An increased cell cycle gene network determines MEK and Akt inhibitor double resistance in triple-negative breast cancer

Triple-negative breast cancer (TNBC) is an aggressive subtype of breast cancer with poor clinical prognosis and limited targeted treatment strategies. Kinase inhibitor screening of a panel of 20 TNBC cell lines uncovered three critical TNBC subgroups: 1) sensitive to only MEK inhibitors; 2) sensitive to only Akt inhibitors; 3) resistant to both MEK/Akt inhibitors. Using genomic, transcriptomic and proteomic datasets of these TNBC cell lines we unravelled molecular features associated with the MEK and Akt drug resistance. MEK inhibitor-resistant TNBC cell lines were discriminated from Akt inhibitor-resistant lines by the presence of PIK3CA/PIK3R1/PTEN mutations, high p-Akt and low p-MEK levels, yet these features could not distinguish double-resistant cells. Gene set enrichment analyses of transcriptomic and proteomic data of the MEK and Akt inhibitor response groups revealed a set of cell cycle-related genes associated with the double-resistant phenotype; these genes were overexpressed in a subset of breast cancer patients. CDK inhibitors targeting the cell cycle programme could overcome the Akt and MEK inhibitor double-resistance. In conclusion, we uncovered molecular features and alternative treatment strategies for TNBC that are double-resistant to Akt and MEK inhibitors.

studies in TNBC and other malignancies 13 . Novel insights in predicting the response and providing alternatives to drug-resistant tumours are thus essential in facilitating the clinical development of such inhibitors in TNBC.
In this study, comparing the efficacy of various kinase inhibitors with multiple targets, we found that TNBC cell lines respond differentially to MEK and Akt inhibitors. By assessing TNBC cell line groups that had different response patterns to these inhibitors, we have shown that Akt or MEK inhibitor sensitivity can be distinguished by the presence of mutations in the PI3K pathway and activity of the PI3K and MAPK pathway. By performing a large-scale gene set enrichment analysis, we found that Akt and MEK inhibitor double resistance is associated with elevated expression of cell cycle genes. On the other hand, MEK and Akt inhibitor double-resistant cell lines were more, but not exclusively, sensitive to inhibitors of this network, such as dinaciclib and flavopiridol, which may constitute alternative treatment options for this TNBC subgroup. Our results thus shed light on a gene network associated with resistance to MEK and Akt inhibitors and provide potential markers of response as well as alternative treatments.

Results tnBc cell lines cluster into three groups based on responses to MeK and Akt inhibitors.
To explore heterogeneity in targeted therapy responses, we evaluated the proliferative responses to 378 kinase inhibitors in 20 TNBC cell lines. To capture the heterogeneity in TNBC disease, these cell lines comprised multiple TNBC classes, including the basal-like, mesenchymal-like and luminal androgen receptor subtypes 15 . The TNBC cell lines largely demonstrated resistance to these inhibitors, but some inhibitors differentially affected the proliferation of the cell lines (Fig. 1A, Supplementary Dataset 1). In particular the responses to inhibitors of MEK and Akt were heterogeneous (Fig. 1A, black box, Supplementary Fig. S1), classifying the TNBC cell lines into three subgroups (Fig. 1B): cell lines were either resistant to Akt inhibitors, but sensitive to MEK inhibitors (Group 1, Akt-i-resistant); resistant to MEK inhibitors, but sensitive to Akt inhibitors (Group 2, MEK-i-resistant); or resistant to both Akt and MEK inhibitors (Group 3, Akt-i/MEK-i double-resistant). At the used concentration of 1 µM, we did not find a clear clustering of the related Raf and PI3K inhibitors (Supplementary Dataset 1 and Fig. S2). For example, while Raf inhibitors TAK-632 and RAF265 resulted in similar effects on the TNBC cell lines as the MEK inhibitors, Raf inhibitor vemurafenib induced opposite responses.
To further validate the Akt-i/MEK-i response classifications, we treated the TNBC cell lines with a dose range of the various MEK and Akt inhibitors ( Fig. 1C and Supplementary Fig. S3). MEK inhibition effectively reduced the growth of Akt-i-resistant Group 1 cells at low concentrations (IC50s~10 nM, Supplementary Table S1), but not of MEK-i-resistant Group 2 or Akt-i/MEK-i-resistant Group 3 cells. Whereas Akt inhibitors affected the cell proliferation of MEK-i-resistant Group 2 cell lines (IC50s~0.1-1 µM), Akt-i-resistant Group 1 and Akt-i/MEK-i double-resistant Group 3 cell lines did not respond to these inhibitors (IC50s > 3.16 µM).

Activity and mutations of PI3K-MAPK pathway members distinguish between Akt-i-resistant
and MeK-i-resistant cell lines. To understand the differential sensitivity to Akt and MEK inhibitors, we evaluated the basal phosphorylation levels of signalling components involved in the PI3K and MAPK pathways. Akt-i-resistant Group 1 cell lines expressed significantly higher levels of p-MEK compared with MEK-i-resistant Group 2 cell lines (P = 0.0003, Fig. 2A,B). In contrast, Group 2 cell lines expressed higher levels of p-Akt compared with Group 1 cell lines (P = 0.0072). Levels of p-Akt and p-MEK thus distinguish between Group 1 and 2 cell lines, suggesting that these groups exhibit a greater dependence on MAPK-or PI3K-mediated signalling, respectively, which may therefore account for the differential sensitivity of these groups to MEK and Akt inhibitors. Akt-i/MEK-i-resistant cell lines showed an average trend of increased p-Akt compared with Group 1 (P = 0.0315) and increased p-MEK compared with Group 2 (P < 0.0001) cell lines, respectively. However, Akt-i/ MEK-i-resistant Group 3 cell line SUM159PT had low levels of p-MEK, indicating that this average trend is not completely discriminative. Therefore, p-Akt and p-MEK levels alone cannot discriminate between Akt-i/MEK-i double-resistant and MEK-or Akt-i-sensitive cell lines.
Next, we evaluated the mutations present in the relevant genes in the three groups to further understand the differential phenotypes. All MEK-i-resistant Group 2 cell lines with high levels of p-Akt had mutations in PTEN, PIK3CA and/or PIK3R1 (6/6) (Supplementary Table S2). These mutations were completely absent in the Akt-i-resistant Group 1 cell lines (0/7). Akt-i/MEK-i-resistant Group 3 cell lines had a mixed mutational profile (5/7). In addition, Group 1 cell lines exhibited a higher frequency of BRAF mutations upstream of MEK (2/7), whereas these were absent in both Group 2 (0/6) and Group 3 (0/7) cell lines. Mutations in KRAS or HRAS did not differentiate between any of the groups (Group 1: 2/7, Group 2: 1/6 and Group 3: 2/7). Altogether, these features could distinguish between Akt-i and MEK-i resistance, but do not explain Akt-i/MEK-i double resistance.

Crosstalk between Akt and MEK is not essential for Akt-i/MEK-i resistance.
To further explore this Akt-i/MEK-i double-resistant phenotype, we investigated the role of crosstalk between Akt and MEK pathways. If crosstalk would drive the resistant phenotype, the double-resistant, but not single-resistant, cell lines would show increased levels of active MEK/ERK upon Akt inhibition, and vice versa. Moreover, an improved, synergistic, response would be expected upon combination treatment. Treatment with ATP non-competitive MEK inhibitors selumetinib and PD184352 resulted in accumulation of p-MEK accompanied by inhibition of ERK phosphorylation in all cell lines (Fig. 3A). The extent of inhibition of ERK phosphorylation was similar across the cell line panel, indicating that MEK-i resistance is not related to a lack of target inhibition. The allosteric Akt1/2/3 inhibitor MK-2206 fully prevented Akt phosphorylation in all cell lines, whilst ATP-competitive Akt1/2/3 inhibitor ipatasertib caused accumulation of p-Akt, as binding of this inhibitor to the active site protects Ser473 phosphorylation sites from phosphatases 16 . This implies that resistance to these agents is also not due to a lack of target inhibition. The Akt inhibitors reduced p-mTOR levels in some, but not all cell lines, an event which was independent of Akt-i or MEK-i sensitivity status, suggesting mTOR activity is not critical in conferring www.nature.com/scientificreports www.nature.com/scientificreports/ resistance to such inhibitors. Akt inhibition slightly increased p-MEK levels in Akt-i-resistant Group 1 cell lines, but not in Akt-i/MEK-i-resistant Group 3 cell lines.
Next, we evaluated whether combining MEK inhibitors with Akt inhibitors enhances the growth-inhibitory effects of either drug. Cell lines were treated with a combination of selumetinib (MEK-i) and ipatasertib (Akt-i). The combined treatment of a dose range of selumetinib, together with a fixed concentration of ipatasertib (1 µM), and vice versa, did not overtly enhance the anti-proliferative effects of either agent in most cell lines (Fig. 3B). The combination treatments moderately sensitised SUM159PT to either selumetinib or ipatasertib. For all of the Akt-i/MEK-i-resistant Group 3 cell lines, combination treatments were incapable of reducing cell proliferation to the same extent as in the Group 1 or Group 2 cell lines. Altogether, these results suggest that crosstalk or additive ERK and Akt proliferation signalling programs are not essential for the resistance observed in Akt-i/ MEK-i-resistant Group 3 cell lines. Within these significantly enriched cell cycle-related gene sets, we selected and explored the genes that contributed to the enrichment result (signal-to-noise ratio > 0.1) for both the transcriptomics (142 genes) and proteomics (102 genes) datasets (Fig. 4B,C, Supplementary Table S3). These genes contribute to a variety of cell cycle processes, such as G1/S transition, G2/M transition, cell cycle regulation, ubiquitination and nuclear division, suggesting that the observed cell cycle enrichment is not due to a prolonged state of one specific cell cycle phase. The genes included, amongst others, cyclin-dependent kinases (CDK) 2, 4, 5, 6 and 9, cell-division cycle (CDC) 25 A and polo-like kinases (PLK) 1, 2 and 4.
Next, to examine whether the elevated mRNA expression of these 142 selected cell cycle-related genes was reflected in patient tumours, we exploited expression data from The Cancer Genome Atlas (TCGA). Hierarchical clustering uncovered a specific set of 46 genes that was consistently expressed at higher levels in a subgroup of the breast cancer patients (Fig. 5A). This cluster mainly consisted of tumours from basal-like, HER2-enriched and luminal B PAM50 subtypes. Although most of the TNBC tumours had high expression of these genes, also within the TNBC subset, these genes were differentially expressed (Fig. 5B). The hierarchical clustering revealed three clusters of TNBC tumours, with either high (42%), intermediate (34%) or low (24%) expression of the genes. The 46 genes selected from these differential cluster in patients still also subdivide TNBC cell lines in a similar manner as the original 142 genes (Supplementary Fig. S7).
Elevated cell cycle gene expression network provides novel targets for bypassing Akt-i/MEK-i resistance. Next, we investigated whether the aberrant cell cycle gene expression network found in Akt-i/ MEK-i-resistant Group 3 cell lines could be an alternative target for this subgroup. Pathway analysis of the overexpressed cell cycle genes visualised core components, including CDC25A, CDK2 and CDK4 (Fig. 6A). Moreover, elevated E2F1 and DP1, together with low expression of RB1, were central in this pathway. Targeting the downstream CDKs affected by these transcription factors and CDC25A may therefore have detrimental direct and www.nature.com/scientificreports www.nature.com/scientificreports/ indirect effects on TNBC cell proliferation. Indeed, CDK inhibition by flavopiridol or dinaciclib strongly inhibited the growth of the Akt-i/MEK-i-resistant Group 3 cell lines (Fig. 6B). Contrastingly, Group 1 cell lines HCC1806 and MDA-MB-435s showed a relatively reduced sensitivity to dinaciclib and flavopiridol, with an average 2-3-fold increase in IC50 for both CDK inhibitors (Fig. 6B, Supplementary Table S4). Dinaciclib at 10 nM or 31.6 nM completely inhibited phosphorylation of RB1 at Ser780, an event specifically mediated by the CDK4/6-Cyclin D1 complex, in both HCC1806 and SUM159PT cells, indicative of cell cycle arrest (Fig. 6C). Importantly, flavopiridol at 100 nM moderately reduced p-RB1 levels in double-resistant Group 3 cell line SUM159PT but not in Akt-i-resistant Group 1 HCC1806 cells. Similarly, CDK9/CDK7-mediated phosphorylation of RNA Polymerase II's C-terminal domain (POLR2A) at Ser2/5 was abolished by dinaciclib in both cell lines whereas flavopiridol exclusively reduced POLR2A phosphorylation in SUM159PT cells. Dinaciclib induced cleavage of PARP1 in both cell lines, whilst flavopiridol-induced PARP1 cleavage was solely visible in SUM159PT cells. Concomitant depletion of BH3-only family member MCL-1 and induction of H2AX phosphorylation at Ser139 in double-resistant SUM159PT cells suggested induction of apoptosis and DNA damage response signalling. These effects were absent in HCC1806 cells after exposure to flavopiridol. Notwithstanding this differential sensitivity, the Group 2 cell lines BT20 and HCC70 were equally as sensitive to the CDK inhibitors as Group 3 cell lines (Fig. 6B). These www.nature.com/scientificreports www.nature.com/scientificreports/ data therefore suggest that CDK inhibitors which target the aberrant cell cycle profile are especially, but not exclusively, effective in Akt-i/MEK-i-resistant Group 3 cell lines.

Discussion
Although various potential targets characterise TNBC tumours, no effective clinical targeted therapy is available for this aggressive disease. Using a kinase inhibitor screen in a broad panel of TNBC cell lines, we mapped the responses to inhibition of various kinase targets and found that MEK and Akt inhibitors differentially affected www.nature.com/scientificreports www.nature.com/scientificreports/ TNBC cell proliferation. The ERK and Akt signalling programs are critical in various aspects of cancer biology, and MEK and Akt inhibitors have entered the clinic for various types of cancer 17,18 . Moreover, TNBC is often accompanied by genetic aberrations in upstream modulators of these pathways. As drug resistance to MEK and Akt inhibitors is not fully understood, this study further focussed on exploring the heterogeneity in these responses. We here demonstrate that MEK and Akt double resistance is associated with the elevated expression of a cell cycle gene network, while inhibition thereof effectively reduced cell proliferation in these MEK and Akt inhibitor double-resistant cell lines. www.nature.com/scientificreports www.nature.com/scientificreports/ In this research we have focused on studying the differential responses to MEK and Akt inhibitors, as the screening indicated the most clear and consistent differences between responses to these inhibitors among the TNBC cell lines. Although Raf and PI3K inhibitors are also promising candidates under clinical development, we did not find consistent responses upon treatment with these classes of inhibitors. This might be due to off-target effects of such inhibitors or due to different mechanisms of action. Moreover, the initial screening of the kinase inhibitor library was limited to testing the inhibitors at a single concentration of 1 µM. Therefore, we do not exclude that other classes of inhibitors may also differentially affect TNBC cell proliferation at higher or lower concentrations.
Previous studies have indicated that MEK or Akt inhibitor resistance in cancer can be caused by crosstalk between the PI3K/Akt/mTOR or RAF/MEK/ERK pathways 7,8,19 . Therefore, these studies suggest that www.nature.com/scientificreports www.nature.com/scientificreports/ combination therapy with MEK and Akt (or PI3K) inhibitors will reverse this resistance. In this study, we did not observe induction of the parallel pathways after Akt or MEK inhibition, with the exception of SUM159PT cells. Combination therapy also did not sufficiently sensitise the Akt-and MEK inhibitor-resistant cells to either therapy. These results suggest that crosstalk is not an essential mechanism of combined Akt-i and MEK-i resistance in TNBC. Indeed, even though various studies demonstrate synergistic effects of combination therapy, these did not disrupt cell viability completely 7,8,19 . Importantly, combination therapy with MEK and Akt inhibitors has also not shown promising results in clinical trials, as beneficial effects over single treatment are hardly obtained, instead causing more severe side effects 13,14,20 . For these reasons we sought to find another strategy to overcome the resistance in MEK-i and Akt-i-resistant cells.
Our results demonstrate that Akt inhibitor-resistant cell lines can be distinguished from MEK inhibitor-resistant lines by increased levels of p-MEK and low levels of p-Akt. Additionally, mutations in PTEN, PI3KCA and/or PI3KR1 characterise MEK inhibitor-resistant but not Akt inhibitor-resistant cell lines. Similar observations were made in other cancer types by others [7][8][9] . However, these features do not entirely explain the observed resistance phenotypes in our panel of TNBC cell lines. Here, we further explored the underlying mechanisms responsible for double resistance to both Akt and MEK inhibitors. To this end, we took advantage of our unique transcriptome and proteome dataset of the entire TNBC cell line panel. Using large-scale gene set enrichment analysis, we revealed a 142-gene and 102-protein cell cycle-enriched network associated with Akt and MEK inhibitor double resistance. Importantly, 46 of these genes subdivides all breast cancer patients, as well as TNBC patients, into separate groups based on their gene expression. The majority of TNBC patients had high expression of these genes, which is line with that so far, most TNBC patients are not responsive to MEK and Akt inhibitor treatments 13,14,[20][21][22][23] . Yet, a minority of the TNBC patients may thus be sensitive to MEK or Akt inhibitors, depending on their p-MEK/p-Akt levels. Thus, stratification of patients in clinical trials of MEK or Akt inhibitors based on the features described in this study may reveal beneficial effects that would otherwise be neglected. However, further studies are needed to translate the associations found here to patient treatment responses. Moreover, also the luminal B and HER2-enriched hormone-positive breast cancers have a higher expression of this gene signature compared to the luminal A subtype, suggesting that for hormone-positive breast cancers the PAM50 subtyping may already be predictive. Nevertheless, as the findings in this study are based on TNBC cell lines, this needs to be explored further.
The overexpression of the cell cycle genes may render signals from Akt and MEK redundant. No studies have previously linked such a profile to Akt and MEK inhibitor resistance in TNBC. However, various studies in other cancer types have recently demonstrated the role of the cell cycle in this resistance by confirming synergy between inhibitors of the PI3K/MAPK pathway and inhibitors of the cell cycle, such as CDK and PLK inhibitors [24][25][26][27][28][29][30] . These findings suggest that the enhanced expression and activity of cell cycle machinery components may have an important role in resistance to Akt and MEK inhibitors, not only in TNBC, but also in other cancer types.
Our data demonstrate that inhibition of CDKs that determine cell cycle progression and transcription can effectively kill TNBC cells that are otherwise resistant to Akt and MEK inhibitors. Although our initial screening did not show clear differences in responses to CDK inhibitors at 1 µM, further evaluation at lower concentrations did reveal differential responses. For example, the pan-CDK inhibitors dinaciclib and flavopiridol inhibited proliferation of SUM159T cells that were resistant to both Akt-i and MEK-i. Interestingly, some, but not all, Akt or MEK inhibitor-sensitive cell lines were more resistant to these compounds compared to MEK-i/Akt-i double-resistant cell lines. Thus, these compounds are especially, but not exclusively, effective against this group and provide possible alternative treatments for overcoming Akt and MEK inhibitor double resistance. Notably, both lethal and sub-lethal concentrations of dinaciclib inhibited RB1 and POLR2A phosphorylation in HCC1806 cells, confirming that its relative insensitivity to this agent is not related to inefficient target inhibition, unlike flavopiridol. The superior activity of dinaciclib compared to flavopiridol was also evidenced by the complete abrogation of RB1 and POLR2A phosphorylation at concentrations ten times lower than those required for flavopiridol to elicit a moderately comparable effect, which was only visible in MEK-i/Akt-i double-resistant SUM159PT cells. As dinaciclib and flavopiridol are more effective in the MEK-i/Akt-i resistant cell lines this may suggest that also for MEK-i/Akt-i resistant TNBC patients these treatments could be alternative options. This strategy could potentially be improved by more selective CDK inhibitors that are currently being developed. Nevertheless, the increased sensitivity could expand the therapeutic window to circumvent the side effects that are associated with flavopiridol and dinaciclib.
To conclude, this study describes the molecular features that can discriminate between Akt-i and MEK-i single agent-sensitive subsets of TNBC cells as well as double agent-resistant TNBC cells. For double-resistant cells we uncovered a high expression of a cell cycle gene network. Accordingly, these double-resistant cells are relatively sensitive to inhibitors of CDKs. Moreover, a number of these genes are also differentially expressed in TNBC patients. Since our study describes clinically relevant features that can subdivide TNBC cells into their Akt and MEK inhibitor response profile, these features may also facilitate the prediction of response in patients. Future mechanistic studies may reveal the origin of the aberrant cell cycle network, which could supplement this study by providing additional alternative targets and simplified biomarkers for clinical assessment. Altogether, this and future studies could therefore lead to a biomarker-based therapeutic strategy for treating TNBC patients with Akt, MEK and/or CDK inhibitors.