Ubiquitin Ligase MARCH5 Regulates Apoptosis through Mediation of Stress-Induced and NOXA-Dependent MCL1 Degradation

MCL1 has critical antiapoptotic functions and its levels are tightly regulated by ubiquitylation and degradation, but mechanisms that drive this degradation, particularly in solid tumors, remain to be established. We show here in prostate cancer cells that increased NOXA, mediated by activation of an integrated stress response, drives the degradation of MCL1, and identify the mitochondria-associated ubiquitin ligase MARCH5 as the primary mediator of this NOXA-dependent MCL1 degradation. Therapies that enhance MARCH5-mediated MCL1 degradation markedly enhance apoptosis in response to a BH3 mimetic agent targeting BCLXL, which may provide for a broadly effective therapy in solid tumors. Conversely, increased MCL1 in response to MARCH5 loss does not sensitize to BH3 mimetic drugs targeting MCL1, but instead also sensitizes to BCLXL inhibition, revealing a codependence between MARCH5 and MCL1 that may also be exploited in tumors with MARCH5 genomic loss.


31
Androgen deprivation therapy to suppress activity of the androgen receptor (AR) is the standard 32 treatment for metastatic prostate cancer (PCa), but tumors invariably recur (castration-resistant 33 prostate cancer, CRPC). The majority will initially respond to agents that further suppress AR, 34 but most men relapse within 1-2 years and these relapses appear to be driven by multiple AR 35 dependent and independent mechanisms (1,2), which may include increased expression of anti-36 apoptotic proteins. The anti-apoptotic BCL2 family proteins (including BCL2, BCLXL, and 37 MCL1) act by neutralizing BAX and BAK, and by inhibiting the BH3-only pro-apoptotic 38 proteins that can activate BAX/BAK (primarily BIM) (3). These interactions are mediated by the 39 BH3 domain, and BH3-mimetic drugs can enhance apoptosis by mimicking the activity of BH3-40 only pro-apoptotic proteins and thereby antagonizing the anti-apoptotic BCL2 family proteins 41 (4,5).  and ABT-263 (navitoclax, orally bioavailable analogue of ABT-737) (7) are NOXA upregulation is mediated by the integrated stress response 115 To determine how erlotinib was increasing NOXA transcription we first focused on p53, as 116 NOXA is a major transcriptional target of p53. However, treatment with erlotinib did not cause 117 any change in p53 expression (Figure 2A; Figure S1A), indicating a p53 independent mechanism 118 for increasing NOXA mRNA. The alternative p53-independent pathway that may increase 119 NOXA transcription is the integrated stress response (ISR), which can be triggered by factors 120 including hypoxia, glucose or amino acid depletion, genotoxic stress, and the endoplasmic 121 reticulum stress/unfolded protein response (40). These stresses activate kinases including PERK 122 (in response to endoplasmic reticulum stress), GCN2 (in response to amino acid starvation), and 123 PKR (in response dsRNA and additional cellular stresses), which converge on phosphorylation 124 of eIF2α (30,(41)(42)(43). Consistent with ISR activation, we found that erlotinib rapidly (within 30 125 minutes) increased phosphorylation of eIF2α ( Figure 2B). The phosphorylation of eIF2α causes 126 an increase in translation of the transcription factor ATF4, which can then stimulate the 127 expression of multiple genes to either resolve the cellular stress or drive to apoptosis. Indeed, 128 eIF2α phosphorylation in response to erlotinib was associated with an increase in ATF4 protein 129 ( Figure 2B). 130 With respect to NOXA, ATF4 can directly or as a heterodimer with ATF3 stimulate 131 expression of NOXA (41,43), although this is generally observed after prolonged stress. 132 Nonetheless, an increase in NOXA protein was observed after 60 -90 minutes of erlotinib 133 treatment, and this rapid time course coincided with the increase in ATF4 and decrease in MCL1 134 ( Figure 2B). Moreover, treatment with an ISR inhibitor (ISRIB), which suppresses the effects of 6 eIF2α phosphorylation (44), decreased basal ATF4 and suppressed the erlotinib-mediated 136 increase in ATF4 and NOXA, providing further evidence for this pathway ( Figure 2C). 137 Consistent with these findings, ISRIB suppressed the erlotinib-mediated increase in NOXA 138 mRNA ( Figure 2D), while MCL1 mRNA was unaffected by these treatments ( Figure 2E). 139 Together, these findings indicate that activation of the ISR by erlotinib drives the rapid induction 140 of NOXA, which then promotes MCL1 degradation. 141 142 MARCH5 mediates kinase inhibitor/NOXA-dependent MCL1 degradation 143 We next sought to identify E3 ligases that contribute to kinase inhibitor-mediated and NOXA-144 dependent MCL1 degradation. MCL1 is a substrate for the ubiquitin ligase HUWE1 (MULE), 145 and HUWE1 has been reported to mediate MCL1 degradation by NOXA (29-31). However, we 146 reported previously that while HUWE1 depletion could increase basal MCL1 levels, it did not 147 prevent the increased degradation of MCL1 in response to kinase inhibitors (22). Figure 3A   148 shows that HUWE1 depletion does not affect the erlotinib-mediated increase in NOXA, and that 149 it does not prevent the subsequent decrease in MCL1. 150 NEDD8 conjugation is essential for cullin-dependent E3 ligases to ubiquitylate their 151 substrates. To determine the role of cullin-dependent E3 ligases in MCL1 degradation in 152 response to tyrosine kinase inhibition, we examined whether NEDD8 inhibition could prevent 153 the effect of tyrosine kinase inhibitors. Treatment with NEDD8 inhibitor MLN4924 increased 154 p27 (a known target of cullin-dependent E3 ligase CUL4), but did not increase MCL1 or block 155 the effects of erlotinib ( Figure 3B). Indeed, MLN4924 moderately decreased MCL1 protein, 156 which may be due to an increase in NOXA, whose degradation is mediated by a cullin-dependent E3 ligase (45). MLN4924 similarly failed to prevent the decrease in MCL1 in response to 158 lapatinib (EGFR/ERBB2 inhibitor) ( Figure 3C), indicating that a cullin-independent mechanism 159 is driving the increased MCL1 degradation. 160 We then hypothesized that a cullin-independent E3 ligase that localizes to mitochondria, 161 where MCL1 is mainly located, may promote MCL1 degradation in response to tyrosine kinase 162 inhibition. To assess this hypothesis, we first examined the well-known mitochondria-associated 163 cullin-independent E3 ligase PARKIN, which has been implicated as a ubiquitin ligase for 164 MCL1 (46). However, while PARKIN depletion increased its target protein p62, it did not 165 increase MCL1 or block the effect of erlotinib ( Figure 3D). MARCH5 is another mitochondria-7 associated cullin-independent E3 ligase that has been implicated as a regulator of MCL1 (37,47). 167 Significantly, depleting MARCH5 with siRNA increased basal expression of MCL1 and a 168 known MARCH5 substrate, MiD49, in LNCaP cells ( Figure 3E and Figure S2A). MARCH5 169 depletion did not increase MCL1 mRNA ( Figure S2B), further supporting a posttranscriptional 170 mechanism for increasing MCL1. MARCH5 depletion also increased basal MCL1 in PC3 PCa 171 cells ( Figure 3F) and in additional prostate, breast and lung cancer cell lines ( Figure S2C-H). 172 These results show that MARCH5 is a major mediator of basal MCL1 degradation in epithelial 173 cancer cell lines. transcription. To confirm these findings, we then used CRISPR/CAS9 to delete MARCH5. 180 Consistent with the RNAi results, there was a marked increase of MCL1 expression, as well of 181 the MARCH5 substrate MiD49, in each of three MARCH5 depleted lines ( Figure 3H). 182 Moreover, erlotinib no longer decreased MCL1 in these MARCH5 depleted lines ( Figure 3H). 183 As expected, transient overexpression of exogenous MARCH5 decreased MCL1 in control and 184 MARCH5 depleted cells ( Figure 3I). 185 Interestingly, and consistent with a previous report (47), MARCH5 depletion by CRISPR or 186 siRNA also increased NOXA protein ( Figure 3H and Figure S3A, respectively). MARCH5 187 depletion did not increase, but instead decreased NOXA mRNA ( Figure S3B), indicating this 188 increase in NOXA protein is through a post-transcriptional mechanism. One plausible 189 mechanism is through increased binding to MCL1, as a previous study found that MCL1 could 190 protect NOXA from proteasome-mediated degradation (48). Consistent with this mechanism, the 191 increased levels of NOXA and of BIM in MARCH5 depleted cells coincided with increased 192 binding of these proteins to MCL1 ( Figure 3J). To further assess this mechanism, we treated  199 We also examined the effects on NOXA and BIM of depleting or overexpressing MCL1. 200 Cells with CRISPR-mediated MCL1 depletion had markedly reduced NOXA and BIM, 201 providing further evidence that MCL1 protects both from degradation ( Figure 3L). Conversely, 202 NOXA and BIM were increased in cells that overexpress ectopic MCL1 ( Figure 3M) The above findings indicated that increased NOXA in response to erlotinib was driving the 211 MARCH5-mediated ubiquitylation and degradation of MCL1. Consistent with this conclusion, 212 we found by coimmunoprecipitation that erlotinib treatment, in combination with proteasome 213 inhibition, enhanced the interaction between MARCH5 and MCL1 ( Figure 4A). Importantly, 214 phosphorylation of BIM and NOXA can modulate their interaction with MCL1, suggesting that 215 kinase inhibitors may further be enhancing MCL1 ubiquitylation and degradation through effects 216 on phosphorylation of BIM, NOXA, or MCL1 that modulate NOXA/BIM-MCL1 interactions 217 (49-51). We have previously found that erlotinib did not alter MCL1 phosphorylation at sites that 218 have been shown to enhance its ubiquitylation and degradation (22). To further assess the role of 219 phosphorylation in erlotinib-mediated MCL1 degradation, we used phospho-tag gels and 220 examined the phosphorylation state of these proteins. Erlotinib treatment did not have any clear 221 effects on the phosphorylation of MCL1, BIM, or NOXA in cells cultured in complete medium 222 (FBS) or cultured in medium with charcoal stripped serum (CSS) to deplete steroids ( Figure 4B). 223 Similarly, erlotinib did not alter phosphorylation in MARCH5 depleted cells. As a positive 224 control, EGF stimulation dramatically increased BIM phosphorylation. 225 In parallel with the above experiments, we asked directly whether erlotinib enhances 226 MCL1 interaction with NOXA versus BIM. This was assessed in MARCH5 knockout cells to 227 avoid effects due to increased MCL1 interaction with MARCH5 by erlotinib. Erlotinib treatment 9 did not clearly enhance MCL1 binding of NOXA versus BIM ( Figure 4C). As expected, 229 treatment with S63845 decreased both NOXA and BIM binding to MCL1. 230 We next asked whether there were alterations in MARCH5 expression or activity that may be 231 enhancing its ubiquitylation of MCL1. We first examined effects of erlotinib versus MARCH5 232 depletion on MARCH5 substrates. Treatment with erlotinib again increased NOXA and 233 decreased MCL1, but did not decrease other MARCH5 substrates (MiD49, MFN1, and 234 FUNDC1) ( Figure 4D). Interestingly, while MiD49 was increased in the MARCH5 knockout 235 cells (see also Figure 3E and 3H), MFN1 and FUNDC1 were not altered, indicating that these 236 latter substrates are not undergoing MARCH5-mediated degradation under basal conditions. In 237 any case, this result indicates that erlotinib is not generally enhancing MARCH5 activity. 238 We then asked whether erlotinib alters the mitochondrial localization of MARCH5, or of 239 MCL1. Consistent with previous reports, cellular fractionation showed that MARCH5 was 240 primarily located to mitochondria ( Figure 4E). Treatment with erlotinib for 2 hours (prior to a 241 substantial decrease in MCL1) did not change this localization of MARCH5. Moreover, it did 242 not increase the mitochondrial localization of MCL1, BIM or NOXA, indicating that erlotinib-243 mediated MCL1 degradation is not through increased targeting of these latter proteins to 244 mitochondria. Finally, MARCH5 depletion did not clearly alter the fraction of MCL1 associated 245 with mitochondria. 246 As MARCH5 may be activated by mitochondrial stress, we also asked whether tyrosine 247 kinase inhibition had acute effects on mitochondria that may alter MARCH5 function. To 248 address this we examined mitochondrial respiration in response to erlotinib or lapatinib in 249 LNCaP-derived C4-2 cells, which were more suitable for these studies as they had stronger phosphorylation. The precise basis for this metabolic adaptation, and whether it is linked to 258 activation of a stress response, is not clear. In any case, these findings indicate that MARCH5 is 10 not altered in response to erlotinib, and that its increased degradation of MCL1 is driven 260 primarily by the increase in NOXA.

262
Mitochondria-targeted agents can increase MCL1 degradation by MARCH5-dependent 263 mechanism 264 MARCH5 regulates mitochondrial fission and fusion in response to mitochondrial stress (33-37), 265 suggesting that agents that alter mitochondria functions may enhance MARCH5-mediated 266 degradation of MCL1 by a distinct mechanism. To assess this hypothesis, we examined the 267 effects of a series of mitochondria-targeted agents on MCL1. Actinonin is an inhibitor of the 268 human mitochondrial peptide deformylase that blocks mitochondrial protein translation (52). 269 Four-hour treatment with actinonin decreased MCL1 in LNCaP cells ( Figure 5A). However, it 270 also increased NOXA, suggesting that it may be acting similarly to tyrosine kinase inhibitors 271 through an ISR, rather than by directly through MARCH5. Gamitrinib-TPP is a mitochondrial 272 HSP90 inhibitor and can induce MCL1 degradation in glioblastoma cells (53). Consistent with 273 previous data (54,55), gamitrinib-TPP rapidly decreased MCL1 in LNCaP cells, and this was 274 also associated with an increase in NOXA ( Figure 5B). The pyruvate dehydrogenase/α-275 ketoglutarate dehydrogenase inhibitor CPI-613 is another clinically promising agent that targets 276 mitochondria (56). Similar to actinonin and gamitrinib-TPP, treatment with CPI-613 decreased 277 MCL1 and also increased NOXA ( Figure 5C). 278 Significantly, each of these mitochondria-targeted agents increased ATF4 ( Figure 5C), 279 indicating an ISR mechanism for increasing NOXA. Consistent with this finding, and with a 280 previous report on gamitrinib-TPP (54), treatment with ISRIB impaired the upregulation of 281 ATF4 and NOXA, and the reduction of MCL1, by each of these mitochondria-targeted agents 282 ( Figure 5C). Moreover, depleting NOXA with siRNA prevented the decrease in MCL1 in 283 response to each of these agents ( Figure 5D). Together these findings indicated that the increased 284 MCL1 degradation in response to these agents was being driven by increased NOXA 285 downstream of an ISR. 286 As further evidence for this conclusion, we found that the decrease in MCL1 by these 287 mitochondria-targeted agents was proteasome-dependent, and was not associated with an 288 increase in p53 ( Figure 5E, F). Finally, we used a caspase inhibitor to confirm that these 289 mitochondrial-targeted agents were not increasing MCL1 degradation through release and 290 11 activation of caspases, which can degrade MCL1 ( Figure 5F, Figure S4A). As a positive control 291 for caspase inhibition, we showed that Z-DEVD-FMK could prevent caspase cleavage in 292 response to erlotinib in combination with ABT-737 ( Figure 5F, Figure S4A). 293 We next used siRNA to determine whether MCL1 degradation in response to these 294 mitochondrial-targeted agents was mediated by MARCH5. Depleting MARCH5 markedly 295 increased MCL1 and prevented the MCL1 loss in response to erlotinib and actinonin, although  Figure S5A). In contrast, 313 HUWE1 loss was very rare ( Figure 6C). Interestingly, assessing genomic alterations across 314 cancers, MARCH5 loss appears to be most common in PCa ( Figure 6D). Significantly, this may 315 reflect its genomic location adjacent to PTEN at 10q23, and hence co-deletion with PTEN. 316 Indeed, in the TCGA primary PCa dataset, all cases with deep deletion of MARCH5 also have 317 PTEN deletion ( Figure 6E). In contrast, MARCH5 deletion appears to be occurring independently 318 of PTEN loss in a subset of metastatic PCa.

319
MCL1 amplification and MARCH5 loss generally occur in distinct tumors, although their 320 mutual exclusivity is not statistically significant ( Figure 6F and Figure S5B, C). Relative to 12 MARCH5 and MCL1, oncogenic alterations in the genes encoding NOXA (PMAIP1) and BIM 322 (BCL2L11) are rare ( Figure 6F and Figure S5B, C). Finally, shallow deletions of MARCH5, 323 suggesting single copy losses, appear to be relatively common in PCa, with a higher frequency in 324 metastatic castration-resistant PCa versus primary PCa ( Figure 6G, H, and Figure S5D, E). 325 Together these results support a tumor suppressor function for MARCH5, which may be related 326 to its negative regulation of MCL1.

338
Identical results were obtained with a second MCL1 antagonist (AZD5991) ( Figure 7B). We also 339 examined cells stably overexpressing ectopic MCL1. These cells similarly had marked increases 340 in NOXA and BIM, which were decreased in response to 1 µM S63845 ( Figure 7C) or AZD5991 341 ( Figure 7D), but apoptotic responses again required high drug concentrations. 342 Although the MARCH5 depleted and MCL1 overexpressing cells showed increased 343 apoptosis in response to MCL1 antagonists, it was unclear why (if it was an on-target effect) it 344 should require substantially higher drug concentrations than those needed for release of BIM and 345 NOXA. One contributing factor may be that the BIM and NOXA that is displaced from MCL1 346 by S63845 and AZD5991 appears to undergo rapid degradation, as their levels in the treated 347 MARCH5 depleted or MCL1 overexpressing cells were not dramatically higher than in the 348 parental control cells ( Figure 7A-D). It is also possible that the high levels of NOXA and BIM in 349 the MARCH5 depleted cells and MCL1 overexpressing cells were effectively competing with 350 BAK for MCL1 binding, so that these cells are less dependent on MCL1 (and more dependent on 351 other anti-apoptotic BCL2 family proteins) to buffer BAK. However, arguing against this 13 mechanism, by coimmunoprecipitation we found that MCL1 was binding increased levels of 353 BAK, as well as NOXA and BIM, in the MARCH5 depleted cells and the MCL1 overexpressing 354 cells ( Figure 7E). Alternatively, as MCL1 has a preference for binding BAK versus BAX (57), it 355 is possible that the increased levels of MCL1 are adequate to neutralize BAK even at drug 356 concentrations up to 10 µM. In support of this latter mechanism, we found that BAK was not 357 increased in the MARCH5 knockout cells ( Figure 7F), which may allow the high levels of MCL1 358 to effectively buffer BAK despite treatment with S63845 or AZD5991. 359 In contrast to BAK, in the unactivated state BAX is localized primarily in the cytoplasm 360 and may be buffered mostly by BCLXL and BCL2. Significantly, BAX protein expression was 361 decreased in the MARCH5 knockout cells ( Figure 7F). The decrease in BAX after MARCH5 loss 362 (as well as the decrease in PUMA) suggested that the MARCH5 knockout cells may have 363 decreased capacity to buffer BAX and be very sensitive to acute increases in free BAX, and 364 hence be more dependent on BCL2 or BCLXL. Therefore, we assessed responses to the 365 BCL2/BCLXL antagonist ABT-263 (navitoclax). Significantly, ABT-263 treatment caused a 366 marked apoptotic response specifically in the MARCH5 knockout cells ( Figure 7G). As we 367 reported previously (22), ABT-263 could induce apoptosis in control parental cells in 368 combination with S63845, but the addition of S63845 only minimally enhanced apoptosis in the 369 ABT-263 treated MARCH5 knockout cells ( Figure 7H). The BCL2 specific antagonist  (venetoclax) was not effective, indicating that the efficacy of ABT-263 is due to BCLXL 371 inhibition ( Figure 7I).

372
Of note, a previous study similarly found that MARCH5 knockdown could increase 373 MCL1 and sensitize to BCLXL inhibition, and suggested that increased NOXA was suppressing 374 the antiapoptotic activity of MCL1 (47). While this increased NOXA may be a factor, our data 375 indicate that the increased MCL1 in MARCH5 knockdown cells is sequestering substantial levels 376 of both BAK and BIM (see Figure 7E). To explore other mechanisms, we examined the Avana 377 CRISPR screen dataset through the Broad DepMap site (https://depmap.org) to identify cell lines 378 that were dependent on MARCH5 and genes that have most similar patterns of dependency (58). 379 Interestingly, the gene that was most co-dependent with MARCH5 was MCL1 ( Figure 7J,K). 380 Conversely, the gene most co-dependent with MCL1 was MARCH5. This strong co-dependency 381 was also observed in screens with another CRISPR library ( Figure S6A, B                 (B and C) Top 5 genes correlated with MCL1 dependency score (B) or genes correlated with MARCH5 dependency score (C) in cancer cells from CRISPR-CAS9 screens using GeCKO libraries.
Correlation with MCL1 dependency score Correlation with MARCH5 dependency score r = 0.614 GeCKO CRISPR library