Proteomic profiling reveals CDK6 upregulation as a targetable resistance mechanism for lenalidomide in multiple myeloma

The immunomodulatory drugs (IMiDs) lenalidomide and pomalidomide are highly effective treatments for multiple myeloma. However, virtually all patients eventually relapse due to acquired drug resistance with resistance-causing genetic alterations being found only in a small subset of cases. To identify non-genetic mechanisms of drug resistance, we here perform integrated global quantitative tandem mass tag (TMT)-based proteomic and phosphoproteomic analyses and RNA sequencing in five paired pre-treatment and relapse samples from multiple myeloma patients. These analyses reveal a CDK6-governed protein resistance signature that includes myeloma high-risk factors such as TRIP13 and RRM1. Overexpression of CDK6 in multiple myeloma cell lines reduces sensitivity to IMiDs while CDK6 inhibition by palbociclib or CDK6 degradation by proteolysis targeting chimeras (PROTACs) is highly synergistic with IMiDs in vitro and in vivo. This work identifies CDK6 upregulation as a druggable target in IMiD-resistant multiple myeloma and highlights the use of proteomic studies to uncover non-genetic resistance mechanisms in cancer. Acquired resistance to immunomodulatory drugs is common in multiple myeloma patients, but rarely attributed to genetic alterations. Here, proteomic, phosphoproteomic and RNA sequencing analysis in five paired pre-treatment and relapse samples reveals a CDK6-regulated protein resistance signature.

M ultiple myeloma is a genetically heterogeneous malignancy of plasma cells. The immunomodulatory imide drugs (IMiDs) lenalidomide and pomalidomide are a mainstay in treating multiple myeloma 1 . Although the combination of IMiDs with other drugs like proteasome inhibitors, antibodies, corticosteroids, and high-dose chemotherapy can induce remissions in most patients, almost all patients eventually relapse due to acquired resistance of the multiple myeloma cells to one or several of the drugs 1 . IMiDs bind to cereblon (CRBN), that together with DDB1, CUL4A, and ROC1 forms the CRBN-CRL4 E3 ubiquitin ligase and modulate the substrate specificity of the enzyme 2 . This leads to ubiquitination and proteasomal degradation of the lymphoid transcription factors Ikaros (IKZF1) and Aiolos (IKZF3), which regulate expression of other genes such as IRF4 and MYC, and are essential for the proliferation and survival of multiple myeloma cells [3][4][5][6][7] . Pomalidomide is the most potent of the approved IMiDs, both in regard to IKZF1 and IKZF3 degradation, as well as clinical activity, and is therefore a preferred treatment for relapsed multiple myeloma 3,8 . Sequencing studies in relapsed multiple myeloma and functional screens identified acquired genetic alterations in members of the CRBN-CRL4 E3 ligase complex that completely abrogate lenalidomide and pomalidomide activity as an IMiD-specific resistance mechanism in 10-20% of relapsed patients [9][10][11][12][13] . In single cases, IMiD-resistance was found to be caused by IKZF1 mutations at the critical degron region, which blocks IMiD-induced IKZF1 degradation 10 . DNA sequencing of heavily pre-treated multiple myeloma patients identified additional recurrent mutations and aberrations enriched at relapse including homozygous inactivation of tumor-suppressor genes TP53, RB1, FAM46C, BIRC3, TRAF3 [14][15][16] . However, only few of these aberrations have been directly linked to the activity of IMiDs or other drugs used in multiple myeloma 17 . Furthermore, inactivating mutations in tumor-suppressor genes are in general not amenable to pharmacologic interventions. Gene expression profiling (GEP) has found an enrichment of the GEP70 prognostically high-risk signature in relapsed cases 15,18,19 . Like genetic alterations, this signature was not associated with a specific treatment. In aggregate, these previous studies imply that genetic alterations alone do not fully explain the occurence of drug resistance in multiple myeloma. In addition, protein abundance and activity frequently cannot be inferred from RNA expression analyses due to post-transcriptional regulation mechanisms in general 20 , and in cancer in particular, due to complex compensation effects of genetic alterations on the protein level [21][22][23] . Proteomic profiling in cell lines and pooled patient samples has been successfully applied to study drug resistance mechanisms in patients with hematological disorders such as FLT3 inhibitor-resistant acute myeloid leukemia 24 or bortezomib refractory multiple myeloma 25,26 .
Here, we apply quantitative proteomic analyses in paired, longitudinal primary multiple myeloma samples and identify CDK6 upregulation as a non-genetic resistance mechansim for IMiDs in multiple myeloma that can be overcome by pharmacologic intervention.

Results
Quantitative proteomic analysis identifies deregulated protein abundance levels in relapsed multiple myeloma. To identify deregulated proteins in relapsed multiple myeloma, five patients with available longitudinal bone marrow samples were included in our study. Patients progressed during (N = 4) or shortly after (N = 1) lenalidomide-comprising treatment ( Supplementary  Fig. 1). Paired bone marrow samples obtained pre-treatment and at relapse were lysed, trypsin digested, labeled with isobaric tandem mass tags (TMT) and analyzed with quantitative mass spectrometry (Fig. 1A).
In total, we quantified 6094 proteins with at least two peptides without missing values across all samples (Supplementary Data 1). Using an FDR cutoff of 0.10, we found 130 proteins upregulated and 228 proteins downregulated in the relapse vs. pre-treatment samples (Fig. 1B). The top six upregulated proteins ranked by FC and FDR were TRIP13, RRM1, NCAPD2, NCAPH, MORF4L1, and CDK6 (Table 1), and the six most downregulated proteins were UPRT, DNAJC1, FCRL2, AUH, HYI and HID1 (Table 2). Although all five patients received lenalidomide and dexamethasone during their treatment, we did not detect changes in proteins involved in the mechanism of IMiDs (CRBN, DDB1, IKZF1, IKZF3, IRF4, BSG) 2,3,27 or the glucocorticoid receptor NR3C1 (Supplementary Fig. 2A). However, several of the top upregulated proteins have been previously found to be implicated in multiple myeloma and were further pursued: the ATPase TRIP13 promotes progression of B-cell malignancies 28 and is part of the validated GEP70 and GEP5 multiple myeloma gene expression high-risk signature 29,30 ; RRM1 plays a role in DNA synthesis and repair, is essential for multiple myeloma cell proliferation and its expression is linked to shorter survival 31 ; the cell-cycle regulator CDK6 is dysregulated in multiple myeloma and CDK6 inhibitors have shown activity in multiple myeloma in early clinical trials [32][33][34] . Protein analyses by western blot in an independent cohort with four samples obtained from patients at first diagnosis and nine samples from patients obtained at relapse confirmed that CDK6, TRIP13, and RRM1 proteins are more frequently detected in the relapse samples as compared to the pretreatment samples (Fig. 1C, Supplementary Fig. 1).
Comparison of proteome, phosphoproteome, and RNA expression analyses. To determine whether the differential expression of the proteins was accompanied by changes in RNA expression levels, we performed RNA sequencing of the five paired samples analyzed by proteomics (Supplementary Data 2). The top upregulated RNAs at relapse versus pre-treatment samples were ADGRG3, FCAR, CAMP and G0S2. Only two of the downregulated RNA transcripts, PAIP2B and ZBTB20 had an FDR below 0.1 ( Supplementary Fig. 3A).
The general correlation of protein and RNA expression changes between pre-treatment and relapse among all protein/RNA pairs was weak with a median Pearson correlation coefficient (PCC) of 0.34 with a high degree of variation (range −1 to 1) ( Supplementary  Fig. 3B). Of the top upregulated proteins, the mitosis regulatory protein TRIP13 showed the highest level of correlation for RNA/ protein expression (PCC = 0.84), followed by NCAPH (0.84) and NCAPD2 (0.67). RRM1 and CDK6 had an RNA to protein correlation of 0.6 and 0.39, respectively (Fig. 1D).
In addition to analyzing the global proteome, we also performed an immobilized metal affinity chromatography (IMAC) phosphopeptide enrichment with 9 of the 10 samples. We detected 24,796 phosphopeptides derived from 5698 proteins ( Supplementary Fig. 3C, Supplementary Data 1). In total, 134 phosphopeptides passed the 0.12 FDR significance cutoff. The majority of the proteins, that the significant phosphopeptides originated from, were also detected in the global proteome analysis of patient samples (92 out of 112). However, only 15 of the significant phosphopeptides belonged to proteins that were also significantly regulated on the global protein level.
The complementary nature of the different datasets was also reflected by highly significant single sample gene set enrichment analysis (ssGSEA) signatures observed in the phosphoproteomic data and, to a lesser extent, in the proteomic data (Supplementary Fig. 3D). SsGSEA revealed upregulation of cell cycle-related, replication, and chromosome maintenance signatures in relapse samples and downregulation of phosphorylation, ATP synthesis, N-glycan biosynthesis and unfolded protein response signatures. The gene signatures significantly enriched in the proteomic and phosphoproteomic data were not corroborated by the RNA sequencing data, indicating post-translational regulatory mechanisms.
Protein expression of CDK6, TRIP13, and RRM1 is independent of CRBN. As impaired CRBN-CRL4 E3 ligase activity due to mutation, deletion or downregulation leads to altered IMiDsensitivity in multiple myeloma cell lines and patients 9-11 , we examined its status in the five patients included in our proteomic analyses. No difference in CRBN RNA, protein or phosphorylation levels were observed between pre-treatment and relapsed patient samples (Fig. 1D). Four out of five patient samples were analyzed also by exome sequencing and none was found to harbor mutations in members of the CRBN-CRL4 E3 ligase complex 17 . In our independent patient sample cohort, we did not observe any correlation between CRBN and CDK6, TRIP13, or RRM1 protein levels (Supplementary Fig. 2B). Consistently, CRISPR/Cas9-mediated knockout of CRBN in myeloma cell lines did not alter expression levels of these proteins ( Supplementary Fig. 2C). Furthermore, no association was observed between genetic alterations and CDK6, TRIP13, or RRM1 protein expression in patient samples.
CDK6 protein is upregulated in in vitro generated, lenalidomideresistant multiple myeloma cells. In order to mimic IMiD resistance in vitro, we cultured MM.1S and LP-1 cells in the presence of different concentrations of lenalidomide. Cells cultured in the presence of 100 nM lenalidomide for several weeks had enhanced levels of CDK6 protein and were partially resistant to lenalidomide, highly consistent with the findings in lenalidomidetreated myeloma patients ( Supplementary Fig. 4A).
In contrast, short-term treatment with lenalidomide or proteasome inhibitors for up to 72 h had no effect or even decreased protein levels of CDK6, showing that their expression is not directly induced by the drugs (Supplementary Fig. 4B, C).
Overexpression of CDK6 impairs IMiD sensitivity. To investigate whether upregulation of CDK6, TRIP13, and RRM1 protein levels are causally linked to drug resistance, we induced their expression in MM.1S and OPM2 cell lines using lenti-and retroviral expression vectors ( Fig. 2A, C, Supplementary Fig. 5A). TRIP13 overexpression did not alter drug sensitivity in MM.1S cells ( Supplementary Fig. 5B-F). In contrast, both CDK6 and RRM1 overexpression in MM.1S and OPM2 cells reduced sensitivity to lenalidomide as well as to pomalidomide (Fig. 2B, D, Supplementary Fig. 5B, C). The effect of CDK6 was kinasedependent since the introduction of a kinase-dead mutant CDK6 K43M 35 was not able to rescue cells from lenalidomide exposure (Fig. 2D). High CDK6 expression levels slightly enhanced sensitivity towards melphalan and dexamethasone in one cell line each ( Supplementary Fig. 6). No effect was observable for bortezomib. In aggregate, these data imply that high CDK6 and RRM1 expression selectively reduce IMiD sensitivity in multiple myeloma cell lines.
CDK6 kinase inhibition sensitizes multiple myeloma cells to IMiDs. Given that CDK6 upregulation was found in lenalidomideresistant patients and induced expression reduced lenalidomide-sensitivity, we next tested the effects of the CDK6 inhibitor palbociclib 36,37 in multiple myeloma cell lines. Palbociclib had no to moderate activity in multiple myeloma cell lines with 5 out of 10 responding (Fig. 3A, C, Supplementary Fig. 7A). However, palbociclib markedly enhanced the anti-multiple myeloma effects of IMiDs when both drugs were combined with high synergy scores ( Fig. 3A, B, Supplementary Figs. 7A, 8A, B, E). This effect was observed across all cell lines expressing CDK6 at various levels ( Supplementary Fig. 7B). Remarkably, this included multiple myeloma cell lines that are naturally insensitive to IMiDs like L363 (Fig. 3C, D) and AMO-1, and synergy was observed at low drug concentrations corresponding to plasma levels in treated patients 38 . In acquired lenalidomide-resistant cells with increased CDK6 protein levels, as well as in CDK6 overexpressing cells, the addition of palbociclib restored IMiD-sensitivity to levels similar as in parental cells (Fig. 3E, F). These data show that palbociclib treatment increases the sensitivity to lenalidomide and pomalidomide in multiple myeloma cells. In contrast, combined treatment of palbociclib and melphalan, bortezomib or dexamethasone was mostly additive ( Supplementary Fig. 8C, D).
Bifunctional PROTACs degrading CDK6 and IKZF1/3 possess intramolecular synergy. We next investigated an alternative way to inactivate CDK6 using protein degradation. Proteolysis targeting chimeras (PROTACs) are bifunctional molecules which comprise two linker-connected moieties that simultaneously bind a target protein and an E3 ubiquitin ligase 39 . Like IMiDs, PRO-TACs hijack E3 ubiquitin ligases and induce ubquitination and degradation of the target protein. We and others have recently described PROTACs that effectively target CDK6 for proteasomal degradation through hijacking the CRBN-or von Hippel-Lindau (VHL) E3 ligase (Fig. 4A, B) 40,41 . We tested the anti-proliferative effects of the CDK6-selective, CRBN-recruiting PROTAC BSJ-03-   41 . In contrast, combination of IMiDs with a CRBNhijacking, CDK6-specific PROTAC (BSJ-03-123) showed antagonistic effects that are likely due to the competition for the CRBN E3-ligase (Supplementary Fig. 9). We next tested a CRBN-hijacking, pomalidomide-based PROTAC, YKL-06-102, that retains the activity of pomalidomide and potently induces both degradation of CDK6 and IMiD neosubstrates IKZF1 and IKZF3 40 . YKL-06-102 significantly reduced viability in all ten multiple myeloma cell lines tested, including those with a low IMiD sensitivity ( Fig. 4E and F, Supplementary Fig. 10). These results show that CRBN-hijacking PROTACs targeting CDK6, IKZF1, and IKZF3 simultaneously are highly effective in multiple myeloma cells through intramolecular synergy.
Combination treatment of pomalidomide and palbociclib is highly effective in vivo. To test whether the combination of IMiDs with CDK6 inhibition has therapeutic efficacy in vivo, we conducted a study in the MM.1S xenograft model (Fig. 5A). MM.1S cells were injected subcutaneously and mice were randomized for treatment groups after 19 days when tumors reached 200 mm 3 . Treatment was performed for 17 days, with pomalidomide and palbociclib being orally administered at 5 and 50 mg/ kg, respectively. Pomalidomide and palbociclib as monotherapy significantly delayed tumor growth, while combination therapy reduced tumor volumes below detection limits after 2 weeks (Fig. 5B, Supplement Fig. 11). After cessation of treatment, tumor growth resumed, indicating that prolonged treatment with both drugs is necessary to prevent multiple myeloma relapse. The potent suppression of tumor growth in the combination group translated to a significant improvement in surivival as compared to mice that recieved pomalidomide or palbociclib alone (Fig. 5C).
Intraperitoneal application of the CDK6/IKZF1/IKZF3-degrading PROTAC YKL-06-102 at a maximal applicable dosage of 5 mg/kg per day (due to low solubility) significantly delayed tumor growth as compared to control treated mice, yet the effect was not better than pomalidomide or palbociclib alone, likely due to lower bioavailability in vivo ( Supplementary Fig. 11).
The synergistic effect of IMiDs and CDK6 inhibition is independent of RB1 and cell cycle progression. RB1 is one of the major substrates of CDK6 and palbociclib treatment resulted in reduced phosphorylation of RB1 and G1 cell cycle arrest in MM cell lines, consistent with previous studies in cancer 42,43 . Heterozygous chromosome 13q/RB1 deletions are among the most frequent genetic alterations in multiple myeloma and complete loss is observed in heavily treated patients, implying that it contributes to drug resistance 14 . We therefore tested whether RB1 knockout through CRISPR/Cas9 affects sensitivity to pomalidomide, palbociclib or the combination treatment with CDK6 inhibition or degradation in MM.1S cell line. In line with studies in breast cancer, RB1 knockout reduced sensitivity to palbociclib ( Supplementary Fig. 12). However, the synergistic effects of CDK6 inhibition and IMiDs were retained in RB1 knockout cells, demonstrating that the sensitization to IMiDs is independent of functional RB1.
CDK6 inhibition reverses a relapse-associated protein signature. To investigate the effects of CDK6 inhibition and the basis for the synergy with IMiDs in multiple myeloma, we  Fig. 13A, B).
When we compared the significantly regulated proteins in patient samples and drug-treated MM.1S cells, we observed a striking pattern where proteins upregulated in patient samples at  75 . Cell viability is normalized to respective DMSO conditions. Data represent the mean ± SD of biological triplicates. One-way ANOVA is applied. P values are displayed as follows: n.s. = P > 0.05; *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001; ****P ≤ 0.0001. Source data are provided as a Source Data file. relapse were downregulated by CDK6 inhibition or degradation in MM.1S cells and vice versa (Fig. 6B, Supplementary Data 4). In detail, we found an overlap of 37 proteins upregulated in the relapse samples and downregulated by CDK6 inhibition (cluster a), as well as 54 proteins significantly downregulated in relapse and upregulated by CDK6 inhibition (cluster c) ( Supplementary  Fig. 13C). Among the CDK6 regulated proteins included in the signature were the two top upregulated proteins in relapse, TRIP13 and RRM1 30,31 . Other proteins upregulated in relapse and targeted by CDK6 inhibition are comprised of cell cyclerelated genes (CDK2, MCM3/5, NCAPH, NCAPD2, PDCD2L, USP16), DNA damage regulators (PAXIP1, RAD18, and MLH1) and transcriptional/ epigenetic regulators (DNMT1, BTF3, EDRF1, KEAP1, MORF4L1). Upregulation of the genome integrity safeguard PAXIP1 links CDK6 to a factor that has been identified to be highly selective for myeloma cell survival in the Dependency Map studies 44  We also observed lower levels of IKZF1/3 and MYC in the combined treatments as compared to single treatments alone. (Fig. 6C, Supplementary Fig. 14A, B). The enhanced inactivation of essential multiple myeloma transcription factors likely contributes to the synergistic effects of IMiDs and CDK6 inhibition.
Identification of CDK6 substrates by phosphoproteomics in multiple myeloma. To identify CDK6 substrates in multiple myeloma, we performed phosphoproteomics analyses in MM.1S cells treated with palbociclib or the CDK6-specific PROTAC BSJ-03-123. We identified 29,811 phosphopeptides across all samples of which 122 phosphopeptides on 96 proteins were significantly downregulated after 3 h of palbociclib or BSJ-03-123 treatment (0.1 FDR). In addition, phosphorylation status of 558 proteins (1022 phosphopeptides) not regulated on global protein level was specifically downregulated after 24 h palbociclib treatment, with an overlap of 45 proteins to the 3 h timepoint, which we deemed particularly interesting, since these phosphosites were downregulated at an early time point and in a sustained manner (Supplementary Data 3, Supplementary Fig. 15A). The effect of palbociclib and the CDK6-specific PROTAC BSJ-03-123 on phosphopeptide levels was highly correlated ( Supplementary  Fig. 15B). Identified phosphorylation targets comprised the known CDK4/6 substrates RB1, RBL1, RBL2, and CDKN1A as well as several proteins that were deregulated in relapse samples (DNTM1, GMPS, KLHDC4, NCAPD2, NCAPH, GLYR1, NOP56), providing a functional link between CDK6 kinase function and CDK6-regulated protein levels. In addition, lower phosphorylation levels were observed in palbociclib treated cells in several transcriptional regulators including RNF169 and ZBTB38, as well as known CDK6 interactors and substrates JUN and FOXM1 (Supplementary Data 3) 46,47 . However, for many proteins deregulated on the protein level, including RRM1 and TRIP13, we did not detect lower phosphorylation levels after CDK6 inhibition, indicating that these proteins are not direct CDK6 substrates. RT-qPCR analyses revealed that TRIP13, RRM1 and other proteins tested had reduced mRNA levels after CDK6 inhibition or inactivation, implying CDK6-dependent transcriptional regulation as has been previously reported (Fig. 6D) 46 .

Discussion
Drug resistance is one of the biggest challenges in cancer therapy. Genetic alterations revealed in comprehensive sequencing studies do not sufficiently explain the emergence of resistance in most cases, implying non-genetic mechanisms 48 . Here we applied an integrated proteomics and transcriptomics approach in primary multiple myeloma matched pre-treatment/resistant samples that identified and validated a targetable CDK6 governed protein resistance signature. Only a few proteomic studies of multiple myeloma have been previously published 26,49,50 . Our dataset containing over 6000 proteins and 20,000 phosphopeptides represents a deep proteomic study of longitudinal samples from relapsed multiple myeloma patients. It provides proof of principle for feasibility and clinical relevance of in-depth proteomic analysis in multiple myeloma. We observed a low correlation of protein and RNA levels implying a high degree of post-  transcriptional and post-translational regulation in multiple myeloma. Indeed, the most upregulated and functionally relevant proteins identified by our proteomics approach would have been missed by RNA sequencing. This is in line with previous studies in other tumors highlighting that protein level and activity cannot necessarily be inferred from RNA expression 21,51,52 .
Unexpectedly, proteins known to be involved in pathways targeted by routinely used myeloma drugs like CRBN, IKZF1, proteasome subunits, or the glucocorticoid receptor were not found deregulated at relapse. However, several of the proteins we identified as significantly deregulated have been previously implicated in multiple myeloma. TRIP13, an AAA-ATPase involved in genome stability 53 , is among the top-ranking genes included in the GEP70, a highly validated high-risk RNA expression signature in multiple myeloma 18,29,30 . TRIP13 overexpression is a driver of B-cell malignancies 28 and is functionally linked to drug resistance in multiple myeloma and other cancers 30 , and inhibition of TRIP13 is toxic to multiple myeloma cells 54 . RRM1, together with RRM2, forms the ribonucleotide reductase heterodimeric tetramer that is involved in DNA damage repair 55 . Multiple myeloma cells were shown to be dependent on RRM1 and high RRM1 RNA expression is associated with an adverse outcome 31 . In line with this, we found RRM1 overexpression to reduce IMiD sensitivity in multiple myeloma cells.
CDK6, together with CDK4, is a vital regulator of the cell cycle through regulation of RB1 42,56-58 . CDK6 has been found  Our data imply that the effects of CDK6 in multiple myeloma depend on its kinase function but are independent of RB1, the CDK4/6 substrate which is most relevant for other cancers. Our proteomic analyses in multiple myeloma cell lines revealed that CDK6 governs many of the proteins deregulated at relapsed/ resistant disease, either through direct phosphorylation of proteins or by indirect alterations of the transcriptional activity of genes including TRIP13 and RRM1, and is thus a master regulator of a relapse-associated program. The CDK6-dependent relapse signature contains proteins connected to DNA damage repair, cell cycle, and metabolic pathways including electron transport chain, energy transfer, and fatty acid metabolism. Antagonizing this signature by CDK6 inactivation in conjunction with IMiDs resulted in a significant and fast downregulation of the essential transcription factor MYC, further highlighting the synergistic effects of IKZF1/3 and CDK6 inactivation. While many of the relapse-associated proteins are regulated by CDK6, the mechanism that drives elevated CDK6 protein expression itself is not clear. We couldn't find any associations with genetic alterations and the lack of correlation with RNA expression implies post-transcriptional and/ or post-translational mechanisms. This, as well as the exact downstream functions of CDK6 in resistant multiple myeloma, needs to be evaluated in detail in future studies.
Several CDK4/6 inhibitors are approved by the FDA and EMA for advanced breast cancer and are being investigated in hematologic malignancies including acute lymphoblastic leukemia 60 and acute myeloid leukemia 33,61,62 . In multiple myeloma, a phase I/II clinical trial that combined palbociclib with bortezomib and dexamethasone showed that this regimen is feasible and has antitumor activity 63 . Here, we show that combining CDK6 inhibition with IMiDs is particularly synergistic in multiple myeloma. Although both drugs cause cytopenia as an unwanted side-effect, our data suggest that low dosages of palbociclib may be sufficient to sensitize multiple myeloma cells to IMiDs. PROTACs have recently entered clinical trials 64 and PROTACs with combined IKZF1/IKZF3/CDK6 activity provide an attractive alternative with single drug synergistic effects. CDK6 protein levels may provide a biomarker to identify patients who would benefit most from this therapeutic combination. In conclusion, our results identify CDK6 as a master regulator in treatmentresistant relapsed multiple myeloma and provide a strong rationale for further investigating CDK6 inhibition together with IMiDs in multiple myeloma.

Methods
Study cohort. Five multiple myeloma patients with paired pre-treatment and relapsed samples were included in the proteomics study. Additional 13 samples from an independent cohort of newly diagnosed or relapsed/refractory multiple myeloma patients were analzyed with western blotting. Patient characteristics and treatment are summarized in Supplementary Fig. 1. All samples were obtained from the iliac crest of patients and were CD138+ enriched by MACS (Miltenyi, Cologne, Germany). For patients #1, #3, #4, and #5 whole-exome sequencing data is reported in a previous study 17 . All patients provided written informed consent according to the Declaration of Helsinki and the study was approved by the institutional review board (IRB) of Ulm University.
Treatment of MM1S for proteomic and phosphoproteomic analysis. MM.1S WT cells were subjected to respective treatments for either 3 or 24 h in biological duplicates. Pomalidomide treatment was administered at a final concentration of 10 µM, while palbociclib, BSJ-03-123, and YKL-06-102 were administered at 1 µM. Control cells were treated with solvent only. Cells were pelleted and washed three times with cold 1× PBS.
Proteomics and phosphoproteomics sample preparation. Samples for global proteome and phospho-proteome profiling were analyzed with isobaric TMT 21 . Samples were lysed at 4°C with urea lysis buffer (8 M urea, 50 mM Tris (pH 8), 150 mM NaCl) supplemented with protease inhibitors (2 µg/ml aprotinin, 10 µg/ml leupeptin 1 mM phenylmethylsulfonylflourid) and phosphatase inhibitors (10 mM NaF, phosphatase inihibitor cocktail 1 and 2, Sigma Aldrich). Extracted proteins were reduced with 5 mM dithiothreitol for 1 h and alkylated with 10 mM iodoacetamide for 45 min in the dark. Sequencing grade LysC (Wako) was added at a weight to weight ratio of 1:50. After 2 h, samples were diluted 1:4 with 50 mM Tris-HCl pH 8 and sequencing grade trypsin (Promega) was added at a weight to weight ratio of 1:50. Digestion was completed overnight and subsequently acidified samples were desalted with Sep-Pak C18 cc Cartridges (Waters). Dried samples were resuspended in 10 mM HEPES (pH 8.5) and peptide concentration was determined. For relative quantification of proteins in multiple myeloma patient samples, 10 µg peptides for each sample were labeled with 10-plexing TMT (TMT-10; Thermo Fisher Scientific) and combined into one TMT plex. For quantification of phosphopeptides in multiple myeloma patient samples, 150 µg of peptides for the samples 1-9 and 23 µg of starting material for sample 10 were labeled with TMT-10 reagents and combined into a second plex. TMT-channel annotation of patient proteomic data is available in Supplementary Data 1. For global and phosphoproteome analysis of MM.1S drug-treated cells, 200 µg peptides of each sample were labeled with 11-plexing TMT (TMT-11; Thermo Fisher Scientific) and the samples were randomly distributed into two TMT plexes. For the MM.1S experiment, an internal reference sample composed of equal amounts of peptide material from all samples was included in each TMT 11-plex to provide a standard for relative quantification. TMT-channel annotation of MM.1S proteomic data is available in Supplementary Data 3. Combined TMT samples were fractionated offline into 24 fractions using high-pH reversed phase chromatography. For analysis of the global proteome, 500 ng peptides of each fraction were analyzed with LC-MS/MS. The remaining material was pooled into 12 fractions for each plex and subjected to phosphopeptide enrichment with immobilized metal affinity chromatography (IMAC) automated on an AssayMap Bravo System (Agilent) 65 .
Liquid chromatography mass spectrometry. Mass spectrometry raw data was acquired on a Q-Exactive HF-X connected to an EASY-nLC 1200 system (both Thermo Fisher Scientific). Each HpH fraction was measured individually as a single LC-MS injection. Samples were separated online on a 25 cm column packed inhouse with C18-AQ 1.9 µm beads (Dr. Maisch Reprosil-Pur 120). A gradient of mobile phase A (0.1% formic acid and 3% acetonitrile in water) and mobile phase B (0.1% formic acid, 90% acetonitrile in water) was used to separate the samples online at a flow rate of 250 µl/min. Mobile phase B was ramped from 4% to 30% in the first 88 min, followed by an increase to 60% B in 10 min and a plateau of 90% B for 5 min. Temperature of the column was kept constant at 45°C. MS data was acquired in data-dependent acquisition and profile centroid mode. MS1 scans were acquired at 60,000 resolution, scan range of 350-1500 m/z, automatic gain control (AGC) target value of 3e6 and maximum injection time (IT) of 10 ms. The 20 most abundant ions were picked for fragmentation with normalized collision energy (NCE) set to 32 and 0.7m/z isolation window. MS2 scans were acquired at 45,000 resolution, fixed first mass 120 m/z, AGC target value of 3e5 and maximum IT of 86 ms. Ions with charge state 1, 6 or higher were excluded from fragmentation and dynamic exclusion was set to 30 s. LC-MS parameters for phosphoproteomic analysis was the same with the exception of MS2 maximum IT that was set to 120 ms.
Mass spectrometry raw data searching with MaxQuant. Raw data was analyzed with MaxQuant (Version 1.6.3.3) 66 and searched against the human reference proteome (UP000005640) downloaded from UniProt in 01/2017 (https://ftp.uniprot.org/ pub/databases/uniprot/previous_releases/) and default protein contaminants included in MaxQuant. PIF filter was set to 0.5. Fixed modifications were set to carbamidomethylation of C. Variable modifications were set to M-oxidation and acetylation of protein N-termini including neo protein N-terms after cleavage of first methionine. A maximum of 5 modification per peptide were allowed. N-terminal acetylation and M-oxidation were used in protein quantification (unmodified counterpart discarded). Unique and razor peptides were used for quantification. TMT correction factors supplied by the manufacturer were applied.
Mass spectrometry data analysis. Analysis of MaxQuant output was performed with R(4.0.3) and R studio (Version 1.3.1093) statistical software environment. Statistical analysis in R was aided by the ProTIGY application provided by the Broad institute on GitHub (https://github.com/broadinstitute/protigy) 67 . Protein group files were filtered for reverse hits, potential contaminants and proteins only identified by site. Resulting proteins were filtered for proteins identified with at least two peptides and at least one unique peptide. Phosphoproteomic data was analyzed in a separate MaxQuant run and filtered for valid values across all samples. Due to limited protein amounts of the patient 5 diagnosis sample, patient 5 was excluded from the analysis of the phosphoproteomic data. For the analysis of multiple myeloma patient samples, log2 ratios (relapse/diagnosis) of proteins and phosphopeptides were calculated for each patient and normalized with median-MAD normalization. The processed data was analyzed with a two-sided moderated one-sample t-test (limma package) 68 . Prior to applying multiple testing correction with the Benjamini-Hochberg method, data was subjected to reproducibility filtering to remove outliers (global proteome 99% confidence interval; phosphoproteome 95% confidence interval with linear mixed effect models). As expected, the phosphoproteome showed higher variability compared to the global proteome, which is more stable towards dynamic modifications. We therefore applied a stronger outlier filter and more relaxed cutoff for the phosphoproteomic data. A cutoff of 0.1 and 0.12 FDR was implemented for global proteome (Fig. 1B) and phosphoproteome ( Supplementary Fig. 3C) analysis of patients samples respectively. For the analysis of MM.1S drug treated cells, the log2 ratios (sample/internal standard) were calculated and normalized with median-MAD normalization. The resulting ratios were subsequently normalized to the control sample and analyzed with a two-sided moderated one-sample t-test. Resulting p-values were corrected for multiple testing with the Benjamini-Hochberg method. MM.1S phosphoproteomic data was subjected to reproducibility filtering (99% confidence interval) with Bland-Altman filtering to remove outliers prior to p-value correction. For MM1.S derived data, a 0.1 FDR significance cutoff was applied. The Phosphosite plus (PSP) database was used for identifying known CDK6 targets 69 . The processed proteomics data as well as TMT channel annotation is available in supplementary data files. The mass spectrometry proteomics data and search results have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository 70 with the dataset identifier PXD021265.
RNA-sequencing. Library preparation was performed from 100 ng of input total RNA using the TruSeq Stranded Exome RNA Kit (Illumina, San Diego, CA, USA) according to the manufacturer's instructions. The pooled RNA libraries were sequenced on an Illumina HiSeq2000 with 50 bp single-end reads with an average coverage of 36.6 × 10 6 reads per sample. RNA-Seq data were aligned and quantified with STAR 71 and mRNA reads were identified using an in-house analysis pipeline detecting exons in a shuffled order. To increase comparability with proteomics data, TPM RNAseq data was processed like TMT-proteomics data and median-MAD normalized log2 ratios (relapse/diagnosis) were analyzed with a two-sided moderated one-sample t-test. RNA transcripts were matched with the protein groups file and Pearson correlation of individual genes to proteins was calculated across all 10 samples (median normalized values). Processed RNAseq data is available in the supplements. Gene set enrichment analysis was performed using GSEA and the Molecular Signatures Database (MsigDB) 72 . RNA-sequencing data is available on Gene Expression Omnibus (accession number GSE162403).
Fluorescence in situ hybridization (FISH). Fluorescence in situ hybridization was performed on plasma cells isolated from patients. Genetic regions of interest specific for the diagnosis for multiple myeloma and their translocation partners were detected. FISH was performed according to standardized protocols using commercially available probes (Abbott Laboratories, Chicago, United States; MetaSystems, Altlussheim, Germany) 73 .
Tumor-bearing mice were randomized after 19 days when tumors reached 200 mm 3 and treatments were started. Pomalidomide was administered at 5 mg/kg while palbociclib was administered at 50 mg/kg. Both drugs were administered daily via oral gavage. YKL-06-102 was administered daily at 5 mg/kg via intraperitoneal injections. Treatment was performed for 17 days and thereafter, remaining animals were observed for another 11 days. Mice were euthanized when tumor volume reached 1200 mm 3 . The animal study was conducted in compliance with the United Kingdom Coordinated Committee on Cancer Research guidelines and have been approved and authorized by the Landesamt für Gesundheit und Soziales, Berlin, Germany (approval No. G 0333/18).
Generation of lenalidomide-resistant cell lines. MM.1S and LP-1 myeloma cells were cultured for over 3 weeks in the continuous presence of lenalidomide at 10, 100 and 1000 nM. Lenalidomide-resistance was confirmed with drug sensitivity assay with lenalidomide treatment.
Real-time quantitative PCR (RT-qPCR) were performed with SYBR Green Master Mix (Thermo Fisher Scientific, Waltham, USA) and StepOnePlus Real-Time PCR System. qPCR standard protocol was 95°C for 5 min, followed by 45 cycles of 95°C for 10 s and 60°C for 30 s. Expression fold change was calculated with 2 −(ΔΔCt) .
Cell viability assay. Cells were seeded in 96-well or 384-well plates with respective treatments and plates were incubated at 37°C for 96 h. Cell viability readout was measured using CellTiter-Glo ® Luminescent Cell Viability Assay (Promega, Madison, USA) and measured with POLARStar Omega plate reader (BMG Lab-Tech, Ortenberg, Germany). All conditions were normalized to the dimethyl sulfoxide-(DMSO) (Sigma Aldrich, Taufkirchen, Germany) treated control. Data represents the mean ± SD of biological triplicates.
Software and statistical analysis. Mass spectrometry raw data was searched with MaxQuant (Version 1.6.3.3) 66 . RNA-Seq data were aligned and quantified with STAR 71 . Analysis of proteomics and RNAseq data was performed with R(4.0.3) and R studio (Version 1.3.1093) statistical software environment. Statistical analysis of omics data in R was aided by the protigy application provided by the Broad institute on GitHub (https://github.com/broadinstitute/protigy). Statistical and graphical analyses of in vitro and in vivo experiments were performed with Prism version 8 and 9.1.0 (GraphPad Software, San Diego, CA, USA). Statistical differences were analyzed by one-way ANOVA, unpaired t tests with Welch's correction and log-rank Mantel-Cox tests. P values are displayed as follows: n.s. = P > 0.05; *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001; ****P ≤ 0.0001. For the generation of synergy maps, SynergyFinder was used, and ZIP and Bliss reference models were utilized 75 .
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The mass spectrometry proteomics data and search results generated in this study have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository 70 with the dataset identifier PXD021265. The human reference proteome (UP000005640) was downloaded from UniProt in 01/2017 (https://ftp.uniprot.org/pub/databases/ uniprot/previous_releases/). The RNA-sequencing data generated in this study are available on Gene Expression Omnibus under accession number GSE162403. Processed proteomics data are available in Supplementary Data 1 (patient proteomics data) and Supplementary Data 3 (MM1S proteomics data); processed patient RNAseq data is available in Supplementary Data 2. The remaining data are available within the Article, Supplementary Information or Source Data file. Source data are provided with this paper.