Targeting CDK2 overcomes melanoma resistance against BRAF and Hsp90 inhibitors

Abstract

Novel therapies are undergoing clinical trials, for example, the Hsp90 inhibitor, XL888, in combination with BRAF inhibitors for the treatment of therapy-resistant melanomas. Unfortunately, our data show that this combination elicits a heterogeneous response in a panel of melanoma cell lines including PDX-derived models. We sought to understand the mechanisms underlying the differential responses and suggest a patient stratification strategy. Thermal proteome profiling (TPP) identified the protein targets of XL888 in a pair of sensitive and unresponsive cell lines. Unbiased proteomics and phosphoproteomics analyses identified CDK2 as a driver of resistance to both BRAF and Hsp90 inhibitors and its expression is regulated by the transcription factor MITF upon XL888 treatment. The CDK2 inhibitor, dinaciclib, attenuated resistance to both classes of inhibitors and combinations thereof. Notably, we found that MITF expression correlates with CDK2 upregulation in patients; thus, dinaciclib would warrant consideration for treatment of patients unresponsive to BRAF-MEK and/or Hsp90 inhibitors and/or harboring MITF amplification/overexpression.

Keywords: MITF; CDK2; Hsp90 and BRAF inhibitors; melanoma; proteomics.

Figure 1. Different cell responses upon treatment with BRAF and Hsp90 inhibitors

1. Cell viability measured on a panel of melanoma cells upon 72‐h treatment with dabrafenib (BRAFi) (±SD is plotted; n = 3).

2. Cell viability measured on a panel of melanoma cells upon 72‐h treatment with XL888 (Hsp90i) (±SD is plotted; n = 3).

3. Analysis of apoptotic cells in SK‐Mel 24 and SK‐Mel 28 by annexin V after 72‐h treatment with BRAFi (1 μM dabrafenib) or Hsp90i (200 nM XL888) alone and BRAFi (1 μM dabrafenib) and Hsp90i (200 nM XL888) combined treatment (±SD is plotted; n = 3).

4. Cell viability for SK‐Mel 28 (left panel) and SK‐Mel 24 (right panel) was measured after 72‐h treatment with other Hsp90 inhibitors (aside from XL888), such as AUY022, BIIB021, novobiocin, and 17‐DMAG. (Discrepancies in the concentration–response profiles for SK‐Mel 28 to XL888 between (B and D) can be attributed to the viability surrogate that is measured. In the case of (B), the MTS assay, quantifying metabolic activity, is used, whereas (D) is based on readout of ATP using CellTiter‐Glo. In both assays, incomplete killing occurs at the concentrations of drugs used) (±SD is plotted; n = 3).

5. ZIP model to evaluate the combined effect BRAFi and Hsp90i in SK‐Mel 28 (±SD is plotted; n = 3).

6. ZIP model to evaluate the combined effect BRAFi and Hsp90i in SK‐Mel 24 (±SD is plotted; n = 3).

Figure EV1. TPP platform employed in this study

1. Workflow of the TPP platforms (“lysate” and “intact cells”) used to measure the proteome and phosphoproteome thermal stability upon drug treatment using a TMT approach.

2. The comparison of the proteome thermal stability upon different conditions (+/− drug) enables to identify the drug targets.

3. The two layouts “lysate” and “intact cells” provide complementary information regarding the nature of the targets (primary drug targets or secondary targets; Franken et al, 2015).

4. Venn diagrams of the entries retrieved by the comparisons in different settings. Experiments were performed in two biological replicates.

5. Validation of Hsp90 and CDC37 thermal shift by Western blot.

6. The TPP workflow enables the comparison of the thermal stability of proteome and phosphoproteome of resistant versus sensitive cells to the Hsp90i XL888.

Figure EV2. TPP and phospho‐TPP protein interaction maps in sensitive and resistant cells

Protein interaction map built using Cytoscape 3.2 and Reactome as plugin (see Appendix) and the statistical significant entries generated from the comparison of the proteome and phosphoproteome of drug versus DMSO for sensitive and resistant cells in both “lysate” and “intact cell” layouts.

Figure EV3. Proteome thermal stability of resistant versus sensitive cells

Protein interaction map built using Cytoscape 3.2 and Reactome as plugin (see Appendix) and the statistical significant entries generated from the comparison of the proteome and phosphoproteome thermal stability of resistant versus sensitive cells different settings.

Figure EV4. Validation of the TPP and proteomics findings

1. Melting curves of some hits generated by the comparison of the baseline thermal stability of SK‐Mel 24 and SK‐Mel 28 in lysate and intact cell settings.

2. Validation of pPAK4 thermal shift by Western blot.

3. Principal component analysis (PCA) of the proteomics (left panel) and phosphoproteomics (right panel) results in different settings.

4. Western blot analyses of SK‐Mel 24 and SK‐Mel 28 upon BRAFi treatment show phosphorylation and activation of pERK.

5. Protein expression levels of the shared kinases statistically significant regulated in SK‐Mel 24 and SK‐Mel 28 upon treatment with BRAFi‐Hsp90i/DMSO.

6. Cell viability measurements ± doxycycline (72–96 h) for the non‐targeting scrambled shRNA (NT CTL), CDK2, and MITF conditional knockdown cell lines in DMSO (±SD is plotted; n = 3).

7. Western blots of protein expression levels of CDK2 in A375 DR1, ESTDAB 37, M026.X1.CL, and MNT‐1 DR100 upon treatment with DMSO, BRAFi, and Hsp90i at 72 h (left panel). Band intensities for the quantification of CDK2 expression levels in different cell lines in different settings were normalized against the mean of GAPDH, and DMSO treatment was used as reference (right panel).

8. Western blot of protein expression levels of AKT1 in SK‐Mel 24 and SK‐Mel 28 in different settings at 48‐h treatment.

Figure 2. Proteomics and phosphoproteomics findings

1. Scheme of the different settings (BRAFi = 1 μM dabrafenib; Hsp90i = 200 nM XL888; BRAFi+Hsp90i = 1 μM dabrafenib + 200 nM XL888) employed in this study after 48 h treatment.

2. Venn diagram of the upregulated protein entries in sensitive (SK‐Mel 24) and resistant (SK‐Mel 28) cells upon treatment with 200 nM XL888 (Hsp90i/DMSO) after 48 h (n = 3). The 240 unique entries for SK‐Mel 28 are highlighted in red, among which nine are protein kinases (upper panel). Volcano plot generated by the comparison between Hsp90i/DMSO in SK‐Mel 28 after 48 h (n = 3; lower panel).

3. Venn diagram of the upregulated phosphopeptides in sensitive (SK‐Mel 24) and resistant (SK‐Mel 28) cells upon treatment with 200 nM Hsp90i (Hsp90i/DMSO) at 48 h (n = 3). The 534 unique phosphopeptides for SK‐Mel 28 were analyzed by KEA. This bioinformatics analysis predicted CDK2 and GSK3β as upstream active kinases.

4. Effects on the cell viability after 72 h of the inhibitors (and their combinations) that target the potential entries reported in the text for SK‐Mel 28 (BRAFi = 1 mM dabrafenib; Hsp90i = 200 nM XL888; CDK2i = 200 nM dinaciclib; GSK3βi = 2 μM CHIR‐99021 HCl; PAK1/2i = 2 μM FRAX597; PAK4i = 2 μM PF‐3758309; STAT1i = 2 μM Fludarabine; CDK4/6i = 2 μM palbociclib). The red arrows highlight the settings were the cell viability falls below 50% upon drug treatment (±SD is plotted; n = 3).

5. The same rationale used in (A) was exploited for the proteomics analysis of the effects of BRAFi (1 μM dabrafenib) treatment (upper panel) after 48 h (n = 3). Volcano plot generated by the comparison between BRAFi/DMSO in SK‐Mel 28 after 48 h (n = 3; lower panel).

6. The same rationale used in (A) was exploited for the proteomics analysis of BRAFi‐Hsp90i combined therapy after 48 h (n = 3).

7. Overlap of the downregulated (upper panel) and upregulated (lower panel) kinases at proteomics level unique for SK‐Mel 28 in different settings at 48 h (n = 3). In red, the only shared “druggable” upregulated kinase CDK2 is highlighted.

8. Western blot analysis confirms the upregulation of CDK2 in different settings (upper panel). Band intensities were normalized against the mean of β‐actins, and lane 1 was used as reference (lower panel).

Figure 3. Validation of CDK2 as driver of melanoma resistance against Hsp90i and BRAFi

1. Cell sensitivity to CDK1 (Ro 3306), CDK2 (K03861), CDK5 (Roscovitine), and CDK9 (LDC000067) inhibitors ± 200 nM XL888 (Hsp90i) at 72 h in SK‐Mel 28 was analyzed by MTS assay (±SD is plotted; n = 3).

2. The induction by doxycycline of the conditional shRNA knocks down CDK2 expression levels at 72 h in SK‐Mel 28.

3. Cell viability assay of the conditional CDK2 knockdown cell upon Hsp90i (200 nM XL888; left panel) and BRAFi (1 μM dabrafenib; right panel) at 72 h in SK‐Mel 28 (n = 3; 72–96 h ± doxycycline and 72 h inhibitors; t‐test P‐value < 0.001) (±SD is plotted; n = 3).

Figure 4. Dinaciclib overcomes drug resistance in multiple cell lines

1. Cell viability was measured on a panel of 11 BRAF‐mutated cell lines in different settings at 72 h (CDK2i = 200 nM dinaciclib; Hsp90i = 200 nM XL888; CDK2i+Hsp90i = 200 nM dinaciclib + 200 nM XL888; MEKi = 100 mM trametinib; BRAFi = 1 μM dabrafenib; MEKi+BRAFi = 100 mM trametinib + 1 μM dabrafenib; MEKi+BRAFi+CDK2i = 100 mM trametinib + 1 μM dabrafenib + 200 nM dinaciclib; MEKi+BRAFi+Hsp90i = 100 mM trametinib + 1 μM dabrafenib + 200 nM XL888) (±SD is plotted; n = 3).

2. Cell viability was measured in two NRAS‐mutant cell lines upon different drug treatments at 48 h (±SD is plotted; n = 3).

3. Apoptotic status was measured in two NRAS‐mutant cell lines upon different drug treatments at 48 h (±SD is plotted; n = 3).

Figure 5. MITF upregulation leads to Hsp90i resistance

1. Overexpressed entries (in circles) that positively regulate the transcription of MITF and its downstream transcriptional targets (in red; P < 0.05; fold change ≥ 1.5) observed in our proteomics dataset (n = 3).

2. Western blot analyses confirmed our bioinformatics prediction analyses by ChEA. Note the inverse correlation between pERK and MITF expressions in SK‐Mel 28. The color of the cell pellets of SK‐Mel 28 upon treatments at 48 h with DMSO, BRAFi, and Hsp90i is shown in the lower panel.

3. The knockdown of MITF causes downregulation of CDK2 expression levels in SK‐Mel 28.

4. Cell viability assay of the SK‐Mel 28 MITF conditional knockdown upon Hsp90i (200 nM XL888; left panel) and BRAFi (1 μM dabrafenib; right panel) treatments at 72 h (n = 3; 72–96 h ± doxycycline and 72 h inhibitors; t‐test ***P‐value < 0.001) (±SD is plotted; n = 3).

5. Western blot analyses show the upregulation of MITF and its transcriptional targets CDK2 and the melanotic marker DCT upon 200 nM XL888 treatment at 72 h in A375 and M029.R.X1.CL.

6. Model of cancer plasticity that leads to resistance to both BRAFi and Hsp90i treatments by switching between different signaling pathways.

Figure 6. Analysis of CDK2 and MITF expression in CCLE and TGCA databases

1. Plot of CDK2 versus MITF mRNA abundance (log₂ TPM) among all melanoma‐derived cell lines in the CCLE. Cell lines included in this study have been labeled.

2. Plot of CDK2 versus MITF mRNA expression in skin cutaneous melanoma (SKCM) patient samples from TCGA.

3. IHC images of matched patient material show strong staining of MITF and CDK2 (co)expression in melanoma tissues (left panels). Normal tissues show low or no (co)expression in skin tissue (right panels). The data were kindly provided by the Protein Atlas Project publicly available (scale bar is 100 μm) (www.proteinatlas.org).