The Unfolded Protein Response: A Novel Therapeutic Target for Poor Prognostic BRAF Mutant Colorectal Cancer

Mutations in BRAF at position 600 from valine (V) to glutamic acid (E) occur in ∼10% of colorectal cancers. BRAF^V600E mutations, which account for more than 80% of all BRAF mutations (1), lead to sustained MAPK signaling and are associated with poor survival and drug resistance (2, 3). Trials investigating the effect of BRAF inhibition given as monotherapy have failed in BRAFMT colorectal cancers (3). The more recent trials with BRAF/EGFR double-therapy or BRAF/MEK/EGFR triple-therapy have shown some increased response rates but at the cost of increased toxicity (4, 5).

The endoplasmic reticulum (ER) is recognized as a major site for protein synthesis/folding (6). Increased protein synthesis/misfolding rates that exceed the capacity of protein chaperones contribute to the development of ER stress. ER stress is sensed by three upstream transmembrane proteins, Inositol-Requiring Protein-1 (IRE1), Protein kinase RNA-like ER Kinase (PERK), and Activating Transcription Factor-6 (ATF6). Under basal conditions, the ER chaperone GRP78 constitutively binds to the three sensors, thus preventing their activation. Upon induction of ER stress, sequestration of GRP78 by unfolded proteins results in activation of the ER-specific unfolded protein response (UPR), which attempts to reduce protein load on the ER and increases its folding capacity. PERK and IRE1α become activated following dimerization and auto-phosphorylation (7, 8). Dissociation of GRP78 from ATF6 results in translocation of the receptor to the Golgi where it is cleaved by SP1 and SP2 proteases into an active transcription factor (9). Unresolved ER stress can lead to apoptosis through upregulation of the pro-apoptotic transcription factor CHOP, resulting in downregulation of the anti-apoptotic protein BCL-2 and upregulation of pro-apoptotic proteins such as DR5 and the BH3-only proteins PUMA, NOXA, and BIM (10–12).

In this study, we have used publicly available microarray datasets from colorectal cancers patients to identify novel pathways associated with the BRAFMT subgroup with the poorest outcome. RNAi, cellular and mechanistic assays, and xenograft studies indicate that inducers of acute ER stress can be a novel treatment strategy for poor prognostic BRAFMT colorectal cancers.

Carfilzomib, ACY-1215 and trametinib were purchased from SelleckChem, AZD6244 (13) from AstraZeneca, and Z-VAD-FMK (14) from Calbiochem. HA15 was generated in house (15). HA15 was synthesized from commercial N-(4-(3-aminophenyl)thiazol-2-yl)acetamide and commercial dansyl chloride, according to the literature (WO2014/072486). siRNAs targeting caspase-8, caspase-9, HSPA5, and CHOP were purchased from Qiagen and Invitrogen, respectively. The ON-Targetplus siRNA library was obtained from Dharmacon. The BRAFV600E plasmid was a gift from Prof. Marais (London, UK; ref. 16).

Authentication and culture of LIM2405, HT-29, VACO432/VT1, RKO, COLO205, COLO320, and CACO-2 colorectal cancer cells have previously been described (17, 18). All cells were passaged for a maximum of 2 months, after which new seed stocks were thawed. LIM2405 cells were a gift from Dr. Whitehead (Vanderbilt University, Nashville, TN) in 2010 (19). VACO432, VT1, and RKO were provided by Prof. Vogelstein (Johns Hopkins University School of Medicine, Baltimore, MD) in 2012. HT-29 (2001), CACO-2 (2005), COLO205 (2012), and COLO320 (2012) cells were obtained from the ATCC (Authentication by short tandem repeat profiling/karyotyping/isoenzyme analysis). DiFi cells were obtained from Dr. Montagut (Barcelona, Spain; ref. 20).

Western blotting has previously been described (18). Anti-pEIF2α^S51, anti-EIF2α, anti-ATF4, anti-GRP78, anti-IRE1α, anti-acetylated-α-tubulin, anti-BCL2, anti-BCLXL, anti-PUMA, anti-BID, anti-BIM (Cell Signaling Technology), and anti-pIRE1α^S724 (Abcam) were used in conjunction with a HRP-conjugated anti-rabbit secondary antibody (Amersham). Anti-caspase-8 (12F5; Alexis), anti-CHOP (Cell Signaling Technology), anti-ATF6 (Abcam), anti-MCL1 (BD Pharmingen), and anti-NOXA (Abcam) mouse monoclonal antibodies were used in conjunction with a horseradish peroxidase–conjugated anti-mouse secondary antibody (Amersham).

Apoptosis was evaluated using propidium iodide (PI) staining to determine the percentage of cells with DNA content <2N (18). For annexin/PI analysis, cells were harvested and analyzed according to the manufacturer's instructions (BD Biosciences).

Caspase-8 or caspase-3/7-GLO reagents (25 μL; Promega) were incubated with 5 μg of protein lysate diluted in cell culture medium in a total volume of 50 μl for 45 minutes at room temperature. Luciferase activity was measured using a luminometer.

Cell viability was determined using 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT; ref. 18).

The PROTEOSTAT Aggresome Detection Kit (Enzo Life Sciences Ltd.) was used to detect aggregated protein cargo, according to the manufacturer's instructions. Images were obtained using the Leica SP5 confocal microscope.

Click-iT L-homopropargylglycine (HPG) Alexa Fluor 488 Protein Synthesis Kit (ThermoFisher Scientific) was used to label newly synthesized protein following the manufacturer's protocol. Detailed methodology can be found in the supplementary methods.

Q-PCR analysis was performed using the LightCycler 480 probes master mix (LightCycler 480II; Roche).

siRNA transfections were carried out using Hiperfect (Qiagen) as previously described (18).

Initial data analysis was carried out using the R Statistical Package (version 3.2.1). The data used in this study were obtained from 566 Affymetrix U133 Plus 2.0 colorectal cancer patient transcriptional profiles accessed through the NCBI GEO accession number GSE39582 (ref. 21; discovery data set) and from the 359 Almac stage II colorectal cancer DSA colorectal cancer patient transcriptional profiles, available in the ArrayExpress database (E-MTAB-863/E-MTAB-864; Supplementary Fig. S1). BRAFV600E mutational status within GSE39582 was assessed by allelic discrimination using TaqMan probes (21). Mutational data for BRAF was unavailable for the Almac dataset, and we used a highly specific classifier to produce a BRAFMT surrogate status (Supplementary Methods). Within the untreated stage II/III GSE39582 and the stage II E-MTAB-863/E-MTAB-864 datasets, 31 and 26 patients were identified as BRAFMT, respectively. A genelist was created of those genes that were differentially expressed between poor and good prognostic BRAFMT colorectal cancer subgroups, as described further in the Supplementary Methods.

This was carried out using ingenuity pathway analysis (IPA; Qiagen Bioinformatics), using the gene lists differentially expressed between poor and good prognostic BRAFMT colorectal cancer patients.

In vivo studies were conducted as previously described using 6- to 8-week-old, female BALB/c nude mice (Envigo; ref. 18). In the initial tolerability study, three healthy mice received ACY-1215 30 mg/kg/day i.p. (Day 1–Day 5 and Day 8–Day 12) alone or in combination with Carfilzomib 6 mg/kg i.p. (Day 1/Day 3/Day 5/Day 8/Day 10/Day 12). In the efficacy study, ACY-1215 30 mg/kg/day i.p. alone or with Carfilzomib 6 mg/kg 3/week i.p. was administered to BALB/c nude mice with VACO432 tumors. Each treatment group contained seven animals. Mice were sacrificed and tumors were excised on day 15. All animal experiments were carried out according to UKCCCR guidelines under licence PPL2704. In vivo experiments were carried out in accordance with the Animals (Scientific Procedures) Act, 1986, and approved by the Department of Health, Social Services and Public Safety, Northern Ireland.

Student t tests and two-way ANOVA were calculated using the GraphPad software (Prism5). Two-way ANOVA test was used to determine the significance of change in levels of apoptosis between different treatment groups. Significant changes had P values <0.05 (*, P < 0.05; **, P < 0.01; ***, P < 0.001; ns, not significant). The nature of the interaction between carfilzomib and ACY-1215 was determined using the method of Chou and Talalay (18).

To identify novel pathways associated with the BRAFMT colorectal cancer subgroup with the poorest outcome, we generated differentially expressed gene lists between poor and good prognostic stage II/III BRAFMT colorectal cancer tumors, using the publicly available transcriptionally profiled GSE39582 colorectal cancer dataset (1.3-fold cut-off; P < 0.05; Supplementary Fig. S1; ref. 21). To identify pathways that are deregulated in poor prognostic BRAFMT colorectal cancer, pathway analyses were carried out using both the significantly upregulated and downregulated genes in the poor prognostic BRAFMT subgroup compared to the good prognostic BRAFMT subgroup (Supplementary Tables S1 and S2). These results showed that cholesterol biosynthesis, the geranylgeranyldiphosphate biosynthesis and mevalonate pathway, G-protein–coupled receptor signaling and the UPR, were the top 5 pathways identified from the significantly upregulated gene list in poor prognostic BRAFMT colorectal cancer (Supplementary Table S1).

We also identified the UPR as a pathway associated with the BRAFMT subgroup with poor outcome, using a second untreated stage II colorectal cancer dataset (E-MTAB-863 and E-MTAB-864; Supplementary Fig. S1). In this study, high-risk patients were previously defined as those with metastatic cancer recurrence within 5 years of primary surgery (22). We generated differentially expressed gene lists between poor and good prognostic stage II BRAFMT colorectal cancer tumors (two-fold cut-off; P < 0.05; Supplementary Fig. S1). IPA analysis, using the significant differential gene list between poor and good prognostic stage II BRAFMT colorectal cancer, revealed that the UPR was significantly associated with the poor prognostic BRAFMT subgroup (Supplementary Table S3).

To identify key functional genes/targets for BRAFMT colorectal cancer, we used an RNAi screening approach targeting proteins that lie at nodal points in the top 20 cancer cell specific pathways identified from the significant upregulated gene list in poor prognostic stage II/III BRAFMT colorectal cancer (GSE39582; Supplementary Table S1). The effect of downregulating each of these proteins on cell viability was tested in the BRAFMT VACO432, RKO, LIM2405, and HT-29 colorectal cancer cells, using an ON-Targetplus siRNA library against 19 targets (Supplementary Table S4). Notably, only one of 19 siRNAs had a significant inhibitory effect on survival in all four BRAFMT cell line models, and this was HSPA5, the gene encoding the master regulator of UPR, GRP78 (Supplementary Table S4; Fig. 1A). This was confirmed using additional siRNA sequences against HSPA5 (Supplementary Fig. S2A). These results are the first to demonstrate the importance of GRP78 and the UPR as a potential novel target for BRAFMT colorectal cancer with poor clinical outcome.

Based on the results from the RNAi screen, we next assessed the involvement of GRP78 and the UPR in regulating survival of BRAFMT colorectal cancer cells. We used a small molecule inhibitor against GRP78, HA15 (15), and the BRAFMT VACO432 colorectal cancer cell line, and its isogenic VT1 clone with a disrupted BRAF^V600E allele (Fig. 1B; ref. 23). A previous study in melanoma has shown that HA15 induces dissociation of GRP78 from PERK, IRE1α, and ATF6 complexes. Treatment with HA15 for 24 hours resulted in phosphorylation of Serine 51 of EIF2α and increases in ATF4 and CHOP, and this was associated with apoptosis induction as indicated by PARP cleavage, caspase-9 cleavage, and increased caspase-3/7 activity in the BRAFMT VACO432 cell line but not in the WT VT1 clone (Fig. 1B; Supplementary Fig. S2B). Increased phosphorylation of IRE1α on Ser724 and ATF6 cleavage were also observed 48 hours following treatment with HA15 in the BRAFMT cells. Similar results were obtained in the BRAFMT HT-29 cell line, but not in the BRAFWT DiFi, CACO-2, and COLO320 colorectal cancer cell line models (Fig. 1C; Supplementary Fig. S2C and S2D). Treatment with HA15 also resulted in significant reduction in cell viability in the panel of BRAFMT colorectal cancer cells (Fig. 1C). As expected, treatment of BRAFMT cells with the pan-caspase inhibitor Z-VAD-FMK abolished the effect of HA15 on PARP cleavage; however importantly, caspase inhibition failed to prevent increased IRE1α phosphorylation and induction of ATF4 and CHOP, indicating that ER stress activation is upstream of apoptosis induction (Supplementary Fig. S2C). Collectively, these data indicate that GRP78 inhibition induces ER stress and apoptosis in BRAFMT colorectal cancer.

To define the relative importance of the extrinsic and intrinsic apoptotic pathways in mediating HA15-induced apoptosis, we used siRNA specifically directed against caspase-8 (extrinsic pathway) or caspase-9 (intrinsic pathway; Fig. 1B). Apoptotic cell death (PARP and caspase 3 cleavage) following HA15 treatment was completely abrogated following caspase-8 silencing. In addition, the increased caspase-9 p35/37 cleavage products (indicative of activation of the intrinsic cell death pathway) that were observed following HA15 treatment were also completely abrogated following caspase-8 silencing. In contrast, caspase-9 silencing resulted only in a partial rescue of the PARP cleavage observed following HA15 treatment. This suggests that the cell death induced by HA15 proceeds via a caspase-8-mediated activation of the caspase-9-dependent intrinsic apoptotic pathway (24). To investigate the mechanism of apoptosis further, we assessed expression levels of pro- and anti-apoptotic proteins following HA15 treatment in the BRAFMT VACO432 cell line and its isogenic VT1 clone (Fig. 1C). Notably, expression of the BH3-only proteins NOXA and PUMA, which control the intrinsic cell death pathway, and the death receptor DR5, which activates the extrinsic apoptotic pathway, were markedly upregulated, whereas expression of anti-apoptotic BCL-2 family members BCL-2 and MCL-1 were downregulated following treatment with HA15 in the BRAFMT VACO432 cell line (Fig. 1C). No changes in DR4 expression levels were observed following treatment with HA15 treatment (Supplementary Fig. S2E). Similar results were obtained in the BRAFMT HT-29 cell line. CHOP promotes both the transcription of DR5 (25) and PUMA (26) and the downregulation of BCL-2 expression (27) contributing to the induction of apoptosis. To determine whether the ER stress induced by HA15 is responsible for apoptosis, we used siRNA against DDIT3 (gene encoding CHOP), a gene that mediates ER-stress-induced apoptosis (Fig. 1D). Notably, CHOP silencing abrogated HA15-induced PUMA levels and downregulated MCL-1 levels, and decreased HA15-induced PARP and caspase-3 cleavage in BRAFMT colorectal cancer cells, indicating that the ER stress-mediated activation of CHOP is responsible for HA15-induced cell death. Silencing of CHOP also partially reversed the decreased cell viability following HA15 treatment in BRAFMT colorectal cancer cells (Fig. 1D).

MEK/ERK pathway activation has been shown to enhance protein translation/synthesis via phosphorylation of the translation initiator eIF4E, thereby exceeding ER protein folding capacity and resulting in ER stress (28). Consistent with this, we found marked increased basal levels of peIF4E^S209 (Fig. 2A, left) and nascent protein production (Fig. 2A, middle) in the BRAFMT VACO432 cell line with activated MEK1/2–ERK1/2 pathway, compared to its WT clone (VT1). In addition, constitutive levels of the ER stress proteins ATF4, CHOP, and the active (spliced) form of XBP1 (sXBP1), (indicators of activation of the PERK and IRE1α UPR branches) were also higher in the BRAFMT VACO432 cell line, compared to its WT clone (Fig. 2A, right). DDIT3 and ATF4 mRNA levels were also higher in the VACO432 cell line compared to the VT1 clone (Fig. 2A, right).

To further investigate the relative importance of the BRAF/MEK pathway in regulating sensitivity to the ER stress activator HA15, we used the MEK1/2 inhibitor AZD6244. Treatment of BRAFMT cells with AZD6244 potently inhibited peIF4E^S209 levels (Fig. 2A, left), decreased nascent protein production (Fig. 2B, right), and resulted in a potent reduction in basal ATF4 and CHOP levels (Fig. 2B, left). Importantly, AZD6244 treatment also resulted in a potent downregulation in HA15-induced ATF4 and CHOP levels and this was associated with decreased levels of apoptosis induction following HA15 treatment (Fig. 2C). Similar results were obtained with the MEK1/2 inhibitor trametinib (Supplementary Fig. S2F and S2G). Moreover, calculation of combination index (CI) values confirmed antagonism between trametinib and HA15 in the BRAFMT HT-29 cell line (Supplementary Fig. S2G). Collectively, these data would suggest that oncogenic BRAF and activated MEK1/2-ERK1/2 signaling results in enhanced protein synthesis and chronic ER stress, rendering BRAFMT colorectal cancer cells susceptible to apoptosis following treatment with acute ER stress activators such as HA15.

To test our hypothesis that BRAFMT colorectal cancer cells are vulnerable to modulation of the UPR, we used two inhibitors of protein degradation pathways: the proteasomal inhibitor Carfilzomib (CFZ; 29) and the aggresome inhibitor ACY-1215 (HDAC6-selective inhibitor; ref. 30). These compounds are in clinical development and have been found to result in overload of misfolded/damaged proteins and ER stress. Initially, we evaluated the effects of CFZ and ACY-1215 on viability of the parental BRAFMT VACO432 colorectal cancer cell line and its isogenic VT1 clone, and a panel of BRAFMT (LIM2405, HT-29, COLO205, RKO) and WT (CACO-2, COLO320, and DiFi) colorectal cancer cells. Sensitivity to both CFZ and ACY-1215 was markedly higher in the BRAFMT VACO432 cell line (5.9 nmol/L ± 2.3 and 1.36 μmol/L ± 0.52, respectively) compared to its WT clone (18.46 nmol/L ± 2.9 and 9.63 μM ± 2.95, respectively), and this was associated with increased apoptosis as determined by PARP cleavage and increased sub-G₁ levels in the VACO432 cells compared to the VT1 cells (Fig. 3A and B; Supplementary Fig. S3A). ACY-1215 caused an increase in acetylation of α-tubulin, a marker of HDAC6 inhibition. There was however no clear pattern of increased sensitivity to CFZ or ACY-1215 in the panel of nonmatched BRAFMT and WT colorectal cancer cells (Supplementary Fig. S3A).

Next, we investigated the effects of combination treatment of CFZ with ACY-1215 on survival of BRAFMT/WT colorectal cancer cells. Combined treatment of CFZ with ACY-1215 resulted in potent increases in apoptosis as indicated by increased PARP cleavage and caspase-3 processing, caspase-8 and -3/7 activity assays and flow cytometry in the BRAFMT VACO432 cell line but not the WT VT1 clone (Fig. 3B; Supplementary Fig. S3B and S3C). In addition, transient overexpression of BRAFV600E led to increased CHOP expression levels and was associated with increased PARP processing and caspase 3/7 activity in response to combined ACY-1215/CFZ treatment in the BRAFWT VT1 cells (Fig. 3B). Importantly, similar results were obtained in a wider panel of BRAFMT (HT-29, LIM2405, RKO, COLO205), but not BRAFWT colorectal cancer cell line models (COLO320, CACO-2, DiFi cells; Fig. 3C and D; Supplementary Fig. S3C). Caspase-dependent apoptosis following CFZ/ACY-1215 treatment was assessed using the pan-caspase inhibitor Z-VAD-FMK, which attenuated apoptosis (Supplementary Fig. S4A). Silencing of caspase-9 partially reduced PARP and C3 cleavage following CFZ/ACY-1215 cotreatment, suggesting a predominant role of the intrinsic apoptotic pathway in regulating CFZ/ACY-1215-induced cell death (Supplementary Fig. S4B). Calculation of combination index (CI) values confirmed strong synergy between CFZ and ACY-1215 in all BRAFMT colorectal cancer cells (Supplementary Fig. S4C). Taken together, these data would indicate that BRAFMT CRC cells show increased sensitivity to dual targeting of protein degradation pathways.

To investigate the mechanism underlying the observed synergy between CFZ and ACY-1215, we used the PROTEOSTAT dye, a fluorescent dye that facilitates the detection of aggregated protein cargoes within aggresomes and other inclusion bodies (31). Treatment of the BRAFMT VACO432 cells with combined CFZ/ACY-1215 resulted in a marked accumulation of aggregated proteins, compared to the effect of either agent alone (Fig. 4A). Aggregation of proteins following combined CFZ/ACY-1215 treatment was markedly less induced in its BRAFWT clone. Moreover, the combination of ACY-1215 with CFZ increased the accumulation of ubiquitinated proteins, in particular in the BRAFMT VACO432, COLO205, and RKO cells, but not in the BRAFWT VT1 and DiFi cells (Fig. 4B; Supplementary Fig. S5).

Proteasomal and aggresome inhibition has been shown to induce ER stress (32, 33). To explore the role of the UPR in mediating apoptosis following combined CFZ/ACY-1215 treatment, we assessed expression and phosphorylation levels of the pro-apoptotic UPR proteins CHOP and JNK, 24 hours following treatment with CFZ and ACY-1215 (Fig. 5A and B). We found marked increased expression levels of CHOP in all BRAFMT but not BRAFWT cells following combined CFZ/ACY-1215 treatment. In addition, increased pJNK^T183/Y185 was also observed in CFZ/ACY-1215-treated BRAFMT HT-29, LIM-2405, RKO, and COLO205 cells (Fig. 5A). We also determined the kinetics of activation of the PERK, IRE1α, and ATF6 branches of the UPR following cotreatment with CFZ/ACY-1215 and found marked increases in ATF4 and CHOP levels, as early as 6 hours and 12 to 24 hours respectively following cotreatment with CFZ/ACY-1215 in all BRAFMT colorectal cancer cells (Fig. 5C). Increased IRE1α and JNK activity was observed 6 hours and 12 to 24 hours, respectively, following treatment with CFZ/ACY-1215, whereas ATF6 cleavage occurred 12 to 24 hours following treatment with CFZ/ACY-1215, in particular in the BRAFMT COLO205 cells (Fig. 5C).

To investigate whether activation of the UPR induced by combined CFZ/ACY-1215 treatment is responsible for the observed apoptosis, we used siRNA specifically directed against CHOP. Notably, silencing of CHOP completely abrogated the synergistic induction in caspase-3 and PARP cleavage following combined CFZ/ACY-1215 treatment in BRAFMT colorectal cancer cells. Collectively, these results would suggest a causal role for the UPR pathway in the cell death following combination of CFZ with ACY-1215 in BRAFMT colorectal cancer (Fig. 5D).

To extend these in vitro findings, we next assessed the therapeutic efficacy of combined ACY-1215 and CFZ in the BRAFMT VACO432 xenograft model. ACY-1215 and CFZ were administered at 30 mg/kg/day i.p. and Carfilzomib 6 mg/kg 3/week i.p., the maximum tolerable dose determined in our initial dose escalation study (Supplementary Fig. S6A). Although both CFZ and ACY-1215 slowed tumor growth, the CFZ/ACY-1215 combination led to a supra-additive reduction in growth in VACO432 BRAFMT xenograft model (Fig. 6A, left). Furthermore, combined CFZ/ACY-1215 treatment significantly reduced tumor weight, compared to the effect of each treatment alone (Fig. 6A, middle). Furthermore, caspase-3 cleavage was observed in the VACO432 tumors following treatment with CFZ in combination with ACY-1215, indicative of apoptosis induction (Fig. 6A, right). The combined treatment was also tolerated in this efficacy study, with only minor weight loss occurring in both vehicle and combination arms (5.96% and 6.1%, respectively; Supplementary Fig. S6B). These findings indicate that dual targeting of protein degradation pathways using CFZ and ACY-1215 may be highly effective to treat BRAFMT colorectal cancer.

A number of research groups have identified three to six molecular subtypes within stage II/III colorectal cancer, using gene expression profiles and unsupervised classification (21). Recently, the Colorectal Cancer Subtyping Consortium (colorectal cancerSC) has integrated these independent classification systems into four consensus molecular subtypes (CMS 1–4; ref. 34). More than 70% of BRAF^V600E mutant cases were assigned to CMS 1 (with strong association with MSI), whereas 17% and 6% of mutant BRAF patients were classified into CMS 4 (cancers which display mesenchymal features with high stromal infiltration) and CMS 3 (cancers with metabolic deregulation), respectively. To better characterize the heterogeneity within BRAFMT colorectal cancer tumors, a recent study performed unsupervised classification of gene expression profiles derived from treated/untreated stage I/II/III BRAFMT colorectal cancer and identified two distinct subgroups: BM1, associated with high KRAS/mTOR/AKT/eEBP1, EMT activation and immune infiltration and BM2 which displays cell-cycle checkpoint dysregulation (35). In this study, we performed for the first time supervised clustering of BRAFMT gene expression profiles, to identify novel pathways and targets associated with the BRAFMT subgroup with poorest outcome. These studies identified a number of biological processes that were deregulated in the BRAFMT subgroup with poorest outcome, including the UPR.

Activation of different arms of the UPR have been reported in several solid tumors (36), as well as hematologic malignancies (37) and have been identified as a poor prognostic marker in these tumors. In colorectal cancer, activation of IRE1α-XBP1 UPR branch has been associated with increased proliferation, invasion, and poor prognosis (38). Using a siRNA screening approach and multiple BRAFMT cell line models, we identified that GRP78, the master regulator of the UPR, was important for maintaining the viability of BRAFMT colorectal cancer cells. Moreover, using an inhibitor of GRP78, which dissociates GRP78 from PERK, IRE1α, and ATF6 (15), the differential dependency of BRAFMT and WT cells on GRP78 for survival was further demonstrated. GRP78 belongs to the heat shock protein 70 (HSP70) family, facilitates protein folding, targets misfolded proteins for proteasomal degradation, and regulates ER stress (39). Overexpression of GRP78 has been observed in many tumors, including colorectal cancer (40) and has been associated with resistance to chemotherapy (41). The relative overexpression of GRP78 highlights the potential for exploitation of GRP78 as a therapeutic target in colorectal cancer. Increased ERK signaling has been shown to activate protein synthesis, which may result in an imbalance between the ER protein load and folding capacity, resulting in chronic ER stress (42). Indeed, we found that BRAF and MEK regulated levels of eIF4E^S209, nascent protein synthesis, and ATF4/CHOP and sXBP1 expression levels in BRAFMT colorectal cancer cells. In contrast to a study from Jiang and colleagues in BRAFMT melanoma (43), our study showed that MEK inhibition decreases chronic ER stress and protects BRAFMT cells against apoptosis induced by acute ER stress. Furthermore, inhibition of CHOP, a transcription factor that controls the development of programmed cell death in response to acute ER stress (44), prevented apoptosis induced by HA15, indicating the causal role of acute ER stress in HA15-induced apoptosis in BRAFMT colorectal cancer. These data are the first to show that BRAFMT colorectal cancer is addicted to the UPR for survival and that UPR activators, such as HA15, result in acute ER stress and apoptosis in BRAFMT colorectal cancer. However, to date there are no existing treatments in routine clinical use that specifically target GRP78.

Proteasome inhibitors (PI) represent a class of agents that exert their anti-cancer effects by inducing accumulation of misfolded poly-ubiquitinated proteins and ER stress (32). There are many PI at various stages of clinical development and, to date, bortezomib, carfilzomib, and ixazomib have been approved by the Federal Drugs Administration for use in multiple myeloma (45, 46). In contrast to bortezomib, carfilzomib is an irreversible inhibitor, characterized by a preferential inhibitory potency against the chymotrypsin-like activity of the proteasome, with potent and persistent inhibition (47). Contrary to a previous study in BRAFMT colorectal cancer (48), we did not find increased sensitivity to carfilzomib across the BRAFMT colorectal cancer cell line panel, compared to our BRAFWT cells. Selective targeting of HDAC6, a class IIb Histone Deacetylase, results in acetylation of α-tubulin which disrupts binding of the HDAC6-tubulin-dynein motor complex to misfolded protein aggregates and inhibits aggresome-mediated protein degradation, leading to ER stress (33). In addition, a recent study has shown that HDAC6 inhibition results in acetylation of GRP78 and disruption of the GRP78/PERK complex (49). We found that dual targeting of protein degradation pathways, using CFZ and ACY-1215, resulted in marked accumulation of ubiquitinated protein aggregates, caspase-dependent apoptosis and strong synergy in BRAFMT colorectal cancer cells, but not BRAFWT cells. Mechanistically, we found that combined CFZ/ACY-1215 treatment resulted in marked increased expression of pEIF2α^S51/ATF4 and pIRE1α^S724/pJNK^T183/Y185, culminating in increased levels of the pro-apoptotic CHOP. Furthermore, we found that siCHOP protected BRAFMT colorectal cancer from apoptosis following CFZ/ACY-1215 treatment, demonstrating the causal role of the UPR in increased apoptosis following combined CFZ/ACY-1215 treatment in BRAFMT colorectal cancer. Importantly, we also demonstrated that CFZ in conjunction with ACY-1215 was highly effective at blocking the growth and inducing apoptosis in a BRAFMT colorectal cancer xenograft.

Combinations of proteasome and HDAC inhibitors have been translated into clinical trials on the basis of dual inhibition of the aggresome and proteasome pathway, in particular in hematologic disease. The VANTAGE 095 phase IIb trial of vorinostat (MK-0683) and bortezomib showed a signal of activity in patients with relapsed/refractory multiple myeloma (50). More recent studies using the selective HDAC6 inhibitor ACY-1215 and bortezomib or the more potent PI CFZ have shown strong synergy in preclinical models of lymphoma and multiple myeloma, respectively (49, 51). This is the first study showing that dual aggresome/proteasome inhibition could be a promising treatment strategy for BRAFMT colorectal cancer patients. This study does not exclude that other poor prognostic subtypes, for example subgroups of BRAFWT/RASMT colorectal cancer may also be associated with chronic ER stress and therefore could potentially benefit from acute ER stress inducers.

In conclusion, using a systems biology approach, we have identified the UPR as an important, novel and druggable target for the BRAFMT colorectal cancer subgroup with poorest clinical outcome. We have shown that BRAFMT colorectal cancer induces chronic ER stress, is addicted to the UPR for survival and that UPR activators result in acute ER stress and apoptosis in BRAFMT colorectal cancer (Fig. 6B). From a clinical perspective, the substantial tumor growth inhibition observed in our xenograft study support the evaluation of UPR activators (e.g., aggresome/proteasome co-inhibition or more specific ATF4/CHOP activators (52)) in clinical trials for patients with metastatic BRAFMT colorectal cancer.

No potential conflicts of interest were disclosed.

Conception and design: N. Forsythe, A. Refaat, P.G. Johnston, S. Van Schaeybroeck

Development of methodology: N. Forsythe, A. Refaat, H. Emam, V. Popovici, P.G. Johnston, S. Van Schaeybroeck

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N. Forsythe, A. Refaat, A. Javadi, H. Khawaja, J.-A. Weir, W.L. Allen, F. Burkamp, M.J. Labonte, P.G. Johnston, S. Van Schaeybroeck

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N. Forsythe, A. Refaat, H. Khawaja, W.L. Allen, V. Popovici, P.V. Jithesh, C. Isella, M.J. Labonte, P.G. Johnston, S. Van Schaeybroeck

Writing, review, and/or revision of the manuscript: N. Forsythe, A. Refaat, W.L. Allen, F. Burkamp, V. Popovici, P.V. Jithesh, M.J. Labonte, I.G. Mills, P.G. Johnston, S. Van Schaeybroeck

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Khawaja, F. Burkamp, P.G. Johnston, S. Van Schaeybroeck

Study supervision: P.G. Johnston, S. Van Schaeybroeck

Funding was supported by Cancer Research UK (C212/A13721, to P.G. Johnston); Cancer Research UK fellowship (C13749/A13172, to S. Van Schaeybroeck); and MErCuRIC, funded by the European Commission's Framework Programme 7, under contract No. 602901 (to S. Van Schaeybroeck). We thank Enzo Medico, Andrea Bertotti, and Dan Longley for providing bioinformatic and in vivo support, respectively.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.