Inhibition of CDK8/19 Mediator kinase potentiates HER2-targeting drugs and bypasses resistance to these agents in vitro and in vivo

We have used the Kaplan–Meier plotter (23) to investigate correlations between CDK8 RNA expression (based on microarray data with long-term followup) and RFS in HER2⁺ BrCa patients, stratified into groups that did or did not receive treatment after sample collection. Treated patients showed much shorter RFS if their tumors belonged to the upper tertile for the expression of CDK8 (hazard ratio [HR] = 1.87, P = 3.4e-05), but there was no correlation with RFS among untreated patients (SI Appendix, Fig. S1). To compare the predictive power of CDK8 relative to other genes, the same analysis was repeated for all the genes in HER2⁺ BrCa-treated patients and the genes were ranked based on their achieved HR values. CDK8 ranked in the top 1% (#77 of 10,091) of genes by the correlation of its higher expression with shorter RFS. We then used Pharos analysis (24) to identify “druggable” genes, categorized as Tchem (small molecules are known to modulate the protein) or Tclin (approved drugs exist for this target). RFS correlations for the top 50 druggable genes are shown in SI Appendix, Table S1. CDK8 ranked #15 of 1,377 Tchem targets and above all but one of 498 Tclin targets. We have also tested RFS correlations for CDK19 (the paralog of CDK8) and CDK8/19-interactive proteins CCNC, MED12, and MED13. CCNC and MED13 showed strong RFS correlations in treated HER2⁺ BrCa patients (ranking in the top 5% of all genes), with a weaker correlation for CDK19 but no significant correlation for MED12 in concordance with previous findings for this gene in other BrCa subtypes (20). As with CDK8, no significant survival correlations for other CDK module subunits were observed among untreated patients (SI Appendix, Fig. S1). These observations suggest that CDK8 with its interactive genes may play a unique role as a determinant of treatment response in HER2⁺ BrCa patients.

We have tested the interactions between two anti-HER2 drugs, the small-molecule kinase inhibitor lapatinib and the monoclonal antibody trastuzumab, and two chemically distinct selective CDK8/19i, senexin B (10) and SNX631 (also known as 15u) (25). SI Appendix, Table S2 provides the chemical formulas and compares the activities of senexin B and SNX631 in a battery of cell-free and cell-based assays for CDK8/19 inhibition (26). SNX631 was 6 to 10 times more potent than senexin B in all the assays except for the DiscoverX active-site–dependent competition binding assay. The latter assay uses recombinant CDK8 and CDK19 proteins without their cyclin partner CCNC; as previously discussed, the absence of CCNC has differential effects on the binding affinities of different CDK8/19i (19). Five HER2⁺ BrCa cell lines were treated for 7 d with HER2 and CDK8/19i individually and in fixed-ratio combinations (Fig. 1 A–E). These cell lines included parental HCC1954 (HCC1954-Par) cells (Fig. 1A) and their derivative HCC1954-Res selected for acquired resistance to lapatinib (27) (Fig. 1B), JIMT-1 (Fig. 1C), SKBR3 (Fig. 1D), and BT474, the only cell line in this panel that is both HER2⁺ and ER⁺ (Fig. 1E). Synergy was determined on the basis of combination index (CI) values calculated using CompuSyn (28). SKBR3, BT474, and, to a lesser extent, HCC1954-Par are inherently sensitive to lapatinib, the other cell lines being relatively resistant. Trastuzumab had little effect on in vitro growth of most of these cell lines, except for BT474. CDK8/19i alone had only minor or no effects on the growth of most cell lines except for BT474; the effect of CDK8/19 inhibition in such ER⁺ cells is mediated through the effect of CDK8 on the transcriptional activity of ER (10). Combining HER2 and CDK8/19i increased the growth-inhibitory effect in every case (Fig. 1 A–E). Moreover, these effects were synergistic in all cases (as indicated by CI values less than 1.0), except for the additive effect of trastuzumab + senexin B combination in SKBR3 cells. Remarkably, synergy was observed not only in lapatinib-sensitive cell lines but also in cells with inherent or acquired resistance to HER2-targeting drugs. Lapatinib-selected HCC1954-Res cells showed cross-resistance to the small-molecule HER2 inhibitor neratinib relative to HCC1954-Par, but the addition of senexin B potentiated the effects of neratinib on both cell lines, reversing the acquired resistance (SI Appendix, Fig. S2).

We have also tested the effects of CDK8/19i on the development of adaptive lapatinib resistance in a previously described short-term single-step adaptation procedure (29). Both SKBR3 (Fig. 1F) and BT474 cells (Fig. 1G) were growth-inhibited by 250 nM lapatinib after 1 wk of treatment, but cell colonies growing in the presence of the drug became apparent after 4 to 8 wk. After 16 wk, extensive cell growth was observed despite the presence of lapatinib (Fig. 1 F and G), indicating drug adaptation. While senexin B alone had only a moderate growth-inhibitory effect, the addition of senexin B to lapatinib almost completely abrogated cell growth even after 16 wk (Fig. 1 F and G). Hence, combining CDK8/19i with HER2-targeting drugs has a synergistic or additive effect and both overcomes and prevents the development of resistance to the latter agents.

We have carried out RNA-Sequencing (RNA-Seq) analysis of HCC1954-Par cells treated with dimethyl sulfoxide (DMSO) (control), senexin B alone, lapatinib alone, or lapatinib + senexin B combination for 24 h. We have identified differentially expressed genes (DEGs) by the following criteria: expressed at Fragments Per Kilobase of transcript per Million (FPKM) >1 in at least one condition; fold change (FC) >1.5 between two conditions; and false discovery rate (FDR) <0.05. Lapatinib increased the expression of 224 and decreased the expression of 195 DEGs relative to the control, whereas senexin B up-regulated 61 and down-regulated 32 genes relative to the control and up-regulated 67 and down-regulated 40 DEGs relative to lapatinib. The effects of senexin B on all 419 lapatinib-regulated DEGs are shown in SI Appendix, Fig. S3A. Senexin B reversed lapatinib-induced changes in 12 DEGs (5 down-regulated and 7 up-regulated) and enhanced lapatinib-induced up-regulation of 7 other DEGs (SI Appendix, Fig. S3B). Interestingly, 2 of 12 DEGs that showed reversal of lapatinib-induced changes following the addition of senexin B, ALPP and ETV5, are among 5 genes that were up-regulated in common upon selection for trastuzumab resistance in SKBR3 and BT474 cells (30), suggesting that CDK8/19 inhibition may suppress transcriptional pathways associated with resistance to HER2-targeting drugs.

We have used gene set enrichment analysis (GSEA) (31) to determine which of the 50 hallmark pathways were differentially affected by lapatinib alone and by lapatinib + senexin B combination. GSEA plots for the most prominently affected pathways are shown in Fig. 2. Among these, the PI3K/AKT/mTOR pathway geneset was weakly affected by lapatinib alone but showed a strong negative correlation for the combination-treated samples; the effects on differentially affected genes of this pathway are shown in SI Appendix, Fig. S3C and Table S3. A different impact was observed with two overlapping genesets that are regulated by interferon γ (IFNγ) and IFNα. These pathways were strongly enriched in lapatinib-treated samples but showed a negative correlation when senexin B was added to lapatinib (Fig. 2); the effects on the combined DEGs of these pathways are shown in SI Appendix, Fig. S3D and Table S4. Based on GSEA analysis, we undertook functional analysis of the role of the most affected pathways on the synergistic interactions between HER2- and CDK8/19-targeting drugs.

To test the effect of PI3K/AKT/mTOR pathway inhibition on the synergy between HER2 and CDK8/19i, we used two chemically distinct PI3K inhibitors: pictilisib, a pleiotropic inhibitor of different PI3K isoforms, and PI3Kα-specific inhibitor alpelisib. The effect of different PI3K inhibitor (PI3Ki) concentrations on 7-d growth of HER2⁺ BrCa cell lines HCC1954-Par, HCC1954-Res, and JIMT-1 are shown in Fig. 3A (for pictilisib) and Fig. 3B (for alpelisib). Pictilisib and (to a lesser extent) alpelisib strongly inhibited the growth of all three cell lines. To analyze the effect of PI3K inhibition on the synergy between HER2 and CDK8/19i, we selected the concentrations of 250 nM pictilisib and 500 nM alpelisib, which inhibited cell growth by ∼25% or less after 7 d (Fig. 3 A and B) but significantly suppressed PI3K activity as indicated by AKT phosphorylation (Fig. 3 C and D). We then tested the effects of lapatinib, SNX631, and their fixed-ratio combinations on the same three HER2⁺ BrCa cell lines in the presence of 250 nM pictilisib or 500 nM alpelisib. Despite their different structures, both PI3Ki had essentially the same effects on the responses of all three cell lines to lapatinib and SNX631 (Fig. 3 E and F) relative to the effects of these drugs in the absence of PI3Ki (Fig. 1 A–C). PI3Ki produced a strong sensitization to lapatinib, with no apparent sensitization to SNX631. Furthermore, SNX631 and lapatinib no longer showed synergy in the presence of pictilisib or alpelisib, as indicated by CI values >1 (Fig. 3 E and F). Notably, the effects of CDK8/19i were not associated with any significant changes in AKT phosphorylation by SNX631 alone or in combination with pictilisib (Fig. 3C) or alpelisib (Fig. 3D).

INFγ and INFα signaling are regulated in part through STAT transcription factors, including STAT1 and STAT3 (32, 33). The activity of these factors is modulated in a complex manner by serine phosphorylation at position 727 (34, 35). CDK8 was shown to be capable of phosphorylating STAT3 (8, 35), a transcription factor implicated in breast carcinogenesis (36), at S727; this phosphorylation was reported to enhance the transcription-stimulating activity of STAT3 (37). STAT1 S727 phosphorylation was also found to be affected by CDK8 (8), in addition to other kinases (19). In contrast to STAT3, STAT1 has been identified as a tumor suppressor in BrCa but S727 phosphorylation of STAT1 may counteract its tumor-suppressive activity (38). We have investigated the effects of lapatinib, trastuzumab, and CDK8/19i on STAT1 and STAT3 phosphorylation at S727, in SKBR3 cells (where senexin B was used as the CDK8/19i) (Fig. 4A) and in HCC1954-Par, HCC1954-Res, and JIMT-1 cells (using SNX631 as CDK8/19i) (Fig. 4B). CDK8/19i alone decreased STAT1 S727 phosphorylation in all four cell lines. Combining lapatinib and CDK8/19i further decreased this phosphorylation in all the cell lines, whereas lapatinib alone also had noticeable effects in SKBR3 and JIMT-1 cells. Trastuzumab alone did not reduce STAT1 S727 phosphorylation in any cell line tested, but phosphorylation at this site was further reduced by combined trastuzumab–CDK8/19i treatment, as compared to CDK8/19i alone, in three of the tested cell lines. STAT3 S727 phosphorylation was also decreased by CDK8/19i (albeit to a lesser extent than STAT1 phosphorylation). Combined lapatinib–CDK8/19i treatment further decreased phosphorylation at this site in all four cell lines but combining trastuzumab and CDK8/19i augmented the decrease in phosphorylation only in SKBR3 and HCC1954-Res cells. Based on these observations, we investigated the role of STAT1 and STAT3 in the resistance to lapatinib and SNX631.

We carried out CRISPR-Cas9 knockout of STAT1 and STAT3 in HCC1954-Par, HCC1954-Res, and JIMT-1 cells. STAT3 knockout in HCC1954-Par and HCC1954-Res cells led to overexpression of STAT1, as apparent compensatory mechanism, but no reciprocal compensation was observed in these cells after STAT1 knockout (Fig. 4C). Such compensation was not seen after STAT1 or STAT3 knockout in JIMT-1 (Fig. 4C). We also generated HCC1954-Par and JIMT-1 cells with knockouts of both STAT1 and STAT3 (Fig. 4C); however, we were unable to obtain viable HCC1954-Res cells with the knockout of both genes.

We then tested the effects of STAT1 and STAT3 knockouts on lapatinib sensitivity in HCC1954-Par and JIMT-1 cells. The knockout of either STAT1 or STAT3 alone had no effect on lapatinib sensitivity (Fig. 4D) but the knockout of both STATs sensitized both cell lines to lapatinib (Fig. 4E), indicating a cooperative effect. The knockout of STAT1 + STAT3 slightly sensitized HCC1954-Par but not JIMT-1 to SNX631 alone (Fig. 4F), while reducing the synergy between lapatinib and SNX631 in HCC1954-Par (CI = 0.954) seemingly abolished such synergy in JIMT-1 (Fig. 4G; CI could not be calculated in JIMT-1 due to minimal effect of SNX631). These results implicate STAT1 and STAT3 as cooperating factors in lapatinib sensitivity and as variable determinants in the synergy between HER2- and CDK8/19-targeting agents.

We have previously found that the effects of Mediator kinase on metastatic growth of colon cancer are mediated by a microRNA (miR) (39). Some oncogenic miRs have been implicated in HER2 BrCa drug response, as well as in STAT and PI3K/AKT/mTOR signaling. In particular, miR-21 is up-regulated by and targets STAT3 (40, 41); and decreased miR-21 was correlated with better survival of HER2⁺ BrCa patients (42). Silencing of miR-21 also confers sensitivity to tamoxifen and fulvestrant in BrCa through inhibition of the PI3K/AKT/mTOR pathway (43). Another microRNA, miR-221, confers lapatinib resistance in HER2⁺ BrCa (44) and gefitinib resistance in cervical cancer (45) and has been implicated as a regulator of both PI3K/AKT/mTOR (45) and STAT pathways (46, 47). We used qPCR to measure the expression of miR-21 for both its guide strand (miR-21-5p) and passenger strand (miR-21-3p) and of miR-221 (guide strand) expression in parental and resistant HCC1954 cells. The cells were untreated or treated with lapatinib, SNX631, or their combination. As shown in Fig. 5A, all three miRs were up-regulated by lapatinib or SNX631 individually in both cell lines, but no up-regulation was observed upon treatment with the drug combination.

A prominent target of miR-21 is the tumor suppressor gene BTG2, which is involved in cell differentiation, proliferation, apoptosis, and other cellular functions (48). qPCR analysis showed that BTG2 was strongly up-regulated in both HCC1954-Par and HCC1954-Res cells by the combination of lapatinib and senexin B, with much weaker effects of individual drugs (Fig. 5B). To determine whether BTG2 expression contributes to the response to HER2 and CDK8/19i, we used shRNA transduction to decrease BTG2 expression (Fig. 5C). BTG2 knockdown made these cells more resistant to lapatinib, SNX631, and their combination (Fig. 5D), suggesting that induction of this tumor suppressor may mediate the antiproliferative effect of HER2 and CDK8/19i.

We have tested the effects of in vivo treatment with lapatinib and SNX631 in HCC1954-Par and HCC1954-Res xenografts in NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ (NSG) mice. Mice were randomized into four groups when the average tumor volume reached ∼100 to 150 mm³ and treated with vehicle, SNX631, lapatinib, or SNX631 + lapatinib combination. Fig. 6A shows the effects of the treatments on tumor volumes (Left), final tumor weights (Middle), and mouse body weights (Right) for HCC1954-Par, and Fig. 6B shows the same data for HCC1954-Res xenografts. Lapatinib significantly inhibited tumor growth in both models, although its effect was stronger in HCC1954-Par than in HCC1954-Res tumors, as expected. Remarkably, SNX631 alone significantly decreased tumor size and tumor weights in both models, despite its weak effect in vitro against the same cells (Fig. 1 A and B), indicating in vivo–specific roles of CDK8/19 in HER2⁺ tumors. The combination of lapatinib and SNX631 exhibited the strongest tumor growth inhibition in both models, significantly enhancing the effects of lapatinib and SNX631 (Fig. 6 A and B). There was no toxicity of the treatments based on mouse body weights and cage-side observations. (Fig. 6 A and B).

We carried out extensive immunohistochemical (IHC) and immunofluorescence (IF) analysis of HCC1954-Par tumors from all treatment arms; representative images are shown in SI Appendix, Figs. S4–S6. Hematoxylin and eosin (H&E) staining (SI Appendix, Fig. S4A) showed extensive tumor necrosis in all arms, including control, in agreement with previous characterization of this xenograft model (49). Both individual drugs and their combination strongly decreased staining for proliferation marker Ki67 (Fig. 6C and SI Appendix, Fig. S4B). Apoptosis (as measured by Terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) staining), was increased by lapatinib and lapatinib + SNX631 but not by SNX631 alone (Fig. 6D and SI Appendix, Fig. S4C). HER2 expression was not detectably altered by any treatments (SI Appendix, Fig. S4D) and ERα expression remained almost undetectable in all arms (SI Appendix, Fig. S4E).

Further, we analyzed the expression of STAT1 and STAT3 phosphorylated at S727, as well as total STAT1 and STAT3 expression. Although lapatinib alone appeared to decrease both total STAT1 and pSTAT1-S727 immunolabeling, the ratio of pSTAT1-S727 to tSTAT1 was not significantly different between lapatinib and control arms. On the other hand, SNX631 alone and in combination with lapatinib significantly decreased the ratio of pSTAT1-S727 to tSTAT1 (Fig. 6E and SI Appendix, Fig. S5 A and B). Both drugs, individually and in combination, decreased immunolabeling for pSTAT3-S727 and (to a lesser extent) for total STAT3 (Fig. 6F and SI Appendix, Fig. S5 C and D). These results indicate that inhibition of pSTAT3-S727 and pSTAT1-S727 may be regarded as a potentially mechanistic pharmacodynamic marker of CDK8/19 inhibition in HER2⁺ BrCa therapy.

Potentially significant effects of SNX631 + lapatinib combination were observed in the analysis of stromal elements of the tumors. In particular, lapatinib treatment significantly increased αSMA immunolabeling, a marker of tumor recruitment of stromal fibroblasts, but such an increase was not observed when lapatinib was combined with SNX631 (Fig. 6G and SI Appendix, Fig. S6A), suggesting that CDK8/19 inhibition may suppress this potentially tumor-promoting effect of lapatinib. Staining for arginase-1 (ARG1), a marker of alternatively activated (M2) macrophages that can promote tumor aggressiveness (50), was strongly decreased by SNX631 and lapatinib alone, with a slightly greater suppression of this tumor-promoting stromal component in combination-treated tumors (Fig. 6H and SI Appendix, Fig. S6B). The recruitment of endothelial cells (assessed by staining for CD31) was decreased by lapatinib and combination treatment but not by SNX631 alone (SI Appendix, Fig. S6C).

HCC1954 lapatinib-resistant cells were generated as described (27). sgRNA knockouts were carried out by lentiviral transduction of sgRNAs cloned in lentiCRISPR v2 vectors Addgene #52961 (from Feng Zheng) and Addgene #98293 (from Brett Stringer), followed by selection of transduced cells with puromycin or blasticidin, respectively. The effects of different compounds and their combinations on cell growth were measured by Sulphorhodamine B (SRB) assay after 7 d of treatment (or by acid phosphatase assay after 5 d of treatment in the case of neratinib). Synergy analysis was based on CI values calculated using CompuSyn software (28).

Expression of individual mRNAs or miRNAs in total RNA samples was measured by qRT-PCR. For miRNA analyses, total RNA was polyadenylated on the 3′ end prior to reverse transcription. RPL13A was used as the reference gene for qRT-PCR. RNA-Seq was carried out on Illumina HiSEq 3000/4000 platform for 2 × 150 bp reads, averaging ∼35 million reads per sample. DEG analysis was performed in R using the edgeR package. The GSEA analysis was performed using Broad Institute GSEA software. The RNA-Seq data in Gene Expression Omnibus is available under accession No. GSE191050. For Western blot analysis, cell lysates were prepared using protease and phosphatase inhibitors. Protein bands were quantified with ImageJ.

Female NSG mice (aged 6 wk) were inoculated subcutaneously into the right flank with 5 × 10⁶ HCC1954-Par or HCC1954-Res cells in 50% Matrigel. Tumors were measured with calipers and mice were weighed twice weekly. Once average tumor volume reached ∼150 mm³, mice were randomized to four treatment groups: control, SNX631 alone, lapatinib alone, or SNX631 + lapatinib. Lapatinib ditosylate was administered by daily oral gavage at 100 mg/kg. SNX631 was administered in a 500-ppm medicated diet (for the HCC1954-Par study) or in a 250-ppm medicated diet plus daily oral gavage supplement (5 mg/kg SNX631) (for the HCC1954-Res study). On the final day of the study, tumors were excised and weighed, and then fixed in 10% formalin and stored in 70% ethanol at 4 °C until processing. All mouse studies were approved by the University of South Carolina Institutional Animal Care and Use Committee.

Immunostaining was carried out on 10-μm sections cut from formalin-fixed paraffin-embedded tissue blocks. For immunofluorescence, confocal imaging fields were selected that were within two field widths of the tumor boundary to avoid necrotic areas deeper in the tumor. A total of 4 to 10 fields per section were imaged. ImageJ scripts were used to convert original czi files into tiff images and create maximum intensity projections (MIPs). Image thresholds were set and verified using randomly selected subsets of MIPs.

Differences between groups were validated by two-way/three-way analysis of variance (ANOVAs) with multiple comparisons, followed by t tests when validated, using GraphPad Prism 9 software. Kaplan–Meier survival analysis was carried out using www.kmplot.com on a total of 3,955 BrCas of Affymetrix microarray data, by selecting HER2⁺ samples, further stratified as untreated or treated (“untreated excluded”). High expression of CDK8, CDK19 (CDC2L6), CCNC, MED12, and MED13 was defined as the upper tertile. The same analysis was repeated in all HER2⁺ untreated excluded samples for all genes present in the gene arrays, using the JetSet best probeset. Druggable genes belonging to Tchem or Tclin categories (24) were identified using www.pharos.nih.gov.