Inducible transcriptional condensates drive 3D genome reorganization in the heat shock response

Mammalian developmental and disease-associated genes concentrate large quantities of the transcriptional machinery by forming membrane-less compartments known as transcriptional condensates. However, it is unknown whether these structures are evolutionarily conserved, capable of stress-inducible gene activation or involved in 3D genome reorganization. Here, we identify inducible transcriptional condensates in the yeast heat shock response (HSR). HSR condensates are biophysically dynamic spatiotemporal clusters of the sequence-specific transcription factor Heat shock factor 1 (Hsf1) with Mediator and RNA Pol II. Uniquely, HSR condensates drive the coalescence of multiple Hsf1 target genes, even those located on different chromosomes. Binding of the chaperone Hsp70 to a site on Hsf1 represses clustering, while an intrinsically disordered region on Hsf1 promotes condensate formation and intergenic interactions. Mutation of both Hsf1 determinants reprograms HSR condensates to become mammalian-like: constitutively active without intergenic coalescence. These results suggest that transcriptional condensates are ancient and flexible compartments of eukaryotic gene control.


INTRODUCTION
Eukaryotes have evolved specialized mechanisms to ensure robust expression of critically important genes such as those activated to specify cell lineage and in response to stress. Mammalian cells utilize clusters of transcription factor binding sites, termed "super-enhancers" (SEs), to drive high-level gene expression by concentrating a large fraction of transcriptional activators and coactivators such as the Mediator complex (Hnisz et  To date, transcriptional condensates have only been described at stably expressed loci in mammalian cells (Sabari et al., 2020). Whether they regulate activation of highly-expressed inducible genes, such as those activated by environmental stress, and whether they are conserved features of eukaryotic transcription remains unknown. In fact, in the single known example of biomolecular condensation of a stress responsive transcription factor -phase separation of Heat Shock Factor 1 (Hsf1) in human cells -the condensates have been shown to be anti-correlated with transcriptional activation of the heat shock response (HSR) and localized away from canonical HSR genes (Gaglia et al., 2020;Jolly et al., 1997). Moreover, it remains unclear how transcriptional condensates conduct gene control in a three-dimensional (3D) context and whether they influence  (Kainth et al., 2021). Indeed, the HSR shares several key features with mammalian SEs, such as recruitment of high levels of the transcriptional machinery at the associated genes and extensive DNA looping interactions. A key distinction is that, as opposed to the stable gene expression enforced by SEs, HSR activation in yeast is induced in a remarkably rapid yet transient manner: large increases in occupancy of Hsf1 and the transcriptional machinery at HSR genes accompany transcriptional activation within 60 seconds of temperature upshift and return to pre-heat shock levels within 30 to 60 minutes (Chowdhary et  Here we provide experimental evidence implicating biomolecular condensation as the underlying mechanism of 3D gene control during the yeast HSR. We show that Hsf1, together with Mediator and RNA Pol II, forms transcriptional condensates at HSR genes in response to heat shock. These condensates are inducible, highly dynamic and promote intergenic chromatin interactions among genes in the Hsf1 regulon. We identify the N-terminal IDR of Hsf1 and a C-terminal binding site for the chaperone Hsp70 as molecular determinants that promote condensation with intergenic interactions and inhibit cluster formation, respectively. Finally, we generate a separation-of-function mutant of Hsf1 by mutating both of these determinants that forms constitutive condensates but fails to drive intergenic interactions. This double mutant displays a strong growth defect at elevated temperature, suggesting that formation of constitutive condensates that lack intergenic interactions comes at a fitness cost in this system. Together, our work reveals that transcriptional condensates are a conserved feature of eukaryotic gene control that can be induced and dissolved as a means to reversibly reconfigure the 3D genome.

Inducible high-level occupancy of the transcriptional machinery at HSR genes
To determine whether yeast HSR genes -defined as genes under the control of Hsf1 -share key features with genes that drive lineage specification and disease progression in mammalian cells, we assessed the extent to which HSR genes concentrate the transcriptional machinery. Using a consistent bioinformatic pipeline, we reanalyzed chromatin immunoprecipitation sequencing (ChIP-seq) data of Hsf1, Med15 (subunit of the Mediator complex known to interact with Hsf1) and Rpb1 (subunit of RNA Pol II) across the genome under basal and . We found strong enrichment of all three factors at HSR genes upon heat shock and observed that the Hsf1 and Med15 peaks colocalize at heat shock elements (HSEs), the known binding sites for Hsf1 ( Figure 1A, B, S1A-C). Rank order analysis of protein-coding genes (n=5983) by change in Med15 occupancy during acute heat shock revealed that all HSR genes are in the top quintile, with the majority residing above inflection point of the curve ( Figure 1C). To determine if Hsf1 binding is sufficient to recruit high levels of Mediator upon heat shock, we performed ChIP of Hsf1 and Med15 at the non-HSR gene BUD3 in a wild type strain and a strain harboring a chromosomal integration of a high-affinity HSE at the BUD3 locus (Chowdhary et al., 2019). While neither Hsf1 nor Med15 were detectable at the wild type BUD3 locus, a brief heat shock induced massive recruitment of Hsf1 and Med15 to BUD3 with the engineered HSE ( Figure S1D), suggesting that Mediator is robustly and selectively recruited by Hsf1 at its gene targets. Together, these data demonstrate that HSR genes disproportionately concentrate the transcriptional machinery upon activation.

Hsf1, Mediator and RNA Pol II colocalize in subnuclear clusters upon heat shock
We hypothesized that HSR genes engage such a large fraction of the transcriptional machinery by spatially concentrating the requisite factors within the nucleus. Therefore, we investigated the spatiotemporal distributions of Hsf1, Mediator and RNA Pol II in live single cells upon heat shock ( Figure 1D). In unstressed cells, endogenously tagged Hsf1-mVenus was predominantly nuclear and diffusely localized. However, Hsf1 formed semi-discrete subnuclear clusters within 6 min of heat shock that persisted for 20-25 min, after which Hsf1 returned to its diffuse localization pattern. To rule out photobleaching artifacts and to quantify how frequently Hsf1 clusters appeared within a population of cells, we performed microscopy of cells that were subjected to instantaneous heat shocks for 0, 2.5, 5, 10, 25 and 90 min and then fixed ( Figure S2A). This analysis revealed the presence of Hsf1 clusters in >80% cells subjected to either 2.5, 5, or 10-min heat shock. Consistent with the live imaging, few cells evinced clusters when induced for 25 min or longer ( Figure S2B). We typically observed 4 clusters of Hsf1 per individual nucleus, although the number varied between 2 and 5 over the course of heat shock ( Figure 1D, E). Analysis of diploid cells expressing Hsf1-GFP gave virtually identical results; neither kinetics nor number of clusters changed significantly, indicating that the number of Hsf1 clusters does not correlate with ploidy ( Figure S2C). To distinguish whether redistribution of Hsf1 into clusters is a physical consequence of elevated temperature or a regulatory consequence of Hsf1 activation, we imaged Hsf1-mVenus expressing cells following treatment with azetidine-2-carboxylic acid (AZC), a proline analog that incorporates into nascent proteins, impairs protein folding and activates Hsf1 (Trotter et al., 2001;Zheng et al., 2018).
Following AZC treatment, Hsf1 formed intranuclear clusters in a majority of cells ( Figure S2D). Therefore, Hsf1 clustering appears to be driven by activation rather than heat per se.
We next tested whether Mediator and RNA Pol II also formed intranuclear clusters during heat shock by imaging haploid cells expressing endogenously-tagged Med15-mCherry or Rpb3-mCherry. In the absence of heat shock, Med15 and Rpb3 were diffusely localized within the nucleus. However, like Hsf1, both formed an average of 4 subnuclear clusters in >80% of cells upon heat shock that emerged and dissolved with similar kinetics to Hsf1 ( Figures 1D, E and S2A, B). Moreover, live cell 3D imaging of cells co-expressing Med15-mCherry and Rpb3-mCherry with Hsf1-mVenus revealed that the clusters formed by both factors co-localized with Hsf1 during heat shock with a fractional overlap of at least 70% (Figures 1F, G and S2E). These data provide evidence of spatiotemporal interactions between Hsf1, Mediator and RNA Pol II inside living cells following heat shock.

Hsf1 clusters activate transcription of HSR genes
In human cells, Hsf1 also forms subnuclear clusters upon activation, but these clusters do not reside at protein coding genes and are therefore not sites of HSR activation (Gaglia et al., 2020;Jolly et al., 1997). To determine whether HSR genes associate with the Hsf1/Med15/Rpb3 clusters in yeast, we created three individual strains bearing 3ʹ-flanking arrays of 256 lacO-repeats at the HSP104 locus -a canonical HSR gene bound by these factors ( Figure 1B) -and co-expressing LacI-GFP with either Hsf1-, Med15-or Rpb3-mCherry ( Figure   1H). Live-cell imaging revealed that under basal conditions, HSP104-lacO 256 minimally overlapped with Hsf1-, Med15-and Rpb3-mCherry but consistently co-localized with these factors during acute heat shock ( Figure 1H and S2F). Across a population of cells, the average fluorescence intensities of Hsf1, Med15 and Rpb3 were enriched at the center of the HSP104-lacO 256 focus ( Figure 1I). These results indicate that the Hsf1, Mediator and Pol II clusters associate with a canonical HSR gene.  A) Metagene plots of Hsf1 (top), Med15 (middle) and Pol II (Rpb1; bottom) normalized ChIP-seq reads of Hsf1-dependent genes in S. cerevisiae (n=42) centered over ORFs with 1 kb surrounding region. Shown are means +/-SE (shaded region). NHS, non-heat shock; HS, heat shock; CPM, counts per million reads. B) Hsf1, Med15 and Pol II ChIP-seq densities as well as nascent mRNA counts (NAC-seq) at the HSP104 locus. C) Fold change in Med15 occupancy across Pol II-transcribed genes (0 to -1 kb from +1 ATG) upon heat shock, arranged according to their rank order. D) Fluorescence imaging analysis of cells expressing Hsf1-mVenus, Med15-mCherry or Rpb3-mCherry before (NHS, 24-26ºC) and following HS (39ºC) for the times (t) indicated. A single cell is followed throughout the heat shock time course. Row 1: images of cells expressing Hsf1-mVenus. Representative planes are shown. Blue line highlights cell boundary. Scale bar is 2 µm. Row 2: enlarged view of the nucleus (see yellow box in Row 1). Shown is the 3D volumetric rendering of each nucleus; x (red), y (green) and z (blue) axes are indicated. Rows 3 and 4: 3D volumetric rendering of nuclei of cells expressing Med15-mCherry and Rpb3-mCherry, respectively. To directly test whether Hsf1 clusters are sites of active transcription, we employed the MS2 RNA imaging system to visualize nascent transcript production (Hocine et al., 2013;Lenstra and Larson, 2016). We inserted 24 repeats of the MS2 stem loop into the genome in the 3ʹ untranslated region of the HSP12 gene, another HSR target ( Figure S1C), in a strain expressing the MS2 coat protein fused to mCherry (MCP-mCherry) and co-expressing Hsf1-mVenus. We observed no HSP12-MS 24 fluorescence under non-heat shock conditions, indicating a lack of transcription of HSP12 mRNA, while heat shock induced the transient appearance of a concentrated focus of nascent HSP12-MS 24 mRNA that colocalized with Hsf1-mVenus ( Figures 1J and S2G). The average fluorescence intensity of Hsf1 was enriched at the center of the HSP12-MS2 24 mRNA focus across a population of acutely heat-shocked cells ( Figure 1K). These data demonstrate that, upon heat shock, Hsf1 forms subnuclear clusters with the transcriptional machinery that mark sites of nascent mRNA production.  Figure 2A). The demixing phase is marked by formation of Hsf1 clusters during the first 6-7 min of heat shock induction. In the intermixing phase (8-27 min), Hsf1 clusters rearrange continuously within the nucleus, demonstrating the exchange of contents among the clusters. In the remixing phase (28-32 min), the contents of the clusters dispersed to the pre-heat shocked diffuse form.

Hsf1 clusters exhibit properties of biomolecular condensates
To further probe the material properties of Hsf1 clusters, we tested their sensitivity to an aliphatic alcohol, minutes to disintegrate the nucleolus and NPC with 3% 1,6-HD, similar effects were evident in less than 4 minutes when 10% 1,6-HD was applied. Moreover, membrane blebbing was prominent following addition of 10% 1,6-HD. Treatment with 5% 1,6-HD resulted in dissolution of the nucleolus and NPC after 7 minutes with no signs of membrane blebbing until after 20 minutes ( Figure S3B). We also scored the effects of 2,5-hexanediol (2,5-HD), a less hydrophobic alcohol with the same atomic composition as 1,6-HD. 2,5-HD had minimal effect on nucleolar and NPC integrity when used at lower concentrations (3 or 5%). However, at 10%, the effects of 2,5-HD were the same as observed with 1,6-HD ( Figure S3A, B). Based on these observations, we identified 5% for less than 20 minutes as the optimum concentration and treatment duration for both hexanediols.
To determine the effect of hexanediols on Hsf1 clusters, we first heat shocked cells for 8 minutes to allow cluster formation and then treated cells expressing Hsf1-mVenus with either 5% 1,6-HD or 2,5-HD. Hsf1 clusters were dissolved within 4 minutes upon addition of 1,6-HD but were unaffected by 2,5-HD treatment and mimicked cells that were only heat-shocked ( Figure 2B, C). We observed similar effects on Med15 clusters, except they were even more rapidly dissolved upon 1,6-HD treatment ( Figure S3C, D). Together, these data indicate that the transcriptionally active Hsf1 clusters display properties characteristic of biomolecular condensates: they continuously intermix and exchange contents, and they are potentially held together by weak, hydrophobic interactions. We will hereon refer to them as HSR condensates.

HSR condensates drive intergenic interactions among HSR genes
HSR genes engage in selective intra-and inter-chromosomal interactions upon heat shock-induced activation   and an enlarged 3D view of the nuclear region indicated within the yellow box. Bottom: enlarged 3D micrographs of a representative cell subjected to heat shock for the times (t) indicated (axes: x (red), y (green) and z (blue)). d, 3D distances measured between signal centroids. F) Distribution of 3D distances between lacO-tagged HSP104 and tetO-tagged TMA10 gene loci in cells subjected to heat shock for times indicated (n=13 for NHS; n=20-30 for HS time points). Bottom: Heat map depicting percentage of cells with tagged gene loci within indicated 3D distances binned at intervals of 0.4 µm. G) Intergenic contacts (solid arcs) between indicated Hsf1 target gene pairs as determined by TaqI-3C. Values indicate normalized interaction frequencies. Genes are segmented as UAS, promoter, mid-ORF and terminator (from blunt to arrowhead direction). Data are derived from two independent biological replicates; qPCR =4. Depicted are means +/-SD (shaded region around solid arcs). H) Distribution of 3D distances between tagged gene loci under conditions as indicated. Cells were left untreated or pre-treated with either 2,5-or 1,6-hexanediol followed by no heat shock or heat shock for 6 min and imaged at times (t) indicated. n =13-22. ****P <0.0001; ***P <0.001; ns (not significant), P >0.05 (calculated using two-tail t test).
fold upon acute heat shock, coincident with their increased transcription. To complement this molecular assay, we directly imaged intergenic coalescence of HSR genes in live cells by utilizing a heterozygous diploid strain bearing operator arrays chromosomally linked to HSR genes, HSP104-lacO 256 and TMA10-tetO 200 , that co-expresses LacI-GFP and TetR-mCherry (Chowdhary et al., 2019) ( Figure 2E). Under non-heat shock conditions, HSP104-lacO 256 and TMA10-tetO 200 were spatially separated, as indicated by a large 3D distance between the two loci (d =1.6 µm). This is consistent with the respective locations of HSP104 and TMA10 on the far left and right arms of chromosome XII, as well as their physical isolation from each other due to the nucleolar barrier between them. However, upon heat shock, HSP104-lacO 256 and TMA10-tetO 200 came in close proximity within 5 minutes of thermal exposure, coalescing into a diffraction-limited single focus after 10 minutes that ultimately To determine if HSR condensates are involved in driving intergenic coalescence, we perturbed Hsf1 clusters and assayed for coalescence using the TaqI-3C and live single cell imaging analyses. First, we induced Hsf1 clusters by subjecting cells to brief heat shock treatment for 2.5 minutes before adding 1,6-HD or 2,5-HD and performed TaqI-3C ( Figures 2G, S3E). We detected strong chromatin interactions between HSR genes in cells heat-shocked for 2.5 minutes. The 3C interactions were only moderately reduced in the presence of 2,5-HD, including intragenic interactions within the HSR genes. By contrast, addition of 1,6-HD reduced the contact frequencies to background levels. Next, we pre-treated cells with 1,6-HD or 2,5-HD prior to heat shock and performed single cell imaging of the strain bearing fluorescently tagged alleles of HSP104 and TMA10. The coalescence of HSP104-lacO 256 and TMA10-tetO 200 was abolished in cells that were pretreated for 5 min with 5% 1,6-HD before heat shock but occurred normally in cells treated with 5% 2,5-HD, matching the cells with no alcohol pretreatment ( Figure S3F). Quantification revealed that the mean 3D distance decreased from >1.0 µm to <0.7 µm upon heat shock over a population of cells subjected to either 5% 2,5-HD or no pre-treatment, whereas the loci remained separated by >1.0 µm in 1,6-HD-treated cells ( Figure 2H). These results indicate that 1,6-HD not only dissolves the clusters of Hsf1, but also disrupts intergenic interactions between HSR genes.

Hsp70 binding represses Hsf1 cluster formation
Do the molecular mechanisms that control Hsf1 activity also regulate HSR condensate formation? Hsf1 activity is repressed when it is bound by the chaperone Hsp70 ( Figure 3A Figure 3A, B). We expressed GFP-tagged Hsf1-ce2AAA as the only copy of Hsf1 and imaged its localization. In contrast to the diffuse nuclear localization of wild type Hsf1, we observed constitutive clustering of Hsf1-ce2AAA in unstressed cells that persisted during heat shock ( Figure 3C).
To determine whether the constitutive clusters formed by Hsf1-ce2AAA are transcriptionally active, we measured mRNA levels of representative HSR genes in cells expressing wild type Hsf1 and Hsf1-ce2AAA ( Figure   3D). In agreement with the known repressive role for Hsp70 binding, we detected a significant increase in the basal mRNA levels of some targets in Hsf1-ce2AAA cells compared to cells expressing wild type Hsf1. However, the basal levels of most target transcripts in the mutant were lower than their respective levels in wild type cells following heat shock, suggesting that the constitutive clusters formed by Hsf1-ce2AAA are less active than those formed by wild type Hsf1 upon activation. Further supporting a lack of full activation in the mutant, both Med15 and Rpb3 were diffusely localized under basal conditions in Hsf1-ce2AAA cells and only formed clusters with Hsf1-ce2AAA upon heat shock ( Figure 3F). Moreover, HSR genes showed no detectable intergenic interactions as measured by TaqI-3C under basal conditions in Hsf1-ce2AAA cells, but both intra-and intergenic interactions were induced upon heat shock to the same degree as in wild type cells (Figures 3E and   S4B). Thus, mutation of the CE2 binding site for Hsp70 results in constitutive Hsf1 cluster formation, but this is not sufficient to trigger formation of HSR condensates that drive intergenic interactions.    (Figures 4B and S4C). To assess their roles in intergenic interactions and HSR condensate formation, we separately removed both regions.

The N-terminal domain of Hsf1 promotes formation of HSR condensates
Since the NTA represses Hsf1 basal activity, we expected its removal would mimic mutation of the CE2, resulting in constitutive clustering. However, Hsf1-DNTA-GFP failed to form clusters even upon heat shock ( Figure   4C, D). Moreover, Rpb3 and Med15 also could not re-distribute into clusters in the Hsf1-DNTA mutant, suggesting that interaction of Hsf1 with Mediator and Pol II is required for clustering during heat shock ( Figure 4C,   D). These data further support the interpretation that clustering of Hsf1, and additionally Med15 and Rpb3, is driven by regulatory activation rather than heat per se. At the molecular level, we found that while Hsf1-∆NTA was able to robustly bind to the promoter of the target gene HSP104 and recruit RNA Pol II upon heat shock, it was unable to recruit high levels of Med15 ( Figure 4F). Since the rate of inducible transcription of HSP104, and consistently the intragenic interactions within HSR genes ( Figure S4A), was only modestly reduced in the

Reprogramming Hsf1 to mimic mammalian transcriptional condensates
HSR condensates share biophysical properties and molecular constituents with mammalian transcriptional condensates that form at SEs, yet HSR condensates are distinct in their inducible and transient kinetics and their ability to drive intergenic interactions. Since loss of CE2 resulted in constitutive Hsf1 clustering and partial de-repression of Hsf1 transcriptional activity, and loss of the NTA reduced intergenic interactions and also partially de-repressed Hsf1, we reasoned that combining these mutations would fully de-repress Hsf1 and reprogram Hsf1 to be mammalian-like: constitutively clustered but lacking intergenic interactions. To test this, we generated a double mutant, Hsf1-∆NTAce2AAA, GFP-tagged it and expressed it as the only copy of Hsf1 from the endogenous HSF1 promoter. Indeed, Hsf1-∆NTAce2AAA-GFP formed constitutive clusters and was fully de-repressed, driving transcription of some target genes under basal conditions to levels exceeding their expression during heat shock in wild type cells ( Figure 4H, I). In contrast to the constitutive clusters we observed in the Hsf1-ce2AAA single mutant, Hsf1-∆NTAce2AAA-GFP colocalized in clusters with Med15-and Rpb3-mCherry under basal conditions ( Figure 4J). Despite this high transcriptional activity and clustering, Hsf1-∆NTAce2AAA was unable to drive intergenic interactions among HSR genes, even during heat shock ( Figure 4K). Importantly, Hsf1-∆NTAce2AAA cells showed impaired growth, especially at elevated temperature, suggesting that the ability to form inducible intergenic transcriptional condensates promotes fitness during stress ( Figure 4L). Thus, the CE2 and NTA are determinants that endow HSR condensates with inducibility and the capacity to drive intergenic interactions, features that distinguish them from the transcriptional condensates described at SEs in mammalian cells ( Figure 4M).  A) Cartoon depiction of Mediator recruitment at HSR genes via dual Hsf1 activation domains. B) Domain map and prediction of disorder in Hsf1 by IUPRED. C) Live imaging of cells expressing Hsf1ΔNTA -GFP and Med15-mCherry under NHS or HS conditions. D) Live imaging of cells expressing Hsf1ΔNTA -GFP and Rpb3-mCherry under NHS or HS conditions. E) Transcription rates of HSP104 in HSF1 + and HSF1-ΔNTA isogenic strains during a heat shock time course, as deduced from RT-qPCR. F) Hsf1, Med15-myc13 and Rpb1 ChIP analysis at the HSP104 locus in wild type and Hsf1-ΔNTA strains heatshocked for 7.5 min. Depicted are means + SD. **, P <0.01 (calculated using two-tail t-test). G) Intergenic contacts (solid arcs) between indicated Hsf1 target gene pairs in 7.5 min-HS conditions as determined by TaqI-3C. H) Representative 3D rendered micrographs of cells expressing Hsf1-GFP and Hsf1-ΔNTA-ce2AAA-GFP under NHS and HS (axes: x (red), y (green) and z (blue)). I) mRNA expression measured by RT-qPCR of representative Hsf1-regulated genes in HSF1 + and HSF1-ΔNTA-ce2AAA strains under NHS and 10 min-HS conditions. **P <0.01; *P <0.05 (calculated using multiple unpaired t tests) J) Live imaging of cells co-expressing Hsf1ΔNTAce2AAA -GFP with Med15-mCherry (top) or Rpb3-mCherry (bottom) under NHS conditions. K) Intergenic contacts (solid arcs) between indicated Hsf1 target gene pairs following 10 min-HS conditions as determined by TaqI-3C. L) Spot dilution analysis of HSF1 + and HSF1-ΔNTAce2AAA cells. Plates were incubated at 30ºC or 37ºC for 30 h. M) Schematic depiction of HSR condensation. In unstressed cells, Hsp70 binds to Hsf1 via the CE2 domain and inactivates Hsf1. Accumulation of misfolded proteins sequesters Hsp70 and frees Hsf1 which then binds to its gene targets and forms initial clusters. The activated Hsf1 recruits Mediator via the NTA domain. Hsf1, together with Mediator and RNA Pol II, forms dynamic transcriptional condensates (HSR condensates) that localize multiple Hsf1 target genes engaging in specific interactions with each other. The HSR condensates with intergenic coalescence promote high transcriptional activity, maximal gene expression and cellular fitness during stress.

DISCUSSION
Here we have shown that activation of the HSR in budding yeast involves the formation of inducible biomolecular condensates that localize multiple target genes into subnuclear compartments with high transcriptional activity. These HSR condensates form and dissolve over the course of heat shock and contain biophysically dynamic clusters of the transcription factor Hsf1 along with Mediator and RNA Pol II ( Figure 4M). Our discovery of HSR condensates in yeast demonstrates that transcriptional condensates exist outside of the mammalian lineage and suggests that these structures are likely an ancient and conserved feature of eukaryotic gene control.  (Figure 4C, D, G), it is likely that cooperative binding among the IDRs of Hsf1, Mediator and RNA Pol II ( Figure S4G, H) could serve to nucleate the HSR condensates and bring HSR genes in 3D proximity ( Figure 4M). The other possibility is that Hsf1 initiates the condensation. This initiation mechanism could entail cooperative binding of Hsf1 trimers to multiple HSEs of canonical gene targets (Erkine et al., 1999), with each activator molecule then tethering or engaging one or more polymerases and/or Mediator subunits (Baek et al., 2021). These local multivalent interactions between Hsf1 and Mediator at individual HSR genes could then intermix with other such local assemblies leading to intergenic coalescence. Finally, HSR condensates may provide several advantages to cells under stress ( Figure 4M). First, condensation serves to concentrate large quantities of the transcriptional machinery to enable robust induction of Hsf1-regulated genes critical to counteracting the stress. Second, the coalescence of multiple target genes separated by large distances on and between chromosomes could reduce search time for repeated transcriptional firing events and coordinate transcriptional bursting across the HSR regulon. Third, spatially coordinated transcription of multiple genes could promote mRNA clustering into "transperons" that may bypass normal mRNA quality control processes and increase efficiency of post-transcriptional steps of gene expression such as nuclear export (Nair et al., 2021;Zander et al., 2016). Fourth, HSR condensates may enable target messages to avoid cytosolic stress granules and thereby promote their privileged translation (Zid and O'Shea, 2014), perhaps by recruiting translation factors or mRNA modifying enzymes. HSR condensates represent the founding example of inducible intergenic transcriptional clusters, but given these potential advantages, it may be the case that analogous structures have evolved to execute other stimulus-regulated transcriptional programs within budding yeast and beyond.

Yeast Strains
For tagging Hsf1, Med15 and Rpb3 with mCherry, PCR amplicons with mCherry coding sequence and homology to 3'-ends of either HSF1, MED15 and RPB3 were amplified from plasmid pFA6a-mCherry-hphMX6. The amplicons were transformed into the respective strains for in-frame insertion of mCherry. Primers used in strain construction are listed in Table S6. For Myc x 13 tagging of Med15, plasmid pFA6a-13Myc-His3MX was used as template to obtain MYC13-HIS3 amplicon with homology to 3'-end of MED15. This amplicon was transformed in strains DPY144, DPY417 and DPY418 to obtain strains ASK215, ASK216 and ASK217, respectively.
The diploid strain ASK741 was created by crossing a MATα derivative of strain DSG200 (Chowdhary et al.,

2019) with MATa DSG200.
Plasmids pNH604-HSF1pr-HSF1-GFP and pNH604-HSF1pr-HSF1(147-833)-GFP were used as templates for quick change PCR (Primers, Fw subCE2-AAA and Rv subCE2-AAA) to create CE2->AAA mutation in WT and ∆NTA HSF1, respectively. These plasmids were linearized with Pme I (New England Biolabs) and transformed in strain DPY034 for integration at the TRP1 locus to obtain strains DPY1805 and DPY1818. Loss of parental HSF1 plasmid was confirmed by growth on 5-FOA media.
A complete list of strains is provided in Table S1. PCR primer sequences are provided in Table S6.

Culture Conditions
For microscopy, cells were grown at 30°C in SDC (synthetic dextrose complete) media to early-log density (A 600 = 0.4-0.5).
For 3C, ChIP and RT-qPCR analyses, cells were grown at 30°C in YPD (yeast extract-peptone-dextrose) to a mid-log density (A 600 = 0.65-0.8). A portion of the culture was maintained at 30°C as non-heat-shocked (NHS) sample while the remainder (heat-shocked (HS) sample) was subjected to an instantaneous 30°C to 39°C thermal upshift for the indicated duration.
For spot dilution analysis, cells were grown to stationary phase in YPD media. Master suspensions were prepared by diluting the saturated cultures to a uniform cell density (A 600 =0.3) and were transferred to a 96-well microtiter plate. These were then serially diluted (five-fold). 4 μl of each dilution were transferred onto solid YPDA plates. Cells were grown at either 30º or 37ºC for 30 h.

Chromosome Conformation Capture
TaqI-3C was conducted essentially as previously described (Chowdhary et al., 2017;2020;Chowdhary et al., 2019). Briefly, cells were cultured to a mid-log density (A 600 = 0.8) at 30°C. They were either maintained at 30°C or heat-shocked at 39°C for 10 min (or as indicated), and then crosslinked with formaldehyde (1% final concentration). For 3C assay involving hexanediol treatment, cells were heat-shocked at 39°C for 2.5 min followed by treatment with either 2,5-or 1,6-hexanediol (5% final concentration), and then crosslinked with formaldehyde. Crosslinked cells were harvested and lysed in FA lysis buffer (50 mM HEPES pH 7.9, 140 mM NaCl, 1% Triton X-100, 0.1% Sodium deoxycholate, 1 mM EDTA, 1 mM PMSF) for two cycles (20 min each) of vortexing at 4°C. A 10% fraction of the chromatin lysate was digested using 200 U of Taq I (New England Biolabs) at 60°C for 7 h. Taq I was heat-inactivated at 80°C for 20 min. The digested chromatin fragments were centrifuged, and the pellet was resuspended in 100 μl of 10 mM Tris-HCl (pH 7.5). The Taq I-digested chromatin was diluted 7-fold to which 10,000 cohesive end units of Quick T4 DNA ligase (New England Biolabs) were added. Proximity ligation was performed at 25°C for 2h. The ligated sample was digested with RNase at 37°C for 20 min and then Proteinase K at 65°C for 12 h. The 3C DNA template was extracted using phenol-chloroform and then precipitated.
The interaction frequencies were quantified using qPCR. Quantitative PCR was performed on a CFX Real-Time PCR system (Bio-Rad) using Power SYBR Green PCR master mix (Fisher Scientific). Sequences of 3C primers used in this study are provided in Table S2. Normalization controls were used to account for the following: (i) variation in primer pair efficiencies; (ii) primer dimer background; (iii) variation in the recovery of 3C templates; and (iv) to ensure a ligation-dependent 3C signal. For detailed algorithms to calculate normalized 3C interaction frequencies, see below and (Chowdhary et al., 2020).

Chromatin Immunoprecipitation (ChIP)
ChIP was conducted essentially as previously described (Chowdhary et al., 2019). Briefly, the cells were heatshocked at 39°C for 7.5 min (or maintained at 30°C) and crosslinked with 1% formaldehyde. Cells were then harvested and subjected to glass bead lysis in lysis buffer (50 mM HEPES pH 7.5, 140 mM NaCl, 1% Triton X-100, 0.1% Sodium deoxycholate, 1 mM EDTA, 2 mM PMSF, and 250 µg/ml cOmplete™, EDTA-free Prote- ChIP occupancy signals were quantified using qPCR. Sequences of ChIP primers used in this study are provided in Table S4. The ChIP DNA quantities were deduced from interpolation of a standard curve generated using genomic DNA template. The qPCR signal for each primer combination was normalized to the corresponding signal from the input DNA. The input DNA control was incorporated to correct for variation in the recovery of ChIP DNA templates.

Reverse Transcription-qPCR (RT-qPCR)
Cells were cultured to a mid-log density (A 600 = 0.8) at 30°C, and were either maintained at 30°C or heatshocked at 39°C for times indicated. 20 mM sodium azide was added at appropriate times to terminate transcription. Cells were then harvested and subjected to glass bead lysis in presence of TRIzol (Invitrogen) and chloroform for 10 min at 4°C. Total RNA was precipitated in ethanol. A fraction of total RNA (~20 µg) was treated with DNase I (RNase-free; New England Biolabs) at 37°C for 15 min. DNase I was heat-inactivated at 75°C for 10 min. RNA was purified using the RNA clean and concentrator kit (Zymo Research). 2-3 µg of the purified RNA template and random hexamers were used for preparing cDNA (Superscript IV first-strand synthesis system; Invitrogen).
The cDNA reaction mix was diluted 2-fold, and 2 µl of the diluted cDNA template was used for qPCR reaction.
Sequences of primers used for RT-qPCR analysis in this study are provided in Table S5. Relative cDNA levels were quantified using the ∆∆Ct method (see Chowdhary et al., 2017). qPCR signal from SCR1 Pol III transcript was used as a normalization control. This accounted for variation in the recovery of cDNA templates.
Relative fold change per minute in mRNA expression was calculated by dividing mean mRNA levels (derived from two independent biological samples) for a given time point by previous time point, and then by the time elapsed in minutes.

Fluorescence microscopy
For live-cell imaging, cells were grown at 30°C to early log phase (A 600 = 0.5) in synthetic dextrose complete (SDC) medium. An aliquot of living cells was immobilized onto concanavalin A-coated glass bottom dish. Fresh SDC medium was added in the dish before imaging. Images were taken on Leica TCS SP5 II STED-CW laser scanning confocal microscope (Leica Microsystems, Inc., Buffalo Grove, IL) equipped with GaAsP hybrid detector. Samples were subjected to heat shock (39°C) by heating the objective using Bioptechs objective heater system and by controlling the temperature of incubator chamber enclosing the microscope. For experiments involving hexanediol treatment, cells were heat-shocked and imaged for ~8 min. Immediately after this, the SDC media in the dish was replaced with pre-warmed (39°C) SDC supplemented with either 2,5-or 1,6-hexanediol (5% final concentration), and cells were imaged at 39°C for the times indicated. For fixed-cell imaging, cells were grown at 30°C to early log phase in YPD, subjected to instantaneous heat shock at 39°C for the indicated times, and then fixed in 1% formaldehyde for 7 min. Cells were harvested and the pellet was washed with 1X phosphate-buffered saline (PBS) (pH 7.4). An aliquot of cells was immobilized onto concanavalin A-coated glass bottom dish, and then imaged using Leica TCS SP5 II STED-CW laser scanning confocal microscope (Leica Microsystems, Inc., Buffalo Grove, IL). Images were acquired and processed as above.
For colocalization analyses, the fractional overlap metric scores (Manders' colocalization coefficients) were calculated using JACoP plugin implemented in ImageJ (Bolte and Cordelieres, 2006;Schneider et al., 2012). The average intensity plots were generated using the plot profile feature in ImageJ. Intensities of pixels were obtained along the line path (as indicated in Figure 1G) for each z-section. Intensities across nine individual z-sections were combined and the average was plotted for each channel.
For analysis of gene coalescence, nine z-planes with an interplanar distance of 0.5 µm, covering the entire depth of the nucleus, were inspected for location of tagged genomic loci. The relative nuclear positions of the tagged loci were assessed by measuring 3D distances between them. The 3D distances were measured as 3D polyline lengths between the signal centroids (green and red spots). A cell was scored positive for coalescence if the 3D distance between the centroids of green and red spots was between 0.3 and 0.7 μm. 3D distance measurements, 3D rendering and visualization were performed in arivis Vision4D.

Super-resolution microscopy
Yeast cells were grown at 30°C to early log phase (A 600 = 0.5) in synthetic dextrose complete (SDC) medium. An aliquot of living cells was immobilized onto concanavalin A-coated glass bottom dish. Fresh SDC medium was added in the dish before imaging. Samples were subjected to heat shock (39°C) by heating the objective using Bioptechs objective heater system and by controlling the temperature of incubator chamber enclosing the microscope. High-resolution Stimulated Emission Depletion (STED) images were acquired on Leica TCS SP5 II STED-CW laser scanning confocal microscope (Leica Microsystems, Inc., Buffalo Grove, IL). Emission depletion was accomplished with 592 nm STED laser. Images were deconvolved as above, rendered (render mode: volumetric) and visualized in arivis Vision4D.

Hsf1, Med15 and Rpb1 ChIP-seq data analysis
Hsf1 ChIP-seq data were obtained from GSE117653 . Med15 data are from PRJNA657372 Wiggle files were generated by normalizing to the library size (--normalizeUsing CPM) using the bamCoverage function of deepTools2 and visualized in IGV genome browser.

Analysis of average relative localization of factors
Foci of LacO-tagged HSP104 gene locus or MS2-tagged HSP12 mRNA locus were manually identified in a specific z plane and centered within a square of 12 x 12 pixels. Distribution of signal from other channel, depicting distribution of secondary factor, was gathered from the corresponding 12 x 12 squares in the same z plane. A composite text image was created by computing average of centered images from roughly 30 cells per condition. The text images were used to generate contour plots. The contour plots were created using the Plotly function (Plotly Technologies Inc., 2015) in R (R Core Team, 2020). The intensity minima and maxima, spanning the entire range of signals in the given channel, were split into 800 steps.

Quantification of 3C
TaqI-3C data was quantified as described in (Chowdhary et al., 2020;Chowdhary et al., 2019). The percent digestion efficiency was determined by amplifying a region across Taq I restriction site using a pair of convergent primers (sequences provided in Table S3). The percent digestion efficiencies were determined for each primer combination and incorporated into the following formula: Here, Ct R is the cycle threshold (Ct) quantification of the digested only (DO) or undigested (UND) templates, and Ct ARS504 the cycle threshold quantification of the ARS504 locus (a region lacking a Taq I site).
For measurement of normalized frequency of intragenic or intergenic interactions, the Ct values for digested only (DO 3C ) and ligated (Lig 3C ) templates for crosslinked chromatin, and digested only and ligated genomic DNA (Lig gDNA and DO gDNA , respectively) were incorporated into the following formula:  templates; also applicable to the gDNA control. (v): Normalized Frequency of Interaction is defined as the ratio of ligation-dependent signals of 3C and gDNA control templates after correcting for differences in their digestion efficiencies.

Statistical tests used
Student's t test (two-tailed) was used to calculate statistical significance between all pairwise comparisons (as assumptions of parametric distributions were fulfilled), except in Figure 2C, one-way ANOVA followed by Tukey's post hoc analysis was used. Each pairwise comparison is done with means of two independent biological samples (N=2) +SD. n.s., P>0.05; *, P<0.05; **, P<0.01; ***, P<0.001.