Noncanonical SQSTM1/p62-Nrf2 pathway activation mediates proteasome inhibitor resistance in multiple myeloma cells via redox, metabolic and translational reprogramming

Multiple Myeloma (MM) is a B-cell malignancy characterized by the accumulation of clonal plasma cells in the bone marrow, with drug resistance being a major cause of therapeutic failure. We established a carfilzomib-resistant derivative of the LP-1 MM cell line (LP-1/Cfz) and found that the transcription factor NF-E2 p45-related factor 2 (Nrf2; gene symbol NFE2L2) contributes to carfilzomib resistance. The mechanism of Nrf2 activation involved enhanced translation of Nrf2 as well as its positive regulator, the autophagy receptor sequestosome 1 (SQSTM1)/p62. The eukaryotic translation initiation factor gene EIF4E3 was among the Nrf2 target genes upregulated in LP-1/Cfz cells, suggesting existence of a positive feedback loop. In line with this, we found that siRNA knockdown of eIF4E3 decreased Nrf2 protein levels. On the other hand, elevated SQSTM1/p62 levels were due at least in part to activation of the PERK-eIF2α pathway. LP-1/Cfz cells had decreased levels of reactive oxygen species as well as elevated levels of fatty acid oxidation and prosurvival autophagy. Genetic and pharmacologic inhibition of the Nrf2-EIF4E3 axis or the PERK-eIF2α pathway, disruption of redox homeostasis or inhibition of fatty acid oxidation or autophagy conferred sensitivity to carfilzomib. Our findings were supported by clinical data where increased EIF4E3 expression was predictive of Nrf2 target gene upregulation in a subgroup of patients with chemoresistant minimal residual disease and relapsed/refractory MM. Thus, our data offer a preclinical rationale for including inhibitors of the SQSTM1/p62-Nrf2 pathway to the treatment regimens for certain advanced stage MM patients.


INTRODUCTION
Multiple myeloma (MM) is a B cell malignancy characterized by the clonal expansion of neoplastic plasma cells in the bone marrow. The introduction of novel agents such as proteasome inhibitors and immunomodulatory drugs has prolonged the lives of patients with MM over the past decade [1]. The first-in-class proteasome inhibitor bortezomib provided proof-of-principle that proteasome inhibition is an important therapeutic target in MM, leading to its FDA approval as frontline therapy in June 2008 [2]. Despite overall response rates of 100% in the setting of newly diagnosed disease [3], most MM patients ultimately relapse because the MM cells develop resistance to the treatment [4,5]. Carfilzomib, a second-generation proteasome inhibitor that exhibits enhanced selectivity for the proteasome [6], received full approval from the FDA for the treatment of patients with relapsed/refractory MM in January 2016 [7]. However, the overall response rate to carfilzomib in the pivotal phase 2 clinical trial was less than 25% [8]. Furthermore, in a subsequent phase 3 study -in which overall survival was the primary endpoint [9] -carfilzomib monotherapy failed to significantly improve the survival of relapsed/refractory MM patients compared to those who received best supportive care (10.2 months vs 10.0 months) (ClinicalTrials.gov Identifier: Priority Research Paper www.impactjournals.com/oncotarget NCT01302392). These results indicate that the majority of MM cells that became resistant to bortezomib were also resistant to carfilzomib. Clearly, to extend the life expectancy of patients with this disease, it is essential to characterize the mechanisms conferring resistance to proteasome inhibitors.
To begin to understand the underlying processes that might be relevant to clinical carfilzomib resistance in MM, we previously established carfilzomib-resistant derivatives of MM cell lines, KMS-11/Cfz and KMS-34/Cfz [10,11]. In both cases, prosurvival autophagy was shown to contribute to carfilzomib resistance mediated, in part, via transcriptional upregulation of the SQSTM1 gene encoding sequestosome 1/p62 (SQSTM1/p62) [11]. SQSTM1/p62 is a multifunctional scaffold protein that interacts with various signaling molecules and serves as a ubiquitinbinding cargo receptor connecting the proteasomal and autophagic protein degradation pathways [12]. Another important function of SQSTM1/p62 is activation of the transcription factor nuclear factor-erythroid 2 (NF-E2)related factor 2 (Nrf2; gene symbol NFE2L2) in the Keap1-Nrf2 signaling pathway [13,14]. The Keap1-Nrf2 pathway maintains cellular redox homeostasis by inducing antioxidant and detoxification genes and by modulating energy metabolism [15]. A growing body of evidence implicates activation of the Keap1-Nrf2 pathway as a contributor to therapy resistance [16] but, to our knowledge, its role in conferring resistance to carfilzomib in MM cells has not been described.
The Keap1-Nrf2 pathway is tightly regulated at multiple levels, the precise details of which are currently being elucidated [17]. Under basal conditions, Nrf2 monomers are sequestered in cytoplasmic complexes through two-site binding to Keap1 homodimers which target Nrf2 for continuous ubiquitination and proteasomal degradation [18]. In response to oxidative stress, the Keap1 molecules are modified such that ubiquitination of Nrf2 is inhibited and it remains bound to the complex [19,20]. In a variation on this theme, overexpressed SQSTM1/p62 binds to and competitively prevents Keap1 from interacting with one of the Nrf2 sites, effecting a conformational change in the complex which leads to impaired ubiquitination of associated Nrf2 [13,14]. In both scenarios, newly synthesized Nrf2 escapes cytoplasmic sequestration by Keap1 and is free to translocate to the nucleus and regulate gene expression. Of relevance in this regard, a number of reports have indicated that Nrf2 expression is subject to diverse mechanisms of translational control [21][22][23][24][25].
Nrf2 activity can also be modulated by other mechanisms [26][27][28][29]. For example, nuclear translocation and activation of Nrf2 is enhanced by phosphorylation by the PERK protein kinase [26,27]. PERK is activated upon accumulation of misfolded/unfolded proteins in the endoplasmic reticulum which results in the induction of an "unfolded protein response" (UPR) [30]. The best characterized function of PERK on UPR induction is to provide a protective advantage to the cell by attenuating global protein translation via inhibitory phosphorylation of eukaryotic translation initiation factor-2α (eIF2α) [31,32]. Phosphorylation of eIF2α also results in the preferential translation of certain mRNAs containing upstream open reading frames (uORFs), the prototypical example of which is activating transcription factor 4 (ATF4) [33]. During this phase of the UPR, ATF4 and Nrf2 coregulate transcription of some cytoprotective genes [34,35]. However, if proteostasis is not restored, ATF4 induces a cell death program involving the CCAAT/enhancerbinding protein homologous protein (CHOP) transcription factor [36].
Here we report the establishment of a new carfilzomib-resistant derivative of the LP-1 MM cell line, LP-1/Cfz, in which carfilzomib resistance was characterized by induction of prosurvival autophagy as well as Nrf2 pathway activation associated with elevated SQSTM1/p62 levels. Unlike carfilzomib-resistant KMS-11/Cfz and KMS-34/Cfz cells, increased SQSTM1/p62 levels were not due to transcriptional upregulation of the SQSTM1 gene [11]. Rather, the higher levels of SQSTM1/ p62 were due to increased translation dependent in part on activation of the PERK-eIF2α axis. Comparative analysis with KMS-11/Cfz cells revealed Nrf2 target gene induction as well but only LP-1/Cfz cells were sensitized to carfilzomib by inhibition of the PERK-eIF2α signaling cascade. Additionally, LP-1/Cfz cells exhibited increased Nrf2 synthesis associated with elevated expression of Nrf2 targets involved in translation initiation, in particular, EIF4E3 encoding an atypical eukaryotic translation initiation factor family member recently demonstrated to mediate context-specific translation in diffuse large B-cell lymphoma [37,38]. Moreover, gene set enrichment analysis (GSEA) of gene expression profiling data from MM patient samples showed that increased EIF4E3 expression was predictive of Nrf2 activation in some chemoresistant minimal residual disease and relapsed/ refractory MM cases. These findings have elucidated further complexities of the proteostasis network in MM cells and provide preclinical rationale for therapeutic development of SQSTM1/p62-Nrf2 inhibitors as a means to overcome proteasome inhibitor resistance in a subgroup of advanced stage MM patients.

GSEA identifies Nrf2 pathway activation in carfilzomib-resistant MM cell lines
The carfilzomib-resistant LP-1/Cfz cell line was established by continuous culture of the LP-1 MM cell line [39] in stepwise increasing concentrations of the www.impactjournals.com/oncotarget drug (4 nM to 12 nM) over an 18 week period according to a previously published protocol used to derive the carfilzomib-resistant MM cell lines, KMS-11/Cfz and KMS-34/Cfz [11] ( Figure S1). As observed for KMS-11/Cfz and KMS-34/Cfz cells, LP-1/Cfz cells retained resistance to carfilzomib even when tested after removal of selective pressure for approximately 8 weeks. Gene expression profiling was performed on LP-1/Cfz and parental LP-1 cells, and GSEA was used to identify differentially overrepresented pathways and processes associated with carfilzomib resistance shared between LP-1/Cfz cells and the KMS-11/Cfz and KMS-34/Cfz cell lines [11].
We first applied GSEA to examine gene sets from the C3:TFT (transcription factor targets) subcollection of the Molecular Signatures Database (MSigDB). Binding site motifs for the "cap 'n' collar" transcription factors NF-E2 (V$NFE2_01) and Nrf2 (V$NRF2_Q4) that recognize a similar AP1-like core consensus sequence (TGA(G/C/T)TCA) [40] were significantly enriched in genes with increased expression in LP-1/Cfz and KMS-11/Cfz cells ( Figure S2). Although NF-E2 expression is restricted to erythroid cells, Nrf2 is broadly expressed and activated in response to stress [41,42]. In line with Nrf2 activation in both carfilzomib-resistant MM models, significant enrichment of Nrf2 target genes (NFE2L2.V2 gene set) was observed when GSEA was applied to the C6 (oncogenic signatures) collection of MSigDB ( Figure 1).
GSEA detects coordinated expression changes in groups of genes by taking all genes into account, even if individual representatives of the groups exhibit modest changes [43]. In a second, complementary approach, we employed fold change (FC) cutoffs to compare similar numbers of differentially expressed genes in LP-1/Cfz (FC ≥ 1.7; 896 probe sets) and KMS-11/Cfz cells (FC ≥ 1.4; 887 probe sets). Transcription factor binding site *Identified by oPOSSUM-3 Single Site Analysis using JASPAR CORE vertebrate profiles (minimum information content = 8 bits; matrix score threshold: 85%) and ranked by Fisher score. Search regions: 5,000 bp upstream and 5,000 bp downstream of transcription start sites. The NFE2L2 binding site motif was identified in both LP-1/Cfz and KMS-11/Cfz upregulated gene sets. Fisher scores reflect the number of genes containing the predicted TFBS.
motif discovery was carried out using the oPOSSUM-3 Single Site Analysis tool and the JASPAR CORE vertebrate database [44]. NFE2L2 was the highestscoring motif in KMS-11/Cfz cells and the only motif in common among the top ten ranked transcription factor binding sites overrepresented in the promoter regions of the upregulated genes in both carfilzomib-resistant MM models (Table 1). This result was notable because an Nrf2 ChIP-seq dataset was one of the data sources used to validate the performance of the oPOSSUM-3 program [41,44]. The Nrf2 ChIP-seq dataset comprised basal (Nrf2 -/-) and inducible (Keap1 -/-) direct binding targets of Nrf2 in mouse embryo fibroblasts [41]. We combined this dataset with Nrf2 direct binding targets identified by ChiP-seq experiments in human lymphoblastoid cells after treatment with the dietary isothiocyanate, sulforaphane [40]. Using GeneSpring analysis software, we found significant enrichment of these Nrf2 direct binding targets in the differentially expressed genes in LP-1/Cfz cells (102 out of 896 probe sets; P < 10 -31 ) and KMS-11/Cfz cells (107 out of 887 probe sets; P < 10 -31 ) (see Table S1 for the probe set lists). Moreover, there was also significant overlap of the differentially expressed genes in the carfilzomibresistant MM cells and genes that were downregulated in human lymphoblastoid cells after sulforaphane treatment -LP-1/Cfz cells (72 out of 896 probe sets; P < 10 -20 ) and KMS-11/Cfz cells (68 out of 887 probe sets; P < 10 -20 )which may represent indirect Nrf2 targets (see Table S2 for the probe set lists). Strikingly, different Nrf2 target genes were upregulated in LP-1/Cfz and KMS-11/Cfz cells. Prototypical cytoprotective Nrf2 targets, as exemplified by NQO1 [15], were significantly upregulated in KMS11/ Cfz cells [11] but not in LP-1/Cfz cells ( Figure 1; Table  S1). These findings suggested differential levels of Nrf2 activation and/or that the Nrf2-interacting partners might be different in the two carfilzomib-resistant MM model systems. In support of the latter possibility, GSEA indicated activation of a prosurvival ATF4 target gene response in LP-1/Cfz cells ( Figure S3A, S3B). GSEA also showed enrichment of HER2/ERBB2-related signatures in LP-1/Cfz cells when the C2:CP (canonical pathways) subcollection of MSigDB was queried ( Figure  S3C, S3D). Both factors were previously reported to cooperate with Nrf2 and modulate specificity of Nrf2 target gene activation [35,36,45,46]. In particular, a number of the Nrf2 targets upregulated in LP-1/Cfz cells (indicated in Figure 1A) were previously shown to be directly coregulated by ATF4 [34,47]. Regardless of the different transcriptional outcomes, taken together, the data implicated activation of Nrf2 pathways in the acquisition of carfilzomib resistance in both MM models. In the following experiments, we concentrated mainly on elucidating the molecular mechanisms and potential functional significance of Nrf2 pathway activation in LP-1/Cfz cells.

The autophagy-related gene GABARAPL1 is an Nrf2 target upregulated in carfilzomib-resistant MM cells
We verified the microarray data for selected Nrf2 targets in LP-1/Cfz cells by real-time reverse transcription polymerase chain reaction (qRT-PCR) assay and western blot analysis ( Figure S4; see also Figure 2C and Figure  9C). We noted that several of the novel Nrf2 targets are involved in translational control mechanisms; for example, EEF1A2, EIF4E3, RND3/RhoE (Table S1A) and FAM129A/Niban (Table S2A). EEF1A2 is a translation elongation factor gene that was previously shown to promote survival of mouse plasmacytoma cells [48], whereas EIF4E3, RND3/RhoE and FAM129A/Niban encode proteins that participate in various facets of capdependent translation initiation (described in more detail below) [38,49,50].
Of the validated genes, we focused initially on the autophagy-related gene GABARAPL1 [51] since it was also significantly upregulated in KMS-11/Cfz cells (Table S1B). Indeed, inspection of the enriched V$NFE2_01 transcription factor motif gene set indicated that GABARAPL1 was the top-ranked upregulated Nrf2 target in common between LP-1/Cfz and KMS-11/ Cfz cells (Figure 2A). In this regard, it was noteworthy that the KEGG autophagy gene set was among the significantly enriched gene sets in the MSigDB C2:CP (canonical pathways) collection identified by combined pairwise comparison of LP-1/Cfz and KMS-11/Cfz versus parental LP-1 and KMS-11 cells, and GABARAPL1 was the top-ranked gene in this instance as well ( Figure 2B). To investigate whether GABARAPL1 is an Nrf2-binding target in LP-1/Cfz and KMS-11/Cfz cells, we performed ChIP-qPCR analysis on the promoter region of the GABARAPL1 gene.
An evolutionarily conserved Nrf2 motif was identified in the GABARAPL1 promoter region by the ConTra v2 transcription factor binding site motif discovery program using position weight matrices from both the JASPAR CORE and TRANSFAC database libraries [52] ( Figure S5A). Of note, this motif coincided with an NF-E2 ChIP-seq binding site identified in K562 erythroid cells by the ENCODE project ( Figure S5B). Using primers flanking this site, we found specific enrichment of Nrf2-precipitated chromatin in comparison to the unrelated antibody control precipitation in both MM cell lines, increased binding of which was observed in the carfilzomib-resistant derivatives ( Figure 2D). This result is congruent with previous findings of substantial overlap of functional Nrf2 and NF-E2 binding sites at the genomewide and individual gene levels [40,42]. Thus, the data support the notion that GABARAPL1 is a direct Nrf2 target gene in LP-1/Cfz and KMS-11/Cfz cells.
GABARAPL1 is involved in selective autophagy through its interaction with ubiquitin-binding cargo  Table S1A for expression changes determined from the microarray data. D. Increased binding of Nrf2 to the GABARAPL1 promoter region indicated in Figure S5 in LP-1/ Cfz and KMS-11/Cfz cells as determined by ChIP-qPCR. E. Fluorescence histograms of LP-1/Cfz and parental LP-1 cells stained with the Cyto-ID autophagy detection reagent. F. Cells were treated with the indicated concentrations of carfilzomib for 72 hours in the absence or presence of chloroquine (10 μM) and cell viability was determined by alamarBlue assay. *, P < 0.001 vs carfilzomib alone (n = 3). receptors such as SQSTM1/p62 [53] and it is essential during the late stages of autophagosome maturation [54]. Previously, we showed that increased autophagic flux contributes to carfilzomib resistance in KMS-11/Cfz cells [11]. Increased staining intensity with the Cyto-ID autophagy detection reagent revealed that LP-1/Cfz cells also had higher steady state levels of autophagosomes ( Figure 2E), indicating that activation of the Nrf2-GABARAPL1 axis correlates with increased autophagic activity [55]. We also previously demonstrated that inhibition of autophagy by chloroquine treatment, which increases lysosomal pH and blocks autophagosomelysosome fusion, sensitized KMS-11/Cfz cells to carfilzomib [11]. Cotreatment with chloroquine likewise diminished carfilzomib resistance in LP-1/Cfz cells ( Figure 2F), indicating that prosurvival autophagy contributes to acquired carfilzomib resistance in this MM cell line similarly to KMS-11/Cfz cells.

LP-1/Cfz cells exhibit increased antioxidant capacity due to altered intermediary metabolism
As mentioned above, Nrf2 targets directly involved in reactive oxygen species (ROS) detoxification were not significantly upregulated in LP-1/Cfz cells by comparison to parental LP-1 cells. However, examination of the microarray data revealed that a number of these prototypical Nrf2 target genes (including NQO1, GCLC, GCLM, GPX4, GSR, GSTM1) are already highly expressed in LP-1 cells. NQO1 ((NAD(P)H:quinone oxidoreductase 1), GSR (glutathione reductase) and other antioxidantassociated enzymes regulated by Nrf2 require NADPH as a reducing cofactor [15]. Accordingly, Nrf2 also regulates several NADPH-generating enzymes. Specifically, Nrf2 facilitates NADPH production by directing carbon flux through the pentose phosphate pathway [56]. Consistent with the hypothesis that Nrf2 is activated in LP-1/Cfz as well as in KMS-11/Cfz cells, NADPH levels were increased concomitant with upregulation of KEGG pentose phosphate pathway genes in both models ( Figure  S6).
It is becoming increasingly appreciated that Nrf2 also affects multiple aspects of intermediary metabolism indirectly [15], including mitochondrial fatty acid oxidation (FAO) [57]. Of note, during the metabolic shift towards FAO, activation of autophagy hydrolyzes lipids into fatty acids for fuel [58]. Considering that FAO is an important source of NADPH in leukemic cells [59], we examined whether rates of FAO were increased in carfilzomib-resistant LP-1/Cfz and KMS-11/Cfz cells. Increased FAO was observed in LP-1/Cfz versus parental LP-1 cells ( Figure 3A, 3B). By comparison, basal rates of FAO were already elevated in parental KMS-11 cells and there was no further increase in KMS-11/Cfz cells ( Figure  3C, 3D).
Considered together, the gene expression profiling and functional assays supporting increased NADPH production and/or upregulation of genes that detoxify ROS suggested that LP-1/Cfz and KMS-11/Cfz cells would exhibit enhanced Nrf2-mediated antioxidant capacity. To examine this, we measured ROS generation using the fluorescent redox-sensitive dyes CM-H 2 DCFDA, which is specific for hydrogen peroxide, and MitoSOX Red, which is selective for superoxide [60]. Cells were treated with varying concentrations of carfilzomib and ROS levels were quantified by flow cytometry. Lower levels of ROS were generated in LP-1 cells compared to KMS-11 cells under all conditions; at the highest carfilzomib concentrations, both hydrogen peroxide ( Figure 4) and superoxide ( Figure 5) levels were reduced by more than 70% in the carfilzomib-resistant derivatives versus their parental counterparts.
To determine whether antioxidant defense conferred protection to carfilzomib in these MM models and whether there was a contribution of FAO, the cells were cotreated with carfilzomib and either (S)-4-carboxyphenylglycine, an inhibitor of xCT (a subunit of the x c cystine antiporter involved in glutathione homeostasis encoded by the SLC7A11 gene; see Figure 1A) [34] or etomoxir, an inhibitor of carnitine palmitoyltransferase-1 (the rate limiting enzyme in FAO; see Figure 3) [61]. Cotreatment with either inhibitor resulted in enhanced sensitivity of the resistant cells to carfilzomib (Figure 6), indicating that the glutathione-based antioxidant system contributes to carfilzomib resistance in these MM models and that FAO is a potential source of reducing equivalents underlying the increased ROS defense.
Finally, to directly test whether Nrf2 activity contributes to carfilzomib resistance, we transfected LP-1/Cfz and parental LP-1 cells with two previously validated Nrf2 siRNAs and a control siRNA with no known mammalian homology [62,63]. We confirmed that Nrf2 mRNA was knocked down ~45-65% by qRT-PCR. This was accompanied by an ~30-45% decrease in GABARAPL1 mRNA levels ( Figure 7A). As illustrated in Figure 7B, both Nrf2 siRNAs sensitized LP-1/Cfz cells to carfilzomib.
Kampmann and colleagues recently reported that knockdown of the 19S proteasome subunit genes desensitized U266 MM cells to carfilzomib [65]. Because the 19S regulator delivers substrates to the 20S catalytic core, they hypothesized that a loss in 19S function may lead to the selective accumulation of certain proteins. SQSTM1/p62 was one of the proteins that accumulated upon knockdown [65]. Congruent with their results, GSEA indicated that many of the proteasome 19S subunit genes were downregulated in LP-1/Cfz cells ( Figure  S7). Therefore, it is possible that this mechanism may contribute to increased SQSTM1/p62 levels in LP-1/ Cfz cells. However, as described below, we found that elevated protein synthesis was a contributing factor to the observed differences in SQSTM1/p62 levels. In this context, it is worth mentioning that knockdown of the 19S proteasome subunit genes in U266 MM cells was not associated with substantially increased 20S chymotrypsinlike protease activity selectively inhibited by carfilzomib [65]. Similarly, we did not detect a significant increase in 20S chymotrypsin-like protease activity in LP-1/Cfz cells (data not shown), which was consistent with only a slight (~1.3-fold) increase in mRNA of the PSMB5-encoded constitutive β5 subunit targeted by carfilzomib ( Figure  S7).
Interestingly, we did not observe enrichment of IRE1 or ATF6 expression signatures corresponding to the other branches of the UPR in LP-1/Cfz cells. In fact, GSEA indicated that UPR signaling was attenuated (HALLMARK_UNFOLDED_PROTEIN_RESPONSE downregulated; Figure S8C). Therefore, the mechanistic basis for PERK activation in LP-1/Cfz cells is unknown. However, alternative signals for PERK activation have recently been uncovered [28,29,66,70]. In particular, Gupta and colleagues described a noncanonical mechanism of PERK-Nrf2 activation in the absence of an endoplasmic reticulum stress response that results from an epithelial-to-mesenchymal transition (EMT) [28,70]. Along these lines, enrichment of an EMT-like expression signature (HALLMARK_EPITHELIAL_ MESENCHYMAL_TRANSITION) was indicated by GSEA in LP-1/Cfz cells ( Figure S8D), and decreased cell surface expression of E-cadherin compared to parental LP-1 cells ( Figure S8E) is consistent with this as a contributory mechanism [71,72].

Induction of Nrf2 target genes in LP-1/Cfz cells creates a positive feedback loop promoting eIF4E3-driven Nrf2 translation
Although Nrf2 levels are primarily governed by interaction with Keap1, accumulating evidence has highlighted the importance of de novo translation [22][23][24]. Therefore, we were prompted to compare Nrf2 synthesis in LP-1/Cfz and parental LP-1 cells. For these analyses, the cells were pretreated with MG-132 to prevent Nrf2 proteasomal degradation. Under these conditions, we found significantly higher levels of Nrf2 in LP-1/Cfz cells ( Figure 9A). We next tested the effects of 4EGI-1, an inhibitor of cap-dependent translation that prevents binding of eukaryotic initiation factor 4G (eIF4G) to eIF4E family members (the best characterized of which is eIF4E1) within the eIF4F translation initiation complex [73]. Inhibition of the eIF4E/eIF4G interaction eliminated the differences in Nrf2 levels between LP-1/Cfz and parental LP-1 cells ( Figure 9B). The data thus indicated more active cap-dependent synthesis of Nrf2 in LP-1/Cfz cells. This result was contrary to the GSEA prediction that canonical eIF4F-mediated translation initiation was downregulated in LP-1/Cfz compared to parental LP-1 cells ( Figure S8A) and suggested the possible involvement of an alternative eIF4F cap-binding complex containing another member of the eIF4E family [74]. Along these lines, it has recently been appreciated that eIF4E3 interacts with eIF4G and competes with eIF4E1 to form a novel eIF4F cap-binding complex [37,38]. Because this Nrf2 target was upregulated at the mRNA level in LP-1/Cfz cells (Table S1A), we were interested in whether this was reflected at the level of eIF4E3 protein. We found that the increased EIF4E3 mRNA levels in LP-1/Cfz cells were accompanied by elevated eIF4E3 protein levels, whereas LP-1/Cfz and parental LP-1 cells had similar levels of eIF4E1 ( Figure 9C).
Phosphorylation of eIF4E1 by mitogen-activated protein kinase interacting kinases (MNK) 1 and 2 stimulates translation of a subset of mRNAs encoding proteins involved in cancer development and progression [75]. Gartenhaus and colleagues found that treatment of diffuse large B-cell lymphoma cells with an MNK inhibitor, CGP57380, reduced eIF4E1-driven translation and resulted in a compensatory increase in eIF4E3-driven translation [38]. To determine whether Nrf2 translation might be regulated via eIF4E1-eIF4E3 interplay, we investigated the effect of MNK inhibition on Nrf2 protein levels in LP-1/Cfz and parental LP-1 cells. We found that after CGP57380 treatment Nrf2 levels increased in both cases, with the levels in parental LP-1 cells approaching those in LP-1/Cfz cells ( Figure 9B). These findings supported a role of eIF4E3 in the elevated synthesis of Nrf2 and the carfilzomib-resistant phenotype of LP-1/ Cfz cells. To directly test this, we transfected LP-1/Cfz and parental LP-1 cells with two specific EIF4E3 siRNAs and a negative control siRNA. Figure 9D shows that Nrf2 protein levels decreased with eIF4E3 protein knockdown. Moreover, eIF4E3 knockdown sensitized LP-1/Cfz cells to carfilzomib ( Figure 9E).
We also examined whether the protein levels of two other novel Nrf2 targets involved in the control of cap-dependent translation initiation -RND3/RhoE and FAM129A/Niban -corresponded with the differential mRNA levels observed in LP-1/Cfz and LP-1 cells ( Figure  S4). FAM129A inhibits PERK-mediated phosphorylation of eIF2α [50]. Additionally, FAM129A and Rnd3 reciprocally modulate the activity of 4E-BP1, an inhibitor of eIF4E1 [49,50]. Briefly, the dephosphorylated form of 4E-BP1 binds to eIF4E1, preventing binding of eIF4G and formation of the eIF4F complex. mTORC1 phosphorylates 4E-BP1, leading to its dissociation from eIF4E1 which allows eIF4F formation and translation initiation [75]. FAM129A positively affects mTORC1 phosphorylation of 4E-BP1, thereby facilitating eIF4E1-driven translation [50]. In contrast, Rnd3 inhibits mTORC1 phosphorylation of 4E-BP1 which prevents 4E-BP1 disassociation from eIF4E1 and therefore inhibits eIF4E1 function [49]. This mechanism only pertains to eIF4E1 since 4E-BP1 does not bind to eIF4E3 [74]. FAM129A protein levels were decreased in LP-1/Cfz cells corresponding with its diminished mRNA expression. Conversely, Rnd3 protein levels were increased in LP-1/Cfz cells concomitant with increased mRNA expression ( Figure 9C). Collectively, these results are consistent with a potential positive feedback process associated with acquisition of carfilzomib resistance in LP-1/Cfz cells wherein initial activation of Nrf2 leads to direct and indirect regulation of target genes -including transcriptional upregulation of EIF4E3 -that facilitate enhanced eIF4E3-driven translation of Nrf2.

Increased EIF4E3 expression predicts Nrf2 target gene activation in minimal residual disease and relapsed MM patient samples
To assess the clinical relevance of our findings, we analyzed three independent publicly available gene expression datasets of MM patients. We first examined the gene expression profiles of chemoresistant minimal residual disease (MRD) and matched diagnostic samples from MM patients included in the GEM2010MAS65 clinical trial (GEO accession number GSE70399) [76]. Notably, EIF4E3 expression was found to be increased in 5 out of 7 MRD samples (FC = 1.25; P = 0.013). When GSEA was applied to these 5 cases, significant upregulation of Nrf2 target genes (NFE2L2.V2 gene set) was observed in the persisting MRD cells ( Figure 10A). As was found for LP-1/Cfz cells, UPR signaling was attenuated (HALLMARK_UNFOLDED_PROTEIN_ RESPONSE downregulated; Figure 10B). Moreover, GSEA predicted an inverse relationship with an enrichment pattern corresponding to genes regulated by EIF4E1 overexpression (Figure 10C).
We also examined patient-paired relapse and diagnostic samples from 17 MM patients treated with various regimens (GEO accession number GSE36824) [77]. EIF4E3 expression was increased in 4 out of 17 cases of disease progression. GSEA indicated that Nrf2 target genes (NFE2L2.V2 gene set) were upregulated during disease course in these 4 cases ( Figure 11A). As for MRD cells with increased EIF4E3 expression, GSEA predicted lack of UPR stress in the relapsed MM cells (HALLMARK_UNFOLDED_PROTEIN_ RESPONSE downregulated; Figure 11B). These relapsed MM cells also had enrichment of an EMT-like expression signature (HALLMARK_EPITHELIAL_MESENCHYMAL_ TRANSITION; Figure 11C). Moreover, genes upregulated through activation of the mTORC1 complex were downregulated in these 4 cases (HALLMARK_ MTORC1_SIGNALING; Figure 11D). In addition, GSEA again predicted an inverse correlation with an EIF4E1specific gene expression signature ( Figure 11E). Of the 17 samples, GABARAPL1 expression was increased in 5 cases during disease progression, two of which also had increased EIF4E3 expression. Performing GSEA on the combined EIF4E3 plus GABARAPL1 dataset (7 out of 17 cases) indicated activation of Nrf2 signaling and yielded the same enrichment patterns for the other pathways identified for the 4 cases with increased EIF4E3 expression (data not shown).
On the other hand, when GSEA was performed on the 10 relapsed MM patient samples that did not exhibit increased EIF4E3 or GABARAPL1 expression, the exact opposite patterns of enrichment were obtained. In these cases, a positive correlation with EIF4E1-upregulated genes was observed during disease progression ( Figure  11F) concomitant with a prediction of increased mTORC1 signaling ( Figure S9A). Although a UPR was predicted ( Figure S9B), Nrf2 target genes were downregulated ( Figure S9C) and a reverse EMT-like expression signature (i.e., mesenchymal-to-epithelial transition) was suggested ( Figure S9D). Therefore, these parameters appeared to separate the relapsed MM samples into two distinct groups: one, "LP-1/Cfz-like", characterized by increased EIF4E3 (and/or GABARAPL1) expression having activation of the Nrf2 signaling pathway, and a second group ("eIF4E1-like") exhibiting the opposite phenotype and pathway activation states in which enhanced eIF4E1driven translation was implied. ncbi.nlm.nih.gov/geo/; GEO accession number GSE70399) for matched MRD and diagnostic plasma cells from MM patients enrolled in the GEM2010MAS65 trial (ClinicalTrials.gov Identifier: NCT01237249). EIF4E3 expression in MRD cells versus baseline was analyzed with the GEO2R web tool (http://www.ncbi.nlm.nih.gov/geo/geo2r/) and was found to be increased in 5 out of 7 samples (average fold change = 1.25; P = 0.013). A. GSEA indicated that Nrf2 target genes were upregulated in MRD cells. Gene set: NFE2L2.V2 (M2870). B. GSEA indicated that genes upregulated during the unfolded protein response were downregulated in MRD cells. Gene set: HALLMARK_ UNFOLDED_PROTEIN_RESPONSE (M5922). C. GSEA indicated that genes downregulated in primary human mammary epithelial cells upon overexpression of EIF4E1 were upregulated in MRD cells. Gene set: EIF4E_DN (M2790). www.impactjournals.com/oncotarget Elevated expression of EIF4E3 is prognostic of poor survival LP-1 MM cells contain the t(4;14) chromosomal translocation that activates a histone methyltransferase encoded by the WHSC1 gene (also known as MMSET) [78], which is consistently associated with poor outcome [79]. Using the UAMS-70 prognostic gene signature, Shaughnessy and colleagues separated WHSC1-positive MM patients into higher-risk and lower-risk subgroups [80]. We were curious to know whether increased expression of EIF4E3 was prognostic for decreased overall survival of WHSC1-positive MM patients. Using the PROGgeneV2 prognostic biomarker identification tool [81] as previously described [11], we performed a Cox proportional hazards regression analysis on 546 newly diagnosed MM patients treated on the UARK 98-026 TT2 and UARK 2003-033 TT3 clinical trials [80]. The hazard ratio (HR) for patients with high versus low WHSC1 expression was 1.3 (95% confidence interval (CI), 1.1-1.54; P < 0.002). As shown in Figure 12A, highlevel coexpression of EIF4E3 resulted in increased risk of mortality (HR = 1.6; 95% CI, 1.17-2.21; P < 0.004), which was higher than the increased risk score predicted by high-level coexpression of EIF4E1 (HR = 1.45; 95% CI, 1.04-2.02; P < 0.03) ( Figure 12B).

DISCUSSION
Despite improved outcomes with the introduction of novel agents such as proteasome inhibitors to MM treatment regimens over the past decade, patients inevitably succumb to their disease because the MM cells become resistant to the drugs [1]. MM cells can develop resistance to proteasome inhibitors through a variety of mechanisms [4]. However, it is still unclear which mechanisms explain clinical proteasome inhibitor resistance.
In this paper we report the establishment of a new carfilzomib-resistant MM cell line derivative, LP-1/Cfz, in which elevated levels of the SQSTM1/p62 ubiquitinbinding cargo receptor were associated with carfilzomib resistance that comprised both prosurvival autophagy and Nrf2 pathway activation. Our results support a model wherein activation of the SQSTM1/p62-Nrf2 pathway -in concert with the PERK-eIF2α/ATF4 axis -directs reprogramming of MM cells, modulating redox and energy homeostasis via elevated FAO and inducing prosurvival autophagy involving GABARAPL1 upregulation [15,[26][27][28][29][34][35][36]. We suggest that the increased SQSTM1/p62 levels allow newly synthesized Nrf2 to escape Keap1mediated sequestering in the cytoplasm [13,14], resulting in the regulation of a subset of its target genes, including EIF4E3, which in turn facilitate enhanced translation of Nrf2 and establish a positive feedback loop. While further mechanistic studies are required to fully define the role played by SQSTM1/p62 in carfilzomib resistance, we demonstrated the direct involvement of Nrf2 by genetic and pharmacologic inhibition and a contribution of PERK using two highly selective small-molecule inhibitors of the PERK-eIF2α axis [67,68].
Previously we reported that acquisition of carfilzomib resistance in KMS-11/Cfz and KMS-34/Cfz MM cells was associated with prosurvival autophagy involving SQSTM1/p62 [11]. Comparative analyses of KMS-11/Cfz cells herein also revealed activation of Nrf2 target genes ( Figure 1B). Carfilzomib resistance in KMS-11/Cfz cells is associated with KLF4-mediated transcriptional upregulation of SQSTM1 expression [11]. Others have described the establishment of a positive feedback loop in which Nrf2 transcriptionally activates SQSTM1 [13,14] (highlighted in Figure S1B). We note that KLF4 is also an Nrf2 target gene (Table S1B), so we envision a model in KMS-11/Cfz cells wherein KLF4 and Nrf2 cooperate to transcriptionally maintain elevated SQSTM1/p62 levels.
GSEA revealed that activation of the PERK-eIF2α axis in LP-1/Cfz cells was not associated with induction of a UPR ( Figure S8C). Gupta and colleagues have described a noncanonical mechanism of PERK-Nrf2 activation in the absence of an endoplasmic reticulum stress response that involves an EMT-like adaptation [28,70]. Enrichment of an EMT-like expression signature was predicted by GSEA for LP-1/Cfz cells ( Figure S8D) and for relapsed MM patient samples with increased EIF4E3 expression ( Figure  11C). Further investigation of an EMT-related mechanism is underway and our preliminary studies are suggestive of crosstalk among Nrf2, ERBB2 and hypoxia signaling pathways [82,83]. Indeed, our ongoing examination of LP-1/Cfz cells indicates that they exhibit intermediate EMT-like characteristics [84]. Of relevance in this regard, a recent study by Orlowski and colleagues identified low levels of TJP1 encoding the epithelial marker tight junction protein 1 (also known as zonula occludens 1) as a determinant of myeloma proteasome inhibitor resistance [85]. During EMT, intercellular junctions are disrupted and TJP1 is coordinately downregulated with E-cadherin [86]. Interestingly, TJP1 expression was decreased in KMS-11/Cfz and KMS-34/Cfz cells but increased in LP-1/Cfz cells (our unpublished results). While Nrf2 has been demonstrated to downregulate E-cadherin expression [72], it has been implicated in the upregulation of TJP1 expression [87]. Intriguingly, Orlowski and colleagues reported that proteasome inhibitor sensitivity associated with increased TJP1 expression was due to suppression of expression of the PSMB8 gene encoding the immunoproteasome β5i/LMP7 (chymotrypsin-like) subunit targeted by carfilzomib and bortezomib as well as the PSMB9 gene encoding the β1i/LMP2 (caspase-like) subunit that is also targeted by bortezomib [85]. In accord with their findings of an inverse relationship with TJP1 expression, both PSMB8 and PSMB9 were downregulated in LP-1/Cfz cells ( Figure S7). In their model, this apparently occurred through TJP1-mediated suppression of EGFR/ERBB2 signaling [85]. As TJP1 also interacts with GABARAPL1 through the CUL3-KBTBD6/ KBTBD7 ubiquitin ligase in the autophagy network [53], we speculate that complex interplay between Nrf2-GABARAPL1/TJP1 and ERBB2 signaling pathways may underlie the carfilzomib-resistant LP-1/Cfz phenotype ( Figure S3C, S3D) [45,46,88,89].
Carfilzomib resistance in the KMS-11/Cfz and KMS-34/Cfz models was characterized by a partial reversal of plasma cell maturation [11]. These results are in line with the findings of others showing that primary MM cells are capable of dedifferentiating into a less mature MM phenotype conferring experimental and clinical drug resistance [90][91][92]. It is notable that LP-1/Cfz cells did not fully recapitulate this pattern. Parental LP-1 cells exhibit a membrane phenotype that is intermediate between late B lymphocytes and plasma cells based on MHC class II antigen expression [39]. Like the KMS-11/ Cfz and KMS-34/Cfz carfilzomib-resistant MM models [11], acquisition of carfilzomib resistance in LP-1/ Cfz cells was accompanied by decreased expression of SLAMF7 encoding the plasma cell-specific CD319 cell surface marker, reflecting a partial reversal of plasma cell maturation (Table S1A). Conversely, MHC class II antigen expression was diminished, which is reminiscent of plasmacytic differentiation (Table S2A). In this latter respect, LP-1/Cfz cells might resemble in part the situation observed in mantle cell lymphoma where bortezomib resistance increased with plasmacytic differentiation [93]. Accordingly, LP-1/Cfz cells may have attained the ''sweet spot'' maturation stage intrinsically less susceptible to proteasome inhibition proposed by Tiedemann and colleagues [92].
Our results indicating a role of Nrf2 activation in experimental carfilzomib resistance parallel those of others wherein Nrf2 target gene activation was associated with poor responsiveness to bortezomib in mouse and human MM models [94,95]. In this regard, both LP-1/ Cfz and KMS-11/Cfz exhibit some cross-resistance to bortezomib ( Figure S1). In that various regimens were used in the treatment of the MM patients that we analyzed [76,77,80], we cannot conclude that increased EIF4E3 expression can be used as a specific indicator of clinical carfilzomib resistance. Nonetheless, the finding that EIF4E3 expression is increased in certain chemoresistant minimal residual disease and relapsed MM patient samples and is predictive of Nrf2 target gene activation, strongly suggests that the Nrf2-EIF4E3 axis and eIF4E3driven translation contributes to MM drug resistance mechanisms in the clinical setting. On the other hand, lack of EIF4E3 expression in relapsed samples was able to distinguish a second group of MM patient samples exhibiting the opposite phenotype and pathway activation states in which enhanced eIF4E1-driven translation was indicated. These findings are consistent with the results of other studies demonstrating a role of eIF4E1 in MM biology and proteasome inhibitor resistance [65,[96][97][98]. It is particularly noteworthy therefore -especially in view of the documented oncogenic activity of EIF4E1 [75] that WHSC1-expressing MM patients with high EIF4E3 expression had less favorable outcomes than those with high EIF4E1 expression ( Figure 12).
In summary, our finding of noncanonical SQSTM1/ p62-Nrf2 pathway activation adds to the growing appreciation from our work and others that carfilzomib resistance in MM can arise via multiple mechanisms [10,11,65,85,99]. MM is heterogeneous in its etiology and progression so these varied results are not entirely unexpected [100]. Collectively, our studies suggest several approaches to sensitize drug-resistant MM cells to carfilzomib [10,11]. In particular, the data presented herein support the development of novel therapies targeting the SQSTM1/p62-Nrf2 pathway for a subgroup of advanced stage MM patients with eIF4E3-driven translation. www.impactjournals.com/oncotarget

Chromatin immunoprecipitation (ChIP)
ChIP was performed on the GABARAPL1 promoter region with 20 µg total chromatin and 5 µg of anti-Nrf2 antibody using the SimpleChIP Enzymatic Chromatin IP Kit (Magnetic Beads) (Cell Signaling, Cat. No. 9003) as previously described [11], and 4% of the precipitated material was used per qPCR reaction. Background ChIP levels were obtained using 5 µg of anti-Keap1 antibody. Primers used were: GABARAPL1 promoter region, forward, CCGTGTCCTTCATCTGACTCC, reverse, TCGCTCCTGAACAGCAACAT. ; Thermo Fisher Scientific Silencer Select) or a negative control siRNA (QIAGEN, ALLStars negative control siRNA, Cat. No. 1027281) using the HiPerFect transfection reagent (QIAGEN). Briefly, 2 x 10 6 cells per ml were seeded into 24 well plates in 100 μl aliquots. Each well received a mixture of siRNAs (750 ng) and HiPerFect reagent (6 μl) in 100 μl serum-free culture medium preincubated for 15 minutes. After 5 hours, the cells were diluted to 6 x 10 5 per ml in complete medium. Cells (3 x 10 5 per ml) were seeded into 96 well plates and treated with a range of carfilzomib concentrations. After 48 hours, cell growth was measured and cells (3 x 10 5 ) were lysed for total RNA isolation and qRT-PCR analysis of Nrf2 and GABARAPL1 mRNA levels or western blot analysis of eIF4E3 and Nrf2 protein levels as indicated. The data were normalized to ACTB mRNA levels. www.impactjournals.com/oncotarget

Cytotoxicity assay
Cells were treated with carfilzomib and agents at the indicated concentrations and cell growth was measured using the alamarBlue cell viability and proliferation reagent (Thermo Fisher Scientific) as previously described [10,11].

Measurement of NADPH
NADPH was measured using an NADP/NADPH Quantitation Kit (Sigma-Aldrich, Cat. No. MAK038) according to the manufacturer's instructions [102].

Measurement of FAO
Oxygen consumption rate was measured using the XF24 Extracellular Flux Analyzer (Seahorse Bioscience) as previously described [103]. FAO was determined by XF Cell Mito Stress Test (Cat. No. 103010-100) using the XF Palmitate-BSA FAO substrate (Cat. No. 102720-100) in the absence or presence of etomoxir according to the manufacturer's instructions.

Cellular fractionation
Nuclear and cytoplasmic extracts were prepared for western blot analysis essentially as described [64].

Confocal microscopy
Immunofluorescence confocal microscopy was performed as previously described [11]. In brief, cells (2.5 x 10 5 ) were centrifuged onto a microscope slide at 1,000 rpm for 5 minutes using a Shandon Cytospin 4 instrument. The cells were then immediately fixed in 3.7% formaldehyde for 5 minutes at room temperature and permeabilized with 0.5% Triton X-100 in phosphatebuffered saline (PBS) for 15 minutes at room temperature. Following permeabilization, the cells were rinsed with PBS and blocked in PBS containing 10% goat serum and 0.01% Triton X-100 for 1 hour at room temperature. The cells were then incubated with anti-Nrf2 antibody diluted to a final concentration of 0.4 µg/ml, in PBS containing 1% goat and 0.01% Triton X-100 for 1 hour at room temperature. The cells were rinsed with PBS and then incubated with Alexa Fluor 488-conjugated goat antirabbit secondary antibody (Thermo Fisher Scientific, Cat. No. R37116) diluted 1:500 in PBS containing 1% goat serum and 0.01% Triton X-100 for 1 hour at room temperature. The cells were rinsed with PBS and mounted with Fluoromount G (Electron Microscopy Sciences). Imaging analysis was performed on a Cell Observer SD spinning disk confocal system equipped with Zen software (Carl Zeiss Microscopy).

Proteasome activity
Proteasome chymotrypsin-like activity was measured by cleavage of a specific luminogenic proteasome substrate (succinyl-leucine-leucine-valinetyrosine-aminoluciferin) using the Proteasome-Glo Chymotrypsin-Like Assay according to the manufacturer's instructions (Promega Corporation, Cat. No. G8660). Luminescence intensities were quantified with a Gemini XPS microplate spectrofluorometer equipped with SoftMax Pro software (Molecular Devices Corp.).

ACKNOWLEDGMENTS
We thank Leif Bergsagel for providing the LP-1 and KMS-11 cell lines. This work was supported by a Grant from the Dr. Cyrus and Myrtle Katzen Cancer Research Center at The George Washington University, a King Fahd Endowment from The George Washington University School of Medicine and Health Sciences, and a generous donation from Marc Cohen.