Targeting LIPA independent of its lipase activity is a therapeutic strategy in solid tumors via induction of endoplasmic reticulum stress

Triple-negative breast cancer (TNBC) has a poor clinical outcome, due to a lack of actionable therapeutic targets. Herein we define lysosomal acid lipase A (LIPA) as a viable molecular target in TNBC and identify a stereospecific small molecule (ERX-41) that binds LIPA. ERX-41 induces endoplasmic reticulum (ER) stress resulting in cell death, and this effect is on target as evidenced by specific LIPA mutations providing resistance. Importantly, we demonstrate that ERX-41 activity is independent of LIPA lipase function but dependent on its ER localization. Mechanistically, ERX-41 binding of LIPA decreases expression of multiple ER-resident proteins involved in protein folding. This targeted vulnerability has a large therapeutic window, with no adverse effects either on normal mammary epithelial cells or in mice. Our study implicates a targeted strategy for solid tumors, including breast, brain, pancreatic and ovarian, whereby small, orally bioavailable molecules targeting LIPA block protein folding, induce ER stress and result in tumor cell death.

T riple-negative breast cancers are negative for the expression of ER-α, PR or HER2, accounting for ~15% of new breast cancer (BC) diagnoses 1 . The absence of expression of these receptors means that effective agents targeting these receptors will have no therapeutic activity in TNBC. Currently, chemotherapy is the primary treatment option for the majority of patients with TNBC. TNBCs are aggressive tumors and have the highest mortality rate among all BC subtypes: 150,000 deaths worldwide were attributed to metastatic TNBC in 2018 alone 2,3 . There is thus an urgent and unmet need for effective targeted therapies in TNBC.
An ideal targeted therapy exploits specific vulnerabilities in cancer cells that are not seen in normal cells. However, TNBC represents a collection of multiple, biologically distinct subtypes categorized by their transcriptional profiles as basal-like (BL1, BL2), mesenchymal (M) or luminal androgen receptor (LAR) subtypes. Despite advances in tumor characterization, the molecular heterogeneity of TNBC and subtype-specific differences in immune cell composition and genetic and pharmacologic vulnerabilities limit the activity of individual targeted therapies.
We previously identified oligobenzamides D2 and ERX-11 that bind to androgen receptor 4 and estrogen receptor (ER-α) 5 , respectively. During the process of lead optimization of these agents, we made the serendipitous discovery of a small molecule (ERX-41) with robust activity against multiple TNBC molecular subtypes. Herein we describe the identification, mechanism of action and molecular target of ERX-41 in TNBC and its applicability to other solid tumors.
Daily PO or intraperitoneal (i.p) administration of ERX-41 at doses up to 100 mg kg -1 showed no clear evidence of toxicity. Pharmacokinetic (PK) studies indicated that ERX-41 was orally bioavailable, with peak detectable plasma levels at 4 h after oral administration (10 mg kg -1 single dose). Additionally, ERX-41 was detectable within 1.5 h in established s.c. MDA-MB-231 xenografts after either PO or i.p. administration (Fig. 2a).
We show that ERX-41 (10 mg kg -1 d -1 PO) significantly inhibited the progression of established MDA-MB-231 xenografts in vivo (Fig. 2b). ERX-41 reduced tumor growth, as shown by extirpated tumor weights and sizes at the end of the study (Fig. 2c,e). Importantly, ERX-41 treatment did not show overt signs of toxicity, as evidenced by unchanged body weights of treated mice (Fig. 2d). ERX-41 significantly reduced growth of D2A1 xenografts in syngeneic mice without affecting body weight (Fig. 2f-h). We note that ERX-41 (10 mg kg -1 d -1 PO) also decreased growth of four distinct TNBC patient-derived xenografts (PDXs) in vivo (Fig. 2i,p) .
Histologic evaluation following ERX-41 treatment showed no significant changes in gross histology of multiple organs including heart, lung, spleen, liver, kidney, uterus and pancreas (Extended Data Fig. 2a). Similarly, in syngeneic mice, no significant histologic changes or immune infiltrates were noted in multiple organs, including spleen (Extended Data Fig. 2a), suggesting that ERX-41 is not immunogenic. In addition, ERX-41 did not affect the proliferation index (Extended Data Fig. 2b), endometrial cell height in the uterus (Extended Data Fig. 2c) or ER-α expression (Extended Data Fig. 2d): these findings are relevant, since the uterus is the organ most sensitive to estrogenic stimulus and antiestrogenic treatment. We then evaluated the effect of ERX-41 on bone marrow plasma cells by flow cytometry (Extended Data Fig. 2e) and ELISpot analysis (Extended Data Fig. 2f). Our analyses indicate that ERX-41 affected neither plasma cell numbers, immunoglobulin (Ig) Igκ expression in plasma cells nor number of IgM or IgG antibody-secreting cells (ASCs).

ERX-41 induces ER stress.
To understand the mechanism of action of ERX-41, we performed unbiased RNA sequencing (RNA-seq) studies in MDA-MB-231 and BT-549 cells (Fig. 3a,b). Gene set enrichment analyses (GSEA) revealed that the top pathways upregulated after 4 h of treatment with ERX-41 were related to induction of ER stress and compensatory unfolded protein response (UPR) pathways (Extended Data Fig. 3a-d). Heatmaps show induction of ER stress and UPR genes in TNBC cells (Fig. 3c). Quantitative PCR with reverse transcription (RT-qPCR) confirmed that canonical ER stress genes, heat shock protein 70 A family member 5 (HSPA5), DNA damage-inducible transcript 3 (DDIT3) and UPR stress sensor-spliced X-box-binding protein 1 (sXBP1) were dramatically upregulated (60-80-fold) in TNBC but not in HMEC cells following ERX-41 treatment (Fig. 3d,f).
We then performed ultrastructural studies using transmission electron microscopy (TEM). ERX-41 induced dramatic ER dilation within 4 h (Fig. 3g,h). Induction of ER dilation by TEM was noted in multiple TNBC but not in normal HMEC cells. Further ultrastructural validation was obtained by Airyscan super-resolution microscopy of live SUM-159 cells stably expressing the ER membrane marker mCherry-RAMP4 ( Fig. 3i-l). In vehicle-treated cells, the peripheral ER network appeared as an intricate network of thin tubules connected by three-way junctions (Fig. 3i), and as a single peak on distance histograms (Fig. 3k). Within 2 h of ERX-41 treatment we observed marked disorganization of the peripheral ER network, characterized by striking elongation and dilation of individual tubules (Fig. 3j,k) (mean ± s.d. tubule width at 0 h, 260 ± 80 nm; at 2 h, 660 ± 320 nm; n (0/2 h) = 102/104 tubules, P < 0.0001) (Fig. 3l). Importantly, since the width of normal ER tubules is ~50-100 nm and thus below the resolving power of Airyscan microscopy (100-200 nm), we expect that the true effect of ERX-41 on tubule width is probably even greater. These data confirm that ERX-41 induces ER stress in TNBC.
We then biochemically confirmed that ERX-41 induces ER stress and downstream UPR pathways via induction of phosphorylated protein kinase R-like ER kinase (p-PERK) and phosphorylated Data presented as mean ± s.e.m., n = 3 biologically independent samples. g,h, Following establishment of MDA-MB-231 TNBC xenografts in the mammary fat pad in vivo, daily administration of 10 mg kg -1 ERX-11-9, ERX-11-16, ERX-11-30 or vehicle control was initiated. Body weights are shown for each set, including for ERX-11-9, ERX-11-16 and ERX-11-30 administered s.c. (g) or PO (h). Data presented as mean ± s.e.m. For tumor volume data, n = 8 tumors per group; significance was determined by two-way ANOVA followed by Tukey's multiple comparisons test. Adjusted P values for the last time points are shown. For body weight data, n = 5 mice per group; significance was determined by one-way ANOVA. i,j, Dose-response curve of ERX-41 in multiple TNBC cell lines using WST-1 assay (i) and CellTiter-Glo assay (j). Data presented as mean ± s.e.m.; n = 3 biologically independent samples.    We confirmed that a single dose of ERX-41 induces ER stress in vivo, as evidenced by enhanced p-PERK and p-eIF2-α staining in TNBC xenografts within 24 h of treatment (Extended Data Fig. 3i-k). Since our PK studies showed that treated tumor tissues have detectable levels of ERX-41 at 24 h (Fig. 2a), these data indicate that ERX-41 selectively induces ER stress.
The functional consequence of ERX-41 induction of ER stress is that global de novo protein synthesis in TNBC, but not in HMEC, cells is blocked by ERX-41, as shown by immunoblots for puromycin-labeled nascent proteins (Extended Data Fig. 3l). These data suggest that ERX-41 can induce uncompensated ER stress,

Molecular target of ERX-41 is LIPA.
Since ERX-41 was derived from ERX-11 (which targets ER-α), we established that ERX-41 did not interact with ER-α. Using a time-resolved measurement with fluorescence resonance energy transfer (TR-FRET) assay, we demonstrated that ERX-41 does not interact with the ER-α ligand-binding domain (LBD), unlike fulvestrant and selective ER-α degraders such as GDC-0810, tamoxifen and ERX-11 (data shown for fulvestrant and GDC-0810) (Extended Data Fig. 4a,b). These data suggest that ERX-11 and ERX-41 have distinct molecular targets.
To identify the molecular target of ERX-41, we performed an unbiased CRISPR-Cas9 knockout (KO) screen in MDA-MB-231 cells. There was significant concordance between two independent experiments of the screen performed at two distinct concentrations of ERX-41, and the top six genes were subjected to a secondary screen in MDA-MB-231 cells ( Fig. 4a and Extended Data Fig. 4c). TNBC cell lines with KO of five of the top six genes-LIPA, SLC5A3, TMEM208, SOAT1 and ARID1A-were generated and evaluated for response to ERX-41 (Extended Data Fig. 4d-h). Of these, KO of LIPA alone (which encodes lysosomal acid lipase (LAL)) was able consistently to abrogate cytotoxic response to ERX-41 ( Fig. 4b and Extended Data Fig. 4i). Knockout of individual ER stress/UPR genes such as PERK and IRE1-α did not affect the ability of ERX-41 to cause cell death, suggesting that multiple ER stress pathways had been activated (Extended Data Fig. 4j,k). While these data do not rule out a role for other identified ERX-41 targets, we were able to show that KO of LIPA in SUM-159 and MDA-MB-436 was able to alter the response to ERX-41 (Fig. 4c,d) in CellTiter-Glo assays in vitro. The altered response of SUM-159 clones with LIPA KO was specific for ERX-41, as shown by similar responses of SUM-159 parental and LIPA KO clones to thapsigargin or paclitaxel (Fig. 4e,f).
Live-cell imaging studies confirmed that KO of LIPA altered the response of SUM-159 cells to ERX-41 with a dramatic decrease in cell death (Fig. 4g,h and Supplementary Video 3). We then confirmed that xenografts of SUM-159 with LIPA KO did not respond to i.p. administration of ERX-41, in contrast to parental SUM-159 xenografts which responded significantly to ERX-41 in vivo ( Fig. 4i,j). Importantly, evaluation of proliferative indices of xenograft tumors showed a decrease in Ki67 staining in parental SUM-159 xenografts but not in SUM-159 LIPA KO xenografts (Fig. 4k,l).
To ascertain that LIPA is associated with ERX-41-induced ER stress, we performed unbiased RNA-seq studies with and without ERX-41 in parental and LIPA KO SUM-159 cells. Principal component (PC) analyses showed that gene expression profiles in these cells tend to cluster independently (Fig. 5a). While volcano plots showed significant alteration in gene expression in parental SUM-159 cells (Fig. 5b), there were no genes significantly altered in LIPA KO SUM-159 cells following ERX-41 treatment (Fig. 5c). Evaluation of canonical genes involved in ER stress and UPR response shows induction of these genes by ERX-41 in parental SUM-159 cells but not in cells with LIPA KO (Fig. 5d). These findings were confirmed by RT-qPCR, showing that ERX-41 induces expression of UPR genes sXBP1 and DDIT3 in parental SUM-159 cells but not in SUM-159 cells with LIPA KO (Fig. 5e Ultrastructural studies using live-cell confocal microscopy show that LIPA KO abrogated ER morphological changes at 2 and 4 h Fig. 2 | ERX-41 has potency against TNBC. a, Following establishment of s.c. MDA-MB-231 xenografts, 10 mg kg -1 single-dose ERX-41 was administered either PO or i.p. Tumor was harvested at 0, 0.5, 1.5, 3, 6 and 24 h after drug administration, and drug levels assayed by LC-MS/MS and graphed. Data presented as mean ± s.e.m.; i.p. group, n = 3 mice; PO group, n = 2 mice. b-e, Following establishment of MDA-MB-231 xenografts in mammary fat pad, daily administration of 10 mg kg -1 ERX-41 or vehicle control was initiated. Tumor volumes were measured using digital calipers and graphed (b). Tumor weights (c), body weights (d) and extirpated tumors (e) at study end are also shown. Data shown as mean ± s.e.m. For tumor volume, n = 10; significance was determined by two-way ANOVA with Bonferroni's multiple comparisons test. Adjusted P values of last time points are shown. For tumor weight (n = 10) and body weight (n = 5), significance was determined by unpaired two-tailed Student's t-test. f-h, After establishment of D2A1 syngeneic xenografts in mammary fat pad, daily administration of 10 mg kg -1 ERX-41 or vehicle control was initiated. Tumor volumes were graphed (f), along with tumor weights (g) and body weights (h). Data shown as mean ± s.e.m. For tumor volume data, n = 8; significance was determined by two-way ANOVA with Bonferroni's multiple comparisons test. Adjusted P values of last time points are shown. For tumor weight (n = 8) and body weight (n = 5), significance was determined by unpaired two-tailed Student's t-test. i-p, Effect of ERX-41 on growth (i,k,m,o) and tumor weight (j,l,n,p) in four TNBC-PDXs compared with vehicle. Data presented as mean ± s.e.m.; TNBC-PDX-1 (i,j), n = 6; TNBC-PDX-89 (k,l), n = 7; TNBC-PDX-96 (m,n), n = 8; TNBC-PDX-98 (o,p), n = 6 mice. For tumor volume data, significance was determined by two-way ANOVA with Bonferroni's multiple comparisons test. Adjusted P values of last time points are shown. For tumor weight, significance was determined by unpaired two-tailed Student's t-test. This experiment was performed once. Numerical source data for a-d,f-p are provided.     . Following daily i.p. administration of vehicle or 10 mg kg -1 ERX-41, tumor size was measured and graphed (i), with mice body weights (j). Data presented as mean ± SEM, n = 5 mice per group. Significance was determined by one-way ANOVA with Tukey's multiple comparisons test. Adjusted P values of last time points are shown. Xenograft tumors were harvested and processed for IHC with Ki67. Representative IHC staining is shown (k) and proliferative indices for each tumor are quantitated and graphed (l). Data presented as mean ± s.d., n = 5 per group. Significance was determined by one-way ANOVA with Tukey's multiple comparisons test. Adjusted P values are shown. Experiment shown in h was done twice independently, with similar results, while that shown in i,j was done once. Numerical source data for a-g,i,j,I are provided.
after treatment (Fig. 5g,h). While changes in ER luminal diameter are more readily apparent with Airyscan imaging, live-cell confocal microscopy allows serial evaluation over time. As noted with Airyscan microscopy, increases in ER tubule diameter from ~200 to ~500 nm at 2 h and to ~620 nm at 4 h were apparent in parental SUM-159 cells after ERX-41 treatment (Fig. 5g,i). In contrast, no significant changes in ER tubule diameter were noted in LIPA KO SUM-159 cells after ERX-41 treatment (Fig. 5h,j). Importantly, SUM-159 cells with stably transduced mCherry-RAMP4 responded to ERX-41 as did parental SUM-159, with IC 50 ~200 nM in WT cells and ~4 μM in LIPA KO cells (Extended Data Fig. 5d).
Immunoblotting revealed that ERX-41 activates PERK (noted by upshifting of the PERK band) and induced p-eIF2-α in parental SUM-159 cells, but not in SUM-159 cells, with LIPA KO  DDIT3  HERPUD1  CCL2  DNAJB9  HSPA5  XBP1  ASNS  ERN1  EIF2AK3  TSPYL2  HSP90B1  WIPI1  DNAJC3  ATF4  HYOU1  DNAJB11  ATF4P3  SYVN1  EDEM1  CALR  SERP1  PDIA6  SEC31A  IGFBP1 PREB LIPA subcellular localization in the ER. The known function of LAL protein as a lysosomal acid lipase relates to its subcellular lysosomal localization. The ability of ERX-41 to induce ER stress prompted evaluation of LAL subcellular localization in TNBC using coimmunofluorescence with both ER and lysosomal markers in SUM-159 cells with overexpressed myc-tagged LAL (Extended Data Fig. 6a). We noted high colocalization of LAL protein with the ER tracker, with a weighted colocalization factor of 1.0 (Extended Data Fig. 6b,c). In contrast, LAL protein poorly colocalized with lysosomal marker LIMP2, with a weighted colocalization factor of 0.17 (Extended Data Fig. 6a-c). The specificity of myc antibody was confirmed by negative staining in parental SUM-159 cells, which have no myc-tagged LAL expression (Extended Data Fig. 6d). These data were supported by biochemical evaluation of subcellular fractions of SUM-159 cells (Extended Data Fig. 6e), which confirmed enrichment of LAL protein in the ER-enriched subcellular fraction. In addition, the sensitivity of glycosylated LIPA to endoglycosidase H (Endo H) and peptide-N-glycosidase F (PNGase F) cleavage supports its ER localization 6 (Extended Data Fig. 6f). If glycosylation occurred in the Golgi, then glycosylated LIPA would be sensitive to PNGase F but not to Endo H cleavage. Since the role of LIPA in the ER has not been previously described, these data support our central finding that LIPA plays a role in ER homeostasis. Taken together, these data suggest a previously uncharacterized function for LIPA in the ER that is targeted by ERX-41 and that induces ER stress in TNBC.
LIPA as a target in TNBC and solid tumors. We then evaluated LAL protein expression in TNBC using tissue microarray (Extended Data Fig. 7a). We found that >80% of primary TNBC tumors had significant and detectable LAL protein expression ( Fig. 6a and Extended Data Fig. 7b); in contrast, normal breast tissue had lower LAL expression (Fig. 6b and Extended Data Fig. 7b). Protein expression was noted to be higher in TNBC tumors than in adjacent normal breast tissue (Fig. 6c,d). Importantly, glycosylation of LAL in TNBC tumors is similar to that observed in TNBC cells (Extended Data Fig. 7c). Analysis of publicly available datasets (The Cancer Genome Atlas (TCGA)) indicated that high LAL expression correlated with significantly worse overall survival outcomes in patients with BC (Fig. 6e). These data suggest that LAL is a viable molecular target in TNBC.
The minimal toxicity of ERX-41 in vivo prompted our evaluation of LAL expression in normal mouse organs by immunohistochemistry (IHC). We noted that LAL expression in multiple mouse tissues, including uterus, liver, kidney, heart, lung, spleen, pancreas and mammary fat pad, was much lower than in tumor tissue (Extended Data Fig. 7d). LAL expression was lowest in heart, pancreas and mammary fat pad, intermediate in liver and kidney and highest in the spleen. Given the role of the spleen in the lymphatic system, our previous analyses showing that ERX-41 had no effect on plasma cells are of relevance. These data suggest that enhanced expression of LAL in tumors may account for their sensitivity to ERX-41.
To ascertain that ERX-41 has activity against primary TNBC tumors, we leveraged our previous experience with ex vivo     (Fig. 6f) 5,[7][8][9] . PDE cultures maintain the native tissue architecture and better recapitulate the heterogeneity of human TNBC in a laboratory setting (Extended Data Fig. 7e). We noted that ERX-41 had significant activity, as evidenced by decreased proliferation (Ki67 staining) and increased apoptosis (cleaved caspase 3 staining) of primary TNBC PDEs (Fig. 6g-i). Immunohistochemical evaluation of UPR markers in these explants showed enhanced p-PERK, CHOP (protein product of DDIT3 gene) and p-eIF2-α staining within 24 h after treatment with ERX-41 (Fig. 6i). These data further validate the potential utility of ERX-41, specifically in patients who would receive the drug. We postulated that ERX-41 would have activity in other tumors with high basal levels of ER stress, such as ER-α + BC, glioblastoma and pancreatic and ovarian cancers. Initial evaluation confirmed responsiveness to ERX-41 in vitro (Extended Data Fig. 8a,c-e). Importantly, we have shown that knockdown of LAL expression in these cell lines using CRISPR abrogated the ability of ERX-41 to induce ER stress (Extended Data Fig. 8b, for ER-α + MCF-7 cells). We also confirmed that ERX-41 has activity against both ER-α + BC PDX (WHIM-20) (Fig. 6j) and an ovarian cancer cell line xenograft (ES2) (Extended Data Fig. 8f-j), and PDX (OCa-PDX-38) (Fig. 6k). These data indicate that ERX-41 has activity against multiple tumors by targeting LAL and inducing ER stress.

ERX-41 targets LIPA in ER and is independent of LIPA lipase activity.
We used cellular thermal shift assays to confirm that ERX-41 binds to the protein product of the LIPA gene, LAL protein.
Our studies indicate that ERX-41 was able to stabilize LAL within the cell and shifted thermal sensitivity, suggesting that ERX-41 binds to LAL (Fig. 7a,b). In contrast, ERX-41 did not affect vinculin thermal stability (Fig. 7c). Our subsequent assays confirmed binding of LAL to ERX-41 (Fig. 7j) and indicate that ERX-41, but not ERX-11, binds to LAL in the cellular context of TNBC.
To study how ERX-41 interacts with LAL, we used in silico molecular docking simulation to evaluate potential binding sites of ERX-41 on LAL (Fig. 7d-i). We note that LAL has a single 239 LXXLL 243 motif and that ERX-41 (shown in green in Fig. 7d-i) could potentially interact with this LXXLL motif (LXXLL motif shown in orange; Fig. 7d-i and Extended Data Fig. 9a). Structurally, the lipase activity of LAL appeared to be spatially distinct from the LXXLL motif: previous studies have identified that the point mutation H274Y in LAL helix 13 abrogates the lipase function of LIPA (Fig. 7i) 10 . ERX-41 showed no inhibition of lipase activity of LIPA while, in contrast, Lalistat 2 (specific inhibitor of LAL lipase activity) attenuated lipase activity in SUM-159 cells (Extended Data Fig. 9b). The specificity of Lalistat 2 for LAL is confirmed by its lack of inhibitory activity on lipase function (from other lipases) in LIPA KO cells 11,12 .
To study the effect of these LIPA domains, we synthesized LIPA plasmid constructs under a constitutive promoter, including wild-type (WT) LIPA (WT-LIPA), H274Y mutant LIPA (H274Y MT-LIPA), ΔLXXLL mutant LIPA (deletion of 238 NLCFLLC 244 ) and L242P mutant LIPA (point mutation of the second L in the LXXLL motif). We then confirmed direct interaction of ERX-41 with LIPA using biotinylated ERX-41 pulldown, which indicated interaction between ERX-41 and proteins encoded by WT-LIPA and H274Y MT-LIPA but not by ΔLXXLL MT-LIPA or L242P MT-LIPA (Fig. 7j). Reconstitution of WT-LIPA (KO + WT) in SUM-159 cells with LIPA KO restored sensitivity to ERX-41, with an IC 50 of 250 nM (Fig. 7k). Interestingly, reconstitution with the lipase-incompetent H274Y MT-LIPA (KO + H274Y) also restored sensitivity to ERX-41 (Fig. 7k). In contrast, reconstitution of lipase-competent ΔLXXLL MT-LIPA (KO + ΔLXXLL) did not restore sensitivity to ERX-41, suggesting that the LXXLL motif is critical for LAL-ERX-41 interaction ( Fig. 7k and Extended Data Fig. 9c,d). Evaluation of the lipase activity of these constructs confirmed that WT-LIPA, ΔLXXLL MT-LIPA and L242P MT-LIPA had lipase activity while H274Y MT-LIPA did not ( Fig. 7i and Extended Data Fig. 9d). Further detailed mutational evaluation of the 239 LXXLL 243 domain indicated that the leucine at the 242 position is critical for ERX-41 activity (Fig. 7j-o and Extended Data Fig. 9d-g). In contrast, the leucine at the 239 position is not critical for ERX-41 binding or activity (Extended Data Fig. 9e-g). These data are validated by the inability of ERX-41 to affect proliferation or induce ER stress in SUM-159 clones with reconstitution of lipase-competent LXXLL motif single mutations (KO + L242P), double mutations (KO + L239P/ L242P) or triple mutations (KO + L239P/L242P/L243P) (Extended Data Fig. 9e-g). These findings are further supported by the ability of WT-LIPA and H274Y MT-LIPA, but not ΔLXXLL MT-LIPA or L242P MT-LIPA, to restore the ability of ERX-41 to induce ER stress at the protein level (shown by activation of PERK in Fig. 7m) and UPR genes at the RNA level (sXBP1 and DDIT3 levels in Fig. 7n,o) in SUM-159 cells with LIPA KO. These data, taken together, indicate that ERX-41 interacts with LAL through residues in its LXXLL domain and that its ability to induce ER stress and cell death in TNBC is independent of the lipase activity of LAL (Extended Data Fig. 9g).
We then evaluated whether LAL localization to the ER is critical for LAL function by using additional LAL recombinants with altered subcellular localization (Fig. 7p-s). We noted that a recombinant LIPA complementary DNA lacking the signal peptide (WT-∆SP) is neither localized to the ER nor glycosylated (Fig. 7q), nor able to restore the ability of ERX-41 to cause ER stress (Fig. 7r) or affect cell proliferation (Fig. 7s). In contrast, a recombinant LIPA cDNA with an added KDEL sequence at the C terminus (WT-KDEL) is  Immunoblot studies in a-c,j,m,q,r were performed twice independently, with similar results. Numerical source data for d,g-i,k and uncropped blots for c are provided. localized to the ER, is glycosylated (Fig. 7q) and restores the ability of ERX-41 to cause ER stress (Fig. 7r) and affect cell proliferation (Fig. 7s). These data indicate that ER localization of LAL is critical for cellular responsiveness to ERX-41 (Fig. 7i, tabulated in Extended Data Fig. 9h). Finally, since mouse LAL has a slightly different LXXLL sequence (VFFLL), we confirmed that this altered sequence in the context of human LIPA or the entire mouse LIPA cDNA is glycosylated and could bind to ERX-41 and restore responsiveness (ER stress and cell death) to ERX-41 in SUM-159 LIPA KO cells (Extended Data Fig.  9i-l). In conjunction with earlier findings that a tumor cell line of murine origin (D2A1) is sensitive to ERX-41 (Fig. 2f-h), these data indicate that the nontoxic characteristics of ERX-41 in the mouse cannot be attributed to species differences in LAL sequences.

ERX-41 targeting of LIPA disrupts protein folding in the ER.
To molecularly characterize how targeting of LAL causes ER stress, we first decided to define the LAL interactome using a recombinant LAL fused to TurboID (Fig. 8a). Importantly, this recombinant LAL could reconstitute sensitivity to ERX-41 in SUM-159 LIPA KO cells in CellTiter-Glo studies (Fig. 8b) and restore the ability of ERX-41 to cause ER stress (Extended Data Fig. 10a). LAL interactors were biotinylated, isolated by streptavidin pulldown (Extended Data Fig. 10b) and identified by unbiased MS analyses in two independent experiments in two different cell clones (Fig. 8c,d). Gene ontogeny (GO) analysis of the 54 LAL interacting proteins (Extended Data Fig. 10c) revealed that four of the top five cellular processes are involved in critical ER protein maturation functions, including protein folding in the ER (Fig. 8e). Cellular component analyses indicated that LAL binders were most commonly localized to the ER (Fig. 8f).
We then used a next-generation proteomic method, DIA-MS (Fig. 8g). Utilizing an unbiased quantitative DIA-MS approach, we identified that short-term (6 h) ERX-41 treatment significantly affected the expression of 189 cellular proteins (Extended Data Fig. 10d,e), including 153 down-and 36 upregulated proteins (PC analyses of proteins provided in Fig. 8h). GO analysis of the 153 ERX-41-downregulated proteins revealed that three of the top five cellular processes were involved in ER protein maturation functions, such as protein folding in ER (Fig. 8i), and that these proteins were localized to the ER (Fig. 8j).
We then combined data from both unbiased proteomic approaches to identify a core set of proteins that were both LAL binders and affected by ERX-41 treatment (Fig. 8k). Again, GO analyses of these 17 proteins confirmed their involvement in protein folding (Fig. 8l) and localization to the ER (Fig. 8m). We then validated the LAL dependence of ERX-41 activity using both LAL-expressing and LAL-KO cells: immunoblots showed that ERX-41 affects the expression of these proteins in LAL-expressing but not in LAL-KO cells (Fig. 8n). These data support the model that ERX-41 binding to LAL in the ER affects expression of several ER localized proteins involved in protein folding, causing significant ER stress/UPR leading to cell death. The lipase function of LAL is neither affected by ERX-41 nor is critical for ERX-41 activity; these findings are modeled in Fig. 8o.

Discussion
Since TNBCs have high growth rates, they have sustained and enhanced demand for de novo protein synthesis, folding and maturation. TNBC tumor growth, metastasis, chemotherapy, hostile environmental conditions-such as hypoxia-and oxidative stress further jeopardize the fidelity of protein folding in their ER, and cause ER stress 13 . Compensatory UPR pathway proteins 78-kDa-glucose-regulated protein (GRP78), PERK and activating transcription factor 6 (ATF6) are overexpressed in TNBC, increased during TNBC progression and are correlated with poor patient survival in TNBC 14,15 . Several genome screens identified components of the ER stress pathway as targets of vulnerability in various cancers, including TNBC [16][17][18] . Although UPR can resolve ER stress and restore homeostasis, unresolved ER stress can be lethal to cells via ER stress-induced apoptosis. However, persistent and severe ER stress kills cancer cells by inducing their autophagy, apoptosis, necroptosis or immunogenic cell death.
In the present study, using a variety of biochemical and ultrastructural studies, we have shown that ERX-41 induces ER stress. Our confocal microscopy results showing ER stress are noteworthy, since ER luminal diameter (50-60 nm) is normally below the resolution of confocal microscopy (100-200 nm). Since ER dilation measured following ERX-41 treatment approached 600 nm, the true fold effect of ERX-41 on ER morphology is profound and significant. We observed that ERX-41 induces ER stress, shuts down de novo protein synthesis, blocks proliferation and induces apoptosis of TNBC in vitro, ex vivo and in vivo. Our results suggest that ERX-41 aggravates this already engaged system in TNBC to exhaust its protective features and cause apoptosis. In normal cells and tissues ERX-41 does not induce ER stress, suggesting that the basal level of ER stress and the compensatory UPR pathway may dictate responsiveness to ERX-41. These data indicate that ERX-41 targets a fundamental vulnerability in TNBC-the high basal level of ER stress-and may be able to overcome the inter-and intratumoral heterogeneity of TNBC.
Knockout of LIPA abrogates the ability of ERX-41 to induce ER stress, UPR, block de novo protein synthesis and cause cell death in vitro and in vivo. Reconstitution of LIPA in this KO system restores the ability of ERX-41 to induce ER stress, UPR and cause cell death. Importantly, reconstitution of a LIPA mutant in this KO system that is incapable of binding ERX-41 does not restore the ability of ERX-41 to induce ER stress, UPR and cause cell death. Our identification of a single-point mutation within LIPA that abrogates responsiveness to ERX-41 represents a gold standard validation that  data for b,e,f,i,j,l,m and uncropped blots for n are provided. FDR, false discovery rate. Tu rb oI D Tu rb oI D LAL is the target of ERX-41. These data establish that the interaction between ERX-41 and LAL is critical for ERX-41-induced cell death. We have also shown that LAL binds to ERX-41 using two distinct assays-the cellular thermal shift assay in intact cells and biotinylated pulldown of LAL from cell lysates. Cocrystallization of LAL with ERX-41 will further inform structural details of this interaction, and are ongoing. Collectively, our data confirm the central role of LIPA in the activity of ERX-41 against TNBC.
Our unbiased proteomic studies offer some insight as to how ERX-41 binding to LAL causes ER stress. We have identified the interactome of LAL, which appears to be primarily composed of ER-resident proteins and proteins involved in critical ER maturation functions such as ER folding. Importantly, our unbiased global proteomic studies indicate that ERX-41 causes a decrease in the expression of several known ER-resident proteins and proteins involved in ER folding. These data suggest that ERX-41 binding to LAL causes downregulation of several ER proteins involved in protein maturation and effectively causes ER stress.
Identification of LIPA as a targetable molecular vulnerability is the critical finding in this study. Since the lipase function of LAL is not targeted by ERX-41, we also noted that Lalistat 2, a known inhibitor of LAL lipase activity, does not induce ER stress or cause cell death in TNBC. Our studies suggest that LAL localization to the ER is critical for ERX-41 activity: a recombinant LAL lacking the signal peptide does not become glycosylated or respond to ERX-41. Our chemistry-first approach enabled the identification of an uncharacterized function of LAL related to its ability to function as a molecular chaperone of ER-resident proteins involved in protein folding. While LAL protein levels in tumors have been inadequately profiled, the expression of LIPA messenger RNA appears to be highest in glioblastoma and breast, ovarian and pancreatic cancers, and our studies indicate that their representative tumors are sensitive to ERX-41 treatment both in vitro and in vivo. Further studies are needed to evaluate whether the LAL expression levels and basal level of ER stress in tumors could serve as biomarkers of cellular response to ERX-41.
In conclusion, this manuscript reveals the important finding of a potent therapeutic agent (ERX-41) with a clear molecular target (LAL) and mechanism of action (disruption of protein folding and induction of ER stress) that may have utility in treating patients with multiple solid tumors.

Methods
Our research complies with all relevant ethical regulations, and was performed in accordance with UTHSA and UTSW IACUC approved protocols.
Chemicals and synthetic procedure. All chemical reagents and solvents were obtained from commercial sources and used without additional purification. 1 H and 13 C nuclear magnetic resonance (NMR) spectra were recorded on a Bruker Avance III HD 600-MHz NMR spectrometer. Chemical shifts (δ) are reported in ppm from an internal standard of residual DMSO-d6 (2.50 and 39.5 ppm in 1 H and 13 C-NMR spectra, respectively). Data are reported as follows: chemical shift (δ), multiplicity (s, singlet; d, doublet; dd, doublet of doublet; t, triplet; q, quartet; br s, broad singlet; m, multiplet), coupling constant (J) in Hertz (Hz), integration. Mass spectra were recorded on a Shimadzu AXIMA Confidence MALDI-TOF mass spectrometer (nitrogen UV laser, 50 Hz, 337 nm) using α-cyano-4-hydroxycinnamic acid as matrix.
The overall scheme of ERX-41 synthesis is shown in Supplementary Note. Synthesis of compound 2:compound 1 was synthesized as previously reported. Trans-4-methylcyclohexylamine (0.73 g, 6.4 mmol) was added to a solution of compound 1 (2.7 g, 3.2 mmol), HATU (1.4 g, 3.7 mmol), DIEA (1.2 ml, 6.9 mmol) and DMF (30 ml). The reaction mixture was stirred at room temperature (RT) for 24 h and then diluted with EtOAc (100 ml) and 0.5 N HCl (100 ml). Layers were separated and the aqueous layer extracted with EtOAc (100 ml). Organic layers were combined, washed with 0.5 N HCl and brine, dried over anhydrous sodium sulfate and concentrated under reduced pressure. The resulting solid was washed with EtOAc and dried in vacuo to yield compound 2 as a white solid (1.75 g). The product was used in the following reaction without further purification.
For synthesis of ERX-41, please refer to the scheme shown in Supplementary Note. Concentrated HCl (30 ml) was added to a solution of compound 2 (1.75 g) and THF (300 ml). The reaction mixture was stirred at RT for 24 h and then concentrated under reduced pressure. The resulting solid was washed with MeOH and dried in vacuo to yield ERX-41 as a light-yellow solid (1.3 g, 57% over two reaction steps). 1  Cell culture. All human BC cell lines were directly obtained from either ATCC, UTSW or the UTHSA ObGyn core and cultured according to ATCC guidelines. All cell lines utilized were free of mycoplasma contamination. Authentication was performed by short-tandem repeat profiling in the UTSW sequencing core facility. All cell lines used in this paper, sources and culture conditions are listed in Supplementary Table 1.
Estrogen receptor binding. LanthaScreen TR-FRET ERα Coactivator Assay (no. A15885, Life Technologies) was performed as per the manufacturer's instructions. Compounds were tested in the range 0.0003-16.66 μM using serial dilutions, with assays in antagonist assay mode. Final assay buffer composition included 3.5 nM ER-α-LBD (GST), 250 nM fluorescein-conjugated coactivator PGC1a peptide, 5 nM terbium (Tb)-labeled anti-GST antibody and 5 nM estradiol. The plate was incubated at RT for 2 h and FRET analyzed on a PHERAstar microplate reader with the settings excitation 340 nm, emission 495 and 520 nm. The emission ratio (520/495) was analyzed and plotted. Curves were generated using a sigmoidal dose-response equation (variable slope) in GraphPad Prism 9.0 software.
Cell viability assays. Cells were seeded in 96-well plates (2 × 10 3 cells per well) 1 day before treatment. Cells were treated with varying concentrations of ERX-41 or ERX-11 analogs for 3-6 days. The effects of ERX-41 and ERX-11 analogs on cell viability were then measured in triplicate or sextuplicate with multiple biological replicates using WST-1 (Promega), MTT as previously described 19 or CellTiter-Glo 2.0 (Promega) assay.
Live-cell imaging. Live-cell images were acquired using a Lionheart FX automated microscope (BioTek). Cells were plated 1 day before experiments. For cell death assay, 0.2 μM SYTOX Green was added to the plate 15' min beforehand. CRISPR screen. The Human Brunello CRISPR knockout pooled library was purchased from Addgene (no. 73178; through D. Root and J. Doench). The library containing lentivirus was transduced into cells in biological replicate or triplicate at a multiplicity of infection of ~0.5 and minimum 500× coverage. Two days after transduction, uninfected cells were removed with puromycin selection. The library containing cells was treated with either vehicle or ERX-41 for 2 weeks. Genomic DNA was extracted from identical numbers of cells. The single-guide RNA cassette was retrieved by PCR, followed by next-generation sequencing (NGS). NGS data were analyzed by MAGeCK 20 .

Live-cell confocal and Airyscan imaging.
To visualize the ER structure, stable SUM-159 cell lines expressing mCherry-RAMP4 were established. pLenti-X1-hygro-mCherry-RAMP4 was purchased from Addgene (no. 118391; through J. Corn) 22 . Live-cell images were acquired with a Confocal Zeiss LSM880 Airyscan. ER tubule width was calculated by drawing 1-px-wide line scans perpendicular to the long axis of individual ER tubules. In MATLAB, peak intensity along each line scan was determined and the distance between half-maximum intensity on either side of the peak was measured. The code for this analysis is available at https://github.com/andmoo91/HalfMaxScript. Lentivirus production. Lentiviral constructs (lentiCRISPR v.2 for KO, pWPI for overexpression), along with helper plasmids Δ8.9 and VsVg, were transfected into HEK293T cells using polyethylenimine (PEI, 1 mg ml -1 ; Polysciences). Medium was changed the following day. Lentivirus was collected after an additional 48-72 h. Filtered (0.45 μm) lentivirus containing medium was used to infect cells, with 6 μg ml -1 polybrene.
Pulldown assay. Cells were lysed in high-salt lysis buffer (1.25 mM Hepes pH 7.5, 400 mM NaCl, 0.5% NP-40, 5% glycerol and 2 mM MgCl 2 ) supplemented with 1 mM DTT and 1/100 Halt protease inhibitor cocktail (Thermo Scientific) on ice for 15 min, followed by centrifugation at 20,000g for 20 min at 4 °C. The supernatant was mixed with a 1.67× volume of no-salt lysis buffer (1.25 mM Hepes pH 7.5, 5% glycerol and 2 mM MgCl 2 ) supplemented with 1 mM DTT and 1/100 Halt protease inhibitor cocktail (Thermo Scientific). The lysate was incubated with either biotinylated ERX-11 or biotinylated ERX-41 overnight, followed by 1 h of incubation with M-270 streptavidin Dynabeads. After four washes with lysis buffer, samples were eluted by boiling in 2× Laemmli buffer. Bound proteins were resolved on SDS-polyacrylamide gel electrophoresis (SDS-PAGE) followed by immunoblotting.
Immunoblotting. Whole-cell lysates from BC cells were prepared with 2× sample buffer (60 mM Tris-HCl pH 6.8, 2% sodium dodecyl sulfate, 10% glycerol) and, before loading, were mixed with loading buffer (5% 2-mercaptoethanol, 0.1% bromophenol blue) and boiled for denaturation 19 . Proteins were separated by SDS-PAGE and subjected to immunoblot analysis using antibodies. For detection of hyperphosphorylated forms of IRE1-α, we used Phos-tag-based immunoblot 23 . Briefly, SuperSep Phos-tag gels (Fujifilm) were used to separate protein samples. Phos-tag gels were washed 3× for 20 min with transfer buffer with 10 mM EDTA before transfer. For detection of nascent translated proteins, puromycin (final concentration 10 μg ml -1 ) was added 30 min before harvesting of cells. Samples were subjected to immunoblotting, and nascent proteins were detected by anti-puromycin antibody.
RT-qPCR. Total RNA was prepared from BC cells using the RNeasy Mini Kit (Qiagen) according to the manufacturer's protocols. Subsequently, total RNAs were reverse transcribed into cDNA using iScript Reverse Transcription Supermix (Bio-Rad) and with SYBR Green on an Illumina Real-Time PCR system. Primers utilized are listed in Supplementary Table 2.
Cellular thermal shift assay. Cellular thermal shift assays were performed as previously described 24 . Briefly, cells were treated with DMSO or 10 μM ERX-41 for 30 min then trypsinized, pelleted and resuspended in PBS supplemented with Halt protease inhibitor cocktail (Thermo Scientific), with DMSO or 10 μM ERX-41. Resuspended cells were aliquoted into PCR strips. Cells were incubated in a thermal cycler (Bio-Rad) at gradient temperatures for 3 min, followed by incubation at 25 °C for 3 min. Cells were snap-frozen in liquid nitrogen and subjected to two freeze-thaw cycles. Samples were briefly vortexed and centrifuged at 20,000g for 20 min at 4 °C. Cleared cell lysates were mixed with a 1/3 volume of 4× Laemmli sample buffer. After boiling, cell lysates were resolved in SDS-PAGE followed by immunoblotting.
LIPA staining in TNBC tissue microarray and primary normal breast tissue. TNBC tissue microarray (TMA) was generated by the UTSW Pathology Laboratory from 51 patients with high-grade TNBC. All cases were treatment naive. This study was approved by UTSW Institutional Review Board (no. STU-032011-117). TNBC diagnoses were based on IHC staining with image quantitation of ER, PR and HER2, and were confirmed by a board-certified BC pathologist (Y. Peng) at UTSW. Cerebellar tissue was included on a TMA slide as negative control. Slides cut from tissue blocks were immunostained for LIPA (1:500). An additional 20 slides were obtained from breast-reduction surgeries from women without known BC. Stained slides were scored manually per tissue core independently by a pathologist who was blinded to clinical data. Immunostaining data were registered semiquantitatively in staining intensity (0, no staining; 1, weak staining; 2, moderate staining; and 3, intense staining; representative staining examples ranging from 0 to 3 are provided).

TNBC PDE studies.
For PDE studies, excised tissue samples were processed and cultured ex vivo as previously described 5,[7][8][9] . Deidentified patient tumors were obtained from UTSW Tissue Repository after institutional review board approval (no. STU-032011-187). Inclusion criteria included women with previous histologic confirmation of TNBC and who were undergoing surgical extirpation or biopsy of their primary tumor. Previous treatment with chemotherapy and/or radiation was allowed. Exclusion criteria included concurrent or previous diagnosis of other malignancies or previous evidence of ER-α + or HER2 + BC. All cases were reviewed by UTSW Tissue Repository in advance, and patients provided consent for their tissue to be used for laboratory research. Only deidentified information was shared with the laboratory. None of the laboratory personnel had access to additional patient information. No attrition was noted. Briefly, fresh tumor samples were incubated on gelatin sponges in culture medium containing 10% FBS, followed by treatment with either vehicle or 2.5 μM ERX-41 for 72 h. Representative tissues were fixed in 10% formalin at 4 °C overnight and subsequently processed into paraffin blocks. Sections were then processed for IHC analysis.
Immunofluorescence. Cells were fixed with ice-cold methanol and blocked with 5% normal serum and 0.3% Triton X-100 PBS. After overnight incubation with primary antibodies at 4 °C and four washes with 0.3% Triton X-100 PBS, samples were incubated with secondary antibodies at RT for 1 h. After four washes with 0.3% Triton X-100 PBS, cells were further washed with Hank's balanced salt solution plus calcium and magnesium. Samples were then incubated with 0.2 μM ER Tracker Red (Molecular Probes) at 37 °C for 30 min. Cells were further washed with Hank's balanced salt solution plus calcium and magnesium before mounting. Samples were imaged with a confocal microscope (Zeiss LSM 880). TEM. DMSO-or ERX-41-treated cells were fixed with 2.5% glutaraldehyde (Electron Microscopy Sciences) in 0.1 M sodium cacodylate pH 7.4 buffer (Electron Microscopy Sciences) for 20 min at RT. After rinsing with 0.1 M sodium cacodylate pH 7.4 buffer, cells were further fixed with 1% osmium and 0.8% K 3 Fe(CN 6 ) in 0.1 M sodium cacodylate pH 7.4 buffer. After prestaining with 4% uranyl acetate in 50% ethanol, cells were dehydrated with series concentrations of ethanol (50-100%). After transitioning from propylene oxide to resin and embedding in Embed 812 resin (Electron Microscopy Sciences), cells were located using light microscopy and trimmed out. Sections (60-70 nm) were cut, mounted on formvar-coated grids and viewed with a transmission electron microscope (Tecnai G2 spirit, FEI) equipped with a LaB6 source using a voltage of 120 kV.
Lipase activity assay. Lipase activity assays were performed as described 25 . Briefly, a 0.345 mM substrate solution was prepared from 1.2 ml of 13.3 mM 4-MUP and 42 ml of 100 mM sodium acetate buffer pH 4.0, 1.0% (v/v) Triton X-100 and 3.0 ml of 0.5% (w/v) cardiolipin. For enzyme reactions, 50 μl of substrate in buffer solution, 40 μl of diluted cell lysate, 10 μl of DMSO and either 30 μM Lalistat 2 or 30 μM ERX-41 were combined in a black, 96-well plate. Plates were sealed with an adhesive aluminum film and incubated in an incubator at 37 °C for 1 h. Reactions were terminated using 200 μl of 150 mM EDTA at pH 11.5. Plate fluorescence was read immediately with a synergy H1 fluorescence microplate reader (BioTek) using a 365-nm excitation filter and a 450-nm emission filter. LIPA lipase activity was calculated by subtracting the enzymatic activity of inhibited (with Lalistat 2) reaction from that of the uninhibited (without Lalistat 2) reaction.