BV173 cells were obtained from DSMZ (Cat#ACC-20, RRID:CVCL_0181) and cultured in 20% FBS, RPMI media (with 20 mM HEPES and L-glutamine; Sigma Aldrich, R7388) at 37°C, 5% CO₂. BV173 cells were grown to a density of 1 × 10⁶ cells/mL prior to all treatments. For immunoprecipitation a total of 1 × 10⁸ cells were used for each experiment.

JQ1-CRBN, JQ1-VHL, and JQ1-MDM2 were all synthesized as previously described (18–20). JQ1 (SML0974) and pomalidomide (P0018) were purchased from Sigma-Aldrich. Treatments were performed in true biological triplicate. Cells were treated for FACS analysis (10 nM, 100 nM; 1–6 h), western blotting (0–1000 nM, 0–16 h), and immunoprecipitation of MHC-I complexes (0.1 −100 nM, 0–6 h). Initial immunoprecipitation-MS screening runs resulted in focused treatments: using PROTAC compounds at 10 nM, for 0–6 hours. All cellular treatments were at 37°C, 5% CO₂ using BV173 cells at 1 × 10⁶ cells/mL density.

Primary antibodies anti-BRD2 (Bethyl, A302-583A), anti-BRD3, (Bethyl, A302-368A), and anti-BRD4 (Bethyl, A301-985A) were used at a 1:1000 dilution in Odyssey® Blocking Buffer (LI-COR, 927-40000). Anti-actin (Sigma-Aldrich, A2066) was used at a 1:5000 dilution. IRDye® 800CW goat anti-mouse (LI-COR, 926-32211) was used as secondary antibody at 1:5000 dilution. Blots were scanned using the LI-COR Odyssey® CLx scanner. For SDS-PAGE 30 μg of protein was used in each lane. SDS-PAGE-separated proteins were transferred to PVDF membranes for western blotting. Quantification of band intensity was performed in ImageJ using the area under the curve (AUC) from the protein of interest, normalized to actin AUC.

1 × 10⁸ BV173 cells were treated with JQ1, pomalidomide, JQ1-VHL, JQ1-MDM2, JQ1-CRBN, or DMSO. After treatment, cells were lysed in 2% CHAPS buffer (120 mM NaCl, 50 mM Tris-Cl pH 8.0, 2% CHAPS). Cells were ultracentrifuged at 100,000 × g for 1 h. Clarified supernatant was incubated with BB7.2 (BioLegend, 343302) for 1 h prior to incubation with protein A beads (Dynabeads™, Invitrogen,10008D) for 1 h. Beads were washed with 1% CHAPS buffer (120 mM NaCl, 50 mM Tris-Cl pH 8.0, 1% CHAPS) and 1X PBS. The remaining lysis supernatant was incubated with W6/32 (BioLegend, 311428) followed by a second incubation with protein A Dynabeads™. Complexes and peptides were eluted with 0.1% TFA prior to peptide separation using a 10 kDa cutoff spin filter (Pall Life Sciences, OD010C34). Peptides were desalted using C18 ziptips (Pierce, 87784) prior to injection onto MS instrumentation.

Data-dependent acquisition of peptides was carried out using a Fusion Lumos mass spectrometer (Thermo Fisher Scientific) equipped with either an Easy nano LC 1200 system (Thermo Fisher Scientific) or Ultimate 3000 RSLCnano (Thermo Fisher Scientific). Mobile phases utilized for the Easy nano LC 1200 separation of peptides were 0.1% (v/v) formic acid (Thermo Fisher Scientific, Cat. No. 85170) in water (buffer A) and 0.1% (v/v) formic acid in acetonitrile (ACN) (Thermo Fisher Scientific, Cat. No. 85174) (buffer B). Samples were loaded onto an Acclaim PepMap 100 trap column (75 μm i.d. × 20 mm, Thermo Scientific, Cat. No. 164946) packed with 3 μm C18 resin and an Easy Spray analytical column (75 μm i.d. × 250 mm, Thermo Fisher Scientific, Cat. No.ES802) packed with 2 μm/100Å C18 resin. The column temperature was maintained at 45°C for the duration of each sample analysis. Peptides were eluted during a 95 min gradient with the following segments: an initial hold at 5% B for 2 min, 5–40% B for 80 min, and 40–100% B for 2 min, followed by an 11 min column wash and re-equilibration for 20 min at 5% B. For those samples analyzed using an Ultimate 3000 RSLCnano, the method incorporated additional loading buffers of 0.05% (v/v) TFA in water and 0.05% (v/v) TFA in ACN. The RSLCnano used an Acclaim PepMap 100 trap column (75 μm i.d. × 20 mm, Thermo Fisher Scientific, Cat. No. 164535) packed with 3 μm C18 resin. The remaining mobile phases and analytical column used with the RSLCnano were identical to the Easy nano LC 1200. Using the RSLCnano, peptides were eluted with a gradient comprised of the following segments: an initial hold at 2% B for 5 min, 2–40% B for 85 min, followed by a hold at 90% B for 10 min, and a return to 2% B with a final hold of 15 min. Despite the longer cumulative chromatographic separation, MS data was only acquired over 95 min to mimic the Easy nano LC 1200 separation. The Fusion Lumos full scan resolving power was set to 60,000 and acquired from 300 to 1750 m/z. An MS1 automated gain control (AGC) target of 4 × 10⁵ ions was used with a maximum injection time of 50 ms and a lock mass of 445.12 m/z. Peptides were selected for fragmentation if they fell between an intensity range from 1 × 10⁴-1 × 10¹² and charge states +1–5. Peptides with +1 charge states were only sampled if they fell within a 750–1750 m/z range, while all +2–5 charge states were sampled. Each peptide fitting these criteria was sampled twice and then excluded from MS acquisition for 60 s within a 15 ppm precursor tolerance. Total MS cycle time was set to 5 s. Peptides were selected using a 1.2 m/z quadrupole isolation window and fragmented using collision-induced dissociation (CID) with 29% normalized collision energy. Resulting peptide fragments were analyzed with rapid scans in the ion trap mass analyzer using an AGC target of 3 × 10³ ions and a 35 ms maximum injection time.

Peptide identifications were made using Byonic™ (Protein Metrics) software against the Human Uniprot database (downloaded on 8/10/2015) with a 20 ppm precursor mass tolerance, and 0.7 Da fragment ion mass tolerance. Peptide identifications were made with a 1% FDR. After compiling all identified peptides from all instrument runs, peptide identifications were filtered down to those made with +/– 5 ppm precursor mass tolerance. Furthermore, due to multiple discrete searches of the immunopeptidome raw files, the composite peptide list was searched using the NCBI BLAST algorithm (https://blast.ncbi.nlm.nih.gov) to determine consensus protein identifications across the data set. With these constraints applied, we further restricted the final peptide list to include peptides identified with >2 PSM.

For label-free quantification of BET peptide abundance, 29 housekeeping peptides were selected for use in normalizing signals across data acquisitions. These peptides were shared across all samples and had a range of observed abundances within each sample. A small FASTA file containing only these housekeeping peptides and the observed BET peptides was created and used to search against all data files to accommodate import into Byologic (Protein Metrics™). Peptide identifications were manually verified in Byologic, and the extracted ion chromatogram area under the curve (XIC AUC) was generated for this subset of peptides as a measure of abundance. For each data file, normalization was performed using the average XIC AUC of the set of 29 housekeeping peptides. Normalized XIC AUC values for each peptide and condition were plotted as averages of all available observations +/– the standard deviation (SD). Single observations were annotated and assigned average experimental error (SD).

BV173 cells were plated at a density of 1 × 10⁶ cells/mL and treated with 10 and 100 nM of each PROTAC compound (or DMSO) for 0, 1, 3, or 6 h. After treatment, cells were washed with FACS buffer (1X PBS, 1 mM EDTA, 1% FBS, 0.1% NaN₃) and incubated with either BB7.2 or W6/32 using 1 μg antibody per 100 μL of cell suspension for 30 min at 4°C. Cells were washed with FACS buffer prior to incubation with secondary antibody (AlexaFluor488, donkey anti-mouse, Thermo Fisher, A-21202) at 4°C for 30 min. Cells were washed 2X with FACS buffer, prior to analysis on FACSCanto-II flow cytometer (BD Bioscience). FlowJo® software was used for data analysis. Cells were gated for live populations only. The median fluorescence intensity (MFI) was extracted and plotted for each sample as the average MFI +/– SD.

Gene ontological annotations were assigned using the Panther classification system (pantherdb.org) (28). To determine overrepresentation of specific cellular components within our dataset against the entire human proteome we used Fisher's Exact with FDR multiple test correction, available as a component of the Panther system. Annotation visualization was performed with Proteomaps software, using the number of unique peptides for each source protein as a measure of significance (www.proteomaps.net) (29). Biological pathway mapping was performed using Ingenuity Pathway Analysis® (Qiagen). Core pathway analysis was performed with our dataset using only human experimentally-determined and high-confidence relationships.

Statistical details of experiments can be found in figure legends. Unless otherwise stated, data is plotted as the mean +/– SD.

JQ1-CRBN, JQ1-VHL, and JQ1-MDM2 were all synthesized as previously described (18–20) and assessed for the capability to degrade BET target proteins in BV173 cells by western blot (Figure S1). Degradation of BRD proteins by these JQ1-PROTACs was observed over a range of concentrations. Time-dependent degradation was also assessed to compare the relative rates of degradation of the BET proteins by each PROTAC within a focused concentration range (Figure S2). The efficiency of protein degradation was consistent with previous reports (18, 20, 30). JQ1-CRBN and JQ1-VHL performed similarly, successfully degrading BRD2, BRD3, and BRD4. However, JQ1-MDM2 was less effective at the degradation of BET proteins, overall.

To understand the contributions of peptide changes to the total immunopeptide pool, we analyzed the effect of PROTAC treatment (at 10 and 100 nM from 0 to 6 h) on the amount of MHC-I on the cell surface using flow cytometry. We stained for the presence of MHC-I using pan-HLA antibody W6/32, as well as, HLA-A^*02 antibody BB7.2 and found no increase in total MHC-I on the cell surface (Figure 2A, Figure S3) indicating that any changes in specific peptides observed were not due to upregulation of MHC-I itself.

We isolated MHC-I peptides from BV173 cells through immunoprecipitation of MHC-I complexes using both BB7.2 (anti-HLA-A2) and W6/32 (anti-pan-MHC-I) antibodies. Peptides eluted from MHC-I complexes were detected by LC-MS/MS, and the resulting spectra were interrogated using Byonic software (Protein Metrics™). Peptides were identified by searching against a human protein database (Uniprot), and the results were filtered to a 1% false discovery rate (FDR). To further standardize the data, peptide identifications that resulted from >5 ppm precursor ion mass accuracy and/or a single peptide spectral match (PSM) were eliminated. After applying these criteria, we retained approximately 7,500 unique peptide sequences from 3,200 protein sources in aggregate across all experiments. The complete list of identified peptides can be found in a supplemental file (Table S1). This prioritized list consisted of peptides ranging in size with the majority (92%) falling between 8 and 14 amino acids in length. We found that PROTAC treatment (or control compound treatment) had no effect on this length distribution within our dataset (Figure 2B). Additionally, across all samples, approximately 60% of identified peptides were 9-mers, which is consistent with previous reports of confident MHC-I isolation (31).

To aid in the conformation that the isolated peptides had features relevant to HLA-binding, we analyzed the motifs of our isolated immunopeptides and found that (1) these motifs were consistent between treated and untreated samples and (2) the identified motifs were appropriate for the HLA alleles present in the BV173 cell line (HLA-A^*02:01, HLA-A^*30:01, HLA-B^*18:01, HLA-B^*15:10, HLA-C^*12:03, and HLA-C^*03:04) (Figure 2C).

The affinity of isolated immunopeptides was predicted using the Immune Epitope Database (http://www.iedb.org/) (32) with the application of Stabilized Matrix Method (SMM) and Artificial Neural Networks (ANN) affinity algorithms to all identified peptide sequences between 8 and 14 amino acids (33–35). From our complete dataset, we observed increased representation of peptides with higher predicted affinity for HLA-A^*02:01. However, we were still able to identify peptides with high predicted affinity for all represented MHC-I alleles within our cell line from those alleles that were available within the IEDB resource. We compared the affinity profile of PROTAC-treated samples to DMSO-treated samples using both SMM and ANN algorithms and found no changes in the distribution of high affinity (0–100 nM IC50) peptides with PROTAC treatment. However, we did observe a small increase in the amount of mid-range affinity (1–10 μM IC50) HLA-A^*02 peptides with PROTAC treatment, with a corresponding decrease in HLA-B^*18 representation within the same affinity range. This effect was observed upon analysis with both SMM and ANN affinity algorithms (Figure S4). Across all other represented alleles, we observed no changes to the distribution of predicted peptide affinity. In total, PROTAC-treated vs. untreated samples maintained the same presentation profile, with a small variance in peptide affinity at mid-range concentrations, in an allele-specific fashion.

The composition of the BV173 MHC-I peptidome was globally assessed using Panther and Proteomaps. The Panther classification system (36) was used for the assignment of gene ontology (GO) terms. Using this tool, we compared MHC-I peptides isolated from each treatment against the human proteome to look for overrepresentation of specific cellular components and biological processes (28). Consistent with previous reports of HLA peptide isolation (37, 38) we observed overrepresentation of peptides that were primarily derived from cellular components such as the protein-DNA complex (p-value = 7.54E-05), nucleus (p-value = 7.98E-47), and ribosome (p-value = 4.14E-10), with an underrepresentation of membrane-bound proteins (p-value = 4.29E-06) in all samples. Across all treated samples (all samples with the exception of DMSO) we observed an increase in glycolysis overrepresentation (p-value = 1.58E-04). With this exception, we did not identify any significant changes in gene ontology between PROTAC-treated and control samples (Table S2).We additionally used Proteomaps software for visualization of ontological distributions within our dataset (Figure S5) (29, 39). Using the number of unique peptides for each source protein as a measure of significance, our Proteomaps representation was similar to previous reports and was not found to vary between PROTAC-treated and control samples.

To examine PROTAC-dependent changes to source protein identifications, we took two approaches. First, we looked at which source proteins were observed only upon PROTAC treatment, but not control treatment. There were 95 source proteins uniquely observed after treatment with all three JQ1-based PROTAC compounds (Figure 3A). We sought to rank these treatment-specific source protein identifications by using the number of PSMs per protein as a measure of prevalence. We were then able to rank the PROTAC-specific proteins from most-to-least likely to be induced by BET PROTAC treatment. Significantly, BRD3 (a BET target protein) ranked as having the most observed PSMs, followed closely by TAF8 (a transcription-related protein). The top 10 most-represented (by PSM count, across 166 replicates total, 122 PROTAC-treated replicates) PROTAC-specific protein identifications are listed in Figure 3B. The complete list of PROTAC-unique source protein identifications can be found in Table S3.

The second approach involved identification of source proteins for which some peptides were observed in controls (prior to treatment), but for which PROTAC-treatment enhanced their presentation through increased peptide coverage. We compared changes in the total number of peptides-per-protein between treated and untreated samples, hypothesizing that proteins already represented in control-treated samples might present additional peptides when stimulated by PROTAC treatment (Figure 3C). For this analysis, we used a ratio of ratios: the number of peptides/protein was calculated for each source protein within each treatment, and this value was compared for each protein between treatments: protein-specific fold change = [peptides/protein]_PROTAC/[peptides/protein]_Controls. From this analysis, we were able to determine specific proteins that had increased representation in PROTAC-treated samples vs. controls (Figure 3D). Significantly, BRD2, Nck-associated protein 1-like, (NCKPL) and a handful of other proteins showed increased representation after PROTAC treatment. The complete list of all source proteins and PROTAC-specific changes in source protein representation can be found in Table S4.

BET-PROTAC compounds induced the presentation of BET-derived MHC-I peptides. PROTAC-specific sequences that we could verify with synthetic peptide MS/MS spectra (Figure S6) and with >2 PSM are listed in Figure 4A. These peptides were not observed in samples prepared from JQ1-, pomalidomide-, or DMSO-treated cells. To verify the uniqueness of these peptides to treatment, the ion chromatograms for the precursor masses (+/– 10 ppm) were extracted and compared across samples. An example for BRD3-specific peptide is shown in Figure S7. We found no evidence of the presentation of these peptides in control samples within the limits of our detection.

When looking at the list of PROTAC-induced BET peptides (Figure 4A), we observed that all three JQ1-PROTAC compounds induced the same two MHC-I peptides: KMDPEVEA (BRD3) and RLAELQEQL, (a peptide originating from a conserved region of the BRD2/3/4 proteins). These two peptides were observed, regardless of E3-ligase recruiting factor, and not in controls. Additionally, JQ1-CRBN and JQ1-VHL both induced the presentation of HQVPAVSSV and LHSAGPPLL from BRD2. We did not observe BRD2 peptides from JQ1-MDM2 treatment. We also noticed that for each BRD protein, PROTAC-induced peptide presentation occurred in close proximity to either the first (BRD2) or second (BRD3, BRD2/3/4) bromodomain (BD) of each protein (Figure 4B).

We analyzed the predicted MHC-I binding affinity for the four identified BRD peptides using both SMM and ANN algorithms and found that they had a range of predicted IC50 values for appropriate alleles. However, no newly detected peptide was predicted to bind with better than 100 nM (IC50) affinity (Figure 5). When the complete sequence of each BRD protein was scanned for predicted HLA-binding 9-mers using the IEDB affinity resource, we noticed that there were several peptide candidates which were predicted to be stronger binders for available MHC-I alleles. These peptide sequences were not selected by the cell for presentation within our limits of detection. Additionally, the endogenously presented BRD2 peptide TAAPPAQPL present in both controls and treated samples had a higher predicted affinity for MHC-I than any of the peptides induced with treatment.

To compare the relative abundances of isolated immunopeptides across different conditions, a label-free MS quantification method was used. In order to normalize signals across all samples, we applied correction factors based on a set of 29 housekeeping peptides, selected due to their consistent representation and range of abundances (Table S6, Figure S8). We measured relative abundances of the two most frequently observed BET peptides, KMDPEVEA, and RLAELQEQL, over a 6 h time course oftreatment with 10 nM of each PROTAC compound (Figure 6). PROTAC-induced BRD2 peptides were not present prevalently enough in our dataset to plot abundance curves. Their sporadic occurrence was consistent with less-robust degradation of BRD2 protein. When comparing the relative abundance of PROTAC-induced peptides over a 6 h time course, we observed that JQ1-VHL and JQ1-CRBN were more efficient than JQ1-MDM2 at producing BET-derived peptides. The corresponding peptides were presented faster and reached maximum abundance within a shorter time frame when the former two PROTAC compounds were applied. We overlaid source protein degradation (measured by western blot, Figure S2) with each peptide abundance plot to compare the rate of degradation with the rate of peptide presentation. We observed that induced peptide abundance reflected source protein degradation in a PROTAC-specific fashion.

We compared the rate of presentation of the BRD3 peptide KMDPEVEA between high (100 nM) and low (10 nM) concentrations of JQ1-VHL (left) and JQ1-CRBN (right) (Figure 7).We additionally compared this presentation to its corresponding source protein degradation (as observed by western blot). In doing so we observed several trends. First, a higher concentration of PROTAC induced a higher response within a 3 h time frame. Second, faster depletion of source protein at higher concentrations resulted in a depletion of presented peptide at later time points, and again, the overall rate of presentation of these peptides reflected the overall rate of degradation of protein induced by each PROTAC.

In order to gain insight into potential unanticipated PROTAC-induced changes, we performed Ingenuity Pathway Analysis (R) (IPA) mapping of proteins that were uniquely represented in our PROTAC-treated dataset, as well as, those proteins that were shown to have increased coverage after PROTAC treatment. After performing a core mapping analysis with only high confidence or experimentally observed associations, we identified the mTOR pathway as being upregulated by PROTAC treatment (p-value = 5.07E-03), as well as, other closely associated pathways. A map of the represented pathway networks and associated p-values is located in Figure S9. Additionally, Myc (p-value = 1.29E-02), CDK4/6 (p-value = 6.32E-04), E2F1 (p-value = 7.82E-04), and E2F4 (p-value = 1.16E-03) were observed to be upstream transcriptional regulators of both PROTAC-unique and PROTAC-enhanced identifications in our dataset. A list of predicted upstream regulators of identifications made within our PROTAC-specific dataset is shown in Figure S10.

The design, study conduct, and financial support for this research were provided by AbbVie. AbbVie participated in the interpretation of data, review, and approval of the publication. The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.