Acquired Resistance to BET-PROTACs (Proteolysis-Targeting Chimeras) Caused by Genomic Alterations in Core Components of E3 Ligase Complexes

This article is featured in Highlights of This Issue, p. 1183

Ubiquitination-mediated proteolysis is a central mechanism of protein homeostasis in cells. E3 ubiquitin ligases bind to defined substrates and mediate ubiquitination and degradation of these proteins. Cullin-RING ubiquitin ligases (CRL) are the largest family of E3 ubiquitin ligases. CRLs are multicomponent complexes that minimally consist of a cullin, a RING finger protein, and a substrate recognition subunit (1). Proteolysis-targeting chimeras (PROTAC) are bifunctional molecules that hijack endogenous E3 ubiquitin ligase to cause ubiquitination and subsequently degradation of proteins of interest (2). VHL and Cereblon (CRBN) are the substrate recognition subunit of the cullin2 (CUL2)-containing VHL–CRL complex and the cullin4-containing CRBN–CRL complex, respectively (3, 4). Recently, it has been shown that PROTACs with robust in vitro and in vivo activities and, in some cases, drug-like pharmaceutical properties can be generated using small-molecule ligands for the E3 ligases VHL and CRBN (5–9). PROTACs targeting the bromodomain and extraterminal domain proteins (BET-PROTAC) are the prototype of these small-molecule PROTACs (5–7). BET-PROTACs trigger rapid and prolonged degradation of BET proteins with exceptional potencies, and exhibit robust antitumor activities in preclinical models of leukemia, lymphoma, prostate cancer, and triple-negative breast cancers (5, 6, 10, 11).

Compared with traditional small-molecule inhibitors, PROTACs offer several advantages. For example, PROTACs can exert more rapid, potent, and durable inhibition of targets, abolish the scaffolding function of a protein, and turn a nonfunctional binder into functional degraders. In addition, the stringent conformational requirement and potential linker interaction between target protein and E3 ligase within ternary complex may allow PROTACs to offer an additional layer of selectivity over small-molecule inhibitors (12–15). These unique properties make PROTACs a promising modality for the development of next-generation therapeutics. However, innate and acquired drug resistance is a common cause of therapeutic failure, particularly for cancer therapy. With the rapid advancement of the PROTAC technology toward therapeutic applications, it would be important to understand whether and how drug resistance to these novel agents may emerge. In this study, using BET-PROTACs with both VHL and CRBN ligands as a model system, we interrogate potential mechanisms of acquired resistance to PROTACs in cancer cell lines.

SKM1, MV4:11, LNCaP, and OVCAR8 cell lines were purchased from ATCC or DSMZ and maintained by a Core Cell Line Facility. All cell lines were tested for Mycoplasma using MycoAlert Detection Kit (Lonza) and authenticated using the Gene Print10 STR Kit (Promega) and maintained in RPMI1640 medium with 10% FBS (Gibco). ABBV-075, ARV-771, and ARV-825 were synthesized at AbbVie by the methods according to (5, 6, 16). All antibodies were purchased from commercial sources as follows: antibodies against BRD2/3/4 from Bethyl; antibodies against c-MYC, PARP, and VHL from Cell Signaling Technology; antibody against CRBN from Thermo Fisher Scientific; antibody against CUL2 from Abcam; and antibody against β-actin from Sigma. O1R- and O3R-overexpressing cells were created by infection with pLOC or pLOC-CUL2 or pLOC-CRBN lentiviral particle (Dharmacon) in the presence of 10 μg/mL of polybrene. Cells were selected with 10 μg/mL of blasticidin. Western blot analysis was performed to confirm expression of these proteins in the cells.

Cells were seeded in 96-well plates and incubated at 37°C in an atmosphere of 5% CO₂. Compounds were added at a series of dilution after overnight incubation. After 3 days incubation, Caspase-Glo3/7 luminescent and CellTiter-Glo Assay (Promega) were performed according to the manufacturer's instruction. Luminescence signal from each well was measured using the EnSpire Luminometer (PerkinElmer), and the data were analyzed using the GraphPad Prism Software (GraphPad Software Inc.).

Total RNA was harvested using the RNeasy Plus Mini Kit (Qiagen) and cDNA was made with the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems). qPCR was performed with TaqMan Universal Master Mix and Probes (Life Technologies), and analyzed using the ΔΔC_t method. Data are presented as mean ± SD. The difference between two groups was evaluated using the two-tailed Student t test. P values less than 0.05 were considered statistically significant.

Cell lysates were prepared in Laemmli Buffer (Bio-Rad). Thirty micrograms of total protein was resolved on a 7%–12% SDS-polyacrylamide gel and probed with corresponding primary antibodies.

Samples were prepared in accordance with the manufacturer's instructions for the KAPA Hyper Prep and NimbleGen SeqCap EZ MedExome kits. Briefly, 100 ng of DNA was sheared using a Covaris M220 sonicator (adaptive focused acoustics). DNA fragments were end-repaired, adenylated, ligated with KAPA dual indexing adapters, and amplified by 11 cycles of PCR with the KAPA hyper prep kit. Samples were pooled into a precapture library and targeted capture was performed using the NimbleGen SeqCap EZ MedExome capture probe set. Captured libraries were then enriched by 14 cycles of PCR and final libraries were evaluated using Qubit (Thermo Fisher Scientific) and Agilent Tape Station and were sequenced on an Illumina NextSeq sequencer using 2 × 75 bp read length. The raw sequencing data is deposited at SRA (submission no. SUB5567117).

Exome sequencing reads were processed through the Sentieon TNscope Pipeline (Sentieon Inc.) for alignment and somatic variant calling (parental sample designated at “normal”, O1R and O3R designated as “tumor”). Quality of sequencing data was assessed using Picard (http://broadinstitute.github.io/picard) and MultiQC (17). SNV and indel variant calls were imported into the VarSeq Software (Golden Helix) for filtering and annotation.

Somatic copy number calling was performed using VarScan v2.4.2 using tumor/normal mpileup from samtools v1.7 as input (18). Parameters used were minimum segment size of 100 bp, minimum coverage of 20 reads, minimum mapping quality of 20, and minimum base quality of 20. LOH in deleted regions were also visually confirmed by plotting minor allele frequencies from germline variant calling with Sentieon joint HaplotypeCaller (Sentieon Inc.).

Each cell line was sampled in triplicate and mRNA library preparation from total RNA was conducted following the manufacturer's protocol for the Illumina TruSeq mRNA preparation kit. Briefly, 1 μg of total RNA was purified by using poly-T oligos attached to magnetic beads then fragmented by divalent cations under elevated temperature. The fragmented RNA underwent first strand synthesis using reverse transcriptase and random primers. Second strand synthesis created the cDNA fragments using DNA polymerase I and RNaseH. The cDNA fragments then went through end repair, adenylation of the 3′ ends, and ligation of adapters. The cDNA library was enriched using 15 cycles of PCR and purified. Final libraries were assessed using the Agilent Bioanalyzer and Qubit (Thermo Fisher Scientific) assay methods then sequenced on an Illumina NextSeq sequencer using 2 × 75 bp read length. The raw sequencing data is deposited at SRA (Submission no. SUB5567117).

RNA-sequencing reads were mapped to the human reference genome (GRCh38) using STAR aligner and genes were quantified using featureCounts for all genes annotated in Gencode v28 (19, 20). Quality of sequencing data was assessed using Picard (http://broadinstitute.github.io/picard) and MultiQC (17). Genes with counts per million less than one in two-thirds of samples or more were considered too lowly expressed and excluded. Differential gene expression (DGE) analysis comparing parental versus O1R and parental versus O3R (n = 3 for each) was then performed using linear modeling in limma with TMM normalization and voom transformation (21) DGE results were plotted using GLIMMA (22). Sashimi plots were prepared using integrated genome viewer (23).

Confirmation of exome copy number calls for CUL2 and CRBN was performed using TaqMan copy number assays. Three predesigned assays each for CUL2 and CRBN (Supplementary Table S1) were ordered along with the reference assays for TERT and RNaseP genes. Experiments were performed according to the manufacturer's instructions using 20 ng DNA per reaction, 2× TaqMan genotyping master mix, and both a target (CUL2 or CRBN) and reference assay (TERT or RNaseP). Each assay combination was run in quadruplicate for parental, O1R, O3R, and a no template control. qPCR was run on the ABI 7500 and analyzed using the CopyCaller Software employing the DDCt method for normalization of copy calls relative to the reference assay and parental samples.

Sequencing primers were designed using Primer3 software (version 1.1.4, http://www.sourceforge.net) and were purchased from Integrated DNA Technologies (Supplementary Table S2). Genomic DNA was amplified by singleplex PCR using the FailSafe PCR System (Epicentre). Thermal cycling was performed with 40 cycles [30 seconds at 98°C; 30 seconds at 62°C (−0.5°C each cycle); 60 seconds at 72°C], followed by 25 cycles (30 seconds at 98°C; 30 seconds at 55°C; and 60 seconds at 72°C). Amplicons were bidirectionally sequenced using Big Dye Terminator version 1.1 technology on an ABI 3130xl System (Applied Biosystems). Sequence analysis was performed using the Sequencher software version 5.4.6 (Gene Codes Corporation).

Drug resistance often arises following chronic exposure to cancer therapeutic agents. To investigate whether PROTACs are subject to similar drug resistance issues, we exposed two AML (SKM-1 and MV4:11) and two solid tumor (LNCaP and OVCAR8) cell lines to increasing concentrations of VHL- or CRBN-based BET-PROTACs over a 4 month period. The establishment of the resistant cell lines is illustrated in Supplementary Fig. S1A. Although all the four cell lines underwent apoptosis after BET-PROTAC treatment, SKM1, MV4:11, and LNCaP cells were much more sensitive to these compounds compared with OVCAR8, likely due to the strong dependency on BET proteins in AML and prostate cancer (refs. 6, 24, 25; Supplementary Fig. S2). While no stable resistant clones were obtained from SKM-1, MV4:11, or LNCaP cells incubated with either of the VHL- or CRBN-based BET-PROTACs, several resistant clones emerged from the OVCAR8 cells. These resistant clones exhibited greater than 40× IC₅₀ increase over the parental cell line for the VHL-based BET-PROTAC ARV-771 (6) or the CRBN-based BET-PROTAC ARV-825 (ref. 5; Fig. 1A). Withdrawing BET-PROTAC from culture media for 2 months did not diminish resistance, indicating that resistance to BET-PROTACs in these cells may result from stable genetic changes (Supplementary Fig. S1B). In addition, the resistant clones maintained similar sensitivities as the parental cell lines to the bromodomain inhibitor ABBV-075 (16), suggesting that the BET bromodomains in these cells retain the ability of binding BET inhibitors (Supplementary Fig. S1C). It is noteworthy that the O1R cells, which are resistant to the VHL-based BET-PROTAC ARV-771, remained sensitive to the CRBN-based BET-PROTAC ARV-825. Conversely, the O3R cells, which are resistant to the CRBN-based BET-PROTAC ARV-825, remained sensitive to the VHL-based BET-PROTAC ARV-771 (Fig. 1A). Consistent with what was observed in the proliferation assays, analysis of PARP cleavage and caspase activation further confirmed that the O1R cells were resistant to ARV-771- but not ARV-825–induced apoptosis, and the O3R cells were resistant to ARV-825- but not ARV-771–induced apoptosis (Fig. 1B and C). Taken together, these results demonstrate that the O1R cells are specifically resistant to the VHL-based BET-PROTAC (ARV-771), and the O3R cells are specifically resistant to the CRBN-based BET-PROTAC (ARV-825).

It has been shown that BET-PROTACs inhibit cell growth and cause apoptosis by inducing degradation of BET family proteins. As expected, treatment of ARV-771 for 16 hours caused degradation of BRD2, BRD3, and BRD4, and inhibited the well-established BET target gene Myc in the parental and the O3R cells in a dose-dependent manner (Fig. 2A). In contrast, ARV-771 failed to induce significant BET protein degradation or Myc inhibition in the O1R cells even at the very high concentration of 3 μmol/L. Similarly, treatment of ARV-825 for 16 hours caused BET protein degradation and Myc inhibition in the parental and O1R cells at as low as 30 nmol/L (Fig. 2A). However, ARV-825 did not degrade BRD2 or BRD4 protein efficiently at up to 3 μmol/L and only partially degraded BRD3 protein in O3R cells at higher concentration (Fig. 2A). This is consistent with previous reports in other cancer types that BRD3 is more readily degraded by BET-PROTACs compared with BRD2 and BRD4 (6, 24, 26). Time course studies further established that extending BET-PROTAC treatment to 48 hours still could not trigger BRD4 degradation in the resistant cells (Fig. 2B). These results collectively demonstrate that the O1R and O3R cells are resistant to BET protein degradation induced by VHL- or CRBN-based BET-PROTACs, respectively. The absence of cross resistance to both VHL- and CRBN-based BET-PROTAC suggests that the proteasome degradation machinery downstream of protein ubiquitination remains functional in these resistant cells.

To unravel the mechanism of resistance to VHL-based BET-PROTAC in the O1R cells, we examined the genomic and transcriptional differences such as acquired mutations, gene expression changes, and copy-number variation (CNV) between the parental and O1R cells using whole-exome sequencing and RNA-seq. No genomic or mRNA expression alterations of BET family proteins or VHL were found in the O1R cells, and the protein level of VHL is comparable in the parental and the O1R cells (Supplementary Fig. S1D). Although the protein levels of BRD4 and BRD3 are slightly decreased in O1R and O3R cells, there is still significant amount of these two proteins, especially BRD4, detected (Supplementary Fig. S1D). Interestingly, the O1R cells were found to possess multiple genomic alterations that impact the gene encoding CUL2, a critical component of the VHL–CRL complex (Fig. 3A). The alterations include a frame-shift mutation predicted to cause nonsense-mediated mRNA decay by introducing a premature stop codon at position 442 in the cullin homology domain; an intronic mutation upstream of exon 12 with concomitant skipping of exon 12; and a large-scale deletion of 42 Mb encompassing the CUL2 gene (Fig. 3B–D). To investigate the copy number of this region, we used SNP minor allele frequencies across chromosome 10 in the parental, O1R, and O3R cells (Supplementary Fig. S3A). The pattern suggested the presence of other chromosomal aberrations preexisting in the parental, making the estimation of exact copy number challenging. Follow-up experiments confirmed the frame shift and the intronic mutations by Sanger sequencing and the copy number loss by TaqMan CNV analysis (Fig. 3E; Supplementary Fig. S3C and S3D). Consistent with the abnormalities at the genomic level, RNA-seq revealed CUL2 to be one of the most significantly downregulated genes in the O1R cells compared with parental cells (Supplementary Fig. S4A). qPCR and Western blot analysis further established the significant reduction of CUL2 mRNA and CUL2 protein in the O1R cells compared with the parental cells (Fig. 4A and B). In contrast, the O3R cells (resistant to the CRBN-based BET-PROTAC) were devoid of these genomic abnormalities in the CUL2 gene, and expression of CUL2 was comparable with the parental cells (Fig. 4A and B).

The hypoxia inducible factors HIF-1A and HIF-2A are physiologic substrates of the VHL–CRL complex. Under normoxic conditions, the VHL–CRL complex mediates HIF-1 ubiquitination and degradation, consequently preventing HIF-1–dependent transcription under normoxia (27). Gene set enrichment analyses revealed significant upregulation and enrichment of the HIF1A and HIF2A pathways under normoxic conditions in the O1R cells compared with the parental cells (Supplementary Fig. S4C). qPCR analysis also demonstrated the upregulation of HIF-1 target genes SOD2, VEGF, and GLUT1 in the O1R cells compared with the parental cells or the O3R cells (Supplementary Fig. S4D). These results collectively support that CUL2 loss in the O1R cells compromises the function of the VHL–CRL complex.

To determine whether CUL2 loss directly contributes to the resistance to VHL-based BET-PROTAC, we created O1R-derived cell lines that express exogenous CUL2 (Fig. 4C). As shown in Fig. 4D and E, overexpression of CUL2 in the O1R cells resensitized the cells to ARV-771 in the cell proliferation assay and restored ARV-771–induced BET protein degradation, demonstrating that CUL2 loss is responsible for acquired resistance in O1R cells.

Whole-exome sequencing and RNA-seq were also carried out to determine the genomic and transcriptional differences such as acquired mutations, gene expression changes, and CNV between the O3R cells and the parental or the O1R cells. While no genomic or expression alterations of BET family proteins were found in the O3R cells, a 12 Mb deletion on chromosome 3 that encompasses the CRBN gene was found in O3R but not in the O1R cells (Fig. 5A). On the basis of SNP minor allele frequencies, copy number was estimated as three copies for chromosome 3p in the parental and the O1R cells and one copy in the region containing CRBN of the O3R cells (Supplementary Fig. S3B). The estimated loss of two copies was consistent with TaqMan CNV analysis, which demonstrated a greater than 50% reduction in genomic DNA level relative to parental (Fig. 5B). Consistent with the CRBN gene deletion, RNA-seq differential expression analysis identified CRBN as one of the most significantly downregulated genes in the O3R cells compared with parental cells (Supplementary Fig. S4B). qPCR and Western blot analysis confirmed significant downregulation of the CRBN mRNA and CRBN protein in the O3R cells (Fig. 5C and D).

To determine whether the reduction of CRBN directly contributes to the resistance to CRBN-based BET-PROTAC in the O3R cells, we created O3R-derived cell lines that express high levels of CRBN (Fig. 5E). As shown in Fig. 5F and G, increasing CRBN expression in the O3R cells resensitized the cells to ARV-825 in proliferation assays and restored ARV-825–induced BET protein degradation.

PROTACs have emerged as a promising novel modality for the development of next-generation therapeutics. In this study, we reported that resistance to both VHL- and CRBN-based BET-PROTACs can occur in cancer cells following chronic treatment. However, unlike what is often observed for other targeted therapeutics, such as kinase inhibitors, cells that were resistant to ARV-771 and ARV-825 did not contain secondary mutations that affect compound binding to the target. Although the protein levels of BRD3 and BRD4 are slightly decreased in O1R and O3R cells, there is still significant amount of these two proteins, especially BRD4, detected. Considering that BRD4 is the major player in c-Myc regulation among the three BRD proteins, this subtle change in BRD4 protein level is not likely to contribute to the acquired drug resistance (28). Rather, resistance to both classes of BET-PROTACs was primarily attributed to genomic alterations that impact core components of the corresponding E3 ligase complexes. In cells that were resistant to ARV-771 or ARV-825, the proteasome degradation machinery also remained intact. The preference of targeting the E3 ligases components over the proteasome machinery for resistance development is intriguing. We suspect that the particular vulnerability of E3 ligase components for resistance development may relate to the redundancy and/or nonessentiality of these components for cell survival, while compromising proteasome function in cells could be lethal or significantly compromise cell fitness. The multiple genomic alterations at the CUL2 locus suggest the existence of strong selection pressure for the cells to abolish CUL2 function in the VHL–CRL complex for resistance development. Interestingly, in cells that are resistant to the CRBN-based BET-PROTAC ARV-825, resistance was primarily attributed to CRBN deletion and no genomic/transcription alterations were observed for CUL4, the CUL2 equivalent of the CRBN–CRL complex. CRBN appears to be a common point of resistance development for CRBN-targeting agents. It has been shown that deletion of CRBN was the primary cause of resistance to iMiDs in myeloma cells (29). The expression level of CRBN has also been identified as a prognostic or predictive biomarker for patients with multiple myeloma receiving iMiDs (30–32). In addition to multiple myeloma, both genetic alterations and differential expression of CUL2 and CRBN have been observed in many other cancer types based on The Cancer Genome Atlas dataset (refs. 33, 34; Supplementary Fig. S5). Interestingly, cell lines with lower CUL2 (Toledo and U266B1) or CRBN (TOV112D and HCT116) expression are less sensitive to ARV-771 and ARV-825, respectively (Supplementary Fig. S6A and S6B). Importantly, siRNA-mediated knockdown of CUL2 or CRBN conferred resistance to ARV-771 and ARV-825 in LNCaP cells, suggesting that there is an association between the expression level of CUL2 or CRBN and sensitivity to VHL- or CRBN-based PROTACs (Supplementary Fig. S6C and S6D). However, the expression level of these two proteins may not be the only determining factor for PROTAC sensitivity. The fact that only the OVCAR8 cell line readily developed resistance to these compounds may be attributed to the genetic background of the parental cells. Consistently, OVCAR8 cells harbor an ATM mutation (p.V613L) and exhibit a mutational signature associated with double-stranded break repair defects with elevated numbers of larger indels (>3 bp; Supplementary Fig. S7). These underlying genetic attributes may play a role in the differential ability of the OVCAR8 cells to gain resistance to the BET-PROTACs compared with other cell lines tested in this study.

In summary, we report here, for the first time, the mechanisms of acquired resistance to PROTACs in cancer cell lines. These results highlight the critical involvement of E3 ligase complexes in resistance development and lay the foundation for future investigation.

Y. Shen has ownership interest (including stock, patents, etc.) in AbbVie. No potential conflicts of interest were disclosed by the other authors.

Conception and design: L. Zhang, B. Riley-Gillis, Y. Shen

Development of methodology: L. Zhang, B. Riley-Gillis

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Zhang, B. Riley-Gillis

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Zhang, B. Riley-Gillis, P. Vijay

Writing, review, and/or revision of the manuscript: L. Zhang, B. Riley-Gillis, P. Vijay, Y. Shen

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Zhang

Study supervision: L. Zhang, Y. Shen

We would like to thank Erin Murphy, Areej Ammar, Elina Dilmukhametova, and Ken Idler from the AbbVie Genomic Technologies team for help in preparing the samples for sequencing analysis. Thank you to Arne Grundstad from the AbbVie Computational Genomics team for help in processing the exome sequencing data. A special thank you to Relja Popovic from the AbbVie Pharmacogenomics team for key discussions guiding the initiation and design of the project.

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.