In the past, several studies have analysed the cellular response to multiple Hsp90 inhibitors in different cell lines and organisms. Genome-wide expression profiling has been employed to monitor the cellular response to Hsp90 inhibition and revealed affected genes and markers for monitoring of effective Hsp90 inhibition in clinical treatment like Hsp70 . However, these transcriptional events are indirect as Hsp90 regulates mostly the protein stability of its clients. Some studies have however used proteomics to monitor protein levels and revealed affected pathways like for example chromatin remodelling or MAPK, WNT, NF-kB and TGF-beta signalling [25, 29]. However kinases, the largest group of clients, were underrepresented in these studies. To close this gap we focused on the analysis of the response to Hsp90 inhibition of a large number of kinases by measuring their protein level changes. We compared the rate of down- or upregulation of our kinase-focussed assay with the two abovementioned previous studies that probed the unfiltered proteome. Schumacher et al. identified 111 differentially expressed proteins of which about 41% were overexpressed and 59% underexpressed . Similar values were found by Maloney et al. (26 proteins in total, 46% over- and 54% underexpressed) . Comparing these values to significant downregulation of 69% to 88% of all identified kinases by our assay argues for a high specificity of kinase level reduction after Hsp90 inhibition.
A higher affinity to Hsp90 inhibitors has been reported in cancer cells . We investigated if there were different kinetics of client degradation between normal and cancer cells in response to geldanamycin. Interestingly, we did not find striking differences arguing that kinetic properties do not contribute to the increased response to Hsp90 inhibition in cancer cells. We only observed slower kinetics in U2OS cells, which may be due to reduced transport of the drug into the cell or differences in metabolism of the drug. For example, it has been shown for 17-allylamino,17-demethoxygeldanamycin (17-AAG), a geldanamycin derivative, that activity of the quinone reductase DT-diaphorase (gene name: NQO1) is positively correlated with the growth-inhibitory activity of 17-AAG, because DT-diaphorase converts 17-AAG into a form more potent for Hsp90 inhibition .
Our kinase-enrichment approach allowed us to monitor the effect of the Hsp90 inhibitor geldanamycin on 144 kinases. From this data we were able to identify 44 high confidence client kinase candidates (Figure 1b) with a strong representation of MAPK and TGF-beta signalling components (Figure 2c). The additional treatment with proteasome inhibitor MG132 allowed us to discriminate between Hsp90 client candidates that likely undergo ubiquitination and proteasomal degradation (their cellular levels are at least partially recovered in presence of MG132) and kinases that are affected by geldanamycin treatment more indirectly - for example via transcriptional regulation if downstream of a client kinase - or via another mechanism (their cellular levels should not be significantly changed in presence of MG132). Whereas a majority of kinases affected by geldanamycin treatment appears to be true client Hsp90 proteins in SW480, only few kinases displays a similar behaviour in Hs68. Based on these results, we propose a list of Hsp90 client kinases (Figure 3). This list regroups 64 kinases and includes many tyrosine kinases (e.g. Fyn, Lyn, Src, Yes, Abl, Arg, Tyk2) or tyrosine-like kinases (ALK2, ALK4, RIPK2, ILK, TGFBR1, MLK3), two phylogenic branches of kinases that have previously been shown to include most of the Hsp90 kinase clients . Ephrin receptors distinguish themselves among tyrosine kinases since - with the noticeable exception of Ephα2 -, they do not appear to be Hsp90 clients, at least in SW480 cells. Among the Serine/Threonine kinase groups, we find well-described Hsp90 clients like CDK2, CDK9, CK2α1, CK2α2 or TBK1 but also some kinases that were thought not to be Hsp90 clients based on their sequence  like CDK5, PKCα, PKCβ or MAPK1/ERK2. Conversely, among kinases that did not appear as Hsp90 clients in this study, we find some kinases that have been previously described as putative Hsp90 clients like GSK3β, JAK1 or FER. Our list of Hsp90 client kinases is therefore significantly different from those that have been proposed so far [27, 31]. Geldanamycin treatment affects many pathways and has major impact on the entire proteome . It is therefore not too surprising that the level of many kinases is significantly modified even if their levels are poorly regulated by Hsp90 machinery, like many kinases from Hs68 cells. This data strongly support the hypothesis that the role of Hsp90 as a kinase chaperone is much less preeminent in healthy primary cells than in a cancer cell type as colon adenocarcinoma SW480 cells. We conclude from our results that in normal cells the majority of downregulated kinases following geldanamycin treatment is driven by indirect, non Hsp90-dependent mechanisms of degradation. In contrast, in cancer cells the majority of kinases appear to be dependent on Hsp90 chaperoning.
Many Hsp90 inhibitors have been developed and several are presently undergoing clinical evaluation . This is the first study that focuses on the impact of Hsp90 inhibition on a broad spectrum of the kinome. Our results reveal an impact of Hsp90 inhibitors on more wide-ranging types of kinases, and hence pathways, than previously thought (Figure 1b and 2c). This is of special clinical interest for the inhibition of feedback loops that often arise in single targeted therapy and that have been acknowledged as a resistance mechanism and escape route for cancers to evade treatment. For example the use of mTOR inhibitors leads to the PI3K-dependent activation of MAPK and Akt signalling, which both are targeted by Hsp90 inhibitors [32, 33]. Our discovery of many new Hsp90 targets further supports the use of Hsp90 inhibitors in combinatorial treatment, especially as a means to suppress feedback loops, because it affects even more processes than anticipated so far.
Many kinases annotated by KEGG as members of TGF-beta signalling identified in our study belong to the bone morphogenetic protein (BMP) signalling pathway. We show for the first time that BMP receptors (BMPR1A, BMPR2, ACTR2, ALK2 and ALK4) exhibit downregulation after Hsp90 inhibition (Figure 1). BMP proteins are members of the TGF-beta superfamily and have important functions including embryonic development, bone formation and tissue homeostasis . BMPs bind to BMP receptors that mediate signals mostly via phosphorylation of SMADs . In cancer, this pathway has been linked for example to bone metastases formation in breast, prostate and lung cancer and control of cell proliferation. The outcome of BMP receptor signalling is strongly cell type- specific and also dependent on which BMPs are present . Therefore the consequences of downregulation of multiple BMP receptors by Hsp90 inhibition could be diverse in different tissues. In prostate cancer BMPR1A and especially BMPR2 downregulation has been correlated with disease progression and activity of BMPR2 has been shown to function in a proliferation suppressive way [36, 37]. Our findings suggest that the use of Hsp90 inhibitors in prostate cancer might, by the downregulation of BMPRs, lead to an unintended promotion of proliferation and metastasis formation, thereby counteracting or attenuating the beneficial effects exerted on other pathways and limiting its clinical use. In line with this, results from a clinical trial with hormone-refractory prostate cancer suggest that Hsp90 inhibitors are no effective agents when used in monotherapy . If this is attributable to their effects on BMPRs remains to be determined. In contrast, in breast cancer BMPR1A activity was shown to promote cell proliferation via SMADs, whereas results for BMPR2 are contradictory [39–41]. This indicates that there might be a more promising therapeutic window for the use of Hsp90 inhibitors to reduce BMP signalling in certain breast cancers, in which setting several Hsp90 inhibitors are currently tested .
Our study identified many kinases of the JNK and p38 MAPK pathways, which are involved in diverse processes like for example stress response, inflammation, cell proliferation, survival and migration. Both pathways are often deregulated in cancer, however the often context-specific oncogenic and tumour suppressive functions impede the prediction of a pharmacological intervention . We identified new Hsp90 targets within the p38, JNK and also Erk5 MAPK signalling cascades on the level of MAPKKKs (MEKK2/3/4, MLK1/3, TAO1/2 and ZAK) and upstream regulatory MAP4Ks (GCK, KHS1, KHS2) and could show an increased downregulation in cancer cells upon geldanamycin treatment for some of them (Figure 1 and 2b). Several inhibitors of p38alpha and JNK have been developed, but have side effects or lacked specificity . We do not propose Hsp90 inhibitors as a single agent treatment in this setting, however it might prove useful for combinatorial treatment with future improved inhibitors against JNK and p38, because they can downregulate several upstream components of the MAPK cascades, likely increasing the efficacy of inhibition. As Hsp90 inhibitors act more specifically on tumours this additional effect would be limited to the target tissue, which likely minimises systemic side effects.