Herein we present the results from a meta-analysis of genome-wide chemical-genetic screens performed on the diploid yeast deletion collections in the absence and presence of the Hsp90 inhibitor macbecin II. Importantly, this robust analysis includes for the first time a screen conducted on heterozygous deletion strains at 15°C, in combination with previous heterozygous deletion pool screens conducted at 30°C and 37°C. This multi-temperature approach allows for explorations of condition-dependent cellular requirements for Hsp90. Of critical significance, it also permits the analysis of how both essential and nonessential half-dosage gene deletions contribute to compromised growth in the face of Hsp90 inhibition, revealing numerous essential genes and pathways that are affected.
When broadly comparing the data sets from the heterozygous deletion screens () we found much less overlap between the different temperature data set ORFs than what was observed for homozygous deletion screens 
. For example, while there was 29% overlap between 30°C and 37°C homozygous data sets, there is only 13.4% overlap amongst the 15°C, 30°C, and 37°C heterozygous data sets. This is perhaps indicative of more distinct temperature-sensitive requirements for Hsp90 under conditions of single copy gene deletion. A further difference between the broad analysis of homozygous and heterozygous deletion strain data sets is that numerous nonessential deletion strains were found to be hypersensitive to Hsp90 inhibition at 37°C in half-dosage, but not when both copies were deleted. This suggests the exciting possibility that these candidates represent labile Hsp90 clients that are dependent upon Hsp90 function under heat-stress conditions. The observed difference in Hsp90 requirement between homozygous and heterozygous states could be due to adaptations that have occurred in the complete absence of the protein versus growth defects that become apparent upon loss of folding maintenance and function of an existing protein. A well-established example of this in yeast is the case of SRP (Signal Recognition Particle), which is required for proper recognition and targeting of nascent secretory proteins to the ER membrane for cotranslational translocation. If SRP-encoding genes are deleted, yeast are still viable (albeit with translocation defects and slowed growth; 
). Yeast instead harboring a temperature-sensitive allele of sec65
, which encodes an essential protein subunit of SRP, are inviable at the non-permissive temperature 
On the other hand, deletion strains that were hypersensitive to Hsp90 inhibition in the homozygous screens, but not when present in single copy, may indicate genetic or indirect, rather than direct physical Hsp90 interactors. For example, there is extensive evidence supporting a role for Hsp90 in intracellular vesicle transport 
. In mammalian cells, loss of Hsp90 recruitment to VAP-33 via the protein TPR1 disrupts intra-Golgi trafficking. The closest yeast homologue of VAP-33 is the ER membrane protein SCS2
, which, like VAP-33, would not be expected to be a direct interactor of Hsp90. Yeast lacking both copies of SCS2
are extremely hypersensitive to Hsp90 inhibition (i.e. in top 5% data sets at both 30°C and 37°C; 
), however, in heterozygous screens the SCS2::scs2Δ
strain is unaffected by Hsp90 inhibition at all three temperatures tested. This supports the supposition that genetic interactions or indirect/mediated physical interactions with Hsp90 are perhaps best detected by the homozygous screens, while direct physical interactors are better revealed by the heterozygous screens. There are, of course, obvious exceptions to this as Sti1p is known to directly bind Hsp90 and the lack of STI1
in both homozygous and heterozygous contexts severely compromises growth when Hsp90 is inhibited. Thus, perhaps a modification to the above prediction is the exception of Hsp90 cochaperones as they both physically interact with Hsp90 and would be expected to exhibit genetic consequences as well since the Hsp90 chaperone cycle is highly dependent on appropriate interactions with its cochaperones. In any case, to gain further insight into these differences it will be of interest to compare the growth of relevant homozygous and heterozygous deletion strains as individual strains, outside of the context of the pools, in the presence of Hsp90-inhibiting drug to rule out non-specific effects. The hypotheses put forth above can also be directly tested by evaluating, for subsets of relevant homozygous and heterozygous deletion strains, if the affected proteins are indeed more likely to physically interact with Hsp90.
Regarding Hsp90 cochaperones, our finding that not all Hsp90 cochaperones, indeed not even those with a shared Hsp90 interacting motif (i.e. TPR domains), exhibit similar patterns of temperature and dosage-dependence () points to the existence of combinatorial Hsp90-cochaperone interactions that could further dictate specificity and functionality of Hsp90. This is supported by recent observations that not only do asymmetric Hsp90-cochaperone complexes containing different TPR-cochaperones bound to each subunit of an Hsp90 dimer exist, but they in fact appear to be favored for forward progression of the Hsp90 chaperone cycle 
Of the three high-confidence Hsp90 cochaperones encoded by essential genes, the CDC37::cdc37Δ
strain is most affected by Hsp90 inhibition at 37°C, while CNS1::cns1Δ
strains are strongly affected only at 30°C. While the latter two exhibit similar patterns and do both contain TPR domains, it has been shown that Sgt1p interacts with Hsp90 via a CS motif (C
). It is unclear whether any significance can be ascribed to these observed differences and similarities. Such an analysis is further complicated by the fact that not all cochaperones are expressed at similar levels, therefore, half-dosage likely means something completely different for each cochaperone under the different temperature conditions.
Predicting the most likely functional interactors or targets of Hsp90 with higher order analyses utilizing GO assignments and interaction networks was much more insightful. As already mentioned, one benefit of conducting these screens with the heterozygous deletion strains is the possibility of directly identifying physical interactors of Hsp90. The network analyses conducted using the heterozygous data sets readily support this notion. The physical interaction network () is strikingly modular, indicating that if Hsp90 directly targets even one subunit of a protein complex, other subunits within that complex will also be affected by loss of Hsp90 function. This points to the existence of Hsp90 substrate “modules” in the cell and indicates a clear partitioning of known protein complexes into Hsp90-mediated or Hsp90-independent. Of note, this indication was previously suggested from generating a secretory pathway-specific physical interaction network from 30°C ad 37°C homozygous and heterozygous data sets, which was primarily populated with modular, cellular transport-related, protein complexes 
GO analyses of heterozygous data sets revealed several expected classifications, such as those related to the secretory pathway or cell cycle regulation, but also many intriguing new enrichments, such as nuclear transport, general transcription, ribosome assembly, and GTPase-related assignments. Importantly, these classifications were evident when limiting the analysis to essential ORFs only, suggesting that they underlie essential cellular requirements for Hsp90.
Regarding the connection between Hsp90 and nuclear transport, it is more likely the case that Hsp90 requires access to the nucleus for some of its activities rather than that Hsp90 is required for nuclear transport to occur. For example, Hsp90 translocates to the nucleus in association with its hormone receptor folding clients and Hsp90 cochaperone p23 assists the remodeling of associated transcriptional complexes 
. Hsp90 has also been shown to enter the nucleus in response to heat shock in mammalian cells 
and in response to starvation in yeast 
. As such, we would expect nuclear transport GO assignees identified in these screens to perhaps be assisting the nuclear transport of Hsp90. Indeed, strains heterozygous for the karyopherin genes KAP95
, recently found to assist the nuclear translocation of Hsp90 in yeast 
, are among those identified in our screens as hypersensitive to loss of Hsp90 function. Of interest, KAP95
both occur in the heterozygous 15°C data set, are highly ranked at 30°C (KAP95
top 7.6%; SRP1
top 16%), and less sensitive at 37°C (KAP95
top 31%; SRP1
top 40%). This suggests that Hsp90 does not require nuclear transport under hyperthermic stress (see also 
), in contrast to what is observed for mammalian cells. Further, it suggests that Hsp90 may require access to the nucleus in more instances than just starvation, for instance, during hypothermic stress.
What might Hsp90 be required for in the nucleus? One likely possibility is a broader role for Hsp90 than previously thought in general transcription, most likely through assisting chromatin remodeling processes. This was first suggested by the above-mentioned paper by Freeman and Yamamoto (2002), further supported by the identification of chromatin remodeling complex components Tah1p and Pih1 as Hsp90-interacting proteins 
, and significantly bolstered in this report by the following: First, chromatin remodeling appears as a significantly enriched GO assignment in our analysis (). Second, our data sets engendered evident physical network modules for (1) mediator complex components, (2) various TAFs, and (3) numerous members of the SWI/SNF chromatin remodeling complex (). Of note, while neither TAH1
homozygous nor heterozygous deletion strains were sensitive enough to Hsp90 inhibition to be included in our data sets for analysis, the associated essential gene RVB2
appears in the heterozygous 15°C data set. A role for Hsp90 in chromatin remodeling has been suggested to explain, at least in part, the dramatic phenotypic variation that can occur when the buffering capacity of Hsp90 is compromised 
. Here we provide robust evidence for a connection between Hsp90 and the SWI/SNF chromatin remodeling complex, five members of which appear in our top 5% data sets (SNF11
), as well as SNF12
ranking in the top 8.3% of the 37°C heterozygous data sets and SWP82
ranking somewhat lower, but exhibiting an identical pattern of sensitivity to SNF11
. Whether Hsp90 and the SWI/SNF complex contribute independently to chromatin remodeling or act in conjunction remains to be determined. One could also envision that loss of appropriate transcriptional regulation or fidelity could similarly contribute to the expression of buffered phenotypes that is observed when Hsp90 function is compromised; the observed growth effects on strains heterozygous for general TAFs and mediator complex components in the presence of Hsp90 inhibitors suggest this as a possibility.
The chromatin remodeling components Tah1p and Pih1p, along with AAA-ATPases Rvb1p and Rvb2p, are also linked to snoRNP biogenesis 
, and a role for Hsp90 in this process was recently confirmed 
. We chose to specifically address a subset of this type of assembly in this report, namely ribosome assembly. Multiple ribosome assembly-related GO assignments were enriched in our analysis and, further, very clear functional modules related to ribosome assembly appear in our physical interaction network. Specifically, in the most minimal physical network (), the two main remaining modules (connected by essential ORF BMS1
) are primarily composed of macro-molecular complex assignments of preribosome, large subunit precursor and 90S
preribosome or small-subunit processome (highlighted in ). Our experimental follow-up on these observations demonstrated that compromising Hsp90 function in vivo
resulted in decreased polysome stability ( C–F). A role for Hsp90 in maintaining ribosome stability in mammalian cells was also previously suggested 
, and mammalian ribosomal proteins were identified in proteomic screens for Hsp90 interactors 
. Altogether, we propose that Hsp90 may play a more significant role in ribosome biogenesis and stability than previously thought.
Lastly, we wish to consider the significant enrichment of GTPases identified in these chemical-genetic screens. Hsp90 has been implicated in Rab GTPase recycling in yeast and mammalian secretory pathways 
, and, more recently, shown to interact with Rab11a to regulate membrane recycling of alpha-synuclein 
. Extensive connections between Hsp90 and the yeast secretory pathway were identified in our previous study, including six Rab family GTPases and five other transport-related GTPases 
. The data and analyses presented here identify 24 GTP binding proteins in total, including six that are essential, that appear in at least one of our screen data sets. In addition to the previously mentioned transport-related GTPases, there are three GTPases involved in cell polarity, six linked to ribosome biogenesis or translation, and three that participate in other cellular processes. Altogether, these data point to a more important and pervasive connection between Hsp90 and cellular GTPases than previously realized and help to explain the far-reaching pleiotropic effects of inhibiting Hsp90 function in eukaryotes.