Previous 3C analyses of yeast chromosomes used purified nuclei, which may result in loss of some interactions due to the rather disruptive nuclei isolation protocol. Therefore, we adapted 3C for use with intact yeast cells [32]. In this method, the cell wall is removed by zymolyase treatment and intact spheroplasts are treated with formaldehyde to induce cross-links between proteins and DNA and proteins and other proteins thereby trapping interacting chromatin fragments throughout the yeast genome. Cross-linked spheroplasts are then solubilized by SDS and Triton X-100. From here the conventional 3C protocol is followed including restriction digestion, DNA ligation, and reversal of cross-links. The identities of interacting fragments are determined through detection of 3C ligation products by semi-quantitative PCR using locus specific primers. In addition, a randomized ligation control is generated which serves as a control for primer efficiency. This template is generated by digesting purified yeast genomic DNA followed by random intermolecular ligation which results in a DNA sample in which every possible ligation product is present in equal molar amounts.

The cross-linking frequency of two loci is determined by PCR using the 3C and the control libraries as templates. Primers are designed that recognize the corresponding ligation product and PCR products are quantified by ethidium bromide staining of agarose gels. We have found that this quantification method reliably measures the relative abundance of ligation products as long as the PCR is performed within the linear detection range [29], [33]–[36]. Figure S1A and B show examples of determination of the linear range of PCR by titrating the template concentration.

To further confirm the interaction between HML and HMR, we performed the reverse experiment, in which crosslinking frequencies were determined between the EcoRI fragment that contains HMR and the same EcoRI fragments including the fragment containing HML (Figure 1B). Again, a peak of crosslinking frequency corresponding to the EcoRI fragment containing HML is observed. Furthermore, specific and prominent interactions were also determined between HML and HMR in MATα cells (Figure S2A and B), indicating the HM-interactions are not specific to one mating type. Further, 3D imaging in MATα cells did not reveal a difference in HML-HMR interactions, nor did 3C analysis of strains deleted for the Recombination Enhancer [40] (Figure S2 Panels C and D). Given the lack of any mating type specific differences in HML-HMR interactions we consider it not likely that any mating type specific proteins or the MAT locus itself plays a role in the interaction between silent mating type loci. Interestingly, we do note that there appears to be a loss of frequent interactions between HML and sites close to it specifically in MATα cells. Although the reason for this is unclear, future comprehensive and chromosome-wide studies can be aimed at analyzing mating type specific differences in the conformation of chromosome III. Here we focused specifically on HML-HMR interactions, which are unaffected by mating type.

Next we tested whether the HM loci also interact with the telomeres of chromosome III. We performed 3C with primers annealing immediately adjacent to the left or to the right telomere (Figure S2E and F). Interactions between the left telomere and other EcoRI fragments along chromosome III revealed frequent interactions between the left telomere and HML (Figure S2E). This is most likely due to the fact that these loci are located close to each other on the chromosome, which typically results in very frequent background interactions [29],[38]. Interactions along the length of the chromosome between the left telomere and other EcoRI fragments were generally lower than observed for HML or HMR. Interestingly, the left telomere interacted also relatively frequently with HMR. The same is true for the right telomere (Figure S2F): this telomere interacted frequently with nearby fragments, including HMR, and with HML. Although the telomeres interacted preferentially with the HM locus located on the opposite side of the chromosome, these interactions are clearly less frequent than the HML-HMR interaction. Contacts between the two telomeres were also less frequent than the interaction between the HM loci (Figure S2F).

As a second approach we created a MATα strain in which the HMR locus was replaced with the KanMX cassette. We determined crosslinking frequencies between the fragment containing HML and fragments along the length of chromosome III including the EcoRI fragments containing and directly flanking the KanMX cassette (Figure 1D third panel). We find that the peak of interaction is no longer observed and that the crosslinking frequency of HML with the fragments containing the KanMX cassette is similar to that of its neighbors. Likewise, when we analyzed interactions between the fragments containing the KanMX cassette with fragments along chromosome III including the fragment containing HML, we no longer detected the peak of interaction at HML (Figure 1D fourth panel). These results indicate that the frequent interaction at HML observed in wild type cells requires the presence of HMR. In addition, this experiment rules out that the interaction is due to the presence of another genomic element in the sub-telomeric region located outside HMR but within the interacting EcoRI fragment.

As a third approach, we determined the role of specific parts of HML and HMR in mediating their interaction. We analyzed the interaction between the heterochromatic loci using a different restriction enzyme, XbaI, as this enzyme cuts inside of HML and HMR to create fragments in which the left and right ends of HML and HMR are contained on separate restriction fragments. 3C primer sequences are listed in Table S1. Interestingly, we find that the XbaI fragment containing the 5′ end of HML interacts preferentially with the fragment containing the 3′ end of HMR (Figure 2A). Similarly, the fragment containing the 3′ end of HML preferentially interacts with the fragment containing the 5′ end of HMR (Figure 2B). Furthermore, the interaction between the 5′ end of HML and the 3′ end of HMR is clearly the most frequent.

These results point to the possibility that the interaction between HML and HMR involves the E- and I-silencer elements that flank HML and HMR on their 5′ and 3′ side, respectively. To test this we repeated the 3C analysis with a frequently cutting restriction enzyme AciI that cuts at multiple locations within the HM loci. We find that the small 814 bp AciI fragment containing the HMR-I element (genomic position 294510–295324) most strongly interacts with the 821 bp fragment containing the HML-E silencer (genomic position 10815–11636), and significantly less frequently with other parts of HML (Figure 2C). Conversely the fragment containing the HML-E silencer interacts most prominently with the HMR-I fragment, as compared to other regions of HMR (Figure 2D). As a control we determined crosslinking frequencies between an AciI fragment located in the middle of HML. This fragment (genomic position 11679–11906) interacted most prominently with the HMR-I fragment and the neighboring AciI fragment located within HMR, but the crosslinking frequencies were lower than that observed between HML-E and HMR-I (Figure 2D), which is as expected for a fragment located just next to the site of interaction. These results strongly suggest that the HML-E and HMR-I silencers, or elements located very close to them, are the sites of interaction. Despite intense efforts, we have not been able to generate a mutant in which HMR-I was deleted without also creating unexplained rearrangements in the locus. Therefore we cannot unequivocally conclude that the interactions between HML and HMR are directly mediated by the silencer elements.

In agreement with the 3C data, colocalization of HML and HMR, as determined by live cell fluorescence microscopy, is largely reduced in sir4Δ and sir3Δ cells, in which only 5% and 7% of distances scored are <250 nm respectively (Figure 3B). As compared to wild-type where in ~60% of the cells HML and HMR are found colocalized or adjacent to each other, in sir4Δ cells the distance between these loci is less than 500 nm in only 21% of cells (Figure 3B). Similarly, in sir3Δ cells <35% of cells scored show HML and HMR colocalized or immediately adjacent to each other (n=500; Figure 3B). The distribution of 3D distances measured in sir3Δ (n=500) and sir4Δ (n=307) cells is clearly shifted to greater distances as compared to wt (n=836). These results indicate that the heterochromatic structure established by the Sir complex or the Sir4p, Sir3p, and Sir2p proteins themselves are critical for the HML-HMR interaction.

To extend our study, we asked whether Sir1p, a protein that interacts with silencer elements flanking HML and HMR via the Rap1p/ORC complex during the establishment of the silent chromatin state [42] participates in HM loci interactions. Interestingly, we find that, similar to sir4Δ, sir3Δ, and sir2Δ mutants, in sir1Δ cells HMR and HML no longer interact as shown by 3C and by 3D microscopy (Figure 3A,B), despite significant residual silencing (see below). Interactions between each HM locus and the telomere on the opposite end of chromosome III are concomitantly reduced (Figure S2G and H).

Given the results obtained with sir1Δ cells we chose to determine the role of Esc2p that has been identified as a protein that can functionally substitute for Sir1p [43]. Although it is currently not known whether Esc2p directly binds the HM loci, there is strong evidence suggesting that Esc2p is directly affecting the HM loci. First, Esc2p was identified as a protein that when targeted to HML can induce silencing [44]. Second, Esc2p has been shown to directly bind Sir2p [44]. Third, overproduction of Esc2p can substitute for Sir1p and aids in the establishment of silencing at HM loci [43]. Deletion of ESC2 has only minimal effects on silencing ([43], and see below). Interestingly, as in sir1Δ cells, we find that deletion of ESC2 completely abolished the specific interaction between HML and HMR (Figure 3A). We conclude that Esc2p plays a critical role in HML-HMR interactions, presumably by directly acting on the HM loci, although we cannot formally rule out a more indirect role.

Furthermore, given the importance of the silencers and silencing proteins for the HML and HMR interaction, we chose to analyze a mutant in which silencing proteins are recruited (albeit to a lesser extent than in wild-type) and assembled at the silencers but in which the Sir complex fails to spread across the silenced loci [45]. The mutant sir2-345 contains a point mutation at residue 345, which results in an Asn-to-Ala substitution. This mutant lacks deacetylase activity which results in a defect in silencing [23]. In a sir2-345 mutant, an interaction between HML and HMR can no longer be detected by 3C (Figure 3A). This indicates that in order for this interaction to occur proper heterochromatin must be formed and that the mere presence of Sir proteins at the silencers is not sufficient to promote HM loci interaction.

Figure 4 shows the radial position of the HM loci for WT and mutant strains in interphase (and not only G1) cells to be directly comparable with 3C studies that involve analysis of non-synchronized cultures (see below). Nuclear positions of the tet^op and lac^op tagged silent HML and HMR loci were visualized using TetR or LacI repressor-GFP fusion proteins in G1 and S-phase cells [47],[50]. Data were acquired in three dimensions to assign the position of the resulting fluorescent spot relative to the GFP-tagged NE (Nup49-GFP) in the focal plane in which it was brightest and in one of three concentric nuclear zones of equal surface. Enrichment of the silent mating type loci near the nuclear envelope in wild-type cells is abolished in a sir4Δ strain [[47]; Figure 4]. Interestingly, we find that this effect is specific for sir4Δ cells: in sir3Δ mutants significant anchoring of both HML and HMR was retained, possibly due to binding of Sir4p to the silencer nucleation site (Figure 4). Since in sir3Δ mutant strains we no longer detected preferential interaction and colocalization of HML and HMR (Figure 3), we conclude that proximity to the NE is not sufficient for their interaction. In addition, we find that in esc1Δ cells, the HM loci maintain their peripheral localization. This suggests that Sir4p containing heterochromatin can associate with the nuclear periphery in an Esc1p-independent manner.

HML and HMR are also both directly associated with the nuclear periphery in a manner that does not require yKu70p. In interphase the position of HMR is unaffected by the absence of yKu70p, while HML's anchoring to the NE is significantly increased ([47], Figure 4). Further, the increase in peripheral localization of the GFP-tagged HML locus in yku70Δ cells is dependent on the presence of the HML locus [47]. Similarly, deletion of HMR reduces the peripheral localization of the right end of chromosome III (KB, unpublished observations). These analyses suggest that peripheral localization of HM loci is not solely due to the close proximity of HM loci to telomeres that are often anchored to the nuclear periphery and further indicate that the HM loci strongly contribute to the peripheral localization of the ends of chromosome III.

Next we analyzed strains deleted for both ESC1 and YKU70 in which both anchoring pathways are abolished. We find that peripheral anchoring of HML and HMR was only slightly but not significantly reduced in interphase cells (Figure 4). Anchoring was somewhat more reduced in G1 than in S phase cells (data not shown). These observations on the radial position of HMR in its native chromosomal location extend those reported by Gartenberg and colleagues who found that deletion of both YKU70 and ESC1 resulted in loss of peripheral localization of an extrachromosomal HMR locus [46]. Our results suggest that chromosomal context plays a role in peripheral localization. We conclude that alternative Sir4p-dependent pathways exist that anchor HM loci to the nuclear periphery.

Next we analyzed HM interactions by 3C. In yku70Δ and in esc1Δ cells, we observed a significant increase in the frequency with which HML and HMR interact as compared to wild type cells (Figure 5A). Deletion of YKU80 did not significantly affect the crosslinking frequency. In yku70Δ esc1Δ double mutants the HML–HMR crosslinking frequency is slightly higher than in either single mutant. These results demonstrate that yKu70p and Esc1p are not required for HML and HMR to specifically interact.

We also analyzed HML–HMR colocalization in these strains by live cell fluorescence microscopy (Figure 5B). In esc1Δ cells 51% of cells measured exhibit HML and HMR colocalization or juxtaposition (n=391), which is comparable to what we observed in wild type cells (P=0.063 wt vs esc1). In yku70Δ cells HML and HMR are found colocalized in 18% of the cells and adjacent to each other in another 28% of cells (n=814). The frequency of co-localization is comparable to wild type, but the distribution of distances between HML and HMR differs from wild type. It appears that in the absence of yKu70p more nuclei display widely separated loci than in wild type (P=6.1e⁻⁶ wt vs yku70Δ). This may be related to the fact that populations of yku70Δ cells display two semi-stable states of silencing: one in which telomeres are delocalized and one in which they remain clustered [51]. Finally, we found that colocalization of HML and HMR in yku70Δesc1Δ double mutants was also comparable to wild type: 52% of the cells measured still show HML and HMR colocalized or immediately adjacent to each other (n=165) (Figure 5B).

We note that the 3C analysis revealed a ~4-fold increase in HML-HMR interactions in all three mutants, but that live cell fluorescence failed to detect an increase in colocalization of these loci. The increased crosslinking frequency as detected by 3C may be due to loss of interactions of HM loci with telomeres, which could result in an increased chance for HM loci to become ligated in the 3C assay or due to a more intimate association that is more easily crosslinked (see discussion).

These analyses show that HML-HMR interactions do not require the known membrane anchors yKu70p/yKu80p and Esc1p. However, their peripheral localization was also not abolished in the absence of both these membrane anchors. Our inability to genetically disrupt peripheral localization of the silent mating type loci prevented us from directly assessing the influence of membrane anchoring on facilitating HML-HMR interactions. Therefore, as an alternative approach, we set out to follow the positions of HML and HMR in wild type living cells over several minutes in order to determine whether HML-HMR interactions can be observed in the interior of the nucleus or only at the periphery (Figure 6). Figure 6A shows a representative movie comprising a series of 50 2D images taken at 10 s intervals of interphase cells on a wide-field Olympus XI inverted microscope. HML and HMR tags colocalized either near the NE as identified by the nuclear pore component Nup49p fused to CFP, or at the center of the nuclei imaged. In addition, colocalization was a transient event, because, after a few minutes, separation of previously colocalized loci was observed (compare time points 9 and 12 or time points 4 and 5). Thus, HML and HMR seem to collide and separate both at peripheral and internal nuclear locations. In order to determine whether HML and HMR were more likely to interact at the nuclear periphery, we followed their position relative to the nuclear center in single nuclei over time taking images in 2D every 1.5 seconds on a confocal LSM510 microscope. Figure 6B summarizes the distances between HML and HMR plotted against the distance of either HML or HMR to the nuclear center at every time point during nine 1–2 min time lapse movies (n=676). We found that in 2D HML is separated <250 nm from HMR (colocalization) in >45% of the time points scored. Moreover, the probability of colocalization was similar in the interior fraction (position of HML or HMR less than 720 nm from the nuclear center, about 2/3 of the positions) and the peripheral fraction of the nucleus. These movies clearly demonstrate that over long periods, interaction between HML and HMR was independent of NE anchoring.

First, we tested a strain in which the CAF-1 complex is disrupted by deletion of CAC1. Cac1p is the largest subunit of the CAF-1 complex and in strains lacking Cac1p HML and HMR are weakly derepressed [26]. We find that deletion of CAC1 did not affect the frequency with which HMR and HML interact (Figure 7A). Next we analyzed a strain lacking Hir1p, a subunit of the HIR protein complex. Deletion of HIR1 has also been reported to result in slight de-repression of HML and HMR [57]. As for cac1Δ strains we find that deletion of HIR1 has no effect on the frequency of the HML-HMR interaction (Figure 7A). Given the known functional redundancy of HIR and CAF-1 complexes, we created a double mutant strain in which both CAC1 and HIR1 are deleted. We find that in this case the prominent interaction between HML and HMR is no longer observed (Figure 7A). These results point to a role of nucleosome assembly in mediating HML-HMR interactions, and show that the HIR complex and CAF-1 complex are functionally redundant in this process. Interestingly, in the double mutant, interaction frequencies along the entire chromosome are two- to three-fold higher than the background interactions we observed for all other strains. This may point to a more flexible chromosome organization.

To further investigate the role of nucleosome assembly, we studied a strain lacking the histone chaperone Asf1p which functions with both the CAF-1 and HIR complex. Recently a role for Asf1p has been proposed for telomere sub-nuclear positioning [60]. Interestingly, we find that in an asf1Δ mutant the interaction between HML and HMR is no longer observed (Figure 7A). We have also analyzed asf1Δ cells by live cell fluorescence microscopy (Figure 7B). As compared to wild-type where in ~60% of the cells HML and HMR are found colocalized or adjacent to each other, asf1Δ cells only have 29% of cells with HML and HMR colocalized or adjacent to each other (n=352; P=2.2e⁻¹⁶ wt versus asf1).

Surprisingly, our analyses also reveal that peripheral localization of HM loci is not reduced upon inactivation of the two previously defined pathways for membrane anchoring of heterochromatin. As shown previously [47], deletion of YKU70 increases the peripheral localization of HML in interphase cells, while not affecting the positioning of HMR. In addition, given that in yku70Δesc1Δ double mutant cells heterochromatin formation near telomeres is disrupted and Sir4p is found uniformly throughout the nucleus [46],[48] but HM interactions are retained (Figure 5), these data also indicate that sequestration of Sir protein in clusters is not required for interactions between HML and HMR. Consistent with the loss of Sir protein clusters, we found that in yku70Δesc1Δ double mutant cells interactions between some telomeres is reduced, as detected by 3C, suggesting that in this mutant telomere clustering is affected, while HM interactions are retained (Figure S3D).

First, two mutants (esc2Δ and asf1Δ) that display a complete loss of interaction between HML and HMR display only very minor defects in silencing as detected by RT-PCR. Similar very minor silencing defects have been detected using reporter constructs. Huang et al. used a GFP reporter gene inserted in HMR to detect GFP expression in WT, sir3Δ and asf1Δ strains [66]. They found that in WT cells HMR is silent and only 0.3% of cells expressed GFP, whereas in sir3Δ cells the locus was mostly derepressed with 99% of cells expressing GFP. Deletion of ASF1 resulted in GFP expression in only 0.9% of cells, indicative of effective silencing. Although we cannot formally exclude the possibility that very small defects in silencing are sufficient to cause loss of HML-HMR interactions, we favor the interpretation that heterochromatin formation and silencing is not sufficient for HM interactions, and conversely that HM interactions are not essential for heterochromatin formation.

Second, genetic evidence indicates that Hir1p and Asf1p act in the same silencing pathway [27], but they display very different effects on HM interactions, suggesting that for the latter process they act in different pathways. In addition, single deletions of HIR1, CAC1 or ASF1 all result in minor silencing defects [26],[27],[66],[67], but only deletion of ASF1 results in loss of HM interactions. Third, the cac1Δhir1Δ double mutant and the asf1Δ mutant display the same loss of HML-HMR interactions, but they have quantitatively very different effects on silencing [27], again pointing to differential dependence of silencing and long-range interactions on these chromatin assembly factors.

The differential effect of deletion of SIR1 on silencing and HM interactions is particularly interesting. In the absence of Sir1p HML and HMR are partially derepressed. This is due to the occurrence of two distinct populations of chromatin states: in one subset of cells the loci are completely repressed, whereas in the other subset they are fully expressed [42]. Our RT-PCR analysis of HMR expression suggests that ~60% of loci remain repressed, whereas ~40% are expressed. However, the 3C analysis showed a complete loss of HML-HMR interactions and the level of colocalization, as determined by live cell 3D imaging was indistinguishable from background levels observed in sir4Δ cells. These results indicate that Sir1p may have a specific role in long-range interactions between heterochromatic loci that is distinct from mediating gene silencing per se.

Taken together, formation of silent heterochromatin at HML and HMR is essential but not sufficient for long-range interactions between HM loci. HML-HMR interactions require Asf1p, Esc2p and possibly Sir1p in a process that is distinct from silencing. Other proteins, such as other SIR complex components may also play a role in that process, but their role in HM interactions is more difficult to assess, as they are also essential for HM silencing.

3D images were captured on a Metamorph driven Olympus IX81 wide-field microscope equipped with a Coolsnap HQ camera and a Polychrome V (Till Photonics). Stacks of 21 images were acquired with a step size of 0.2 µm either at 490 nm for 200 ms for GFP or alternating the wavelength between 435 nm (CFP) and 512 nm (YFP) at every image plane, and exposures of 400 ms per wavelength. A 100×/1.4 Oil Plan-Apochromat objective (Olympus) was used. Position of GFP tagged loci relative to the nuclear rim identified by the Nup49-GFP signal were determined as a percentage of fluorescent spots in one of three concentric nuclear zones of equal surface in the plane bearing the brightest GFP-lacI or tetR-GFP focus using the Pointpicker plug-in for Image J [47],[50]. Only the 10 core focal planes were scored. For 3D distance determination, CFP and YFP images were automatically analyzed using SpotDistance implemented as a plug-in for ImageJ, freely available at: http://bigwww.epfl.ch/spotdistance/ [80]. The measured distances were loaded into R software (www.r-project.org) and the measured distance distributions translated into a box plot. Outliers are defined as 1.5 times the Inter Quartile Range (IQR) and are represented as open circles. Different distance distributions (determined as non-normal using a Kolmogorov-Smirnov test) were scored for significant difference in R using the Wilcox test.

In addition to the above 3C template a randomized control template was also generated which is used to determine PCR amplification efficiency of specific 3C ligation products. This template was created by digesting purified non-crosslinked yeast genomic DNA which is then followed by a random intermolecular ligation. The resulting template was purified by a series of phenol-chloroform extractions and ethanol precipitations. The resulting template was also treated with RNAse A.

Once both the 3C templates and control templates are generated a PCR titration is performed to determine the amount of template to be used in the subsequent PCR reactions as shown in Figure S1A and B. PCR is performed in a 50 µl reaction in a buffer containing 10 mM Tris-Cl pH 8.4, 50 mM KCl, 2.25 mM MgCl₂, 0.5 mM dNTPs and 0.4 µM of each primer. The following PCR program gives quantitative results: 32 times: 1 minute at 95°C, 45 seconds at 60°C, 2 minutes at 72°C. This is followed by 1 minute at 95°C, 45 seconds at 60°C, and 8 minutes at 72°C. PCR products are quantified semi-quantitatively on 1.5% agarose gels in the presence of ethidium bromide. Previous work displays that semi-quantitative PCR and quantitative real-time PCR yield similar results [39],[81],[82]. Multiple primer combinations are used that detect highly abundant 3C ligation products as well as infrequently formed products to be sure that the amount of template used in each reaction is within the linear range. The subsequent PCR reactions are done in triplicate for both the control and 3C template. The crosslinking frequency is then determined by calculating the ratio between the amount of PCR product obtained with the 3C template and the control template. The three resulting crosslinking frequencies are then averaged together and one value is obtained. This is the value that is plotted in interaction graphs. The standard deviation of the mean is determined and is used as the error bars (see Table S3 for examples of crosslinking frequency determination).

Restriction enzymes were chosen to create an equal distribution of restriction fragment cut sites and also to allow for important elements (i.e. HMR-I, HMR-E, HML-E, HML-I, etc) to be on different restriction fragments. Primers were then designed for restriction fragments of interest (see Table S1 for sequences and Figure S1C for diagram of EcoRI primers). Primers were designed unidirectionally unless a unique primer could not be designed on that particular end of the fragment. They were designed to be approximately 28 base pairs long with 50% GC content. Once designed, the primers were tested on a control template and those primers that give aberrant products, multiple products, and abundant primer dimers were re-designed.

3C analyses for wild type and each mutant described above were performed in at least two, and in most cases three independent experiments, with each data point quantified in triplicate. Figure S4 displays some examples of biological repeats of 3C analyses that illustrate the highly reproducible nature of 3C interaction profiles in WT and several mutant strains.

Marian Walhout and Nynke van Berkum are acknowledged for valuable discussions and critical reading of the manuscript. Microscopy analyses were performed at the RIO microscopy facility in Toulouse and the University of Geneva. We acknowledge David Villa for assistance with image presentation and thank Paul Kaufman for advice.