Introduction

Chromosomes form extensive 3D contacts that can occur over long linear genomic distances (Marenduzzo et al. 2007; Fazary et al. 2017; Misteli 2020). A subset of these interactions, such as promoter-promoter or promoter-enhancer looping, are thought to play a central role in gene regulation (Mifsud et al. 2015; Schoenfelder and Fraser 2019). In some cases, multiple genes come together in 3D space to form “multi-gene clusters” (Mir et al. 2018; Lim and Levine 2021; Zhu et al. 2021; Uyehara and Apostolou 2023). These clusters often involve co-regulated genes responding to stress or other types of environmental cues, such as heat shock, starvation, virus infection, or developmental signals (Ferrier and Minguillón 2003; Sawado et al. 2003; McCord et al. 2011; Du and Bai 2017; Zhao and Faryabi 2023). Evidence from imaging and genomic studies suggests that these gene clusters may be related to a phenomenon called “transcription factories”, where RNA polymerase II (Pol lI), mediator, and nascent transcripts coalesce into distinct foci (Bartlett et al. 2006; Faro-Trindade and Cook 2006; Carter et al. 2008; Xu and Cook 2008; Sutherland and Bickmore 2009; Papantonis and Cook 2011; Rieder et al. 2012). Proteomic analysis of transcription factories showed that they are also enriched with other components in the transcription pathway, including transcription factors (TFs), histone modification enzymes, and chromatin remodelers (Lyons et al. 2023). Overall, these observations suggest that some co-regulated genes can physically cluster into sub-nuclear compartments with elevated local concentrations of transcription-related factors.

How multi-gene clusters are formed is not well understood. Since co-regulated genes are often activated by a common set of TFs (Kiran and Nagarajaram 2016; Mir et al. 2018; Tsai et al. 2019; Uyehara and Apostolou 2023), such clustering may be mediated by the DNA-protein and protein-protein interactions of the TFs. Recent studies showed that TFs like Gcn4, OCT4, and TAZ, as well as some co-factors such as the mediator and histone-acetylation reader Brd4, tend to form biomolecular condensates through liquid-liquid phase separation (LLPS) (Whyte et al. 2013; Boija et al. 2018; Sabari et al. 2018; Gibson et al. 2019; Lu et al. 2020; Lee et al. 2021). In LLPS, multivalent interactions among proteins, often through their intrinsically disordered regions (IDRs), allow them to condensate into liquid-like droplets that are separated from the aqueous phase (Banani et al. 2017; Dignon et al. 2020; Peng et al. 2020; Mitrea et al. 2022). The existence of TF / co-factor condensates is supportive of the “transcription factory” model mentioned above where transcription-related factors are locally concentrated through LLPS. So far, LLPS of TFs and co-factors are often studied in vitro using purified systems or in vivo with artificial induction. How these proteins behave in a chromatin context at the endogenous concentration, and more specifically, if they contribute to multi-gene cluster formation, is not very clear.

Another unresolved issue related to multi-gene clusters is their functional role in gene regulation. Earlier studies of gene clusters found a positive correlation between clustering and gene expression level (Li et al. 2012; Whyte et al. 2013; Zhang and Bai 2016; Mir et al. 2018; Zhu et al. 2021; Zhao and Faryabi 2023), but the mechanism of the phenomenon was not clear. This again may be related to TF / co-factor condensates: if a gene cluster is situated within a condensate, higher local concentrations of these factors can potentially enhance expression. This is proposed to be the case at super-enhancers, where the condensates of co-factors like Med1 and Brd4 are thought to drive the high transcriptional activities of the target genes. However, some reports studying TF condensates reveal a complex, sometimes contradictory, relation to gene expression, where the condensates are observed to have positive, negative, or neutral effects on transcription (Lu et al. 2020; Lee et al. 2021). In a recent study of the GAL genes, for example, it is proposed that Gal4 condensation facilitates their recruitment to target genes but does not contribute to gene activation (Meeussen et al. 2023). In general, the relation between TF condensates, multi-gene cluster, and gene regulation needs to be further studied.

Upon met depletion (-met) in budding yeast, Met4 is recruited by several sequence-specific DNA-binding co-activators, including Cbf1, Met31, and Met32, to bind and rapidly activated a subset of genes (Thomas et al. 1992; Lee et al. 2010; Ouni et al. 2010; Carrillo et al. 2012). This activator complex is further stabilized with Met28 (Kuras et al. 1997). In a previous study, we carried out a screen to probe long-distance gene regulation that affected MET3 promoter (MET3pr) activity by inserting an insulated MET3pr-GFP reporter into many genomic loci and measuring GFP intensity in -met (Du et al. 2017). At most locations, the reporter has comparable induction level; however, it shows enhanced activity at a small subset of loci, which were defined as “transcriptional hotspots” (Du et al. 2017). These hotspots tend to be located near endogenous Met4 target genes, and 3C assay detects physical proximity among the hotspots. These results lead to a hypothesis that hotspots occur at loci where MET genes cluster, and such clustering and transcriptional elevation may be both mediated by the condensates of related TFs.

In this study, we test the hypothesis above using a combination of biochemistry, genetics, genomics, and imaging approaches. We demonstrate that Met4 and Met32 form co-localized puncta in the yeast nuclei upon met depletion. Purified Met32 forms condensate with LLPS properties in vitro, with which Met4 can merge. In relation to chromatin, we show that the loci that are associated with the Met TFs co-localize with the Met4 puncta, and at least a subset of Met4 target genes cluster in 3D nuclear space in a Met4-dependent manner. Functionally, the MET3pr-GFP reporter inserted near the endogenous Met4 target sites shows higher enrichment of Met TFs / Pol II and enhanced transcriptional activity (hotspots). A Met4 IDR truncation that reduces Met4 puncta formation selectively affects the reporter activity at hotspots. Collectively, our results support a model where TFs in the met response pathway form LLPS condensates, organize multiple target genes in their vicinity, and enhance gene expression by creating a local environment with elevated concentrations of transcription-related factors. This may represent a common strategy for cells to concentrate limited resources to mount a strong response to stress.

Results

The trans-activator Met4 and its co-factor Met32 form puncta in the nucleus

Our previous study presented evidence that the “hotspots” for MET3pr transcription may form 3D clusters, but the mechanisms of clustering and transcriptional elevation are unknown. Given the recent reports on TF condensation (Sabari et al. 2018; Lu et al. 2020; Zhang et al. 2022; Okada et al. 2023), we hypothesize that these phenomena may be related to condensates formed by TFs in the MET pathway. To test this possibility, we labeled the endogenous Met4 protein with GFP and visualized Met4 in live cells under Airyscan confocal microscopy (Materials and Methods; see Table S1 for strain list). We observed nuclear-localized Met4-GFP under both the repressed (+met) and activated (-met) conditions, with the latter showing higher intensity. Importantly, the Met4-GFP displays puncta-like structures, especially under -met (Fig. 1A & S1A). As controls, we constructed strains expressing a nuclear-localized free GFP or GFP fused to various chromatin-associated factors, including TFs Cbf1 and Reb1, and Sth1, a subunit of RSC nucleosome remodeler, which also bind chromatin with sequence specificity (Cai and Davis 1990; Ju et al. 1990; Donovan et al. 2023). The GFP signals appear more evenly distributed in these control strains (Fig. 1A). To evaluate the distribution more quantitatively, we calculated the coefficient of variance (CV) of the fluorescent intensities among the pixels inside each nuclei (Materials and Methods, Table S2). Uneven distribution of the GFP molecules would result in large variations in the pixel intensities and higher CVs. The CVs of Met4-GFP in both ±met conditions are significantly higher than free GFP and other nuclear proteins (Fig. 1B). Since the strains in Fig. 1A express GFP at different levels, which directly affect CV, we also calculated the Fano number (variance divided by the mean) to control for this variable. Met4-GFP in the -met condition has the largest Fano number (Fig. 1C & S1B), which indicates that in the activated state it forms the most punctate structures. These data support the notion that Met4 forms condensates in the nuclei upon activation.

Figure 1.

We next carried out the same experiment with Met32, the DNA binding co-factor of Met4. We imaged cells containing endogenous Met32 labeled with mCherry and compared its distribution to free mCherry. Unlike Met4, Met32-mCherry is barely visible in +met, and it is significantly upregulated in -met (Fig. 1D & S1C). In the latter condition, Met32-mCherry also shows uneven distribution with high CV and Fano number (Fig. 1E, F & S1D). We also imaged Met31-mCherry, but its concentration is too low to be detected. When Met4-GFP and Met32-mCherry are tagged in the same strain, their fluorescent signals largely overlap (Fig. 1G & S1E). To quantify the overlap, we extracted the GFP and mCherry fluorescence intensities for all the pixels inside each nucleus and calculated their correlation coefficient. As a control, we carried out the same analysis with co-expressed Sth1-GFP and Met32-mCherry, which appear to be less co-localized (Fig. 1G). Indeed, the Pearson correlation between Met4 and Met32 fluorescence on average was 0.75, much higher than that between Sth1 and Met32 (R = 0.19) (Fig. 1H). Overall, the puncta-like structures of Met4 and Met32 and their co-localization support the possibility that these two factors co-condensate under the -met condition.

Met32 forms condensates with LLPS properties in vitro

LLPS is proposed to be a mechanism that leads to TF condensation (Banani et al. 2017; Boeynaems et al. 2018). To test if this mechanism applies to Met4 and Met32, we first evaluated their propensity to phase separate using the FuzDrop algorithm (Hatos et al. 2022). All the Met activators, including Met4, Met31, Met32, and Met28, are predicted to have higher LLPS probabilities than GFP and mCherry (Fig. 2A). Among these TFs, Met4 and Met32 have particularly high scores, which suggests they function as droplet drivers that can spontaneously phase separate (Vendruscolo and Fuxreiter 2022). In addition, both factors are predicted to have large unstructured domains, a feature known to promote LLPS (Fig. S2A & B).

Figure 2.

We next purified Met4 and Met32 to study their condensation properties in vitro. Like many other recombinant TFs, Met4 is prone to degradation. We managed to extract full length Met4 tagged with 6xHis and MBP on the N- and C-terminal, respectively, although some degradation products could not be avoided (Fig. 2B) (Materials and Methods). We imaged the purified Met4 labeled with Alexa488 in various buffers. Met4 remains diffusive up to 5 μM at physiological salt concentrations and forms aggregates at higher density or lower salt concentrations (Fig. 2C). Although these results suggest that Met4 may not inherently undergo LLPS, we cannot rule out the potential effects from impurities (degradation products) and tags on our observations. In contrast, we managed to purify high-quality Met32 fused with mCherry (Fig. 2B). At the physiological salt concentration with no crowding agents, Met32-mCherry forms condensates at various concentrations from 1 to 20 µM (Fig. 2C & S2C). These condensates show properties consistent with LLPS such as droplet fusion (Fig. 2D) and fluorescent recovery after photobleaching (FRAP) (Fig. 2E). The Met32-mCherry condensates have a 64 ± 9.7s half recovery time, comparable to well-established LLPS proteins such as FUS (5-120 seconds), DDX4 (3-200 seconds), and hnRNPA1 (5-120 seconds) (Nott et al. 2015; Shin et al. 2017; McSwiggen et al. 2019).

To further test potential Met4 / Met32 co-condensation, we added purified Met4 and Met32 together in 150 mM NaCl buffer and imaged in both Alexa488 and mCherry channels. The two proteins are mixed into droplets in solution (Fig. 2F). In these co-localized droplets, Met32-mCherry still undergoes FRAP recovery with a slightly slower rate (83 ± 11s half recovery time) (Fig. S2D). These results are consistent with the idea that Met4 and Met32 can co-condense through a LLPS mechanism in vitro, which may contribute to the Met4 / Met32 puncta formation in vivo.

Met4-activated genes co-localize with Met4 puncta

Since Met32 is a sequence-specific DNA binder, the Met4 / Met32 condensates can potentially contact multiple target genes. This leads to our model in Fig. 3A, which predicts that 1) at least some Met4 activated genes should be co-localized with Met4 condensates, and 2) multiple target genes should cluster in 3D. To test these predictions, we first performed chromatin immunoprecipitation with sequencing (ChIP-seq) of related TFs (Met4, Met32, Met28, Cbf1), as well as RNA-seq, to identify genes directly targeted by Met4. As expected, the bindings of Met4, Met32, and Met28 are strongly induced by met depletion, while Cbf1, an abundant pioneer factor (Kuras et al. 1996; Blaiseau and Thomas 1998; Ouni et al. 2010; Carrillo et al. 2012; Yan et al. 2018; Donovan et al. 2019), shows constitutive binding (Fig. 3B & C, S3A & B). Under the induced condition, we identified 34 regions co-bound by all four TFs (Fig. 3C, Table S3), and these binding sites potentially regulate 46 genes (some sites are located in divergent promoters). Met depletion for 2 hrs causes wide-spread changes in the mRNA level with both up- and down-regulation (Fig. 3D, S3C), reflecting the fundamental role of methionine in metabolism (Thomas and Surdin-Kerjan 1997). Importantly, 40 out of the 46 genes co-bound by all four TFs show significant increased expression under the -met condition (Fig. 3D, Table S4), indicating that they are direct targets of Met4 and its co-activators.

Figure 3.

To determine if the Met4 target genes are co-localized with the Met4 puncta, we inserted 196X tetO repeats near two of these genes, MET13 and MET6, which allow us to visualize these loci through their association with tetR-mCherry. We also created control strains with the same array inserted near PUT1 and the mating locus MAT, two loci with no Met4 association. These chromatin dots were imaged together with Met4-GFP under ±met conditions. Visually, MET13 and MET6 loci have higher probabilities to be co-localized with Met4 puncta in -met compared to MAT and PUT1 (Fig. 3E). To test this more quantitatively, we aligned these images with the chromatin dots at the center and averaged the surrounding GFP signals (Materials and Methods). For all four loci, the averaged GFP intensities are higher near the center because the distant pixels are more likely to be outside nuclei boundaries, where Met4 signals are absent (Fig. 3F & G). In +met condition, Met4-GFP has low intensities. Upon activation, we can see localized GFP dots with heightened peak intensities at the MET13 and MET6 loci in comparison to controls (Fig. 3F & G). These results indicate that at least some Met4 target genes are associated with Met4 condensates upon activation, presumably through sequence-specific co-binders like Met32.

Met4-dependent clustering of MET genes upon induction

We next tested the second prediction of the model in Fig. 3A, i.e., Met4 condensates mediate the 3D clustering of target genes. We first carried out a Hi-C measurement under -met condition, which did not show any significant interactions among Met4 targets (Li et al. 2023). Given the limitations of Hi-C in detecting fine-scale interactions (Oluwadare et al. 2019), we used an assay recently developed in our lab, Methyltransferase Targeting-based chromosome Architecture Capture (MTAC) (Li et al. 2023), to identify potential chromosomal interactions with selected Met4 targets (Materials and Methods). In this method, a 256X LacO array is inserted near the gene of interest (viewpoint, or VP) to recruit LacI fused with DNA methyltransferase M.CviPI, which methylates proximal cytosines in the “GC” context in cis and in trans (Fig. 4A). Physical proximity to the VP can thus be evaluated based on the enrichment of the methylation signal in comparison to a control strain containing the same LacI-M.CviPI but no LacO array.

Figure 4.

In a previous study (Li et al. 2023), we performed MTAC assays with a VP near the MET6 gene under ± met condition. This VP 3kb away from MET6 captures highly-specific inter-chromosomal interactions with four other Met4-targeted genes under -met, but not +met, condition. To further illustrate this point, we carried out MTAC with MET13 as the VP (Fig. 4B). Under +met condition, MTAC only detects interactions from the adjacent local chromatin (Fig. 4C, S4A). When the same strain was switched to the -met condition, MTAC captures two far-cis interactions (intra-chromosomal interactions that are over 30kb apart) with two promoter regions near Met4 targeted genes (STR3, MUP1) (Fig. 4C, S4A). Interestingly, these two regions were also found to interact with the MET6 VP (Li et al. 2023). Taken together, these data support the clustering of a subset of Met4-targeted genes upon activation.

The model in Fig. 3A implies that the clustering of MET genes is mediated by Met4 / Met32. To test this idea, we carried out MTAC using the MET6 VP ± acute auxin-induced degradation of Met4 ± met (Fig. S4B & C). For the -met experiment, auxin (IAA) was added 30 mins prior to the met depletion so that Met4 activation would be largely eliminated (Fig. 4D). This would also prevent the accumulation of Met32, as it relies on Met4 for induction (Menant et al. 2006; Lee et al. 2010; Carrillo et al. 2012). Consistent with our previous observations, very few long-distance interactions were found in the +met repressed condition, regardless of the presence or absence of Met4 (Fig. 4E, (1) and (3)). Upon met depletion without IAA addition, we observed significant trans (inter-chromosomal) interactions between the MET6 locus with three other Met4-targeted genes (MUP1, STR3, YKG9) (Fig. 4E, (2)). Importantly, these interactions disappear when Met4 was degraded (Fig. 4E, (4)). These data show that the presence of Met4, and likely Met32, is required for MET gene clustering. This is consistent with the model in Fig. 3A that clustering is mediated by Met4 / Met32 condensates.

Regions near Met TF binding sites constitute MET “transcriptional hotspot” in haploid yeast

We next investigated the functional role of Met4 condensates in gene expression. In our previous screen, we identified the MET13 locus on chr7 as a “transcriptional hotspot”: when the S.kud MET3pr-GFP reporter was inserted into MET13 or neighboring genes, it showed higher GFP expression than most of the other insertion sites upon induction (Du et al. 2017). Combined with the observation that MET13 locus co-localizes with the Met4 puncta (Fig. 3E-G), this suggests a potential link between the higher reporter activity and Met4 / Met32 condensates. However, the initial screen was performed in diploids, while all the experiments above were carried out in haploids. We therefore first tested if MET13 also functions as a transcriptional hotspot in haploid yeast. We again used the S.kud MET3pr for this test, as the sequence of this promoter deviates significantly with the endogenous MET3pr, allowing them to be differentiated in PCR (Du et al. 2017).

We created two haploid strains with the S.kud MET3pr-GFP reporter inserted near the MET13 gene (CWC23 and RSM23) and measured its transcriptional profile (Fig. 5A) (Materials and Methods). For comparison, we also inserted the same reporter into two non-hotspot basal loci ATG36 and LDS2 (Fig. 5A). The GFP intensities from the reporter at CWC23 and RSM23 loci are ∼50% higher (P-value < 1e-4) compared to the control under the -met condition (Fig. 5B). Reporter GFP mRNA levels, as well as Pol II associated with the GFP ORF, are also significantly higher in the former strains (Fig. 5C & D). These results are consistent with our previous observations in diploid cells that the MET13 locus promotes the transcriptional activity of the MET3pr.

Figure 5.

Given the model in Fig. 3A, we reasoned that the enhanced activities of the GFP reporter may be due to increased local concentrations of TFs, which likely lead to higher TF occupancies. To test this idea, we performed ChIP of Met4 and Met32 followed by qPCR on the S.kud MET3pr at the hotspot vs basal genomic locations because it contains a Met32 motif. Consistent with our expectation, we observed higher enrichment of Met4 and Met32 over the S.kud MET3pr at hotspot locations (Fig. 5E & F). Taken together, these findings confirm the existence of “hotspot” loci in haploid yeast and suggest that it may be caused by enhanced binding of Met32 and Met4.

Reporter activities near Met4 binding sites are enhanced over a ∼40kb range

Besides MET13, other loci that colocalize with Met4 puncta, like MET6 (Fig. 3E-F), should also function as transcriptional hotspots. To test this idea, we first verified that MET6 remains associated with Met4 puncta after the reporter insertion. We used the MET6 yeast strain in Fig. 3E and integrated a S.kud MET3pr reporter nearby (Fig. 6A). This reporter contains a frame-shifted GFP (fsGFP) in order not to interfere with the Met4-GFP signal. As controls we integrated the same reporter near two genomic regions not bound by Met TFs (PUT1 and ATG36). We imaged these strains in -met and analyzed them with the same method used in Fig. 3F. Interestingly, despite the fact that S.kud MET3pr is bound and activated by Met4, insertion of this promoter does not increase the local Met4 intensity at MET6 and PUT1 (Fig. 6B & C, ± rep). In comparison with the reporter-containing PUT1 and ATG36 loci, MET6 again shows stronger co-localization with the Met4 puncta (Fig. 6B & C). This finding indicates that S.kud MET3pr alone is insufficient in nucleating a new Met4 condensate or recruiting the integrated locus to an existing condensate.

Figure 6.

We then integrated the original S.kud MET3pr-GFP reporter into MET6, ATG36, and LDS2 (the latter two are controls far from any endogenous Met4 binding sites) (Fig. 6D) and compared their GFP expressions. GFP shows significantly higher intensity at MET6 in comparison with the two basal loci (P-value < 1e-4) (Fig. 6E). To test the generality of this observation, we inserted the reporter into two additional genomic loci associated with Met TFs, MET17 and RMD6, and the reporter shows higher GFP expression in both loci (Fig. 6E). This data supports a model where Met4 condensates associated with the endogenous MET genes enhance the activity of a nearby reporter. This model also predicts that such enhancement effect should be specific to the met pathway, i.e., inducible genes activated by other TFs should not benefit from higher local concentrations of MET activators. We tested this idea by inserting into the same loci a GFP reporter driven by either MET17pr, a promoter in the same pathway, or by GAL1Spr, a promoter controlled by a different set of TFs in another pathway (Rohde et al. 2000; Slot and Rokas 2010). Indeed, MET17pr shows the same trend as MET3pr, while the strengths of GAL1Spr remain constant across all these loci (Fig. 6F). These results demonstrate the pathway-specificity of the transcriptional hotspots.

We next conducted experiments to determine the genetic range of the transcriptional hotspot. We inserted the GFP reporter driven by MET3pr or GAL1Spr at various distances from the Met4 binding site in the MET6 promoter (Fig. 6G). MET3pr-GFP shows higher expression level when inserted up to 30kb upstream and 10kb downstream of the Met4 site. Beyond this region, the MET3pr activity returns to the basal level (Fig. 6H). In contrast, GAL1Spr shows similar activities among these insertion sites (Fig. 6H). These data indicate that, at least near the MET6 gene, the elevated expression of the MET3pr can occur over a ∼40kb region.

The deletion of a disordered region in Met4 reduces puncta formation and reporter expression at transcriptional hotspots

To determine the relation between Met4 condensates and its gene regulatory function more directly, we introduced mutations to Met4 to reduce its propensity to form condensates. Since LLPS is often promoted by IDRs, we first performed PONDR analysis of Met4 (Xue et al. 2010). This analysis reveals three distinct stretches of IDRs interspersed with two previously annotated functional domains: the activation domain (aa95-144), which interacts with mediator, and the auxiliary domain (aa312-375), which interacts with Met31/32 (Fig. 7A) (Kuras and Thomas 1995). AlphaFold also predicts that Met4 is largely devoid of folded structures (Fig. S2A) (Jumper et al. 2021). To probe the effect of these IDRs on Met4 puncta formation, we first truncated individual IDR stretches: ΔIDR1 (aa1-69), ΔIDR2 (aa117-359), or ΔIDR3 (aa397-651). As Met4 is essential for cell viability in the -met media, we constructed strains containing the endogenous unlabeled Met4 with an additional copy of truncated Met4-GFP driven by the native MET4 promoter. This allows us to visualize Met4 mutants without hindering cell growth (Fig. 7B). Deletions of IDR2 and IDR3 both lead to significant reductions in Met4-GFP Fano number, with ΔIDR2 having a slightly more pronounced effect (Fig. 7B & C, S5A). To further specify the part of IDR2 that affects Met4 condensates, we generated shorter truncations within IDR2, ΔIDR2.1 (aa117-135), ΔIDR2.2 (aa136-270), and ΔIDR2.3 (aa271-359). All three truncations result in intermediate GFP Fano number between the full length Met4 and Met4-ΔIDR2 (Fig. 7C, S5B & C). These findings suggest that multiple IDRs contribute to Met4 condensation, and deleting parts of the Met4 IDR can diminish, but not completely eliminate, Met4 puncta formation.

Figure 7.

When the full length Met4 is replaced by Met4 ΔIDR2.3, the strain is viable in the absence of met, allowing us to evaluate the functional consequence of this mutant on Met4 binding and transcriptional activation. Given our model and supporting evidence that Met4 condensates increase local TF concentration at Met4 targeted genes and enhance their expression, we expect a Met4 mutant less prone to condensate will result in diminished Met4 binding and reduced induction. We therefore compared the ChIP-seq and RNA-seq data of the full-length vs truncated Met4. Consistent with the idea above, the Met4 ΔIDR2.3 shows decreased binding across all the previously annotated Met TF binding sites, without generating new ChIP-seq peaks (Fig. 7D & E). Interestingly, the three Met4 binding sites near MUP1, STR3, and YKG9 that cluster with MET6 show significantly more reduction in Met4 ChIP-seq peaks (reduced to 42±4%) than other Met4 sites (61±8%, P-value = 0.02) in the Met4 ΔIDR2.3 strain. Out of the 40 potential target genes, 22 show significant decreases in their mRNA level (Fig. 7F, S5D, Table S4). Importantly, our model also predicts that decreased Met4 condensation should have more impact on target genes that are located near Met4 puncta than those farther away. To test this idea, we again took advantage of the MET3pr-GFP reporter inserted into hotspots and basal loci. We expect the reduced Met4 condensation to have a larger impact on the reporter activity in the former case because the reporter is only associated with Met4 puncta at hotspots (Fig. 6B). Indeed, after substituting Met4 with Met4-ΔIDR2.3 in these strains, we observed strong decreases in the reporter gene activity in hotspots but minimal effect on those in basal regions (Fig. 7G). These findings collectively support a model wherein Met4 condensates cluster target genes, enhance their activity, and give rise to the formation of transcriptional hotspots.

Discussion

In this study, we investigated TF distribution, 3D genome organization, and gene activation in the met response pathway using a combination of genetics, genomics, and imaging approaches. Our findings connect three emerging phenomena in gene regulation: TF condensates, co-regulated gene cluster, and transcriptional hotspot. More specifically, we show that the activator Met4 and its co-factor Met32 form puncta-like structures that co-localize with at least some Met4 target genes. Multiple MET genes on different chromosomes cluster in the 3D nuclei space, and such clustering requires the presence of Met4. Chromosomal loci near endogenous Met4 binding sites, in one case within a 40kb range, are hotspots for MET3pr reporter expression. Mutations that decrease Met4 puncta formation lead to selective reduction of the reporter transcriptional activity at the hotspots.

Condensate formation of Met4 and Met32

Using live-cell imaging, we found that fluorescently labeled Met4 and Met32 from puncta-like structures in yeast nuclei under the -met condition. This is reflected by the higher Fano numbers in Met4 / Met32 pixel intensities in comparison to free GFP, and in the Met4 case, other chromatin-associated factors. High resolution microscopy of yeast TFs is challenging due to the small size of the nuclei. The low intensities of the Met4 / Met32 signals require high excitation for imaging, which also makes them prone to photo-bleaching. Thus, we were unable to measure Met4 and Met32 puncta dynamics in vivo. When purified in vitro, Met32-mCherry readily forms droplets with typical LLPS properties without molecular crowder, while His- and MBP-tagged Met4 remains diffusive at physiological salt concentration. We also found that Met4 partitions into the Met32 droplets when they are mixed, and inside cells, the two factors show significant co-localization. These observations raise the intriguing possibility that Met32 serves as a scaffold to nucleate the co-condensates of Met4 and other Met TFs. This may explain why the Met4 distribution becomes more puncta-like when switched from +met to - met condition, when Met32 becomes available inside the nuclei. However, we cannot exclude the possibility that the absence of Met4 LLPS is due to our in vitro condition (tag, impurity, lack of post-translational modification, etc.). Met4 is also reported to recruit the Mediator and SAGA complex (Leroy et al. 2006), and these factors, together with Pol II and RNA, may also affect Met4 condensate formation in vivo (Henninger et al. 2021).

TF condensates and the 3D genome topology

TF condensation in vivo occurs in a chromatin context, but the relation between TF condensates and 3D genome topology has not been extensively studied. By recruiting multiple TFs, mediator, and transcription machineries into proximity, TF binding loci can potentially serve as nucleation sites for transcription condensates. Looped enhancers and promoters, as well as clustered genes, may also function as scaffold and stabilize these condensates. This is best illustrated at super-enhancers (SEs), where clustered enhancer sites tend to associate with transcriptional condensates and lead to very high activities of their target genes (Hnisz et al. 2013; Sabari et al. 2018). Synthetic LacO array and endogenous microsatellite repeats were also shown to promote local condensation of associated TFs (Chong et al. 2018). Reversely, TF condensates can mediate long-distance chromosomal interactions by contacting multiple genomic sites and therefore inducing their 3D proximity. This idea is supported by the findings here: 1) multiple Met4-targeted genes are co-localized with Met4 condensates, 2) at least a subset of Met4-targeted genes form clusters with long-distance intra- and inter-chromosomal interactions, and 3) such clustering only occurs in -met activating condition in the presence of Met4. Similar observations were made during yeast heat shock response, where the activator Hsf1 coalesces and mediates transient clustering of heat shock genes (Chowdhary et al. 2019). A few other mammalian TFs were also shown to contribute to 3D genome conformation. For example, YAP/TAZ organizes their target sites into clusters that co-localize with the YAP/TAZ/TEAD1 condensates (Lu et al. 2020). NUP98-HOXA9, a chimeric TF related to the pathogenesis of AML, was also shown to mediate long-distance chromosomal loops between NUP98-HOXA9 bound enhancers and oncogenes into SE-like clusters (Ahn et al. 2021).

Interestingly, a MET3pr reporter that is associated and activated by Met4 only co-localizes with Met4 puncta when inserted near native Met4 binding sites. Also, adding MET3pr reporter gene into these sites does not further increase the local Met4 concentration (Fig. 6B & C). These observations indicate that Met4 binding is not sufficient for a gene to nucleate or be recruited into a Met4 condensate. The underlying mechanism of this observation is unclear, but we speculate that some properties of the endogenous MET genes may facilitate their clustering with Met4 puncta. For example, some MET genes, like MET13 and MET6, are associated with nuclear pore complex (NPC) even in the presence of met (Forey et al. 2021). The tethering by NPCs may increase the chance for these genes to come into proximity, enhancing the local density of Met4 target sites to nucleate the Met4 condensates. In contrast, a reporter gene inserted into random genomic locations may not be optimally positioned to contact NPC and/or other Met4 target genes. Further experiments are required to test these ideas. Notably, interaction with NPC was found to be necessary and sufficient to promote clustering of the GAL1-10 alleles (Brickner et al. 2016).

TF condensates and gene expression

The Met4 condensates represent a local environment with elevated Met4 concentrations, which can promote Met4 binding and lead to enhanced gene expression. Consistent with this idea, MET3pr reporter inserted near these Met4 targets, which also co-localizes with the Met4 puncta, shows increased transcriptional activity. Near MET6, such hyper-activity of the reporter occurs over a 40kb range. According to previous measurement (Bystricky et al. 2004), 40kb linear chromatin explores a 3D space with ∼200 nm diameter, which may reflect the size of the Met4 condensates. The elevation of transcription level is pathway-specific as GAL1Spr-GFP inserted into the same loci does not show the same effect. This finding agrees with the “specialized transcription factory” previously proposed (Bartlett et al. 2006) and argues against a model where the transcription elevation solely relies on general transcription factors like mediator or transcription machinery.

The relation between Met4 condensation and transcriptional activity is further supported by Met4 ΔIDR2.3. This Met4 mutant, in comparison to full length Met4, reduces the Met4 puncta formation and the activities of most of its target genes in the genome. Importantly, ΔIDR2.3 does not have strong effect on the transcriptional activity of MET3pr reporter inserted into basal loci, suggesting that this truncation does not directly impact Met4 integrity and its activation function. Instead, it selectively reduces reporter activity at transcriptional hotspots, consistent with the idea that such reduction is due to the disruption of Met4 condensates.

In terms of mechanism, we speculate that 1) high Met4 local concentration inside the condensates may increase the binding rate of Met4, and/or 2) multivalent interactions among Met4, Met32, and possibly other co-activators may slow down the dissociation of Met4 from its target genes. Both ideas are consistent with our observation that Met4 ΔIDR2.3 shows less binding than the full length Met4. The same mechanism has been proposed for other TF condensates, and in some cases, supported by direct kinetic evidence (Trojanowski et al. 2022). It should be noted that more TF binding does not necessarily lead to an increase in gene expression level, for example, transcription may be limited by a step downstream of TF binding. Consistent with this idea, a recent study reports that Gal4 condensation facilitates its recruitment to target genes but does not contribute to gene activation (Meeussen et al. 2023). In human cells, artificially enhanced condensation of a TF Ews::Fli1 causes sequestering of this TF into the nucleolus, resulting in decreased expression of their target genes (Chong et al. 2022). This indicates that the exact relation between TF condensates and gene expression relies on the nature of the multivalent interaction, location of condensate formation, transcription kinetics, etc., and it needs to be analyzed on a case-by-case basis.

TF condensates and stress response

The sulfur-containing amino acid methionine is a key metabolite for cell survival, and its depletion represents a major stress for budding yeast. Interestingly, the TFs that were found to form condensates in yeast, include Met4, Gal4, and Hsf1, all respond to certain types of environmental stress. Similar observations are made in higher eukaryotes. For example, YAP forms condensates in response to osmotic stress (Cai et al. 2019; Lu et al. 2020), and phyB in plants uses altered LLPS properties to sense light and temperature cues (Chen et al. 2022). In contrast, constitutive TFs in yeast, like Reb1, Cbf1, and Sth1, are more evenly distributed in the nuclei (Fig. 1). The condensation of stress-responding TFs may reflect a functional need to rapidly concentrate related resources into a subset of the nuclei space for more efficient usage. In the case of Met4, its condensates indeed allow yeast cells to mount a stronger response to met depletion. In addition, TF condensates may accelerate TF target search and response rate, facilitate synchronized expression of target genes, and/or enhance the activation specificity. These potential condensate functions need to be further explored.

Materials and Methods

Plasmid and strain construction

Standard molecular methods were used to construct budding yeast strains and plasmids (Table S1). The haploid strain was derived from w303a background, and most genetic modifications here involve homologous recombination. More specifically, to create strains with GFP-tagged endogenous Met4, the GFP-HIS3 fragment was PCR amplified from a plasmid (pJL1) with primers containing upstream and downstream homologous sequences of the endogenous MET4 gene. The amplified PCR product was transformed into haploid yeast to insert into the MET4 locus. A similar strategy was used to create strains with mCherry-tagged endogenous Met32, Met4-TAP, Met32-TAP, Met28-TAP and Cbf1-TAP strains. For strains with tetO-tetR-mCherry labeled loci, variations of plasmid pSR11 were created from containing 196X tetO repeats with flanking homologous sequences to targeted sites were linearized and transformed into Met4-GFP expressing haploid strains. For expression of tetR-mCherry, a plasmid (pMY39) containing the tetR-mCherry insert was linearized and transformed into the ADE2 genomic locus. To create Met4 truncated strains, individual plasmids containing truncated versions of Met4 (pJL1) were created and were linearized with restriction digestion. The linearized plasmid was then integrated into haploid yeast directly upstream of the MET4 locus. For endogenous replacement of MET4 with MET4 (ΔIDR2.3), a plasmid variant of pJL1 with MET4 (ΔIDR2.3) was linearized and transformed to replace the endogenous MET4 gene.

Microscopy and image analysis

For imaging of Met4-GFP and Met32-mCherry, yeast cells were grown in SCD + 10X Met liquid media at 30°C to OD660 ∼ 0.2, washed, and then transferred onto 5 ml of SCD - Met media for 2 hrs for induction. Afterwards the cells were put on a SCD-Met agarose pad and put under the confocal microscope for imaging. Confocal microscopy was performed with Zeiss LSM880 scanning laser confocal (Penn State University Huck Life Sciences Institute) with 488 nm Argon laser (541 nm detection) and 594 nm HeNe laser (648 nm detection) using 63x/1.4 plan apochromat objective. For single cell fluorescent images, the images were taken with Airyscan detector and deconvoluted with Zen Black software.

The individual nuclei from single cells were segmented to extract the nuclear boundaries and pixel fluorescent information of single nuclei with NucleiSplicer (Matlab) and subsequent nuclei information was analyzed with NucleiAnalyzer (Matlab). Variance and standard deviation were calculated from the nuclear pixels and plotted with OriginPro (Table S2). The mean GFP intensities were calculated by subtracting the measure GFP intensities by the mean background intensities of unlabeled cells. Coefficient of Variation (CV) values were calculated by dividing the Std. Dev by the mean GFP values for individual cells. The Fano number was calculated by dividing the Variance by the mean GFP. To normalize the Fano number, the mean Fano number of Free GFP was set to a value of 1 and the rest of the conditions were divided by the mean GFP Fano number (Table S2).

Single cell images of yeast cells expressing tetR-mCherry and Met4-GFP were imaged with 10 z-stacks (0.4 µm). The z-stack images of mCherry and GFP channels were analyzed by DotTracker (MATLAB). For individual cells, the program detects the z-stack with the highest mCherry dot intensity and extracts the GFP image from the same z position. These images were aligned with the mCherry dot at the center. The mean intensity projection of the GFP channel was analyzed with ImageJ.

To measure S.kud Met3pr-GFP expression, yeast cells were grown in SCD+10X Met liquid media at 30°C to OD660 ∼ 0.2, washed, and then transferred onto a SCD-Met agarose pad for induction. After 6 hrs, the agarose pad was put under a Leica DMI6000 B fluorescent microscope for imaging. The GFP fluorescent intensity within each cell boundary was quantified using Celltracker4 (MATLAB) (Zou and Bai 2019).

Met4 and Met32 purification

Constructs of Met32-mCherry, and Met4-MBP were inserted into pET28b(+) backbones with an N-terminal 6x-His tag. Met4 and Met32 genes were cloned from budding yeast genomic DNA. MBP-tag was cloned from Addgene CAT#98651 and fusion constructs were assembled using 2xHiFi mastermix (NEB #E2621L) and transformed into DH5a cells (NEB C2987l). A single cysteine sequence was added to the linker between Met4 and MBP to facilitate maleimide conjugation of fluorophores for imaging. For protein expression, all plasmids were transformed into BL21(DE3)-Sigma32 cells (courtesy of Xin Zhang) and induced according to reported protocol for the Sigma32 cell line (Zhang et al. 2014). Bacterial protein expression was carried out with cells grown in LB supplemented with 50 mg/L kanamycin and 100 mg/L ampicillin at 37°C to OD₆₀₀=0.3, induced with 2 g/L arabinose (TCI A0515). Cells were then grown at 30°C until OD₆₀₀=0.6, induced with 1% isopropyl b-D-1-thiogalactopyranoside (IPTG, Research Products International I56000) and grown overnight at 20°C. Cells were resuspended in wash buffer supplemented with 1 mM phenylmethylsulfonyl fluoride (PMSF), and frozen before lysis.

For protein purification, bacterial cells were lysed with sonication and treated with DNaseI and RNaseA according to published protocols for 15 mins (Carrillo et al. 2012). Met32-mCherry was purified using Ni-NTA chromatography (Thermo Scientific HisPur resin #88222) with proteins buffers A (20 mM Tris pH 8.0, 400 mM NaCl, 5 mM imidazole) and B (20 mM Tris pH 8.0, 400 mM NaCl, 250 mM imidazole) with a stepwise gradient of buffer B. Protein was buffer exchanged into buffer C (20 mM HEPES pH 7.5, 400 mM NaCl, 10% glycerol, 1 mM DTT) and concentrated to at least 80 mM using Amicon Pro centrifugal filters 10,000 NMWL (Milipore #ACS501024). The Met4 construct was purified using Ni-NTA chromatography with proteins buffers A and B with a stepwise gradient of buffer B. Ni-NTA elution was immediately loaded onto an amylose column (NEB E8021S) and purified using a stepwise gradient of buffers A and D (20 mM Tris pH 8.0, 400 mM NaCl, 10 mM maltose). Protein was labeled by incubating with 0.8 molar equivalents of Alexa Flour 488 maleimide (Invitrogen A10254). Proteins were buffer exchanged into buffer C and concentrated to less than 50 mM (to avoid aggregation).

Western blot analysis

Met4-MBP samples were run on a 4-20% SDS-PAGE gel (BioRad #4561094). Gels were wet transferred to a PVDF membrane (BioRad #1620177) at 80V for 3 hrs in Tris-Glycine buffer. Transfer was confirmed by Ponceau staining (Sigma P7170-1L). Membrane was blocked with 5% non-fat milk, washed with TBS with 1% Tween 20 (VWR 0777). Membrane was probed with anti-MBP antibody (Thermo Fisher MA527544). Membrane was stripped using PVDF stripping solution (DOT Scientific 65000-500) according to manufacturer instructions and reprobed with anti-6x-His primary antibody (Thermo Fisher MA1-21315) and conducted identically to the first probing. All incubation steps proceeded overnight. Blots were developed using Pierce ECL substrate (Thermo Scientific 32209) with a 2 mins exposure time.

Droplet formation

Proteins were diluted from 400 mM NaCl storage buffer so that the final concentration of the buffer was 150 mM NaCl, 20 mM HEPES pH 7.4, 2% glycerol. Samples were plated on a microscope slide with a hydrophobic spacer (Invitrogen #S24735) between the slide and untreated coverslip. Droplets were allowed to settle for 5 mins prior to imaging. All droplet formation and imaging was conducted at room temperature.

Fluorescent Recovery after Photobleaching (FRAP)

Bleaching was conducted after three frames using the 594 nm laser at full power for 45 iterations. Fluorescence intensity was monitored every 3 seconds for a total of 270 seconds per experiment. Fluorescence intensity levels were extracted using Zeiss Zen Black software. Fluorescence intensity was normalized to percent of pre-bleach intensity then averaged. Error bars show +/- SD of the normalized fluorescence intensity. FRAP experiments were conducted at 20 µM Met32-mCherry. Experiments were replicated with at least three preparations of each protein. Snapshots of the bleaching process were processed using FIJI.

Chromatin Immunoprecipitation (ChIP)

Yeast strains were grown to OD₆₆₀ of 0.4 in 50 mL of SCD-met medium and crosslinked with 1.39 mL of 37% formaldehyde for 20 mins at room temperature. Crosslinking was quenched by adding 2.7 mL of 2.5 M glycine and incubation for 5 mins. Crosslinked cells were centrifuged at 1882 × g for 3 mins at 4°C, and then washed twice with cold 1x Tris-buffer saline (TBS). Cell pellet was then resuspended in 250 μL of fresh FSPP buffer (50 mM HEPES-KOH, pH 7.5, 140 mM NaCl, 1 mM EDTA, 1% Triton X-100, 0.1% sodium deoxycholate, 1% protease inhibitor cocktail, and 1 mM phenylmethylsulfonyl fluoride (PMSF) and vortexed with ∼300 μL glass beads for 2 cycles of 20 mins with an intervening 10 mins in a 4°C cold room. Another 250 μL of fresh FSPP was added to each sample, and the cap and bottom of tubes were punched with a hot needle so that cell lysate could be collected by centrifuging at 836 × g for 5 mins at 4°C. FSPP (0.5 mL) was added to the cell lysate to resuspend the pellet. The cell lysate was then sonicated using a 30 s on 30 s off cycle at 4°C for 7 cycles. Sonicated cell lysate was transferred to 1.5 mL tube and centrifuged at 17,136 × g for 20 mins at 4°C. Two hundred microliters of supernatant was saved as input. Another 400 μL of supernatant was mixed with 600 μL of FSPP buffer and 20 μL of pre-blocked Magnetic IgG beads and TAP antibody (Thermo CAB1001) or Rpb3 antibody (Biolegend 920204) for 10–12 hrs at 4°C. Chromatin was incubated with the beads overnight at 4°C, and then washed sequentially at 4°C with 1xFA-lysis buffer (50 mM HEPES-KOH, pH 7.5, 140 mM NaCl, 1 mM EDTA, 1% Triton X-100, 0.1% sodium deoxycholate), 1xFA-lysis buffer containing 150 mM NaCl, 1xFA-lysis buffer containing 500 mM NaCl, LiCl buffer (0.25 M LiCl, 1% NP-40, 1% sodium deoxycholate, 1 mM EDTA, and 10 mM Tris-HCl pH 8.0), and lastly TE buffer, pH 8.0. Chromatin was then eluted with 400 μL ChIP elution buffer (50 mM NaCl, 50 mM Tris-HCl, pH 8.0, 10 mM EDTA, 1% SDS). Samples were mixed by rotation at 30°C for 30 mins. Beads were pelleted by centrifugation at 17,136 × g and discarded. Five microliters of 20 mg/mL Proteinase K was added to supernatant; 180 μL of 1xFA-lysis buffer, 20 μL of 10% SDS and 5 μL of 20 mg/mL Proteinase K were added into the input samples. ChIP and input samples were then incubated overnight at 65°C to reverse crosslinking. DNA was extracted by phenol:chloroform:isoamyl alcohol (25:24:1) followed by ethanol precipitation. Input samples were then treated with 200 μg RNase A and precipitated with ethanol after adding 20 μg glycogen. Enrichments of target TFs were quantified by quantitative PCR (qPCR). All ChIP experiments were done in biological duplicates.

RNA-seq

Cells were grown in 5 ml of SCD +10x met overnight. 5 ml of fresh SCD +10x was innoculated with 5 µl of saturated culture until OD₆₆₀ ∼0.2. The culture is collected by centrifugation and washed 3x with sterile H2O. The collected culture is then grown in SCD -met media for 2 hrs to reach OD₆₆₀ ∼0.4 and centrifuged at 300g for 4 mins at RT. The pellet was washed twice with sterile H2O and resuspended in 250 µl of RNA lysis buffer (10mM Tris-HCl pH8.5, 5mM EDTA, 2% SDS, 2% stock 2-mercaptoethanol) and placed on a heat block at 85°C for 20 mins mixing every 2 mins. The mixture was centrifuged at 12,000g for 5 mins and the supernatant was transferred to a new tube. The supernatant was then mixed with 1 ml of trizol and heated at 65°C for 20 mins with mixing in between. Standard trizol/chloroform RNA extraction protocol was followed. The extracted RNA was analyzed with Tapestation to check their integrity. RNA with RIN (>7.5) was used for downstream analysis. Messenger RNA was captured using RNA purification beads (NEB). The eluted mRNA was fragmented and denatured for first and second strand synthesis for conversion into cDNA. The sequencing libraries were constructed using conventional protocols (NEB).

Methyltransferase Targeting-based chromosome Architecture Capture (MTAC)

Yeast cells were grown in 50 ml SCD +10x Met media at OD₆₆₀ 0.3. The cells were centrifuged and washed 3 times with autoclaved H₂O before transfer to 50 ml of SCD-Met and induced with 2 nM β-eastradiol for 2 hrs. After induction the cells were centrifuged at 4000g for 5 mins and collected in 800 µl of TE and 200 µl lysis buffer (0.5 M EDTA pH 8.0, 1 M Tris pH 8.0, 10% SDS) and incubated at 65 °C for 30 mins. The cell lysate (1ml) was then sonicated using a 30 s on 30 s off cycle at 4°C for 11 cycles. The sonicated lysate was centrifuged at 12,000g at 4 °C for 20 mins and the supernatant was transferred to a new tube. The supernatant was mixed with 500 µl for Phenol:Chloroform:Isoamyl Alcohol (25:24:1) nucleic acid extraction. The methylated genomic DNA (5 µg) was incubated with 2 µg of anti-methylcytosine antibody in 1X IP buffer on a rotating platform 4°C for 8-12 hrs. Input DNA (1%, 50 ng) was set aside for normalization. The lysate/antibody complex was incubated with protein A/G agarose beads for 2 hrs and washed 3x with 1x IP buffer. The beads were resuspended with 210 µl digestion buffer and incubated on a rotating platform at 55 °C for 2 hrs. The eluted DNA was precipitated in 1 ml of ethanol and 3 µl glycogen (20mg/ml) and eluted in 60 µl of TE.

For strains with Met4 Auxin-induced degradation, we tagged the endogenous Met4 protein with V5-AID (Met4-V5-AID) and introduced OsTIR1 to the cells. Depletion of the Smc2 was achieved by adding auxin (Sigma, I2886) to a final concentration of 500 μM. Strains were incubated with auxin for 30 mins prior to MTAC experiment. Depletion was confirmed by western blot.

Data availability

All sequencing data in this study can be accessed on GEO series GSE252386, which include 1) GSE252384: ChIP-seq raw data (fastq), bed files, and bigWig files, 2) GSE252385: RNA-seq raw data (fastq), counts data (featureCounts), and DESeq2 tables, and 3) GSE252985: MTAC raw data (fastq), bigWig files, counts data (featureCounts) and DESeq2 tables.

Acknowledgements

We thank Dr. Joseph Reese for providing auxin-inducible degron plasmids and yeast strains. We acknowledge all members in the Bai lab for insightful comments on the manuscript. We thank the Microscopy Facility and the CSL Behring Fermentation Facility at Penn State for their help with imaging and protein purification. We are grateful to Dr. Cheryl Keller at the Genomics Research Incubator at the Huck Institutes of the Life Sciences for helping with the genomic assays. We also want to thank the members of the Center of Eukaryotic Gene Regulation at PSU for discussions and technical support. This work is supported by the National Institutes of Health (R35 GM139654 to L.B.) and National Science Foundation (MCB-2016266 to L.B.).

Author Contributions

L.B., J.L., L.S. and Y.L. designed the experiments; J.L. performed most of the experiments and data analysis with help from S.B.; L.S. and Y.L. carried out most of the experiments in Fig. 2 and Fig. 4, respectively; F.Z. contributed to the image data analysis; X.Z. helped with protein purification; J.L. and L.B. wrote the manuscript.

Conflict of Interest Statement

The authors declare no conflict of interest.