Targeting the ATF6-Mediated ER Stress Response and Autophagy Blocks Integrin-Driven Prostate Cancer Progression

In the United States, prostate cancer is the second-leading cause of cancer lethality in men. Prostate cancer death is attributed to metastatic castration-resistant prostate cancer (mCRPC) and multiple alterations in homeostatic signaling networks. Primary therapeutic approaches (androgen deprivation, prostatectomy, and radiotherapy) have failed, and normal signaling networks are altered, including integrins, the surface transmembrane αβ heterodimer receptors, which play a crucial role in prostate tumor progression. Integrins bind to various ligands in the extracellular matrix (ECM), including fibronectin (FN). The alteration of integrin signaling increases cell migration, invasion, proliferation, and cancer cell survival. It is well established that CRPC is associated with the atypical expression of different integrin proteins, especially those containing the α_v subunit (1).

Multiple studies have clarified the link between endoplasmic reticulum stress (ER stress), unfolded protein response (UPR), and cancer (reviewed in ref. 2). Several reports indicate that cancer cell Golgi fragmentation is driven by prolonged, sublethal ER stress (3–5). One branch of UPR, mediated by the activating transcription factor 6 (ATF6), is mechanistically linked to the Golgi. Under normal conditions, cleavage of ATF6 in the Golgi by S1P and S2P proteases is necessary for the transactivation of ATF6-mediated ER stress response. Molecular studies in our laboratory confirm that the dimeric trans-Golgi matrix protein, GCC185, serves as a Golgi retention partner for S1P and S2P proteases. However, Golgi disorganization in advanced prostate cancer cells results in the monomerization of GCC185 and its downregulation. This, in turn, leads to the translocation of S1P and S2P to the ER, where these proteins cleave ATF6 (6). This elegant escape mechanism employed by prostate cancer cells accelerates ATF6-mediated UPR, thus maintaining the fragmented Golgi phenotype of prostate tumor cells. In previous studies, we introduced the “onco-Golgi” concept, postulating that Golgi structural dispersal is associated with specific prometastatic glycosyl epitopes (7). For instance, androgen-responsive LNCaP and 22Rv1 cells have a compact perinuclear Golgi, whereas androgen-refractory, docetaxel-resistant PC-3 and DU145 cells demonstrate disorganized Golgi, which is associated with de-dimerization of the key Golgi matrix protein, Giantin (4). This is associated with the mislocalization of critical O-glycosylation enzymes, which leads to the formation of prometastatic glycan epitopes (4, 8). It was not known whether the same association exists between fragmented Golgi phenotype and activation of N-glycans that promote prostate tumor growth and metastases.

In cells with normal compact Golgi morphology, biantennary N-linked glycan chains can be modified by adding a bisecting GlcNAc. The reaction is catalyzed by N-acetylglucosaminyltransferase-III (MGAT3; Fig. 1A, asterisk). This suppresses further elongation of the N-glycan by N-acetylglucosaminyltransferase-V (MGAT5) or N-acetylglucosaminyltransferase-IV (MGAT4; ref. 9). Enhanced expression of MGAT3 has been found to block MGAT5-mediated glycosylation and suppress metastasis (10). In tumor cells, MGAT5-modified glycans are processed by beta-1,4-galactosyltransferase (β1,4-GalT) and elongated with poly-N-acetyllactosamine (LacNAc), followed by the addition of sialic acid and fucose. The α_v integrins, especially α_vβ₃, α_vβ₅, and α_vβ₆, form an abnormal cell surface repertoire due to the high affinity of their LacNAc to Galectin-3 (Gal-3). At the PM, integrins and pentameric Gal-3 form clusters, termed lattices, which modulate tumor cell behavior, including the cardinal metastatic factors: adhesion to ECM and migration (11). Integrin α_vβ₃ and α_vβ₅ promote prostate tumor dissemination to lymph nodes and distant organs, including bones (12–14), suggesting their potential as prognostic markers. Overexpression of MGAT5 was also detected in other types of tumors (15–17). Importantly, enhanced glycosylation by MGAT5 was observed in mice metastatic xenografts derived from LNCaP and PC-3 cells, as well as in patients with prostate cancer (18). Overexpression of MGAT5 significantly enhanced LNCaP cell invasion (19). Bisecting GlcNAc structures in certain glycoproteins, including integrins, results in their altered expression levels and decreased sorting to the cell surface (20). Conversely, integrins and other ECM-binding proteins glycosylated by MGAT5 demonstrate enhanced representation at the plasma membrane (PM) via interaction with Gal-3, which is associated with a more aggressive tumor phenotype (21, 22). It is unclear what mechanism dictates the downregulation of MGAT3 and subsequent upregulation of MGAT5 in advanced prostate cancer.

In prostate cancer cells, autophagy is involved in the dysregulation of androgen receptor (AR) signaling and may interfere with other pathways, ultimately leading to castration resistance and dissemination of prostate tumors (reviewed in ref. 23). Autophagy inhibition by chloroquine and hydroxychloroquine (HCQ), alone or in combination with other agents, have been shown to overcome chemoresistance (24, 25), reduce prostate cancer cell proliferation, and induce apoptosis (26, 27). Despite our increased understanding of the antitumor effect of HCQ, no mechanism is known that can clearly describe its potential effect on prostate cancer cell metastasis and the expression of integrins.

The primary objective of this study was to examine the impact of fragmented Golgi phenotype and ATF6-mediated UPR on the activation of MGAT5-mediated glycosylation of integrins and their expression at the PM of prostate tumor cells. Using both in vitro and in vivo models of prostate cancer, we discovered exciting links between cell surface expression of integrin α_v and Gal-3 and detected that the internalization of α_v integrins does not require Gal-3. Finally, we found that inhibition of autophagy by HCQ with parallel ablation of ATF6 remarkably decreases the expression of α_v integrins, suggesting a novel effective therapeutic strategy for prostate tumor metastases.

Male athymic nude mice (BALB/c nu/nu, 25–31 g, 5–6 weeks old; The Jackson Laboratory) were housed in filter-top cages in the animal facility at the University of Nebraska Medical Center (UNMC) and provided food and water ad libitum. The mice underwent surgical implantation of control or ATF6 KD PC-3 cells as described in the Supplementary Materials and Methods. Two weeks after implantation, mice were randomly assigned to PBS or HCQ groups and received by intraperitoneal injection of 25 mg/kg HCQ or PBS twice per week (based on a weekly physiologic dosage of 50 mg/kg) for 4 weeks. Body weights were recorded weekly, and mice were sacrificed 6 weeks after implantation. Tumors and serum were collected for IHC, IF, and ALT/AST analysis, and tumor volume was recorded. This protocol was approved by the Institutional Animal Care and Use Committee (IACUC) at UNMC.

The tissue sections from normal prostate and patients with prostate cancer were obtained from US Biomax, Novus Biological, and Creative Bioarray. Also, sections were provided through the Department of Pathology and Microbiology (IRB protocol no. 304–16-EP) at the University of Nebraska Medical Center and the Johns Hopkins University School of Medicine (Prostate Cancer Biorepository Network), and the Vernadsky Crimean Federal University (Russian Federation, IRB protocol no. 98-2019-KFU). Prostate cancer was graded in accordance with the 2014 International Society of Urological Pathology (ISUP) Guideline. Alcohol consumption was stratified as follows: nonalcoholic—less than one drink per month; moderate-alcoholic—five to six beers per week, three to five glasses of wine per week, or two to three 3.4 oz servings of hard liquor per week; heavy-alcoholic–two or more beers per day, one or more glasses of wine per day, or one or more 3.4 oz servings of hard liquor per day.

Statistical analysis was performed using Microsoft Excel, Prism 9.0 (GraphPad), and R Software. Statistical parameters are described in each figure legend. Multiparametric analysis was performed on the basis of median values in R software as indicated using: (i) Tukey method, P_adjusted by Benjamini–Hochberg, (ii) Dunn test (1964) Kruskal–Wallis multiple comparisons, P_adjusted by Benjamini–Hochberg, (iii) pairwise comparison using Wilcoxon rank sum exact test, P_adjusted by Benjamini–Hochberg, or (iv) Pairwise t test, P_adjusted by Benjamini–Hochberg. Other analyses based on mean ± SD included the Mann–Whitney test, Student t tests (unpaired, two-sided, or two-tailed), and one-way or two-way ANOVA. The specific test and parameters used, sample size, and P values are reported in the figures and figure legends.

For IHC, the data of integrin α_v or Gal-3 from each microarray of tissue specimens were sorted by groups (normal or prostate cancer grades), and correlation between all combinations of Abs (Pearson or Spearman r value) was evaluated. If the coefficient was greater than or equal to 0.4, those combinations were considered correlated and used to calculate the median value for each patient, which was then incorporated into the final analysis. After subtracting combinations that showed no correlation and samples with damaged tissues, we analyzed 19 donors and the following numbers for prostate cancer samples: n = 18 for grade 1, n = 33 for grades 2 to 3, and n = 56 for grades 4 to 5.

To evaluate lymphatic or organ metastases statistically in the orthotopic model of prostate tumor, the number of mice with and without lymphatic and/or organ metastasis was counted as “yes” or “no,” as in, yes metastasis or no metastasis. The results were inputted into a table, then, using R-statistical software, a chi-square analysis was done. If the P value was less than or equal to 0.05, then a post hoc evaluation using Fisher exact test was done to compare each group. The P value was adjusted using the Benjamini–Hochberg method.

The remaining methods are detailed in the Supplementary Information, including: Antibodies and Reagents; Cell Culture, EtOH and HCQ Treatment; LC3B Plasmid Transfection; Plasma Membrane Protein Isolation; RNA Isolation and RT-qPCR; Lectin-affinity chromatography, Fibronectin Adhesion Assay; Pulse-Chase Endocytosis Assay; Confocal Immunofluorescence Microscopy and Analysis; 3D SIM and Analysis; Proximity Ligation Assay; Electron Microscopy; Immunohistochemistry, Lectin Histochemistry and Analysis; ALT and AST Activity Assays; Detailed Orthotopic Implantation.

The tissue sections from normal prostate were obtained from US Biomax and Novus Biological. Also, sections were provided through the Department of Pathology and Microbiology (IRB protocol no. 304–16-EP) at the University of Nebraska Medical Center and the Johns Hopkins University School of Medicine (Prostate Cancer Biorepository Network), and the Vernadsky Crimean Federal University (Russia, IRB protocol no. 98–2019-KFU). All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. Informed consent was obtained from all individual participants included in the study. The animal studies were approved by the Institutional Animal Care and Use Committee (IACUC) at UNMC.

The datasets used and/or analyzed during this study are available from the corresponding author upon reasonable request.

Previously, we reported that expression of MGAT3 protein in high passage androgen-refractory LNCaP cells (c-86) is reduced compared with androgen-responsive, low passage LNCaP (c-26) cells (hereafter, LNCaP; ref. 28). In contrast, the MGAT5 expression was higher. In addition, ethanol (EtOH)-induced Golgi disorganization in LNCaP cells was accompanied by translocation of MGAT3 from the Golgi to the ER, whereas intra-Golgi localization of MGAT5 appeared unaffected (28). Previous studies indicated that MGAT3 and MGAT5 reside predominantly within medial-Golgi membranes (29, 30). However, because MGAT3 and MGAT5 compete for N-glycan elongation (Fig. 1A), and activation of MGAT3 leads to the blockage of glycosylation via MGAT5 and reduction of metastasis (19), it is logical that MGAT3-mediated glycosylation has priority over MGAT5 under normal conditions, potentially due to more proximal intra-Golgi localization. To check this hypothesis, we performed a detailed analysis of the sub-Golgi position of these enzymes, using structured illumination superresolution microscopy (SIM) of Golgi stacks, which allows us to achieve two-color 3D imaging at ∼110 nm resolution. In LNCaP cells, MGAT3 colocalization with the cis-Golgi marker, GM130, was higher than with the medial-Golgi marker, Giantin (Fig. 1B and C). The distribution of MGAT5 was opposite to that of MGAT3; more colocalization with Giantin than GM130. These data indicate that MGAT3 is localized in the earlier Golgi compartments than MGAT5, explaining the MGAT3 priority over MGAT5 for modification of proteins carrying N-glycans in Golgi. This cell line finding was further validated in human normal and neoplastic prostate tissues. Indeed, in normal prostate, MGAT3’s colocalization with cis-Golgi marker GRASP65 was higher than that for MGAT5. In contrast, MGAT5 exhibits stronger colocalization with Giantin than GRASP65 (Fig. 1D–G and J). In tumor tissues from patients with prostate cancer, the Golgi is fragmented, and we observed that MGAT3 loses cis-Golgi positioning whereas MGAT5 maintains its localization within the medial-Golgi (Fig. 1H–J). A similar tendency was observed in PC-3 and DU145 cells with fragmented Golgi: MGAT5 displayed strong colocalization with Giantin, but colocalization between MGAT3 and GM130 was weak (Fig. 1K–M). Western blotting of the ER fraction isolated from PC-3 and DU145 cells detected a significantly higher amount of MGAT3 protein than that in LNCaP cells (Fig. 1N). Finally, using mouse anti-GM130 and rabbit anti-MGAT3 antibodies (Abs), we performed Proximity Ligation Assay (PLA) in normal prostate and prostate cancer tissue samples. Significant reduction of MGAT3 and GM130 closeness was documented in tumor cells, indicating non-Golgi positioning of MGAT3 (Fig. 1O and P).

In sum here, MGAT3 demonstrates cis-Golgi localization while MGAT5 is positioned in medial-Golgi, and in advanced prostate cancer cells, MGAT5, but not MGAT3, maintains a strong intra-Golgi signal; this may justify the enhanced integrin expression in prostate tumors. Higher expression of integrin α_v in PC-3 and DU145 compared with nonmalignant prostate epithelial (RWPE-1) cells, low aggressive malignant LNCaP, and 22Rv1 cells supports this observation (Fig. 1Q). Also, it is known that PC-3 and DU145 cells demonstrate higher expression of MGAT5 compared with androgen-responsive cells (19, 31). Interestingly, LNCaP and 22Rv1 cells, which exhibit the lowest expression of integrin α_v, also express little, if any, Gal-3 (Fig. 1R). This aligns with previous observations indicating marginal expression of LGALS3 (Gal-3) mRNA in these cell lines (32, 33).

It is well-known that expression of α_v integrins and Gal-3 in primary prostate cancer and patients with mCRPC is aberrant and positively correlated with Gleason grades (13, 34–37). However, to our knowledge, no prior work convincingly analyzes integrin α_v and Gal-3 co-expression on the prostate tumor cell surface. Here, we performed comprehensive IHC analysis, staining integrin α_v and Gal-3, then measuring their PM signal using Multiplex IHC v3.1.4 algorithms by Halo v3.4 (Indica Labs, Inc.). To clearly distinguish the expression of these proteins at the cell surface, the PM was labeled with one of two different markers, E-cadherin or Na⁺/K⁺-ATPase. The PM stain was then segmented to match the visually observed PM cues most appropriately. Each microarray of prostate tissue from donors (normal, n = 49) and primary patients with prostate cancer with varying grades of tumor (n = 160) was stained with different combinations of anti-integrin α_v, anti-Gal-3, and anti-E-cadherin or anti-Na⁺/K⁺-ATPase Abs (see Supplementary Table S1). Representative images from one Ab combination are presented in Fig. 2A. The H-score of integrin α_v at PM demonstrated no difference between normal and prostate cancer grade 1; however, a significant increase was detected in all other prostate cancer grades (Fig. 2C). Also, cancer patient specimens with grades 4 to 5 exhibited higher integrin α_v PM H-score than their grades 2 to 3 counterparts. The data of PM-specific H-score for Gal-3 were slightly different. Although we quantified a significant increase in prostate cancer grades 2 to 3 compared with normal and grade 1, there was no substantial difference between grades 2 to 3 and 4 to 5 (Fig. 2D). Importantly, colocalization between integrin α_v and Gal-3 at the cell surface was higher in grade 1 and grades 2 to 3 compared with normal. However, patients with grades 4 to 5 demonstrated significantly lower colocalization than grades 2 to 3 (Fig. 2E).

Next, in the same patient cohort, we analyzed the PM expression of different α_v integrins using specific Abs against both α_v and different β subunits. The PM signal for integrin α_vβ₃ in all grades of prostate cancer was significantly higher than that in normal prostate. Interestingly, patients with grades 4 to 5 showed a slight, although still significant, decrease compared with grades 2 to 3 (Supplementary Figs. S1A and S1B). Also, the colocalization between integrin α_vβ₃ and Gal-3 at PM was higher in prostate cancer samples than normal (Supplementary Fig. S1C). The expression of integrin α_vβ₅ was higher in all prostate cancer groups compared with normal; however, no difference was detected between grades (Supplementary Figs. S2A and S2B). Like α_vβ₃, α_vβ₅ colocalized with Gal-3 on the PM more in prostate cancer than in normal donors (Supplementary Fig. S2C). The results of the IHC signal for PM integrin α_vβ₆ were most interesting: in addition to the difference between all grades and normal, there was a significant rise in grades 2 to 3 and 4 to 5 compared with grade 1 (Supplementary Figs. S3A and S3B). Again, the association of Gal-3 with integrin α_vβ₆ at the cell surface was more evident in prostate cancer (Supplementary Fig. S3C). These data indicate that most α_v integrins are highly associated with Gal-3 on the prostate cancer cell surfaces and increase with the cancer grade.

Then, we compared integrin α_v and Gal-3 PM expression in patients with mCRPC (with tissue or bone metastases) and nonmetastatic patients (Fig. 2B). We found that integrin α_vH-score at the cell surface of metastatic samples (both tissue and bone) was substantially higher than that in primary tumors (Fig. 2F). Importantly, patients with bone metastases demonstrated higher H-score than those with tissue metastases. A similar tendency was observed for Gal-3: patients with tissue metastases display less PM-specific Gal-3 than those with bone metastases. However, only bone metastasis cases had a significant increase compared with nonmetastatic patients (Fig. 2B and G). The colocalization between integrin α_v and Gal-3 at PM in bone metastases mirrored results from primary tumors of patients with grades 2 to 3 and 4 to 5 (Fig. 2E). Although this parameter was higher in tissue metastatic patients than in patients with the primary tumor, we detected a significant decrease in bone metastases compared with tissue and primary prostate tumors (Fig. 2H).

The data presented above suggest that the development of metastases is associated with enhanced expression of abnormally glycosylated integrins. We performed a series of lectin IHC staining of tumor samples mentioned above. To evaluate MGAT5-mediated glycosylation of integrin α_v at the cell surface, we co-stained integrin α_v with Na⁺/K⁺-ATPase and phaseolus vulgaris leucoagglutinin (PHA-L) lectin, which binds preferentially to GlcNAc residues on β1–6 branches of tri- or tetra-antennary sugar chains (the product of MGAT5 glycosylation; refs. 38, 39; Fig. 1A). As shown in Fig. 2I and J, colocalization of integrin α_v with PHA-L lectin at PM was significantly higher in both groups of patients with mCRPC compared with nonmetastatic patients. However, the difference between tissue and bone metastases was nonsignificant. Therefore, the development of mCRPC is linked to an increase of MGAT5-modified integrin α_v at the PM of tumor cells.

In previous studies, we reported that ablation of motor protein nonmuscle Myosin IIA (NMIIA) restores compact Golgi in PC-3 and DU145 cells (4). Given that, broadly speaking, Golgi and autophagy are interconnected in many aspects (reviewed in ref. 40), including that Golgi membranes serve as a source for phagophores, we wondered whether Golgi disorganization in prostate cancer cells could be attributed to enhanced autophagy. Thus, we hypothesized that autophagy blockage would restore Golgi morphology. Here, we found that treatment of PC-3 cells with HCQ (50 μmol/L HCQ for 72 hours) converts the Golgi phenotype from fragmented to more compact (Fig. 3A and B). A similar effect of HCQ was observed in DU145 cells, although it required a higher concentration (60 μmol/L HCQ for 72 hours; Supplementary Figs. S4A and S4B). The dosage of HCQ was determined on the basis of the physiologic concentration that, within 72 hours, induced a significant reduction in the number of Golgi membranes but did not induce apoptosis.

In addition, to determine the effect of ATF6-mediated ER stress response on the Golgi, we developed shRNA-mediated stable ATF6α knockdown (KD) in PC-3 cells (Fig. 3C). In ATF6 KD cells, we detected decreased expression of ATF6 target proteins, including ERp72 and GRP78 (Fig. 3D and E). Interestingly, such downregulation of ATF6 did not induce a compensatory response from the other two branches of UPR: PERK and IRE-1. Despite a slight increase in total PERK, the expression of its phosphorylated form was identical to control cells (Fig. 3F and G). Also, we observed a moderate reduction of both total and phosphorylated IRE-1 (Fig. 3H and I). These results support the conclusion that ATF6 KD cells lack excessive ER stress response, which is typically observed in control PC-3 cells and in advanced prostate tumor cells (6). Consequently, Giantin's dimerization in ATF6 KD cells was higher than in control cells, implying the restoration of Golgi morphology (Fig. 3J; refs. 4, 41). Indeed, depletion of ATF6 has an effect close to HCQ: in both PC-3 (Fig. 3A and B) and DU145 cells (Supplementary Figs. S4C–S4E) lacking ATF6α, Golgi appeared more compact and perinuclear compared with cells transfected with an appropriate control shRNA/siRNA. Remarkably, treating ATF6 KD cells with HCQ had the most noticeable impact on Golgi morphology, leading to a significantly smaller number of Golgi fragments than HCQ treatment or ATF6 KD alone (Fig. 3A and B).

To validate these results, we performed electron microscopy (EM) of PC-3 cells treated with HCQ. In control cells, multiple unstacked Golgi fragments were detected, consistent with our previous observations (Fig. 3K, top, indicated as G; ref. 4). On the contrary, in HCQ-treated cells, we found stacks of Golgi cisternae fused in a continuous ribbon structure (Fig. 3K, bottom). Similarly, PC-3 and DU145 cells depleted of ATG5, a key protein for the extension of the phagophore membrane, demonstrate a more compact Golgi phenotype (Supplementary Figs. S4F–S4K). Notably, treatment of PC-3 cells with Bafilomycin A1 (10 μmol/L for 72 hours), which also inhibits autophagosome–lysosome fusion and autolysosome acidification (42), results in a significant reduction of the number of Golgi membranes (Supplementary Figs. S4L and S4M).

Our previous study found that, in PC-3 and DU145 cells, O-glycosylation enzyme core 2 N-acetylglucosaminyltransferase-L (C2GnT-L) mislocalized from Golgi to the ER; however, restoration of Golgi through KD or inhibition of NMIIA resulted in the recovery of its Golgi positioning (4). This stimulated us to examine whether HCQ-induced Golgi reconstruction will bring MGAT3 back to the Golgi. Using 3D SIM, we rigorously analyzed the colocalization of MGAT3 with GRASP65 before and after HCQ treatment of PC-3 cells. As shown in Fig. 3L and M, the colocalization of MGAT3 with GRASP65 was substantially enhanced in HCQ-treated cells, indicating the restoration of MGAT3 intra-Golgi localization (Movies S1 and S2). Importantly, we noticed that recovery of compact perinuclear Golgi is associated with a reduction of Golgi area. This was confirmed by the expression of Golgi scaffold proteins, golgins, which represents the matrix of Golgi membranes. The protein level of cis-golgins, GRASP65 and GM130; medial-golgin, Giantin; and trans-golgins, Golgin-245 and TGN46, was substantially reduced after HCQ (Supplementary Figs. S4N–S4R).

The next question of interest was whether fragmented Golgi membranes in prostate cancer cells feed directly into autophagy, that is, whether prostate cancer cells demonstrate a high rate of Golgiphagy. To check this, we employed the Autophagy Tandem Sensor RFP-GFP-LC3B Kit (Thermo Fisher Scientific, see Supplementary Materials and Methods for details) that allows for the detection of the maturation of the autophagosomes or autophagic vacuoles (AV) to the autolysosome. Briefly, by combining an acid-sensitive GFP with an acid-insensitive RFP, the change from AV (neutral pH) to autolysosome (with an acidic pH) can be visualized by the specific loss of the GFP fluorescence, leaving only a red signal. To monitor Golgi morphology and its incorporation into AVs, PC-3 cells transfected with this LC3B-plasmid were also stained with Alexa Fluor 647 anti-GM130 Ab (Abcam, ab195303). First, we verified the autophagy flux rate by taking the AV ratio (yellow-red and green colocalized) to autolysosome (red). Predictably, because HCQ blocks the fusion of AVs with lysosomes, the area of autolysosomes was significantly reduced in HCQ-treated cells (Fig. 3N and O). This was further validated by W-B analysis of LC3B-II protein in control and ATF6 KD cells treated with HCQ (Fig. 3P). HCQ treatment alone significantly enhanced LC3B-II expression, indicating the blockage of its degradation. Meanwhile, depletion of ATF6 alone had no impact on the level of LC3B-II compared with control cells. However, comparison of control and ATF6 KD cells treated with HCQ indicates even more accumulation of LC3B-II in cells lacking ATF6, suggesting that alleviation of the ER stress response may intensify the effect of HCQ. Next, we analyzed the colocalization of autophagosomes with Golgi membranes. In control PC-3 cells, some autophagosome (yellow) and Golgi membrane (magenta) punctae were merged, indicating that under normal conditions, prostate cancer cells are experiencing steady-state Golgiphagy (Fig. 3N, Ctrl, and Q). Indeed, EM imaging revealed in both PC-3 and DU145 cells multiple phagophores nucleating from Golgi cisternae and AVs localized in the vicinity of Golgi (Supplementary Figs. S5A and S5B). Moreover, Golgi membranes isolated from PC-3 and DU145 cells contain a substantially higher level of early autophagy marker WIPI2 compared with LNCaP cells (Supplementary Fig. S5C). In cells treated with HCQ, Golgi compaction was associated with a significant decrease in its colocalization with autophagosomes (Fig. 3N, HCQ, and Q), implying the blockage of Golgiphagy.

The recovery of MGAT3's intra-Golgi localization after HCQ suggests that the pool of integrins modified by MGAT5 will be reduced. To examine this, we performed lectin chromatography using immobilized PHA-L lectin. The lysates from control and HCQ-treated PC-3 cells were incubated with PHA-L-Separopore 4B beads, followed by extensive washing and elution by GlycoElute Elution Buffer. Integrin α_v immunoblotting of eluates showed a significant reduction of integrins captured by PHA-L, indicating the blockage of glycan modification via MGAT5 (Fig. 4A).

Abrogation of glycosylation via MGAT5 reduces the complex formed between integrin α_v and Gal-3. This, in turn, should limit integrins' retention at the cell surface due to the inability to form clusters. To check this, we performed a series of immunofluorescence (IF) experiments, employing the same strategy as IHC, co-staining integrin α_v with PM marker, Na⁺/K⁺-ATPase. As demonstrated in Fig. 4B and C, the PM intensity of integrin α_v was significantly reduced in HCQ-treated PC-3 cells. These data were further verified by W-B analysis: in HCQ-treated PC-3 and DU145 cells, the level of integrin α_v protein was decreased in the PM fraction (Fig. 4D). Analogously to integrin α_v, we found that after HCQ, Gal-3 exhibits a reduction of IF signal on the cell surface (Supplementary Figs. S6A and S6B). W-B analysis of PM fraction from PC-3 and DU145 cells validated this finding (Fig. 4E). Next, we examined whether the integrin α_v/Gal-3 association was decreased alongside the reduction in PM integrin α_v and Gal-3. In PC-3 cells, we performed PLA using mouse anti-Gal-3 and rabbit anti-integrin α_v Abs and co-stained PM with goat anti-E-cadherin Ab. In control cells, multiple PLA dots were observed at the cell surface; however, their numbers significantly declined in cells receiving HCQ (Fig. 4F and G).

Overall, these data prompt us to conclude that HCQ-induced abrogation of MGAT5-mediated glycosylation impacts the amount of both Gal-3 and integrin α_v at the cell surface. With all this information in hand, we proceeded to evaluate the adhesion potential of these cells to FN. Using polystyrene microtiter plates coated with FN, we measured the adhesion of PC-3 and DU145 cells. The average from three independent experiments strongly indicated a reduction in the adhesion of cells treated with HCQ (Fig. 4H).

It is known that cell surface expression and function of integrins are regulated by their internalization to EE, where they can be recycled back to the membrane to promote cell migration or routed to the lysosome for degradation (43). A study in breast carcinoma cells reported that Gal-3 mediates the endocytosis of β1 integrins (44). Thus, the next question we asked was: what is the role of Gal-3 in HCQ-induced redistribution of α_v integrins?

To address this question, PC-3 cells were stained for integrin α_v, Gal-3, and EEA1, an EE marker. Three-color colocalization analysis of these images revealed more integrin α_v/Gal-3 complexes in EEs after HCQ treatment (Fig. 4I and J, Supplementary Movies S3 and S4). The simplest explanation of these results is that HCQ blocks the fusion of EE with AV (40). However, integrin aggregation in EE could also be caused by the intensification of their internalization. We hypothesize that HCQ-induced deficiency in Gal-3 at PM stimulates internalization of MGAT5-modified integrins, which were delivered to the cell surface before HCQ's effect on Golgi and trafficking. To examine this assumption, we rigorously analyzed the impact of integrin α_v gene silencing on Gal-3 expression and vice versa. For depletion of each protein, we conducted two independent experiments using different siRNAs (see Supplementary Table S2). After integrin α_v KD, the total Gal-3 in lysates was unchanged (Fig. 4K, middle), but Gal-3 on the PM increased (Fig. 4K, bottom), indicating that Gal-3 cannot be internalized without integrins. Indeed, co-staining of Gal-3 and EEA1 in integrin α_v KD cells and quantification of colocalization between them indicates that lack of integrin α_v abrogates relocation of Gal-3 to EE (Fig. 4L and M). Therefore, to be transferred to EE, Gal-3 needs integrins. On the other hand, after Gal-3 KD, total integrin α_v was unchanged in lysate samples (Fig. 4N, middle). However, PM-localized integrin α_v was decreased (Fig. 4N, bottom), indicating that Gal-3 is required for the retention of integrin α_v on the PM.

We concluded that upon deficiency in Gal-3 (in this case, induced by HCQ), integrins are unable to form clusters and are quickly internalized. To prove this, we performed a proof-of-principle pulse-chase experiment. The detail of this experiment is described in the Supplementary Materials and Methods. Briefly, cells were treated with HCQ for 72 hours and then incubated with anti-integrin α_v Ab. The Abs were allowed to internalize for 1 hour, and then cells were stained for Gal-3 and EEA1, followed by secondary antibodies for all three antibodies (Fig. 4O). Then, three-color colocalization (integrin α_v/Gal-3 with EEA1) was measured again. It is important to stress that the IF signal of integrin α_v presented here can be only ascribed to the portion of integrins internalized from the PM together with anti-integrin α_v Ab within the 1-hour incubation. We found that in HCQ-treated cells, merged complexes α_v-integrins/Gal-3 were increased in EE (Fig. 4P). These data suggest that abrogation of Gal-3 trafficking to the cell surface may stimulate the internalization of integrins and Gal-3 that form a complex at the cell surface. Hence, in terms of Gal-3 expression at PM, it appears that HCQ treatment mimics the effect of Gal-3 depletion by intensifying integrin internalization.

These data raise the logical question of whether the decrease of integrins at the cell surface after HCQ treatment is due to altered gene transcription or altered trafficking to and stability at the PM. In both HCQ-treated PC-3 and DU145 cells, we found a significant reduction of integrin α_v mRNA expression by qRT-PCR analysis (Fig. 4Q). However, the total protein level of integrin α_v was identical to the control (Fig. 4R). These suggest a multifaceted effect. Although transcription of integrin α_v mRNA decreases after HCQ treatment, so does degradation as HCQ blocks lysosomal cleavage of autophagosome contents. Thus, the total protein levels remain constant, as less protein is produced, but less protein is degraded. Meanwhile, we see lower overall levels on the PM caused by increased internalization of integrin α_v. Therefore, the reduction of integrin α_v in the PM fraction cannot be ascribed solely to its gene downregulation but rather also to intensified internalization.

Interestingly, the effect of HCQ on PM expression of integrin α_v was close to that in ATF6 KD cells. Still, more reduction was observed in HCQ-treated ATF6 KD cells (Fig. 4S). Similarly, these cells demonstrated the most noticeable decrease of MGAT5 protein amount compared with ATF6 KD or HCQ-treated cells (Fig. 4T).

To further define the subcellular distribution of integrin α_v and Gal-3 after HCQ treatment, we performed double pre-embedding immunogold EM of integrin α_v and Gal-3 using Abs conjugated to 1.4- and 10-nm gold particles, respectively, in PC-3 cells. In control cells, we detected the association of both proteins at the PM as both single merged spots and clusters (Fig. 5A, Ctrl-PM). Also, integrin α_v and Gal-3 joined spots were observed in the cytoplasm, late endosomes (LE), EE, and Golgi, which is surprising because Gal-3 is suggested to be delivered to the cell surface via noncanonical trafficking (45). After HCQ treatment, we could not detect joined integrin α_v and Gal-3 spots in cytoplasm or Golgi. Nevertheless, many merged spots were found in EE. Quantification of integrin α_v and Gal-3 association indicated a significant decrease at the PM but a robust increase in EE (Fig. 5B and C), echoing the results presented above. Thus, enhanced internalization of α_v-integrins and Gal-3 is associated with their segregation at the cell surface.

In another series of single integrin α_v immunogold staining, we validated that treatment of PC-3 cells with HCQ substantially reduces the content of integrin α_v particles at PM (Fig. 5D and E). Then, we corroborated that Gal-3 depletion decreases integrin α_v at the PM, as presented above. In cells lacking Gal-3, the number of integrin α_v-positive particles was reduced at the cell surface (Fig. 5F and G). Importantly, in terms of integrin α_v distribution, we found two different aspects of Gal-3 depletion. On the one hand, Gal-3 KD leads to accumulation of integrin α_v in EE, confirming its enhanced internalization (Fig. 5G and H). Thus, the internalization of α_v integrins is independent of Gal-3. On the other hand, deficiency in Gal-3 impacts integrin α_v trafficking as we detected aggregation of integrin α_v in ER (Fig. 5G and I). Also, we found more integrin α_v in the trans-Golgi network (TGN) and multivesicular bodies (MVB; Fig. 5G), indicating its elevated recycling. In sum here, Gal-3 is required not only for the retention of α_v integrins at the cell surface but also for their anterograde trafficking. Conversely, the restoration of MGAT3-mediated glycosylation of integrins may impact the trafficking of Gal-3 to the cell surface because integrins carrying these glycans do not form a complex with Gal-3.

The link between alcohol consumption and the progression of prostate cancer is well-documented, but the mechanism is not fully understood (46). Our laboratory revealed how EtOH and its metabolites could interfere with the growth of prostate tumors. EtOH-induced Golgi disorganization in androgen-sensitive LNCaP and 22Rv1 cells (i) relocates GSK3β from Golgi to ER, which in turn, activates the HDAC6-HSP90-AR axis (8), and (ii) accelerates ATF6-mediated ER stress response via induction of Golgi dispersal and redistribution of S1P and S2P to the ER (6). Importantly, we have shown that the frequency of alcohol consumption positively correlates with Golgi fragmentation and activation of ATF6 (6, 28).

Considering this information, the next question to address was whether EtOH administration aggravates Golgi disorganization in prostate cancer cells and accelerates integrin α_v expression, thus potentially promoting prostate tumor expansion. Notably, both PC-3 and DU145 cells abundantly express the main EtOH-metabolizing enzymes, alcohol dehydrogenase (ADH1A) and aldehyde dehydrogenase (ALDH2; Supplementary Fig. S7). First, we found that the number of Golgi fragments was significantly increased in both PC-3 and DU145 cells after 72 hours of EtOH treatment (Fig. 6A, B, D, and E). This effect was detected at the dosage of 11.5 μmol/L EtOH, which is within the physiological dosage of alcohol administration that ranges from 1 to 100 μmol/L (47). Thus, this in vitro model falls at a very moderate to low-level dose, yet still elicits Golgi fragmentation, avoiding the cytotoxic effects of higher doses in vitro. Second, the intensity of integrin α_v IF was also enhanced at the PM following EtOH treatment (Fig. 6A, C, D, and F). Interestingly, we detected a strong positive correlation between the number of Golgi fragments and the PM integrin α_v IF signal, demonstrating a nexus between Golgi disorganization and the expression of integrins (Fig. 6G). W-B analysis of PM fraction from EtOH-treated PC-3 and DU145 cells confirmed enhanced retention of integrin α_v at the cell surface (Fig. 6H and I). This was further validated by immunogold staining of integrin α_v and EM imaging. In DU145 cells, clusters formed by integrin α_v-positive punctae were significantly enhanced in EtOH-treated cells (Fig. 6J and K, arrowheads, Fig. 6L). Notably, in both cell lines, EtOH also upregulates MGAT5 (Fig. 6M and N). We further examined whether this in vitro data could be replicated in primary patients with prostate cancer with the same tumor grade, consuming alcohol at different levels. We sorted patients with prostate cancer with grades 2 to 3 into three groups: non-alcoholic, moderate-alcoholic, and heavy-alcoholic, based on the criteria of alcohol consumption history published by our group previously and described in the methods (6, 28). First, we measured integrin α_vH-score at the PM of the prostate tumor cells. We found that this parameter was significantly higher in patients heavily consuming alcohol compared with their non-alcoholic counterparts (Fig. 6O and P). Analysis of integrin α_v and Gal-3 colocalization at PM revealed a significant increase in both alcohol-consuming groups compared with nondrinkers; however, no difference was detected between drinkers (Fig. 6Q). Next, we examined MGAT5 immunoreactivity. Again, we did not observe a substantial difference between patients drinking alcohol at moderate or heavy levels; nevertheless, heavy drinkers demonstrated a higher level of MGAT5 compared with nondrinkers (Fig. 6R and S).

To investigate the clinical significance of our findings, we examined the effect of ATF6 depletion in combination with HCQ treatment using a mouse model of prostate cancer. In nude mice, we performed the orthotopic implantation of control or ATF6 KD PC-3 cells, followed by treatment with PBS (control) or HCQ. Treatment was administered by intraperitoneal injection of 25 mg/kg HCQ in PBS or an equivalent volume of PBS twice per week, based on a weekly physiological dosage of 50 mg/kg (48). After 40 days of tumor growth, mice were sacrificed. Importantly, at this dosage, HCQ did not show hepatotoxicity, as the serum level of both aspartate aminotransferase (AST) and alanine aminotransferase (ALT) did not differ significantly from the control (Supplementary Fig. S8). As shown in Fig. 7A and B, tumor size was substantially reduced after HCQ treatment in mice that received PC-3 control cells. Similarly, the tumors derived from ATF6 KD cells were smaller than the control and even smaller than in the HCQ group. However, mice injected with ATF6 KD cells and treated with HCQ demonstrated the smallest average tumor size, significantly lower than the control, HCQ, or ATF6 KD alone groups. Moreover, in control mice, various macro- and micro-metastases were detected in regional and distant lymph nodes, as well as in the liver and lung (Fig. 7A, Ctrl, arrowheads). In the HCQ-treated or ATF6 KD groups, we could not detect organ metastases; however, regional lymph node metastases were found (Fig. 7A, HCQ and ATF6 KD, arrowheads). Finally, the tumors derived from ATF6 KD cells in mice receiving HCQ did not demonstrate signs of metastasis. Detailed information about the tumor size and the number of metastases is presented in Supplementary Table S3. A significant difference in the occurrence of lymphatic metastases was detected between (i) Ctrl and ATF6 KD-HCQ, (ii) HCQ and ATF6 KD-HCQ, and (iii) ATF6 KD and ATF6 KD-HCQ groups. In addition, a significant difference in the occurrence of organ metastases was detected between (i) Ctrl and HCQ and (ii) Ctrl and ATF6 KD-HCQ groups. This Pearson Chi squared and Fisher exact test statistical analyses are summarized in Supplementary Table S4.

Next, using both IHC and IF of the orthotopic tumor samples, we were able to reproduce our in vitro data. First, we demonstrated the restoration of Golgi after HCQ or ATF6 KD and significant compaction of Golgi in the ATF6 KD-HCQ group compared with HCQ or ATF6 KD groups (Supplementary Figs. S9A and S9B). Second, we validated the lack of ATF6 in ATF6 KD cell-derived tumors (Supplementary Figs. S9C and S9D). Third, PM expression of integrin α_v was reduced substantially in all groups of mice compared with the control (Supplementary Figs. S9E and S9F). Moreover, a robust reduction of this parameter was found in ATF6 KD-HCQ relative to HCQ or ATF6 KD alone. In addition, the PM expression of Gal-3 was significantly lower in ATF6 KD-HCQ than in the control group or ATF6 KD (Supplementary Figs. S9E and S9G).

The recovery of MGAT3's intra-Golgi localization after HCQ (Fig. 3L and M) suggests that integrins will not be modified by MGAT5 after HCQ treatment. To evaluate MGAT5-mediated glycosylation of integrin α_v in orthotopic tumor samples, we co-stained integrin α_v with Cy3-labeled PHA-L lectin. To obtain more detailed information about PHA-L and integrin α_v colocalization, we captured high-resolution Z-stack images. We reconstructed 3D image projection of images using Imaris and quantified the Mander's overlap coefficient B, which reflects the proportion of the red signal (PHA-L) coincident with a signal in the green channel (integrin α_v) over the total red intensity (Supplementary Figs. S10A and S10B). In all groups of mice, Mander's coefficient was substantially reduced compared with the control. Also, ATF6 KD-HCQ samples demonstrated the lowest level of colocalization compared with ATF6 KD or HCQ.

A similar experiment was performed using Phaseolus Vulgaris Erythroagglutinin (PHA-E) lectin, which specifically recognizes complex type N-glycans containing bisecting GlcNAc, the product of MGAT3 (Fig. 1A; ref. 49). Co-staining of the orthotopic tumor samples with PHA-E and integrin α_v revealed significantly higher colocalization in ATF6 KD and HCQ samples. Again, the number of merged PHA-E/integrin α_v spots was the highest in ATF6 KD-HCQ samples (Supplementary Figs. S10C and S10D). Overall, these results mirror those from PHA-L, proving that restoration of MGAT3 glycosylation associates with the reduction of glycan modification via MGAT5.

Here, for the first time, we explain the phenomenon of excessive MGAT5-mediated glycosylation in prostate cancer tumor cells. We detected that the progression of prostate malignancy is associated with the mislocalization of MGAT3, which, otherwise, in cells with unaffected Golgi, blocks MGAT5-mediated sugar modification. Histochemical clinical data indicate that the increased integrin α_v PM expression in patients with high-grade prostate cancer and mCRPC is commensurate with enhanced MGAT5-catalyzed glycosylation. Analogously, patients with high alcohol consumption had significantly higher immunoreactivity for both MGAT5 and PM-specific integrin α_v. Downregulation of MGAT5 and recovery of MGAT3 localization to the Golgi following HCQ treatment caused a reduction of PM-specific integrin α_v in the PC-3 and DU145 cellular models and a substantial decline in FN adhesion. The in vivo data indicate that this expression is attributed to a decrease in glycosylation of α_v integrins via MGAT5 and restoration of MGAT3 glycosylation.

For the first time, we observed that inhibition of autophagy converts fragmented Golgi phenotype to compact. This was accompanied by a reduction of Golgi size, suggesting that Golgi fragmentation in prostate tumor cells is also associated with a significant increase in its volume, probably driven by the intensification of Golgi membrane biogenesis. However, these nascent membranes are unable to fuse due to the ER stress and impaired dimerization of several critical golgins, such as Giantin (4).

So, how does HCQ restore Golgi morphology exactly? Currently, there is no consensus regarding the precise involvement of autophagy in carcinogenesis since its activation and inhibition have each been observed in different cancers (reviewed in ref. 50). However, emerging evidence indicates the critical role of autophagy during metastasis in mCRPC (reviewed in ref. 23). In this study, we observed the close association between Golgi membranes and autophagosomes, which was mitigated by HCQ treatment. Fragmented Golgi membranes serve as a steady-state source of phagophores, so by inhibiting the fusion of AV with EE or lysosomes, HCQ prevents Golgi disorganization (Fig. 7C).

In addition, it is known that Golgi luminal pH in normal cells is slightly acidic (6.0–6.5); however, moderate alkalinization of Golgi has been shown in different cancers and is associated with the fragmentation of Golgi (51, 52). Thus, it is unlikely that chloroquine and HCQ significantly de-acidify the lumen of fragmented Golgi membranes in cancer cells and impair the function of Golgi. In fact, the restoration of Golgi by HCQ not only brings MGAT3 back to the Golgi but also rescues its function because the number of α_v integrins colocalizing with PHA-E lectin was significantly increased.

We showed that in advanced prostate cancer cells, steady-state activation of the ATF6 branch of UPR is critical for the maintenance of dispersed Golgi. Indeed, depletion of ATF6 restores dimerization of Giantin, recovers compact Golgi, and reduces integrin α_v in the PM fraction in both in vitro and in vivo experiments. The effect of ATF6 KD on Golgi is not mediated by autophagy inhibition, as we could not detect changes in LC3BII expression in PC-3 cells lacking ATF6. However, dimerization of Giantin requires PDIA3 chaperone delivered to the nascent Golgi membranes via COPII vesicles (53). At this point, we speculate that ATF6 may transcriptionally downregulate critical proteins required for the formation of the COPII complex. However, this hypothesis requires further investigation.

Importantly, we observed a synergistic effect of HCQ and ATF6 depletion on autophagy inhibition and a more prominent effect on Golgi morphology and expression of integrin α_v at the cell surface (Fig. 7C). Finally, in the mice orthotopic tumor model, the combination of ATF6 KD with HCQ had the most noticeable impact on MGAT5-mediated glycosylation and halted metastasis. It is essential to note here that we could not find any strong correlation between the orthotopic tumor size and the presence or absence of metastases. In some cases, large tumor volume was not associated with any sign of dissemination; in others, small tumors were found to be metastasized. Thus, the lack of metastases in the ATF6 KD-HCQ group cannot be ascribed to the smallest tumor size. Also, in clinical settings, small prostate tumors appear to be more aggressive than large (54).

Intriguingly, we found that in primary prostate tumors with grades 4 to 5, the expression of α_v integrins at PM was higher than that in patients with grades 2 to 3. Also, these groups demonstrate a similar expression of Gal-3 at the cell surface. Nevertheless, colocalization of integrins and Gal-3 at PM was significantly lower in grades 4 to 5 tumors than that with grades 2 to 3. This implies that in high-grade prostate cancer, some fraction of integrin molecules can be expressed at PM independently of Gal-3. Also, in patients with mCRPC, the PM expression of integrin α_v and Gal-3 was substantially higher in samples from bone metastases compared with those from tissue metastases. However, the association of integrin α_v and Gal-3 was lower in bone metastases samples than in visceral. These data suggest that the expression of integrins during the growth and dissemination of prostate tumors may also involve Gal-3-independent mechanisms, for instance, the delivery of integrins to the cell surface via an alternative, noncanonical trafficking. Recently, we reported that alcohol-induced ER stress and Golgi dispersal stimulate the trafficking of one of the hepatic proteins, the asialoglycoprotein receptor, to the cell surface via ER–PM junctions, bypassing canonical (ER→Golgi→PM) anterograde transportation (55). Our preliminary data indicate that in prostate cancer cells, some pool of α_v integrins that bear high-mannose glycans can be detected at the cell surface. Thus, we cannot exclude the noncanonical trafficking of integrins and the contribution of their high-mannose glycans to the metastatic potential of cancer cells.

It was suggested that galectins are non-glycoproteins, which can be exported to the cell surface and excreted by a nonclassical pathway, bypassing the classical biosynthetic ER→Golgi→PM pathway (45). Indeed, our multiple attempts to check the sensitivity of Gal-3 to different glycosidases confirmed that Gal-3 does not bear any N- or O-glycans. Paradoxically, integrin α_v–Gal-3 spots were observed in the medial and terminal Golgi compartments, indicating Gal-3 is integrated into canonical trafficking. IHC and W-B data revealed an interesting bidirectional relationship between integrin α_v and Gal-3. Gal-3 depletion did not affect the total level of integrin α_v but reduced PM content via deposition in EE and ER, substantiating Gal-3 is involved in α_v integrins trafficking, but it is not required for their internalization.

Our unexpected EM finding of the association of integrin α_v and Gal-3 in the Golgi membranes of control PC-3 cells raises the possibility that a complex between these proteins can be formed en route to the cell surface (right after MGAT5-mediated glycosylation of integrin in Golgi) rather than at the cell surface as was suggested previously (11, 56). Importantly, we could not find merged integrin α_v–Gal-3 immunogold spots in Golgi after HCQ treatment, which may be caused by the restoration of MGAT3-mediated integrin α_v glycosylation.

The results of integrin α_v depletion were surprising to us. Although cells lacking integrin α_v showed no difference in the total level of Gal-3, they demonstrated an increase of Gal-3 at the PM. Thus, without integrin α_v, Gal-3 is not internalized and, instead, aggregates at the cell surface. Intriguingly, the expression of Gal-3 in androgen-responsive LNCaP and 22Rv1 cells is negligible, but they still express integrin α_v. However, we would like to stress again that in LNCaP and 22Rv1 cells, Golgi is intact, contrary to PC-3 and DU145 cells. Therefore, the chance of MGAT5-mediated glycosylation of integrins and their potential need for Gal-3 is low. Thus, it appears that the function of Gal-3 and trafficking to the cell surface are directly determined by Golgi disorganization and MGAT5-modified integrins. Also, we cannot rule out that Gal-3 binding to LacNAc-carrying integrin α_v is required not only for its retention at PM but also for the transition from the inactive to the active conformation and subsequent binding to ECM ligands, such as FN (22, 57). These possibilities also await further verification.

Our data allow us to disentangle the effect of HCQ on integrin expression at the PM. On the one hand, Golgi recovery reduces MGAT5-mediated glycosylation, thus blocking the formation of integrin clusters with Gal-3. On the other hand, restoration of MGAT3 in Golgi and subsequent glycosylation of integrins by this enzyme reduce their association with Gal-3, which leads to abrogation of Gal-3 trafficking to the cell surface. Altogether, this results in instability of the integrin α_v–Gal-3 complex and subsequent acceleration of their internalization (Fig. 7D). Thus, we believe that HCQ-induced lack of Gal-3 at the PM is a direct consequence of Golgi reorganization, and it is a major contributing factor leading to a reduction of integrins at the PM.

In sum, we envisage that the results from this study can help to establish several new combinational therapeutic strategies. First, HCQ may be used in parallel with ceapins, a class of small molecules that selectively block the activity of ATF6 without effecting the other UPR pathways (58). Another selective ATF6 inhibitor, Melatonin (59), has shown therapeutic benefits in combination with conventional prostate cancer therapy (reviewed in ref. 60). In addition, given that the major integrin-binding ligand, FN, is heavily glycosylated in Golgi (61–63), it would be interesting to check whether fibroblasts in advanced prostate cancer cells exhibit fragmented Golgi phenotype and whether they employ the same alteration of FN glycosylation as described here for integrins. Then, the critical question to address in the future is whether HCQ and ATF6 ablation can reduce the production of FN by cancer-associated fibroblasts. These approaches may also impact ECM and other microenvironmental factors that mediate prostate cancer progression.

C.A. LaGrange reports personal fees from Manzanita outside the submitted work. No disclosures were reported by the other authors.

A.J. Macke: Conceptualization, data curation, validation, investigation, writing–original draft, writing–review and editing. A.N. Pachikov: Data curation, investigation, writing–original draft. T.E. Divita: Formal analysis, investigation, visualization. M.E. Morris: Investigation. C.A. LaGrange: Resources, supervision, writing–review and editing. M.S. Holzapfel: Visualization. A.V. Kubyshkin: Investigation. E.Y. Zyablitskaya: Investigation. T.P. Makalish: Investigation. S.N. Eremenko: Investigation. H. Qiu: Methodology. J.-J.M. Riethoven: Methodology. G.P. Hemstreet: Investigation. A. Petrosyan: Conceptualization, resources, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing–review and editing.

A. Petrosyan was supported by NIH grant NIAAA R01AA027242. J.M. Riethoven was supported by NIGMS 1P20GM113126. J.M. Riethoven was funded by the Russian Federation State program FZEG-2020–0060. We thank the UNMC Advanced Microscopy Core Facility and Tissue Sciences Facility for help with imaging and sample preparation. Portions of Fig. 7 were created using BioRender.com; subscription provided by the UNMC Center for Heart and Vascular Research.

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