; and as triple-negative if all of the above-mentioned fields were “Negative.”

For statistical analysis, a Student t test was performed between Z-scores on all pairs of indications of TCGA that had both a cancer and normal subtype, with each cancer subtype compared against its corresponding normal. Breast cancer subtypes, as described previously, were considered beyond the annotations from TCGA. P values were computed as one sided with the null hypothesis that the normal subtype had greater expression than the cancer subtype. FDR were computed over all P values using the Benjamini–Hochberg procedure using the q value function from the q value package in R with pi0 set to 1.

In all in vivo studies, 5 × 10⁶ MDA-MB-231 or 2 × 10⁶ HCC1806 tumor cells and their corresponding KO clones or shRNA expressing cells were suspended in Hank's Balanced Salt Solution (HBSS), admixed with 50% Matrigel (Corning) to a final volume of 100 μL, and injected subcutaneously in the right flank of 6- to 8-week old female C.B-17 SCID (MDA-MB-231 model) or SCID.bg (HCC1806 model) mice (Charles River Laboratories).

For efficacy studies, tumors were monitored until they reached a mean tumor volume of approximately 150 mm³ (MDA-MB-231), or 150–400 mm³ (HCC1806). To test efficacy of doxycycline-inducible shRNA in HCC1806 tumor xenografts, animals were randomized into the following treatment groups: (i) 5% sucrose water (provided in drinking water, changed weekly); or (ii) doxycycline (0.5 mg/mL, dissolved in 5% sucrose water, changed 3×/week).

To assess efficacy of compound 18 in combination with anti-VEGFA, animals were randomized into one of the following treatment groups: (i) vehicle control (35% PEG400 and 10% EtOH in water, 100 μL total, intraperitoneally (i.p.), every day) and anti-ragweed 1428 (2 mg/kg, 100 μL, i.p., twice per week); (ii) compound 18 [(41, 42) 30 mg/kg, 100 μL, i.p., every day] and anti-ragweed 1428; (iii) anti-VEGFA [B20-4.1.1 (43), 2 mg/kg, 100 μL, i.p., twice per week) and vehicle control; or (iv) compound 18 and anti-VEGFA.

To analyze tumor xenograft vessel leakiness, 200 μL of FITC-dextran (15 mg/mL in PBS, MW 2,000,000; SigmaAldrich) was injected into the tail vein of mice 2 hours before animals were sacrificed. Xenografts were excised and fixed in 4% paraformaldehyde for storage and further processing.

The tumor graft line HCI-004 was used, which is a patient-derived xenograft (PDX) model derived from a primary TNBC of an infiltrating ductal carcinoma (44). Tumor pieces were surgically transplanted into the right #2/3 mammary fat pad of female 6- to 7-week old NOD SCID mice (Charles River Laboratories). When donor mice had tumors of around 1,500 mm³, animals were humanely euthanized as outlined below and size-matched xenografts of two mice aseptically collected. Xenografts were rinsed in HBSS, sectioned into 2 mm³ pieces and then surgically engrafted as described above in an experimental cohort of female NOD SCID mice. Tumors were monitored until they reached a mean volume of approximately 150 mm³, and to test the efficacy of compound 18, anti-VEGFA, or the combination thereof, animals were randomized into treatment groups as outlined above.

In all in vivo studies, tumor size and body weight were measured twice per week. Subcutaneous and mammary fat pad tumor volumes were measured in two dimensions (length and width) using Ultra Cal-IV calipers (model 54 − 10 − 111; Fred V. Fowler Co.). The tumor volume was calculated using the following formula: tumor size (mm³) = (longer measurement × shorter measurement²) × 0.5. Animal body weights were measured using an Adventurer Pro AV812 scale (Ohaus Corporation). Percent weight change was calculated using the following formula: group percent weight change = [(new weight − initial weight)/initial weight] × 100. To analyze repeated measurements of tumor volumes from the same animals over time, a generalized additive mixed modeling approach was used (45). This approach addresses both repeated measurements and modest dropouts before the end of study. Cubic regression splines were used to fit a nonlinear profile to the time courses of log₂ tumor volume in each group. Fitting was done via a linear mixed-effect model, using the package “nlme” (version 3.1-108) in R version 2.15.2 (R Development Core Team 2008; R Foundation for Statistical Computing). Tumor growth inhibition (TGI) as a percentage of vehicle was calculated as the percent difference between the daily average area under the tumor volume–time curve (AUC) of treatment and control group fits on the original untransformed scale over the same time period using the following formula: %TGI = (1−[(AUC/day)treatment ÷ (AUC/day)vehicle]) × 100. All AUC calculations were baseline adjusted to the initial tumor burden on day 0. As such, positive values indicate an antitumor effect (100% TGI equals tumor stasis and TGI values >100% denote tumor regression). Values in parenthesis indicate the upper and lower boundaries of the 95% confidence interval for the percent difference based on the fitted model and variability measures of the data.

All procedures were approved by and conformed to the guidelines and principles set by the Institutional Animal Care and Use Committee (IACUC) of Genentech and were carried out in an Association for the Assessment and Accreditation of Laboratory Animal Care–accredited facility.

When mice reached endpoint criteria or on the last treatment day, mice were euthanized and xenografts harvested for further analysis. Animals in all studies were humanely euthanized according to the following criteria: clinical signs of persistent distress or pain, significant body-weight loss (>20%), tumor size exceeding 2,500 mm³, or when tumors became ulcerated. Maximum tumor size permitted by the IACUC is 3,000 mm³ and in none of the experiments was this limit exceeded.

All values are represented as arithmetic mean ± SD with at least three independent biological or technical replicates experiments if not otherwise indicated in the figure legends. Statistical analysis of the results was performed by unpaired, two-tailed t test or one-way ANOVA followed by an appropriate post-hoc analysis, including Bonferroni correction to compensate for multiple comparisons. A P value < 0.05 was considered significant, and denoted by *, P < 0.05; **, P < 0.01; ***, P < 0.001. All statistical analyses were performed using GraphPad Prism 6 (GraphPad Software, Inc.). For statistical analysis of ER16, scores refer to the corresponding section in Supplementary data.

Close sequence similarity between the two splice variants of XBP1 mRNA makes their discrimination within standard RNA-seq datasets relatively difficult. Therefore, to gauge more precisely activation of the IRE1α–XBP1s pathway in various cancers, we assembled a gene signature comprising 16 transcripts (dubbed ER16) found to be significantly upregulated in an XBP1s-dependent manner under ER stress (16, 46). These genes displayed correlated expression in tumors in the TCGA database (Supplementary Fig. S1A), confirming their coregulation. Several solid-tumor types, including urothelial bladder, breast, colon, head and neck, kidney, liver, lung, stomach, and uterine cancer, showed significantly elevated ER16 expression as compared with matched normal-tissue controls (Fig. 1A; Supplementary Fig. S1B). Although ER/PR⁺, HER2⁺, and TNBC breast tumors all had significant ER16 upregulation, we focused on TNBC, which urgently requires more effective biotherapies.

To assess tissue expression of IRE1α protein, we developed an IHC assay, based on a newly generated mouse anti-human IRE1α mAb (GN35-18). We validated this antibody as specifically recognizing human IRE1α's kinase domain by using purified recombinant IRE1α protein fragments and cells expressing or lacking IRE1α (Supplementary Fig. S1C–S1F). IHC analysis of 152 TNBC tumors via tissue microarrays revealed significant expression of the IRE1α protein in breast epithelial cells, with IHC scores ranging from 0 (40%), through 1+ (47%), to 2+ (13%; Fig. 1B). Thus, nearly two-thirds of TNBC specimens express the IRE1α protein at levels detectable by IHC.

We also examined ER16 expression in a panel of 34 breast cancer cell lines (Supplementary Fig. S2A; ref. 47). Consistent with TCGA data, ER/PR⁺, HER2⁺, and TNBC lines showed comparable ER16 levels. Further analysis by immunoblot (IB) using anti-IRE1α and anti-XBP1s antibodies revealed detectable baseline levels of IRE1α, but not XBP1s (Fig. 2A). Nevertheless, exposure to the pharmacologic ER stressor Thapsigargin induced significant XBP1s protein amounts in all cases. Other studies detected baseline XBP1s in certain TNBC models (26, 28, 31): we suspect this divergence is due primarily to differences in cell culture conditions, for example, number of passages or confluence. Regardless, although TNBC cell lines express variable levels of IRE1α, this protein appears to be uniformly poised to generate XBP1s in response to ER stress.

While shRNA knockdown of XBP1 attenuated TNBC growth (26, 28), the effect of upstream genetic IRE1α perturbation in this context is not well defined. To examine whether TNBC cells require IRE1α itself, we disrupted the gene via CRISPR/Cas9 technology in MDA-MB-231 and HCC1806 cells; for comparison, we also disrupted XBP1 in MDA-MB-231 cells (Supplementary Fig. S2B and S2C). To control for interclonal variation, we characterized two independent KO clones for each gene.

Whereas MDA-MB-231 cells showed little dependency on either IRE1α or XBP1 for two-dimensional (2D) growth on standard tissue culture plates, they displayed a strong reliance on either gene for 3D colony formation on soft agar and for survival on ultralow adhesion (ULA) plates (Fig. 2B–D). Consistent with the loss of viability, MDA-MB-231 cells harboring IRE1α or XBP1 KO displayed increased proapoptotic caspase-3/7 activity when cultured on ULA plates (Fig. 2E). HCC1806 cells similarly showed little requirement for IRE1α in 2D; strong dependency on IRE1α in 3D (Fig. 2F–H); and elevated caspase-3/7 activity during ULA culture (Fig. 2I). IB analysis of MDA-MB-231 cells revealed markedly elevated activity of IRE1α in 3D versus 2D settings, as evident by increased IRE1α protein abundance and phosphorylation, as well as elevated XBP1s protein (Fig. 2J). These results suggest that TNBC cells rely more heavily on the IRE1α–XBP1s pathway in 3D growth settings, where cell–cell adhesion is more critical for growth and survival. Supporting this idea, an RNA-seq comparison of parental versus IRE1α KO and XBP1 KO MDA-MB-231 cells revealed significant downregulation of numerous transcripts annotated with gene-ontology terms of cell–cell adhesion (Supplementary Fig. S2D).

To interrogate the importance of IRE1α for growth of TNBC tumors in vivo, we first tested the MDA-MB-231 model, which harbors MYC amplification. Albeit with some variation, MDA-MB-231 clones harboring IRE1α KO displayed markedly attenuated subcutaneous xenograft tumor growth for at least 35 days, as did clones with XBP1 KO (Fig. 3A and B; Supplementary Fig. S3A and S3B). Similarly, IRE1α KO clones of the HCC1806 cell line, which also carries amplified MYC, showed markedly abrogated xenograft tumor growth for at least 35 days (Fig. 3C; Supplementary Fig. S3C). Furthermore, doxycycline-inducible shRNA-based depletion of IRE1α or XBP1, but not respective control shRNA, substantially inhibited HCC1806 tumor growth (Fig. 3D and E; Supplementary Fig. S3D and S3E). Importantly, reconstitution of HCC1806 cells harboring shRNA-mediated knockdown of endogenous IRE1α with exogenous IRE1α (^RIRE1α) rescued tumor growth (Fig. 3F; Supplementary Fig. S3F and S3G). These data show that TNBC cells require IRE1α not only for in vivo tumor initiation but also for malignant tumor progression. TNBC cell lines with MYC amplification strongly depend on IRE1α for in vivo growth, extending and reinforcing earlier studies of XBP1s (26, 28) to IRE1α—a more readily druggable target.

Cancer cells often expand their ER to alleviate ER stress caused by intrinsic oncogene activation and by metabolically restrictive TMEs. Indeed, transmission electron microscopy–based analysis of TNBC cells cultured in vitro revealed an abnormally dilated ER, indicating stress adaptation (26). However, it is unknown whether this distended ER ultrastructure is maintained in vivo, and—perhaps more importantly—to what extent such adaptation depends on the IRE1α pathway. To address these questions, we examined size-matched MDA-MB-231 tumor xenografts by backscattered electron–scanning electron microscopy (BSE-SEM), which enables very powerful image magnification, zoomed-in from the whole tissue level to the subcellular ultrastructure level (48) (Fig. 4A). The parental tumor cells displayed a notable dilation of their ER as well as frequent bulging of their nuclear envelope, indicating marked ER adaptation within the TME. Strikingly, IRE1α KO tumor cells showed an essentially complete reversal of these features. Quantification of multiple BSE-SEM images confirmed a significant reduction in the number of cells with distended ER in both IRE1α KO clones as compared with parental MDA-MB-231 tumor xenografts (Fig. 4B). Thus, TNBC cells critically depend on IRE1α to adapt their ER and nuclear envelope to in vivo growth within the TME.

Given this cell-autonomous impact of IRE1α disruption, we turned to investigate whether IRE1α KO in malignant TNBC cells also exerts cell nonautonomous effects on the TME. Consistent with earlier data for XBP1 shRNA (26, 28), IRE1α KO markedly decreased the levels of VEGFA mRNA and protein expression, as well as VEGFA secretion, in MDA-MB-231 tumor xenografts (Supplementary Fig. S4A–S4C). RNA-seq analysis of tumor samples further indicated that multiple genes involved in the promotion of angiogenesis were downregulated upon KO of IRE1α or XBP1 (Supplementary Fig. S4D). We tested and confirmed this observation by qRT-PCR analysis of angiogenin mRNA (Supplementary Fig. S4E). Moreover, IB analysis of MDA-MB-231 xenografts verified that IRE1α KO diminished the abundance of several proangiogenic proteins, namely, fibroblast growth factor receptor 1 (FGFR1), angiopoietin-1, and angiogenin (Supplementary Fig. S4F).

Normalization of the tumor vasculature is critical for the effectiveness of antiangiogenic therapy (49–51). However, whether the IRE1α pathway affects this endpoint is unknown. To address this question, we performed vascular perfusion of tumors with FITC-dextran, followed by microscopic analysis of tumor tissue by IHC (Fig. 4C). Viable parts of IRE1α KO MDA-MB-231 tumors contained significantly less FITC dextran–positive areas than did parental controls, suggesting a marked decrease in leakage of this large molecule from within tumor vessels into the extravascular tissue space. Further analysis of tumors by FACS revealed that the levels of CD31-positive endothelial cells were unaffected (Fig. 4D). In keeping, IHC analysis of pan-endothelial cell antigen (MECA)-32 staining showed only a small, albeit statistically significant, decrease in IRE1α KO versus parental tumors (Supplementary Fig. S4G). In contrast, numbers of Thy1-positive pericytes were significantly increased upon IRE1α KO (Fig. 4E). In agreement with these findings, flow cytometric analysis of HCC1806 tumor xenografts showed an insignificant decrease in CD31-positive endothelial cells, but a significant elevation in Thy1-positive pericytes upon IRE1α KO (Supplementary Fig. S4H and S4I). Together, these results indicate that IRE1α perturbation in the malignant TNBC cells increases the level of pericytes relative to endothelial cells within the TME, thus providing a likely mechanistic explanation for the observed tumor-vascular normalization upon IRE1α disruption.

In light of our findings with genetic IRE1α disruption, we turned to investigate the effect of pharmacologic IRE1α inhibition, which would block the pathway both in malignant and nonmalignant cells within the TME. In previous studies, inhibition of IRE1α at the RNase level attenuated growth of TNBC tumor xenografts, particularly when performed in conjunction with taxane chemotherapy (28, 31). However, the impact of likely more target-selective pharmacologic perturbation of IRE1α at the kinase level on TNBC tumor growth is unknown. To determine this, we used compound 18—previously validated as a selective and effective allosteric inhibitor of IRE1α that acts via binding to the ATP pocket within the kinase domain (41, 42). Given that XBP1s cooperates with HIF1α to upregulate VEGFA (26), and the vascular changes we observed upon IRE1α KO in TNBC cells, we compared the antitumor activity of compound 18 with that of a previously established anti-VEGFA neutralizing antibody, which blocks both murine and human VEGFA (43, 52). In the MDA-MB-231 model, compound 18 treatment caused significant TGI (TGI, 93%; range, 81%–102%). Anti-VEGFA monotherapy showed weaker efficacy (TGI, 62%; range, 43%–76%). Remarkably, treatment with both agents together enabled complete suppression of tumor growth in 6 of 14 mice and tumor regression in 8 of 14 mice (TGI, 121%; range, 112%–131%; P < 0.05 as compared with compound 18 monotherapy; Fig. 5A; Supplementary Fig. S5A–S5C). Similarly, in the HCC1806 model, compound 18 treatment also attenuated tumor growth more substantially (TGI, 78%; range, 69%–85%) than anti-VEGFA monotherapy (TGI, 54%; range, 38%–64%), while combination of both treatments led to complete tumor growth suppression in 13 of 20 mice and tumor regression in 7 of 20 mice (TGI, 109%; range, 105%–113%; P < 0.001 as compared with compound 18 monotherapy; Fig. 5B; Supplementary Fig. S5D–S5F).

Next, we turned to PDX models, which set a higher bar for preclinical validation of potential therapeutic targets. We first determined IRE1α expression by IHC and IB in three PDX TNBC tissue samples (Supplementary Fig. S5G). While all three tumors displayed detectable IRE1α protein, HCI-004, which harbors amplified MYC, had higher levels, comparable with other TNBC samples scored earlier as 2+ (Fig. 1B). IB analysis confirmed that HCI-004 expressed relatively higher levels of both IRE1α and XBP1s (Supplementary Fig. S5H), suggesting stronger IRE1α pathway engagement. We therefore tested this model in vivo, by orthotopic transplantation of tumor tissue into the mammary fat pad of female NOD-SCID mice (Fig. 5C; Supplementary Fig. S5I–S5K). Monotherapy with compound 18 substantially inhibited tumor growth (TGI, 89%; range, 78%–96%), while anti-VEGFA administration showed less efficacy (TGI, 51%; range, 38%–62%). Again, combination treatment was markedly more effective, leading to tumor regression in 14/14 animals (TGI, 117%; range, 110%–126%; P < 0.05 as compared with compound 18 monotherapy). Thus, IRE1α kinase inhibition strongly attenuates TNBC tumor growth in cell line–based subcutaneous as well as patient-derived orthotopic models. While IRE1α kinase inhibition provided substantial benefit, the two modalities cooperated to achieve significantly better tumor growth control or even a reversal of tumor progression.

Consistent with our earlier observations upon IRE1α KO (Fig. 4E), treatment of MDA-MB-231 tumor-bearing mice with compound 18 diminished serum levels of human VEGFA (Supplementary Fig. S5L) and decreased vascular leakage as judged by FITC-dextran staining (Supplementary Fig. S5M). We next explored whether IRE1α disruption also impacts nonvascular gene targets and cells within the TME. RNA-seq comparison of parental and IRE1α or XBP1 KO MDA-MB-231 xenografts revealed transcriptional downregulation of several genes known to be involved in the recruitment of CAFs and MDSCs, namely, Cox-2 (PTGS2), CXCL8, and CXCR4 (Supplementary Fig. S4D; refs. 53–55). Deconvolution analysis of bulk RNA-seq data comparing MDA-MB-231 parental and IRE1α KO xenografts, by referring to reported single-cell RNA-seq gene signatures for CAFs (56), suggested that podoplanin (Pdpn) and platelet-derived growth factor receptor A (Pdgfrα) double-positive CAFs (dubbed CAF2), which are considered to drive the desmoplastic reaction associated with tumor development, are reduced in the TME (Supplementary Fig. S5N). To assess whether this modulation of the TME could contribute to the observed synergy between IRE1α and VEGFA, we examined the expression of Cox-2 (PTGS2), CXCL8, and CXCR4 in the HCC1806 model by comparing RNA-seq data of parental and IRE1α KO xenografts treated with or without anti-VEGFA. Consistent with the results in the MDA-MB-231 model, perturbation of IRE1α—but importantly, not anti-VEGFA treatment—strongly diminished the expression of these three genes (Fig. 5D). In light of these results, we investigated tumor infiltration of CAFs and MDSCs in MDA-MB-231 and HCC1806 tumor xenografts. Flow cytometric analysis of Pdpn and Pdgfrα double-positive cells, as well as IHC analysis of fibroblast activation protein (FAP) staining suggested that compound 18 treatment alone or in combination with anti-VEGFA—but again, not anti-VEGFA treatment alone—significantly decreased tumor levels of CAF2 cells (Fig. 5E–G). Furthermore, flow cytometric analysis also indicated a significantly diminished number of tumor-infiltrating MDSCs following perturbation of IRE1α alone or in combination with anti-VEGFA—but not upon anti-VEGFA monotherapy (Supplementary Fig. S5O and S5P). These results suggest that IRE1α inhibition uniquely disrupts tumor growth by remodeling the TME.

Our work conceptually advances the current mechanistic and therapeutic understanding of the IRE1α pathway in TNBC. In particular, we demonstrate that (1) TNBC cells critically rely on IRE1α to adjust their ER for malignant in vivo growth; (2) Disruption of IRE1α impairs vascular density and decreases vessel leakage within TNBC tumors by increasing pericyte to endothelial cell ratios; (3) IRE1α inhibition decreases Cox-2 (PTGS2), CXCL8, and CXCR4 mRNA, reducing the number of tumor-infiltrating CAF2 cells and MDSCs—cell types conducive to tumor vascularization and malignant progression; (4) Inhibition of IRE1α at the kinase level significantly attenuates TNBC tumor growth as monotherapy and cooperates with anti-VEGFA antibody to cause frequent tumor regression.

Intriguingly, TNBC cells growing in vitro in 3D settings as compared with 2D culture exhibited markedly increased activation of the IRE1α pathway and a much stronger dependency on IRE1α. These data are reminiscent of our recent observations with multiple myeloma cells (42). We hypothesize that both types of cancer cells require IRE1α-supported cell–cell adhesion mechanisms for survival in 3D growth settings. Consistent with this notion, RNA-seq analysis showed that IRE1α or XBP1 KO in TNBC cells downregulates numerous transcripts known to enable cell–cell adhesion, which implicates the IRE1α pathway as a key facilitator of this important cellular function.

In our in vivo experiments, KO or knockdown of either IRE1α or XBP1 substantially inhibited growth of TNBC xenografts, demonstrating a critical role for IRE1α itself in driving TNBC progression (26, 28). Mechanistically, we observed a strong cell-autonomous requirement of IRE1α for ultrastructural ER and nuclear envelope adaptation in TNBC cells. These IRE1α-dependent changes appeared to have major consequences not only directly, for the malignant cells, but also indirectly, for other cell types within the TME. Cell nonautonomous changes caused by IRE1α pathway disruption may be linked to altered folding and secretion of ER-client proteins destined for the cell surface and extracellular space. Our finding that IRE1α KO decreases VEGFA transcript and protein levels as well as VEGFA secretion in TNBC tumor xenografts extends earlier XBP1s data (26) to IRE1α upstream. Furthermore, we obtained evidence suggesting that IRE1α disruption decreases vascular leakage by promoting vascular pericyte coverage—likely a beneficial aspect of antiangiogenic therapy that may improve drug delivery to tumors and decrease tumor invasion and metastasis (49–51). The IRE1α kinase inhibitor used in this study as a tool molecule afforded marked suppression of tumor XBP1s levels and displayed significant single-agent efficacy, reaching approximately 80–90% TGI. This demonstrates for the first time that kinase-based IRE1α inhibition is therapeutically efficacious in models of a solid-tumor malignancy.

In breast cancer clinical trials, anti-VEGFA therapy was not sufficiently beneficial to garner FDA approval (34). In this preclinical study, anti-VEGFA antibody treatment afforded approximately 50%–60% TGI in three distinct TNBC models. IRE1α perturbation showed superior efficacy by exerting stronger changes in the TME, significantly decreasing the number of tumor-infiltrating CAFs and MDSCs. Our RNA-seq analyses of xenograft samples from two distinct TNBC models detected several IRE1α- and XBP1-dependent genes, that is, Cox-2, CXCL8, and CXCR4, which could individually or in concert contribute to the reduction of CAFs and MDSCs upon IRE1α perturbation. CAFs promote angiogenesis and tumor progression in breast cancer via SDF-1/CXCR4 signaling (57). Tumor epithelial expression of Cox-2 correlates with higher CAF numbers in the TME of TNBC xenografts and promotes tumor progression in breast cancer models (53, 58, 59). Furthermore, the Cox2–PGE2 pathway drives angiogenesis and tumor growth independent of VEGFA (59). Importantly, disruption of the Cox2–PGE2 pathway also attenuates MDSC recruitment via the CXCL8–CXCR1/2 and CXCL12–CXCR4 signaling pathways (54, 55). Together, our results suggest that IRE1α inhibition, likely in conjunction with consequent Cox-2 downregulation, disrupts TNBC tumor growth by remodeling the cellular composition of the TME in modes that are both overlapping (i.e., tumor vasculature) as well as complementary (i.e., CAF2 cells and MDSCs) to VEGFA inhibition, enabling synergy between the two therapeutic strategies. These findings support investigating this type of combination approach in the clinic, provided that appropriate safety can be established in suitable preclinical models. Of note in this context, IRE1α kinase inhibition was not associated with overt toxicity in mice, nor did it disrupt viability of primary human hepatocytes and glucose-induced insulin secretion by human pancreatic microislet cultures (42).

In conclusion, we have demonstrated a critical role for IRE1α in adapting the ER of malignant TNBC cells and remodeling their TME to facilitate malignant tumor progression. This work provides a compelling rationale for small-molecule targeting of IRE1α in this yet unconquered lethal disease. Furthermore, IRE1α inhibition has the potential to augment the efficacy of anti-angiogenic therapy—for TNBC, and perhaps beyond. Evidently, the IRE1α pathway is activated in several additional solid-tumor malignancies, which supports exploring whether similar mechanistic-therapeutic concepts apply more broadly to other cancers.

J.M. Harnoss, A. Le Thomas, and O. Guttman are postdoctoral fellows; M. Reichelt is principal scientific researcher; T.D. Wu, Z. Modrusan, and J. Rudolph are principal scientists; S.A. Marsters is principal research project manager II; E. Segal is senior scientific manager; D.A. Lawrence is senior research associate; A. Shemorry, D. Kan, L.M. Crocker, and K. Mesh are senior scientific researchers; and A. Ashkenazi is senior staff scientist at Genentech. M. Merchant is an associate director at Roche/Genentech and has ownership interest (including patents) in Roche. H. Koeppen is senior staff pathologist at Roche/Genentech and has ownership interest (including patents) in Roche shares. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J.M. Harnoss, A. Shemorry, J. Rudolph, A. Ashkenazi

Development of methodology: J.M. Harnoss, M. Reichelt, S.A. Marsters, A. Shemorry, D.A. Lawrence, E. Segal, K. Mesh, M. Solon, H. Koeppen, A. Ashkenazi

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.M. Harnoss, A. Le Thomas, M. Reichelt, O. Guttman, S.A. Marsters, A. Shemorry, D.A. Lawrence, D. Kan, E. Segal, M. Merchant, K. Totpal, L.M. Crocker, Z. Modrusan, H. Koeppen

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.M. Harnoss, A. Le Thomas, M. Reichelt, O. Guttman, T.D. Wu, E. Segal, M. Merchant, K. Totpal, K. Mesh, H. Koeppen, A. Ashkenazi

Writing, review, and/or revision of the manuscript: J.M. Harnoss, A. Le Thomas, O. Guttman, S.A. Marsters, E. Segal, L.M. Crocker, J. Rudolph, H. Koeppen, P. Walter, A. Ashkenazi

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.M. Harnoss, M. Merchant

Study supervision: E. Segal, K. Totpal, J. Rudolph, P. Walter, A. Ashkenazi

This work was supported in part by a Howard Hughes Collaborative Innovation Award to P. Walter, who is an investigator of the Howard Hughes Medical Institute.

We thank Marie Gabrielle-Braun for compound synthesis, Scott Stawicki for antibody generation, Jing Qing for shRNA design, Ben Haley for gRNA design, and Stephen Gould, Bob Yauch, Frank Peale, Daniel Sutherlin, and Ira Mellman for helpful 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.