To assess the intrinsic behavior of heterochromatin spreading and what shapes its precise re-establishment with respect to position and extent of repression (‘fidelity’), we employed transcriptionally encoded fluorescent reporters to read silencing by heterochromatin at a given locus, as previously reported. Several critical improvements over prior systems enable documentation of the spreading reaction at high sensitivity (Bintu et al., 2016; Hathaway et al., 2012; Obersriebnig et al., 2016; Osborne et al., 2009; Xu et al., 2006). First, our system has high signal to noise and minimized delay from epigenetic changes to fluorescent output. We accomplish this using the weak, well-characterized ade6 gene promoter (ade6p) (Allshire et al., 1994; Kagansky et al., 2009) to drive production of bright, fast-folding fluorescent proteins (XFPs) (Al-Sady et al., 2016). Second, our system provides separate sensors for nucleation, spreading, and cellular noise. We used ade6p-driven recoded super-folder GFP (Pédelacq et al., 2006) (‘green’) and monomeric Kusabira Orange (Sakaue-Sawano et al., 2008) (‘orange’) to report on nucleation and spreading, respectively (Figure 1A). A third XFP, an ade6p-driven triple fusion of E2Crimson (Strack et al., 2009) (‘red’, noise filter), is fully uncoupled from heterochromatin and inserted in a euchromatic locus. Here it reports on intrinsic or extrinsic noise that arises from cell-to-cell variation in the content of specific and general transcription factors and also translational efficiency (Figure 1A). To validate this reporter system, we characterized the non-heterochromatic state, via null mutation of clr4 (Δclr4), encoding the only S. pombe H3K9 methyltransferase. We show that in the absence of heterochromatin, expression of the noise reporter (‘red’) correlates well with that of reporters for both nucleation (‘green’) and spreading (‘orange’) (Figure 1—figure supplement 1A,B), especially when all cells in the population are considered without applying a size gate (Figure 1—figure supplement 1B, ρ ~0.83–0.93). This analysis mode is required when cell number is limiting. When a smaller subset is considered where all the cells are of similar size and stage of the cell cycle, the correlation still provides useful noise filtering (Figure 1—figure supplement 1A), which becomes evident when the normalization is applied to clr4+ cells that fall in the size gate (Figure 1—figure supplement 1C). Thus, cellular noise is mitigated by dividing the signals from the proximal ‘green’ and distal ‘orange’ heterochromatic reporters by the signal of the ‘red’, euchromatic reporter (‘green’/‘red’; ‘orange’/‘red’). Together, these elements constitute our heterochromatin spreading sensor (HSS) (Figure 1A).

Figure 1 with 1 supplement see all

Download asset Open asset

We first examined the intrinsic behavior of the heterochromatin spreading reaction in an ectopic context. We constructed the initial ectopic HSS based on a strain where a part of the centromeric ncRNA-driven nucleation element (dh) is inserted proximal to the endogenous ura4 gene (Canzio et al., 2011; Marina et al., 2013). We replaced the ura4+ open-reading frame (ORF) with ‘green’ to track nucleation element-proximal events. Then, to track distal events, we inserted ‘orange’ at one of several sites downstream from ‘green’ (ura4::dhHSS^1kb, ura4::dhHSS^3kb, ura4::dhHSS^5kb ura4::dhHSS^7kb, Figure 1B). The noise filter (‘red’) was inserted between SPBC1711.11 and SPBC1711.12, a bona fide euchromatic region (Garcia et al., 2015). All strains were initially constructed in a Δclr4 background, and we initiated heterochromatin formation by crossing in clr4+. We assessed heterochromatin formation after ~80–100 generations by quantifying the production of ‘green’ and ‘orange’. This period is significantly longer than ~25 generation timeframe required for full formation of a heterochromatic domain (Obersriebnig et al., 2016), ensuring that the population is at equilibrium.

To quantitatively assess the products of heterochromatin formation, we performed steady-state flow cytometry on log-phase cells, which were size-gated for small, recently divided cells (~91% G2, Figure 1—figure supplement 1D and supplemental experimental materials) to remove size- and cell cycle-related effects. At this stage, we only normalize the cells by the ‘red’ noise filter and scale the signal in each channel to Δclr4, giving us a broad overview of the possible expression states of ‘green’ and ‘orange’. We observe no cells that express ‘green’ but repress ‘orange’ (insets, Figure 1C), instead, all cells that are fully or partially ‘orange’ repressed are also robustly ‘green’ repressed. This observation, together with our finding that ‘green’ repression kinetically anticipates ‘orange’ repression (Figure 3—figure supplement 1), is consistent with heterochromatin spreading outward from the ura4::dh nucleator. Considering ‘green’ repression thus a proxy for nucleation, we observed that cells populate a wide range of nucleation states rather than a single state, with the distribution of repressed states varying among the HSS distance sensor strains (ura4::dhHSS^1-7kb, Figure 1C). To specifically examine cells that have fully nucleated, we applied a computational ‘nucleation clamp’ that isolates cells with a ‘green’ signal that is lower than the median plus two standard deviations of wild-type cells containing no XFPs (see Appendix 1-Supplemental Materials and methods). Using ‘orange’ as a spreading proxy, we found spreading to be stochastic in nucleated cells, with some cells exhibiting full repression, but others partial repression or full de-repression (Δclr4, x = 1) of the ‘orange’ spreading sensor. The proportion of cells that are fully repressed by spreading declines linearly with distance (scheme, Figure 1B; data, Figure 1C). Intriguingly, cells that are not fully repressed mostly exhibit intermediate levels of repression, which are neither at values of full repression or de-repression.

We next assessed the nature of these intermediate states in the 3 kb distance reporter strain, where ~30% of cells had maximal repression at the ‘orange’ locus and the remainder had intermediate states ranging from strongly to weakly repressed. Using Fluorescence Activated Cell Sorting (FACS), we gated for successful nucleation in the ‘green’ channel and then binned the ‘orange’ channel for fully repressed (low), intermediate and de-repressed (high) populations (Figure 1D, cartoon). We queried each bin for molecular events associated with heterochromatin formation, using RT-qPCR to determine the expression levels of ‘orange’, and Chromatin Immunoprecipitation (ChIP) to query the presence of the marks H3K9me2 and H3K4me3. These marks are thought to be mutually exclusive, associating with repressed heterochromatin and active promoters, respectively (Noma et al., 2001). The message level of ‘orange’ is tightly repressed in the ‘low’ population (0.05 of max), partially repressed in the intermediate population (0.3 of max), and nearly fully ‘de-repressed’ (0.8 of max) in the ‘high’ population. Thus, cells with intermediate fluorescence also exhibit partial gene repression, demonstrating that fluorescence accurately reports on gene expression (Figure 1D, RT primers indicated in diagram in 1C, solid line). Histone modification levels also correlated well with the HSS signals (Figure 1E, ChIP primers indicated in diagram in 1C, dashed line). The ‘low’ fluorescence population has high H3K9me2 (0.9 of dh, positive control) and low H3K4me3 (0.09 of actin, positive control); the intermediate population had intermediate H3K9me2 (0.49 of dh) and H3K4me3 (0.23 of actin), and the high population had low H3K9me2 (0.2 of dh) and higher H3K4me3 (0.44 of actin). Hence, successfully nucleated cells with intermediate fluorescence also exhibit intermediate amounts of the mRNA for ‘orange’ and histone marks reflecting heterochromatin (H3K9me2) and transcriptional activity (H3K4me3). These results support the notion that intermediate states of repression observed by cytometry represent intermediate states of spreading.

These observations are not due to the particularities of the ectopic site chosen or the behavior of the XFPs, as our results are recapitulated at the his1 locus (his1::dhHSS^3kb, Figure 1—figure supplement 1E), which contains only one gene (rec10) in the ‘spreading zone’, rather than several transcriptional units. Additionally, switching the nucleation and spreading reporter fluorophores produced similar results (Figure 1—figure supplement 1E). These results suggest that ncRNA-driven heterochromatin spreading at ectopic sites is intrinsically stochastic and multimodal, producing intermediate states of repression.

We next examined spreading behavior at the endogenous mating type locus (MAT), which is tightly repressed (Grewal and Klar, 1997; Thon et al., 2002) and a bona fide high-fidelity locus, as it can behave in a bistable manner with stable epigenetic inheritance even when disrupted (Grewal and Klar, 1996). The MAT locus has two known elements shown to recruit the H3K9 methylase Clr4: the cenH element, homologous to the ncRNA-nucleated dh fragment we inserted at ura4 and his1, and the RNA-independent element termed REIII (Jia et al., 2004; Thon et al., 1999). At REIII, two stress-responsive transcription factors, Atf1 and Pcr1, which form a heterodimer (Wahls and Smith, 1994), recognize two DNA-binding sites within REIII, directly recruit Clr4, Swi6/HP1 and histone deacetylases (HDACs) (Jia et al., 2004; Kim et al., 2004) and are required for heterochromatin formation at MAT when cenH is compromised (Noma et al., 2004). We validated that MAT retains its well-documented tight repression following insertion of the HSS, placing the ‘green’ reporter within the cenH nucleator, and the ‘orange’ reporter proximal to the REIII nucleator. Both colors were fully repressed in the large majority of cells (Figure 2B), which is reproduced when the color orientations are reversed (Figure 2—figure supplement 1A). However, for both reporter configurations, the REIII proximal color showed a small proportion of cells that are slightly de-repressed compared to the cenH internal color, consistent with previous findings (Thon and Friis, 1997). We conclude that the HSS can be used to dissect spreading at the MAT locus.

Figure 2 with 2 supplements see all

Download asset Open asset

We then examined spreading in cells nucleated solely by the cenH element. The REIII nucleator was inactivated by deleting the critical cis-acting Atf1/Pcr1-binding sites, to create a strain designated ΔREIII^HSS (Figure 2C). To our surprise, the high fidelity that the MAT locus exhibits in the repressed state (Grewal and Klar, 1996) disappeared. Instead, cenH nucleated spreading in the ΔREIII strain behaved similarly to spreading from the ectopic ncRNA-nucleated strains, showing high stochasticity and predominantly intermediate repression states (Figure 2C). We wanted to address if this stochastic silencing is reflected in weakened heterochromatin assembly. We preformed ChIP for H3K9me2 and H3K9me3, marks signaling heterochromatin assembly (Nakayama et al., 2001) and repression or spreading (Al-Sady et al., 2013; Jih et al., 2017; Zhang et al., 2008), respectively. We found that these marks decline progressively towards the distal ‘orange’ reporter in ΔREIII^HSS (Figure 2D), compared to the wild-type (WT) MAT^HSS. This is consistent with the observed tight repression for WT MAT^HSS (Figure 2B) and weakened silencing at the distal ‘orange’ in ΔREIII^HSS (Figure 2D). It is possible that this difference in spreading results from an altered heterochromatin structure at cenH in ΔREIII^HSS. However, H3K9me2 and me3 accumulation does not differ between ΔREIII^HSS and WT MAT^HSS at the cenH nucleator, or the leftward REII locus (Figure 2D). Thus, the observed behavior of ΔREIII^HSS is consistent with stochastic and multimodal spreading, rather than compromised nucleation at cenH.

To examine heterochromatin formation independent of cenH, we used the historical ΔK strain, where the entire cenH nucleation element is deleted and replaced with a ura4+ reporter (Grewal and Klar, 1996). We introduced the HSS into this context (ΔK^HSS, Figure 2E), placing the ‘green’ reporter proximal to REIII and the ‘orange’ reporter distally, replacing ura4. We then introduced clr4+ by cross and directly cultured colonies derived from germinated clr4+ spores. We found that although ΔK^HSS has very weak nucleation compared to strains with intact ncRNA nucleators, the distribution of cells is sharply bimodal: Cells were either repressed at both reporters (‘OFF’, lower left corner) or de-repressed at both reporters (‘ON’, upper right corner; Figure 2E). We note that isolation of single colonies on nonselective media from original spores of the cross yields mostly ON (ΔK^HSS-ON) or OFF (ΔK^HSS-OFF) colonies, consistent with each state being metastable (Grewal and Klar, 1996; Thon and Friis, 1997). This heterochromatin formation pattern requires REIII, as in 34/34 strains tested, no silencing can be established if Atf1/Pcr1 binding sites are deleted before clr4+ is introduced (Figure 2—figure supplement 2A,B). Additionally, the bimodal behavior does not require the H3K9me-independent gene-repressive REII element (Hansen et al., 2011), as ΔK^HSS REII::LEU2, containing a deletion of REII, behaved similarly to ΔK^HSS (Figure 2—figure supplement 1C), and is further independent of reporter placement (Figure 2—figure supplement 2C,D). We next characterized the molecular signature of the locus. While in our two color plots cells that were repressed in ‘green’ did not show any de-repression in ‘orange’ (Figure 2E, cells in bottom left corner), we wanted to test if the heterochromatic state at these loci correlated with this silencing pattern. Since we can isolate ΔK^HSS-ON and ΔK^HSS-OFF alleles by simple plating of ΔK^HSS cells, we performed H3K9me2 ChIP on both and H3K9me3 ChIP for ΔK^HSS-OFF cells (not detectable for ΔK^HSS-ON). We found that methylation correlates with the repression state (Figure 2E) and importantly, does not significantly differ between ‘green’ and ‘orange’. Together, these result indicate that in ΔK^HSS-OFF cells heterochromatin spreading is continuous across the locus and does not, unlike cenH-triggered spreading, accumulate any intermediates.

Our measurements thus far cannot reveal the dynamics of transitions between states. This requires long-term imaging of cells over a substantial number of generations (>20), which is difficult with traditional microscopy because of cell crowding effects. To deal with this issue, we used the Fission Yeast Lifespan Micro-dissector (FYLM) microfluidic device (Spivey et al., 2017, 2014), which traps the old pole of a rod shaped S. pombe cell at the bottom of a chamber well for its entire lifetime. Sibling cells generated at the new pole by medial fission eventually exit the chamber. We continuously image the old-pole cell with fluorescence microscopy for up to 60 hr (Figure 3A). We note that unlike Saccharomyces cerevisiae, S. pombe does not execute an aging program but rather dies stochastically (Coelho et al., 2013; Nakaoka and Wakamoto, 2017; Spivey et al., 2017). Thus, imaging S. pombe over long timescales avoids the confounding effects of aging on epigenetic behavior (Guarente, 2000; Li et al., 2017). To capture the long-range dynamics of spreading, we imaged approximately one hundred cells of each strain concurrently (see Figure 3—figure supplement 2B for a summary of cell fates in all experiments). For each cell, we imaged all three channels continuously, and performed similar normalizations as for the flow cytometry data (Appendix 1-Supplemental Materials and methods). We first imaged the HSS distance sensor strain (ectopic ura4::dhHSS^3kb). Our ability to observe cells that were initially fully de-repressed allowed us to trace ‘green’ and ‘orange’ repression kinetically. Consistent with linear heterochromatin spread outward of the dh nucleator, we find that ‘orange’ repression is anticipated by repression at ‘green’ (Figure 3—figure supplement 1). While nucleation in this strain is not stable (likely due to ‘green’ being adjacent to, rather than within dh), over time intervals where nucleation does persist, we observed dynamic fluctuations in the distal ‘orange’ color without a fixed temporal pattern (Figure 3—figure supplement 2A and Figure 3—videos 1 and 2), which is not due to the repression state of ‘green’ (Figure 3—figure supplement 2F).

Figure 3 with 7 supplements see all

Download asset Open asset

Next, we analyzed the MAT locus strains and selected cells that maintained nucleation for their entire measured lifespan (Appendix 1-Supplemental Materials and methods). Under this constraint, the three strains exhibit vastly different behaviors (Figure 3B). WT MAT^HSS cells maintained ‘orange’ repression for the majority of their measured lifespans (Figure 3C, Figure 3—figure supplement 2C and Figure 3—video 3). However, we documented transient loss of ‘orange’ silencing for 20% of the cells. (Figure 3B and C). In contrast, while most cells stay similarly nucleated in ΔREIII^HSS (Figure 3D, Figure 3—figure supplement 2D) 83% of the cells imaged experienced at least half-maximal ‘orange’ de-repression at some time points (Figure 3B). For this strain, 30% of the cells transited through the fully ON state (Figure 3B and D, Figure 3—figure supplement 2D and Figure 3—video 4). In fact, cells sampled a wide range of values from OFF to fully ON, indicating that cells do not occupy ON or OFF states exclusively, but adopt intermediate values across time (Figure 3D). Importantly, ΔREIII^HSS cells, just as ura4::dhHSS^3kb cells, fluctuate in their ‘orange’ values, indicating that spreading is unstable and adopts a random walk type behavior. To analyze ΔK^HSS cells, which exist predominantly in fully ‘green’ and ‘orange’ ON state (Figure 2C), we analyzed ΔK^HSS-OFF cells (see above). ΔK^HSS-OFF behaved markedly differently from ΔREIII^HSS: in all the cells analyzed, ‘green’ and ‘orange’ reporters remained OFF throughout the time course (Figure 3B,E and Figure 3—video 5), up to 25 generations, revealing a fundamentally different dynamic behavior between cenH- and REIII-dependent heterochromatin. We note it remains possible that isolation of a ΔK^HSS-OFF colony may bias our analysis against potentially more frequent OFF-ON switching events in the primary mixed population derived from continuous propagation of the germinated spore (Figure 2E). However, since the mixed population resolves spontaneously into ON and OFF states once plated, and OFF cells behave similarly in either the mixed ΔK^HSS or ΔK^HSS-OFF isolated populations (compare Figure 2E and Figure 5C), we believe the stability of ΔK^HSS-OFF is intrinsic to the ΔK MAT locus.

To probe memory capacity (i.e. the ability of cells to retain information of an ancestral state established many generations prior), we compared cells containing an intact MAT locus to those lacking either ncRNA- or REIII-dependent heterochromatin. We established two ancestral states (Figure 4A); one with unperturbed heterochromatin, and a second treated with the HDAC inhibitor trichostatin A (TSA), known to fully disrupt the heterochromatin state ([Hall et al., 2002] and Figure 4—figure supplement 1). Following production of the ancestral states, we grew cells either in rich media alone or in a TSA concentration gradient (0–50 µM) for 25 generations and then measured the fraction of fully nucleated cells that effectively silence the ‘orange’ spreading marker (Figure 4A). Cells exhibit memory if the fraction of the population with full spreading (‘orange'^OFF) depends on the ancestral state, which would be indicated by separation of the unperturbed (light orange) and perturbed (red) lines. In contrast, no memory is indicated by convergence of the two lines (graphs in Figure 4B–D). We further measure a second parameter we term relative ‘resistance’, which is defined as the TSA concentration at which the fraction of cells with ‘orange'^OFF declines to 50% of the no TSA pretreatment value. This value reports on the intrinsic sensitivity to perturbation of the locus formed by spreading.

Figure 4 with 2 supplements see all

Download asset Open asset

As expected, wild-type MAT^HSS exhibited clear memory at 25 generations (Figure 4B), which was still weakly evident even at 35 generations (Figure 4—figure supplement 1C). Among fully nucleated (‘green'^OFF) cells, those that derived from untreated ancestral cells showed a greater fraction of silencing (‘orange'^OFF) than those derived from treated cells throughout the entire TSA gradient, with a half-resistance point of ~2 µM (Figure 4B). Thus, wild-type MAT^HSS memory is robust in the face of perturbations of the heterochromatic state.

In sharp contrast, when spreading exclusively nucleates from cenH (ΔREIII^HSS strain), memory of silencing (‘orange'^OFF) is significantly weaker. Memory collapsed beyond low TSA concentrations (>0.2 µM TSA), with the red and light orange lines coinciding for much of the gradient. Even at 0 μM TSA, history dependence was erased at 35 generations (Figure 4—figure supplement 2C). Interestingly, the half-resistance point was ~0.2 μM, 10-fold lower than that of wild-type MAT (Figure 4C). As cenH-nucleated spreading in ΔREIII^HSS produces little memory capacity and lacks resistance, the memory capacity at MAT does not derive from ncRNA-nucleated spreading. These results are consistent with REIII being required for the memory behavior of WT-MAT.

The ΔK^HSS strain at face value had the widest separation in the behavior of the progeny of TSA pretreated and untreated cells. However, ascribing this behavior directly to memory is complicated by the fact that ΔK^HSS cells are no longer able to re-nucleate if they were ancestrally TSA treated, consistent with previous findings indicating that RNAi factors are required for heterochromatin establishment at MAT (Hall et al., 2002). However, when examining resistance, that is the behavior of cells not ancestrally TSA pretreated, we observe that the REIII dependent ΔK^HSS strain has a half-resistance point of ~3 µM TSA (Figure 4D), similar to the intact locus. This indicates that the increased resistance of the wild-type over ΔREIII^HSS is conferred by REIII. Together these results indicate that REIII is required for epigenetic stability at MAT.

We next studied how REIII contributes to epigenetic stability in the context of a physiological perturbation, such as change in ambient temperature. Consistent with previous reports, we found that ncRNA-nucleated spreading is sensitive to continuous growth at high temperature, likely due to the cytosolic shuttling of RNAi-components (Woolcock et al., 2012; Figure 4—figure supplement 2A). WT MAT behaved in a similarly sensitive manner. In contrast, heterochromatin in ΔK^HSS cells was highly resistant to elevated temperature (Figure 4—figure supplement 2A).

We next probed the ability to remember the heterochromatin state after a transient exposure to elevated temperature, by exposing cells to 38˚C for 10 doublings, followed by return to growth at 32˚C (Figure 4E). As expected from our steady-state experiments above, REIII-dependent heterochromatin (ΔK^HSS cells) is only minimally affected by the perturbation and regains full spreading rapidly (Figure 4F, Figure 4—figure supplement 2F), whereas WT MAT and ncRNA-nucleated (ΔREIII^HSS) strains lose a significant amount of spreading (Figure 4F, Figure 4—figure supplement 2D,E) and nucleation (Figure insets). Both strains regain nucleation at cenH rapidly (1 day after return to 32˚C; Figure 4—figure supplement 2B,C). However, they are discrepant in their kinetics of restoration to the 32˚C extent of spreading, with WT MAT recovering much more rapidly than the strain nucleated exclusively by ncRNA (∆REIII^HSS) (Figure 4F). Indeed, plot fitting reveals a half-life (t_1/2, time to reach 50% of initial state) difference of ~22 hr, or ~9–10 generations between WT MAT and ΔREIII^HSS (Figure 4F). Therefore, REIII- is required for efficient recovery to the fully repressed state after heat perturbation. These data suggest that a central role of REIII is to ensure that epigenetic stability at MAT is maintained in the face of environmental perturbations in the wild.

To address dependence of the epigenetic maintenance of spreading on nucleation following heterochromatin establishment, we examined the behavior of cells following the removal of trans-acting factors required for the initial recruitment of nucleation factors such as Clr4, Swi6/HP1 and HDACs. This experiment is similar to the induced removal of the cis-acting sites in S. cerevisiae (Cheng and Gartenberg, 2000). ∆REIII^HSS and ∆K^HSS-OFF isolate cells (see above, derived from nonselective plating of ∆K^HSS) with established heterochromatin were crossed to mutants disrupting recruitment of nucleation factors at each element (Figure 5A). To impair REIII, we crossed the ∆K^HSS-OFF reporter strain to ∆pcr1 (Noma et al., 2004). To impair ncRNA nucleation, we crossed the ∆REIII^HSS reporter strain to seb1-1, a mutant allele of the Seb1 RNA binding protein. Seb1 functions redundantly with the RNAi pathway in ncRNA nucleation, including binding cenH transcripts, and the mutant allele seb1-1 has defects in triggering nucleation at dh and dg pericentromeric elements (Marina et al., 2013). We focus on Seb1, as RNAi pathway mutants have little discernable effect on MAT when introduced after establishment (our unpublished data and [Hall et al., 2002]), indicating a stronger role for Seb1. Identifiable ∆REIII^HSSseb1-1 and ∆K^HSS-OFF∆pcr1 colonies were grown for flow cytometry analysis immediately following mating and selection. The control cross mutant strains ∆REIII^HSS∆pcr1 and ∆K^HSS-OFFseb1-1∆dcr1 (loss of all ncRNA-nucleation [Marina et al., 2013]) allowed us to assess any effects the trans-factor may have even in the absence of its cognate site of action.

Figure 5 with 1 supplement see all

Download asset Open asset

Strikingly, most ∆K^HSS-OFF∆pcr1 cells remains robustly repressed over 456 hr, around 200 generations (Figure 5B). However, removal of Pcr1 does have a small discernable effect, as the ∆K^HSS-OFF∆ pcr1 strain showed a small population of cells not completely in the OFF state compared to the ∆K^HSS-OFF parent (Figure 5C LEFT). Further, by ~400 hr we detected a small fully ON population absent in the parent. This behavior is broadly consistent with the reported stability of intact ∆K^OFF (switch rate of ~10⁻⁴ generation, Grewal and Klar, 1997; Thon and Friis, 1997), even though our assay appears to show even smaller ON populations. Very small ON populations are more apparent in a growth selection based assay as only the targeted population survives, as opposed to our assay, which captures all cells. We note a formal possilibty remains that selection of OFF colonies yields higher apparent stability. To get a closer view of the behavior of individual isolates from the population, after 288 hr of continuous passage, we streaked for single colonies and measured the resulting populations. While 5/6 isolates behaved like the broader population, we found 1/6 isolates that experienced more severe breakdown in its heterochromatic state (Figure 5C RIGHT). In this isolate heterochromatin collapsed in a manner not ordered with respect to REIII proximity and exhibited ‘green”^ON/‘orange”^OFF cells. In contrast, ∆REIII^HSSseb1-1 lost most spreading at the first measurement point (24 hr, Figure 5E) with progressively increasing de-repression of ‘orange’, but also some loss of ‘green’, over the next 72 hr (Figure 5F). This suggests that the epigenetic inheritance cenH-spreading requires continuous nucleation, at least via the Seb1 pathway, consistent with the behavior at synthetic nucleators (Audergon et al., 2015; Ragunathan et al., 2015).

It is known that REIII recruits the HDAC Clr3 (Yamada et al., 2005), which was later shown to repress the turnover of histones (Aygün et al., 2013). This suggested the intriguing possilibty that unstable epigenetic inheritance in the absence of REIII is linked to elevated histone turnover. To test this idea, we adopted the Recombination Induced Tag Exchange (RITE) system (Verzijlbergen et al., 2010) to assay replication-independent turnover of H3 in ∆REIII^HSS and ∆K^HSS strains (Figure 6A). Tag switching (T7 for HA tag) in log phase growth was induced by administering β-estradiol concurrently with stalling replication with 15 mM hydroxyurea (HU) for 4 hr, during which time cells remain in early S phase (Figure 6—figure supplement 1). We compared the incorporation of T7 at 4 vs. 0 hr between ∆REIII^HSS, ∆K^HSS-OFF and ∆K^HSS-ON strains. First, we examined two euchromatic genes, pyk1 on chromosome 1, and mtd1, which is just outside the MAT locus. H3 turnover at these regions does not differ between the strains (Figure 6B) and is highest in in the strongly expressed pyk1 gene. We next examined sites in the MAT locus that are shared in sequence and genomic position between ∆REIII^HSS and ∆K^HSS (probes indicated in diagram, Figure 6B). We note this includes also REIII, since this locus only differs between the strains by the 14bp containing the two Atf1/Pcr1 binding sites. In contrast to euchromatic loci, we observed that ∆K^HSS-OFF experiences very low or no histone turnover at MAT targets by 4 hr HU compared to ∆K^HSS-ON and ∆REIII^HSS, which experienced levels of H3 turnover more consistent with our euchromatic controls. This in not unexpected for ∆K^HSS-ON, as is it effectively not heterochromatic (Figure 2E), and is consistent with previous results (Aygün et al., 2013). However, the observation that ∆REIII^HSS displays H3 exchange at levels similar to ∆K^HSS-ON and euchromatin suggests that it is memory, rather than heterochromatin formation itself, that requires repressed histone turnover.

Figure 6 with 1 supplement see all

Download asset Open asset

The dominant form of heterochromatin in S. pombe, triggered by ncRNA nucleators, leads to stochastic spreading of both silencing and H3K9 methylation that only occurs in some cells, and forms intermediate states (Figure 1 and Figure 1—figure supplement 1E, Figures 2C and 3D). This is consistent with position effect variegation in genetically disrupted systems (Elgin and Reuter, 2013; Nimmo et al., 1994). Additionally, the linear distance-dependent behavior we observe (Figure 1C) is reminiscent of the continuous spreading model in S. cerevisiae telomeres (Renauld et al., 1993; Talbert and Henikoff, 2006). This behavior of ncRNA spreading is not due to weak nucleation, as repressive histone marks accumulate to the same high extent at cenH in wild-type and ΔREIII and at Atf1/Pcr1 proximal region in ΔK cells. (Figures 1E and 2D and E). In a key result, we find ncRNA-triggered spreading to be epigenetically unstable. This is evidenced by highly dynamic behaviors over time and across generations, little discernable memory, and low resistance to chemical or environmental perturbations (Figures 3–5). Those behaviors are not necessarily predicted by the stochastic induction spreading, given that PEV in flies results in clonally inherited patches (Elgin and Reuter, 2013). This result opens the question how high fidelity can be achieved with ncRNA nucleators at loci that carry critical cell type specification information. The most likely cause for this instability is elevated and near-euchromatic levels of histone turnover (Figure 6B). This implies that while elevated histone turnover is compatible with heterochromatin formation per se, it is incompatible with epigenetic memory.

In contrast to the behavior of ΔREIII, ΔK cells, dependent on REIII for heterochromatin formation (Figure 2—figure supplement 2A&B), do not display stochasticity in spreading (Figures 2E and 3E), and instead repress MAT uniformly across nucleated cells in the population (Figure 2E). Under environmental perturbation, ΔK heterochromatin is extraordinarily resistant (Figure 4F and Figure 4—figure supplement 2A,F) and capable of high memory retention, even in the absence of the REIII-targeted Pcr1 protein, which attracts HDACs and Clr4/Swi6 (Jia et al., 2004; Kim et al., 2004) (Figure 5B,C). This is consistent with previously documented bistable behaviors ascribed to the overall locus (Dodd et al., 2007; Grewal and Klar, 1996). It, however, remains to be resolved whether heterochromatin in ΔK cells is nucleated by REIII and spreads outwards, or is nucleated at multiple sites, yielding apparent uniform heterochromatin formation. A REIII nucleated spreading model is favored by results presented here and by others (Jia et al., 2004; Wang and Moazed, 2017) that demonstrate that Atf1 and Pcr1 proteins or their binding sites are absolutely required for established of heterochromatin in ΔK cells, yet not for ΔREIII cells. However, unlike for cenH, where sufficiency has been clearly shown (Hall et al., 2002), we and others (Wang and Moazed, 2017) do not document significant heterochromatin formation by REIII when it is placed ectopically (Figure 2—figure supplement 1D). Thus, it cannot be differentiated whether the uniform heterochromatin formation in ΔK is the result of cooperation between different yet-to-be identified cis-acting elements, or a special property of REIII-driven spreading. Single site driven spreading of this ‘all or none’ type could be the result of looping, invoked for the polycomb system (Bantignies and Cavalli, 2011), predicted to improve spreading efficiency and memory in fission yeast (Erdel and Greene, 2016), or a unique molecular signature at REIII. For example, REIII recruits the HDAC Clr3 (Yamada et al., 2005), which promotes accumulation of the H3K9 trimethyl state, required for efficient spreading by Clr4 (Zhang et al., 2008; Al-Sady et al., 2013; Jih et al., 2017).

Regulation of histone turnover has been linked to epigenetic memory in fission yeast (Taneja et al., 2017) and has been previously shown to be low at wild-type MAT (Aygün et al., 2013). Hence, the high histone turnover we observe in ΔREIII cells results from unaided cenH-spreading. REIII recruits the HDAC Clr3 (Yamada et al., 2005), which represses histone turnover (Aygün et al., 2013). Our finding that the ΔK^OFF allele features very low histone turnover (Figure 6B), similar to the wild-type locus (Aygün et al., 2013), is thus consistent with REIII acting to repress histone turnover, when in a heterochromatic state. The extraordinary memory of repression we observe in ΔK^OFF likely is explained by this repressed turnover, although we should note it is possible that isolation of ΔK^OFF alleles, while consistent with the literature (for example [Grewal and Klar, 1996; Thon and Friis, 1997]), could bias the population to enhanced inheritance of repression. We speculate that reduced turnover increases retention of H3K9me3 nucleosomes, promoting methylation across nucleosomes by Clr4 via its H3K9me-dependent ‘read-write’ functionality (Al-Sady et al., 2013; Jih et al., 2017; Ragunathan et al., 2015; Zhang et al., 2008), thus facilitating re-establishment in the next generation. H3K9me3 is also directly promoted by Clr3, which is recruited to REIII (Yamada et al., 2005), further favoring reestablishment of methylation.

Repression of histone turnover and resulting epigenetic stability in ΔK strains requires cells to first nucleate and adopt a heterochromatic state (ΔK^OFF, Figure 6B). However, since ΔK cells only nucleate infrequently (Figure 2E), how is REIII able to stabilize heterochromatin in most wild-type MAT cells (Figures 2B, 3B,C and 4B)? The independent action of cenH and REIII elements cannot account for this behavior, hence they must collaborate. We propose that in the context of wild-type MAT, cenH stimulates REIII nucleation (model, Figure 6C). Recent findings indicate that Atf1/Pcr1 are present at REIII even in non-silenced ΔK-type cells (Wang and Moazed, 2017). We speculate that since Atf1/Pcr1 recruits silencing factors such as Clr4 and HDACs (Jia et al., 2004; Kim et al., 2004; Yamada et al., 2005), heterochromatin originating from cenH might stabilize this recruitment. This hypothesis is supported by our observation for nucleation during TSA recovery. Although ΔK^HSS cells very rarely renucleate (Figure 4D), REIII at the intact MAT locus must be active in most cells, as the heterochromatin reformed after erasure has much higher resistance to perturbation than that nucleated from cenH alone (red lines in Figure 4B and C).

Activated REIII in turn stabilizes the MAT locus most prominently when the heterochromatin state is perturbed. We infer this from the difference between the initial challenge and recovery from growth at high temperatures. When initially challenged, heterochromatin spreading at the wild-type MAT locus resembles that of ncRNA-nucleated heterochromatin (Figure 4—figure supplement 2A), suggesting that REIII or other nearby elements plays a minor role under normal circumstances at MAT. However, the heat recovery experiment suggests that changes in the REIII–dependent heterochromatin stabilization or assembly, not cenH nucleation (Figure 4—figure supplement 2B and C), takes on a major role in the accelerated recovery of the collapsed heterochromatin (Figure 4F). Thus, REIII is required under perturbation conditions to protect or quickly re-establish the heterochromatin state (Figure 4B,F and model Figure 6C). The relatively transient distal de-repression events experienced by wild-type MAT cells, which are much more pronounced in ΔREIII cells (Figure 3C and D), further points to REIII acting after stochastic loss of cenH spreading in steady state. It is possible that REIII does so by stabilizing existing heterochromatin via repression of histone turnover when the loss of these structures is sensed, or alternatively, that REIII-dependent structures expand or ‘fill-in’ the void left by collapse of cenH-spreading. In either case, we propose that REIII acts as a failsafe, ensuring the integrity, and ultimately epigenetic memory, of heterochromatin at MAT through perturbations.

In summary, we propose a model whereby the division of labor between cenH and REIII is uniquely suited for a cell type specification locus such as MAT, which requires silencing that is both robust and intergenerationally stable. ncRNA-nucleation is extremely robust but intrinsically too labile and stochastic to reliably control the cell type specification locus, thus requires support from an accessory element. The need for reliable control of cell type specification loci is shared in more complex systems. The nature of equivalent accessory elements to REIII and how they act in these cases remains to be determined.

For standard flow cytometry experiments, cells were grown overnight in rich media (YES) and then diluted in the morning to OD = 0.1 in minimal media plus supplements (EMM complete) and grown 4–6 hr before analysis by flow cytometry. Flow cytometry was performed using Fortessa X20 Dual or LSRII instruments (Becton Dickinson, San Jose, CA). Samples sizes ranged from ~2000 to 100,000 cells depending on strain growth. Compensation was performed using cells expressing no XFPs and single-color controls expressing 1 XFP each. Compensated data was used for all downstream analyses. Fluorescence was detected for each color as described (Al-Sady et al., 2016).

For FACS sorting experiments, cells were grown overnight from OD = 0.025 in YES and then in the morning concentrated into a smaller volume to achieve a flow rate of ~5000 events/second on the cytometer. Sorting was performed using either Aria2 or Aria3u machines (Becton Dickinson). Prior to sorting cells were strained through a 35–40 μm mesh (Corning) to reduce clogs. Sorting criteria included a gate for size (forward (FSC) and side (SSC) scatter), removal of doublets, a gate for ‘green'^OFF (‘green’ signal within the range of an unstained control) and then gated into Low, Intermediate, High ‘orange’ signal defined by the following: Low encompassed signal overlapping that of an unstained control and High encompassed signal overlapping that of the Δclr4 no heterochromatin control strain PAS355. Intermediate gate was set in between Low and High with about 100 fluorescence units of a gap (representing ~2% of the full range of captured fluorescence) to ensure reliable separation. The entire range of fluorescence detected was ~2.5 orders of magnitude. At least 8 × 10⁶ cells were collected for each population for Chromatin Immunoprecipitation and 2 × 10⁶ cells for RT-qPCR. Immediately after sorting, the final populations were subjected to the appropriate treatment for either Chromatin Immunoprecipitation or RT-qPCR. The R scripts for analysis is included as a text file, ‘Source Data 1’.

Cell cycle analyses were performed essentially as described (Knutsen et al., 2011). Briefly, cells were fixed with 70% ethanol, washed with 20 mM EDTA pH 8.0, and treated with RNaseA for 3 hr at 37°C. Immediately before analysis by flow cytometry, 2 μM Sytox Green (Invitrogen) in 20 mM EDTA pH 8.0 was used to resuspend pelleted cells. Cells were excited with a 488 nm laser and Sytox Green signal was detected with a 505-nm longpass filter and a 530/30 bandpass filter. Cell cycle analysis was performed in the FlowJo Software (Tree Star Inc, Ashland, OR) The identification of cell populations and fraction of cells in each cell cycle phase (G2, S, and G1 + M) were determined as described (Knutsen et al., 2011).

Cells were taken from fresh plates, and then grown overnight with shaking (Elmi) in 96-well plates containing 150 μL YES (Day −1). The next day (Day 0), cells were diluted into YES and measured by cytometry. At the end of Day 0, cells were passaged into YES + DMSO (0 μM TSA) or YES + 50 μM TSA overnight. The next day (Day 1), cells were diluted and grown briefly into the same pretreatment conditions and the 50 µM TSA pre-treated cells were checked for complete de-repression by flow cytometry. Complete de-repression was defined as a qualitative overlap of WT and Δclr4 profiles, with no evidence of repression. Both 0 and 50 μM TSA pretreated cells were then diluted into a gradient of TSA of 11 two-fold dilutions from 50 μM along with a twelfth 0 μM (DMSO) point. Cells were measured after ~6 hr and then passaged into the same TSA gradient conditions to continue growth.

The next day (Day 2), cells were diluted from overnight growth into the same gradient as above, measured ~6 hr later by flow cytometry and passaged into the same gradient again overnight. The same protocol was followed for Days 3 and 4. The full experiment was performed twice at different times (biological replicate). Given the lengthy continuous growth, contamination was occasionally observed in <1% of wells. The replicate shown was chosen based on lacking contamination.

Cells were taken from fresh plates, and then grown overnight with shaking (Elmi) at either 32°C or 38°C (Day-1) in 96-well plates containing 200 µL YES medium per well. In the morning, cells were diluted into 200 µL YES and grown ~6 hr at the same temperature before measurement by flow cytometry (Day 0). At the end of Day 0, all cells were all diluted again into YES and grown at 32°C. The next day (Day 1), cells were diluted from overnight growth into YES at 32°C, measured ~6 hr later by flow cytometry and passaged into the same temperature overnight. The same protocol was followed for Days 2, 3, and 4.

HSS strains were crossed to parent strains lacking functional nucleation factors for REIII (Δpcr1) or cenH (Δdcr1 seb1-1). Cross progeny were identified via a random spore approach by growth on selective media 2 or 3 days after plating. Absence of pcr1 or dcr1 open-reading frames was confirmed by PCR. Presence of seb1-1 allele was confirmed by sequencing. Single colonies were grown in 96-well plates at 32°C containing 200 µL YES medium per well. In the morning, cells were diluted into 200 µL EMM and grown ~6 hr at the same temperature before measurement by flow cytometry. Cells were again diluted into 200 µL YES for overnight growth at 32°C and grown and measured similarly the subsequent days. For Δdcr1 and/or seb1-1 strains and their controls, this was continued for four days. For Δpcr1 strains and their controls this was continued for 5 days then resulting cells were plated onto selective media and allowed to grow 48 hr at 32°C. Patches were then passaged in bulk on selective plates every 36–48 hr for 7 additional days. On the 6th day, the passaged ΔK^HSSΔpcr1 cells were additionally struck for singles. On the 8th day, patches of passaged cells and six single colonies of ΔK^HSSΔpcr1 cells were grown in 96-well plates as above and measured by flow cytometry for 5 additional days.

Recombination Induced Tag Exchange (RITE) parent strain (HU2549) was crossed into HSS reporter strains. Resulting isolates were verified by growth on selective media. The cdc-25ts allele was crossed out. RITE was performed essentially as described (Audergon et al., 2015; Svensson et al., 2015) with the following exceptions. Given the labile nature of heterochromatin at elevated temperatures, replication stalling was performed with hydroxyurea as published (Aygün et al., 2013). Cells were grown to saturation overnight in YES supplemented with Hygromycin. In the morning cells were diluted to OD = 0.1 in 50 mL YES+Hygromycin and grown for 4 hr at 30°C, 225 rpm. After 4 hr of growth, 13 mL of cells were pelleted and processed for ChIP as the 0 hr time point. The remaining cells were washed twice in media devoid of Hygromycin and finally resuspended in YES supplemented with 15 mM Hydroxyurea (HU) and 1.5 μM β-Estradiol (ER) and incubated for 4 additional hours at 30°C, 225 rpm. After 4 hr incubation with HU and ER, 10 mL of cells were pelleted and processed for ChIP.

We found that sonication of a small number of cells such as can be collected by FACS leads to a marked increase in background signal from negative control regions that was absent when ChIP was performed with larger log phase cultures (>50 × 10⁶ cells). To address this, ChIP in Figure 1E was performed on each of the FACS sorted populations with the addition of 42 × 10⁶ formaldehyde fixed cells of S. cerevisiae W303 strain as a carrier. ChIP in Figure 2D was performed with 15 × 10⁶ cells of each fission yeast strain and 50 × 10⁶ additional W303. ChIPs for Figure 2E and Figure 2—figure supplement 2B were performed with 80 × 10⁶ cells and no added W303. ChIPs for Figure 6B were performed with no added W303. ChIP was additionally performed on a sample of W303 alone, which only produced signal equivalent to background. S. pombe ChIP samples and W303 cells were fixed and pre-processed for ChIP separately, then mixed together immediately prior to lysis. Cells were cross-linked and lysates prepared for ChIP as described (Canzio et al., 2011) with the following exceptions: After lysis, the chromatin fraction was resuspended in 350 μL lysis buffer and sonication performed using a Diagenode Bioruptor Pico machine at 4°C, with 16–20 rounds of 30 s ON, 30 s rest. ChIP was essentially as described, with the total lysate split into 2–6 equal volumes (after ~8% set aside as input fraction) and ChIP performed in 600–800 μL per sample. Two or three technical replicates were performed across experiments. 1 μL of each of the following antibodies was added per ChIP replicate: anti-H3K9me2 (Abcam ab1220); anti-H3K4me3 (Active Motif 39159); anti-H3K9me3 (Millipore 07–442); anti-T7 (Novagen 69522–3). ChIP samples were agitated on a Nutator overnight at 4°C. Immune complexes were collected for 3 hr with 15–20 μL washed protein A Dynabead slurry (Invitrogen). Washing and downstream processing steps were essentially as described, except ‘wash buffer’ wash was performed once. Samples were purified using a Machery-Nagel PCR purification kit and NTB buffer for SDS containing samples. DNAs were quantified by RT-qPCR (see below). Enrichments were calculated as follows: For Figures 1E, 2D and E IP/input values for amplicons of interest were calculated and normalized to the IP/Input values for positive controls for each antibody, dh for H3K9me2 and H3K9me3 and the actin promoter for H3K4me3. For Figure 2—figure supplement 2B, ChIP signal was normalized to signal from a matched background Δclr4 strain. For Figure 6B IP/input values for the 4 hr time points were normalized to the IP/input values for the 0 hr time point.

After sorting, samples were spun at 5000xg, supernatant decanted, and pellets flash frozen in liquid nitrogen and stored at −80°C. For the Δclr4 strain PAS335, cells were grown into log phase and then cell pellets were isolated in the same fashion. Total RNA was extracted in technical duplicates from the same cell pellets using the ‘MasterPure- Yeast RNA Purification Kit’ (Epicentre), including a 30 min DNAse treatment step post-RNA isolation. Reverse Transcription was performed with SuperScript III RT (Invitrogen), using the supplied protocol and 1.5–2 μg of RNA and an oligo dT primer. Following cDNA synthesis the reaction was treated with RNAse H (New England Biolabs). cDNA samples were quantified by RT-qPCR. For each sorted sample, mKO2 cDNA values were normalized to actin and then divided by the max value calculated similarly from PAS355 (Δclr4).

Real-time quantitative PCR was performed using a BioRad CFX-384 machine. 15 μL reactions were prepared, each containing 7.5 μL of Applied Biosystems SYBR Select Master Mix, 4.5 μL 3.3M betaine, 1.2 μL of 2.5 μM oligo mix, 0.8 μL water, and 1 μL template. The thermocycler protocol was: 2 min at 50°C then 2 min at 95°C followed by 40 cycles of 15 s at 95°C and then 1 min at 60°C followed by a plate read. Lastly a melt curve was generated. Standards were generated with five fold dilutions of genomic DNA containing templates for all PCR products.

Single cells of strains PAS 387, 389, 391 and 244 (see Table 1; E2Crimson under act1 promoter) were captured in microfluidic devices as described (Spivey et al., 2017). Multi-channel fission yeast lifespan microdissectors (multFYLM) contained six independent devices (channels), each of which is capable of capturing up to 392 cells (https://bio-protocol.org/e2783). In brief, the devices were cast in polydimethylsiloxane (PDMS, Sylgard 184, Dow Corning) using conventional soft lithography methods. Master structures were fabricated from P-doped silicon wafers (ID#452, University Wafers) and SU-8 photoresists 3005 and 2010 (Microchem, Westborough, MA). MultFYLMs were cleaned and adhered to glass coverslips (48 × 65 mm #1, Gold Seal), and then connected to syringes (60 mL, Becton-Dickson) containing YES 225 liquid media (Sunrise Science) via PFA tubing and microfluidic fittings (IDEX Health and Science). The multFYLM was maintained at 30˚C in a custom staged-mounted environmental chamber on an inverted microscope (Eclipse Ti, Nikon) equipped with NIS Elements software (Nikon), a 60X air objective (CFI Plan Apo λ, 0.95 NA, Nikon) fitted with an objective heater (Bioptechs), a motorized stage (Proscan III, Prior), and an active feedback-based focusing system (Perfect Focus System, Nikon). An LED lamp (Sola II, Lumencorp) and a scientific-grade CMOS camera (Zyla 5.5, Andor) were used for fluorescent imaging. Multi-color fluorescent imaging of sfGFP, mKO2 and E2Crimson fluorophores was carried out by alternating between three filter sets mounted in a computer-controlled filter ring (Chroma 49002, 49010 and 49015, respectively). To help with the semi-automated cell identification, each channel was imaged every ten minutes via brightfield imaging (100 ms exposure, both in focus and 4 μm below the focal plane). Fluorescent images of each of the three fluorophores were taken every 30 min (150 ms exposure). This illumination scheme was well below the phototoxicity limit, as described previously (Al-Sady et al., 2016). Raw images were saved as uncompressed 16 bit ND2 files and further analyzed using a custom-written image analysis pipeline (see below).

Cells were grown overnight (30°C with 225 rpm shaking) to saturation in YES media, then diluted in YES to an optical density at 600 nm (OD₆₀₀) of 0.1 and allowed to grow for approximately 5 hr to reach an OD₆₀₀ of 0.5. Cells (60 μL at OD 0.5 in YES + 2% Bovine Serum Albumin, BSA) were loaded at the entry port of the multFYLM. After cells entered individual channels, media lines were reattached and YES media was pumped through on a pulse cycle (14 min: 5 µLmin⁻¹, 1 min: 55 µLmin⁻¹) for the entire experiment. This flow regime was optimized to flush out occasional cell clumps that grew at the device inlets and other fluidic interfaces. Four genotypes were imaged simultaneously for 60 hr in each channel of a multFYLM device to ensure identical imaging and growth conditions. In all cases, we only analyze the innermost cell, which was the oldest cell pole (see below). Cells that were ejected or died within the first 12 hr after loading were not included in the downstream analysis.

Single-cell imaging data was processed using an updated version of the custom-written FYLM Critic analysis package (Spivey et al., 2017). The source-code is available via GitHub (https://github.com/finkelsteinlab/fylm; Rybarski et al., 2015; copy archived at https://github.com/elifesciences-publications/fylm). FYLM Critic performs the following automated processing on the raw images: (1) rotation; (2) jitter removal via a cross-correlation algorithm; and (3) generation of kymograph and individual cell images. The latter were used to create videos of individual cells in Fiji (Schindelin et al., 2012). The final outputs of FYLM critic are the position and contour of each dividing cell, as well as the time-dependent fluorescence intensities for each cell. These fluorescence intensities are obtained by averaging the intensity across all pixels that fall within the cell volume, as defined by the bright-field images. This normalization also ensures that the fluorescence intensity is corrected for the size of the rapidly dividing cells. Time-dependent fluorescent intensities were analyzed via custom-written MATLAB scripts (version 2017a Mathworks, available upon request). Background fluorescence from the PDMS device was subtracted using catch tubes that did not receive a cell. The maximum heterochromatin reporter (GFP, mKO2) fluorescence intensity was calculated using ∆clr4 cells in the same reporter construct background. To control for expression variation across the cell cycle, the fluorescence from heterochromatin reporters was also reported as a ratio of the control fluorophore, E2Crimson. Similarly, cells fluorescing in the clamp channel were removed from analysis for MAT-locus-derived strains (see Appendix 1-Supplemental Materials and methods).

Single-cell images generated by the FYLM Critic analysis were compiled into stacked movies using Fiji. Images in bright field and for each color channel were processed separately in batch and then later combined into a vertical stack. For each channel, 0.2% of pixels were allowed to become saturated and pixel values were normalized to the maximum range for the whole sequence in that channel. For bright field, every third image was included to match the imaging frequency of the fluorescent channels. Movies were edited for length to include contiguous imaging sequences without loss of focus and for size to remove non-cellular debris and cells from the opposite side of the channel that entered the field of view. After combining all color channels and bright field, the brightness and contrast were increased for cell 407 to match the red channel brightness of the other strains. Image sequences were saved as uncompressed. avi files with a rate of 15 frames/s.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

We thank Shiv I Grewal, Karl Ekwall, and Hiten D Madhani for their generous gifts of fission yeast strains. We thank Graham A Anderson and Shengya Cao for stimulating discussions, especially on hysteresis, and Brandan La for the initial Matlab scripts for cytometry data analysis. In addition, we thank Carol A Gross for substantial help with writing the manuscript and Jonathan S Weissman and Sigurd Braun for critical comments. This work was supported by grants from the National Institutes of Health (DP2GM123484) and the UCSF Program for Breakthrough Biomedical Research (partially funded by the Sandler Foundation) to BA-S, American Federation of Aging Research (AFAR-020) and the Welch Foundation (F-l808) to IJF and the National Institute of Aging (F32 AG053051) to SKJ. Flow Cytometry data was generated in the UCSF Parnassus Flow Cytometry Core which is supported by the Diabetes Research Center (DRC) grant, NIH P30 DK063720.

© 2018, Greenstein et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.