Remodeling of perturbed chromatin can initiate de novo transcriptional and post-transcriptional silencing

Significance

This study addresses an enigmatic question of how transcriptional and post-transcriptional gene silencing can be initiated de novo in the absence of strong sequence-specific cues. Using the fungus Neurospora crassa as a model organism, we found that both types of silencing can be triggered in mitotic cells by the remodeling of a transiently perturbed (nucleosome-depleted) chromatin state. In this system, the initiation of silencing requires SAD-6, a conserved SWI/SNF chromatin remodeler orthologous to ATRX (Alpha Thalassemia/Mental Retardation Syndrome X-Linked), which has been already implicated in repetitive DNA silencing in fungi, plants, and animals. Thus, the model proposed in this study may underpin a range of gene-silencing phenomena observed in other eukaryotes.

Abstract

In eukaryotes, repetitive DNA can become silenced de novo, either transcriptionally or post-transcriptionally, by processes independent of strong sequence-specific cues. The mechanistic nature of such processes remains poorly understood. We found that in the fungus Neurospora crassa, de novo initiation of both transcriptional and post-transcriptional silencing was linked to perturbed chromatin, which was produced experimentally by the aberrant activity of transcription factors at the tetO operator array. Transcriptional silencing was mediated by canonical constitutive heterochromatin. On the other hand, post-transcriptional silencing resembled repeat-induced quelling but occurred normally when homologous recombination was inactivated. All silencing of the tetO array was dependent on SAD-6, fungal ortholog of the SWI/SNF chromatin remodeler ATRX (Alpha Thalassemia/Mental Retardation Syndrome X-Linked), which was required to maintain nucleosome occupancy at the perturbed locus. In addition, we found that two other types of sequences (the lacO array and native AT-rich DNA) could also undergo recombination-independent quelling associated with perturbed chromatin. These results suggested a model in which the de novo initiation of transcriptional and post-transcriptional silencing is coupled to the remodeling of perturbed chromatin.

The genomes of most eukaryotes contain large amounts of repetitive DNA silenced transcriptionally and post-transcriptionally by diverse processes. While some of these processes are directed by strong sequence-specific signals [for example, Krüppel associated box (KRAB) domain–containing zinc finger proteins recruiting the methyltransferase SETDB1 (1)], others might be induced by the combined effect of multiple weaker interactions, necessitating a threshold number of tandem repeats to initiate silencing (2, 3). This principle may also apply to gene arrays [such as the human D4Z4 array carrying multiple copies of DUX4 (4)] and transgenic repeats, as documented in fungi (5), plants (6), and animals (7, 8). Notably, in mammals, a screen for regulators of transgenic repeat-induced gene silencing (RIGS) identified a number of factors also required for transcriptional (heterochromatin-mediated) silencing of the native gene arrays (9, 10), suggesting that the underlying mechanisms could be conserved.

ATRX (Alpha Thalassemia/Mental Retardation Syndrome X-Linked) is a member of the SWI/SNF(SWItching defective/Sucrose Non-Fermenting) family of chromatin remodelers (11). In a complex with the histone chaperone DAXX, ATRX was shown to deposit the histone variant H3.3 at several types of repetitive sequences (pericentromeric, telomeric, ribosomal, and other repeats) as well as nonrepetitive regions (11). H3.3 deposited by ATRX-DAXX often carries H3K9me3 (11). Overall, ATRX has roles in several chromatin-based processes including transcriptional silencing of repetitive DNA, telomere maintenance, resolution of secondary DNA structures, and regulation of gene expression (11). In addition to DAXX, ATRX is known to interact with other chromatin factors (proteins and long noncoding RNAs) as well as specific histone modifications (H3K9me3 and H3K4me0) (11). ATRX is also known for its ability to recognize and remodel nucleosome-depleted chromatin linked to aberrant transcription, such as the Hsp70 genes during heat shock in Drosophila (12) and GC-rich telomeric repeats giving rise to R-loops (13).

The fungus Neurospora crassa sports several genome-defense processes that involve the de novo initiation of transcriptional and post-transcriptional silencing. First, in vegetative cells of N. crassa, transgenic repeats can become heavily methylated, as the result of potent RIGS (5). Second, during sexual reproduction, duplications of genomic DNA longer than a few hundred base-pairs trigger repeat-induced point mutation (RIP), by which many cytosines in repeats are converted to thymines (14, 15). Interestingly, RIP can be carried out by the same pathway that establishes de novo DNA methylation during RIGS (5, 16).

Concerning the initiation of post-transcriptional silencing, two processes need to be considered. One process, known as quelling, triggers the expression of small interfering RNAs (siRNAs) from repetitive transgenes in vegetative cells (17). Another process, known as meiotic silencing by unpaired DNA (MSUD), takes place in early meiosis and induces the expression of siRNAs from the mismatching loci found at the allelic positions on pairs of homologous chromosomes (18, 19). Whereas MSUD occurs normally without DNA breakage and recombination (20), quelling was proposed to rely on recombinational DNA repair as a mechanism for repeat recognition (21). Interestingly, potent MSUD was shown to require SAD-6, a fungal ortholog of ATRX (22).

A substantial part of the N. crassa genome corresponds to AT (Adenine and Thymine)-rich DNA (produced by RIP), which nucleates H3K9me3 through a process mediated by the SUV39 methyltransferase DIM-5 (23). In its turn, H3K9me3 is recognized by HP1 (Heterochromatin Protein 1), which forms a complex with the cytosine methyltransferase DIM-2 responsible for all DNA methylation in this organism (23). Thus, in N. crassa, AT-rich DNA functions as a hard-wired signal for the formation of constitutive heterochromatin and ensuing DNA methylation (23). Several additional components of this pathway were identified, including histone deacetylases and chromatin remodelers (24, 25). In parallel, a more dynamic type of DNA methylation occurs at some loci associated with antisense transcription (26, 27). Interestingly, this DNA methylation also requires DIM-5 and HP1 (28).

This study was motivated by two earlier results. First, in N. crassa, a long tetO operator array was reported to trigger RIP by the heterochromatin-related pathway, suggesting that it could initiate transcriptional silencing (29). Second, in budding yeast, the association of such arrays with their corresponding repressor proteins was shown to induce transcriptional silencing of a nearby gene (30, 31). That process started with phosphorylation and concomitant depletion of histone H2A, suggesting that the binding of repressor proteins led to chromatin stress and nucleosome loss, and it did so by a mechanism independent of DNA replication (30). The perturbed array was directed to the perinuclear compartment, where it became occupied by Silent Information Regulator (SIR) proteins, the main effectors of heterochromatic silencing in yeast (30). Overall, it was proposed that the tight yet dynamic binding of repressor proteins could accelerate the turnover of nucleosomes and promote the incorporation of histones without certain modifications (such as H3K79me3), which would normally interfere with the recruitment of SIR proteins (30).

We now report that the lacO and the tetO operator arrays (32), when integrated as single-copy constructs and never exposed to RIP, can initiate strong transcriptional and post-transcriptional silencing in somatic cells of N. crassa. At the tetO array, the silencing was induced by the aberrant activities of either TetR-GFP or NIT-2 (a transcription factor of the GATA family). Both situations were associated with locally perturbed chromatin. Transcriptional silencing of this locus was mediated by canonical constitutive heterochromatin. On the other hand, post-transcriptional silencing resembled quelling but still occurred normally in the absence of RAD51 or RAD52, two main recombination factors. Therefore, this process was named recombination-independent quelling (RIQ). All silencing of the tetO array was dependent on SAD-6 (a fungal ortholog of the SWI/SNF chromatin remodeler ATRX), which was required to maintain nucleosome occupancy at the perturbed locus. In addition, we found that two other types of sequences (the lacO array and native AT-rich DNA) could also undergo RIQ associated with perturbed chromatin. These results suggested a model in which the de novo initiation of transcriptional and post-transcriptional silencing is coupled to the remodeling of perturbed chromatin.

Results

Experimental System.

In budding yeast, the binding of bacterial repressor proteins to their corresponding operator arrays can trigger transcriptional silencing of a nearby reporter gene (31). It was interesting to test whether a similar process occurred in N. crassa, an organism with canonical transcriptional (heterochromatin-mediated) and post-transcriptional (RNAi-mediated) silencing (23). To this end, the standard lacO and tetO arrays were used (32). These arrays contained 168 lacO1 and 191 tetO sites (interspersed with random sequences) and were characterized by GC content of ~40% and ~50%, respectively (SI Appendix, Fig. S1A). In addition, the lacO array included three perfect repeats of several hundred base-pairs (SI Appendix, Fig. S1A). The arrays were concatenated on a plasmid and integrated between his-3 and lpl by homologous recombination (Fig. 1A: “Strain A”). A synthetic construct expressing TetR-GFP was subsequently integrated as the replacement of csr-1+ (Fig. 1A: “Strain B”). To protect the arrays from RIP and to eliminate the effects of cytosine methylation on H3K9me3 (23), the genes encoding DIM-2 and RID (a putative cytosine methyltransferase involved in RIP) were deleted in all strains used in this study. The state of chromatin was assayed by ChIP-seq (chromatin immunoprecipitation coupled with high-throughput sequencing) and MNase-seq (micrococcal nuclease digestion coupled with high-throughput sequencing), while expression of small RNAs (sRNAs) was followed by sRNA-seq (small RNA high-throughput sequencing).

Fig. 1.

Repressor-bound tetO array initiates transcriptional and post-transcriptional silencing. (A) Series of strains created by homologous transformations, as indicated (strain names and genotypes are provided in SI Appendix, Table S2). (B) Properties of sRNAs expressed from the repressor-bound tetO array (in Strain B). (C) Changes in the expression of array-derived sRNAs (see text for explanation). Corresponding raw data are plotted in F. (D) Density of ChIP-seq reads over the lacO–tetO reporter and the neighboring genes (per 1 bp, per 1 million mapped reads). Unless noted otherwise, all profiles are plotted to the same scale, corresponding to the default scale bar in the Upper Right corner. Tc: cultures were supplemented with tetracycline (100 μg/μL). Positions of the two neighboring genes (his-3 and lpl) are indicated. The plotted region is 26,000-bp long. (E) Density of sRNA-seq reads over the lacO–tetO reporter and the neighboring genes (analyzed and plotted as in D). (F) Scatter plots showing genome-wide changes in sRNA expression. Number of reads (calculated per 500 bp, per 1 million mapped reads) was augmented by 1 to enable log10 transformation. Tiles overlapping the lacO array, the tetO array, and the reference loci are shown as pink, green, and light-blue circles, respectively. Small RNAs expressed from subtelomeric and AT-rich regions are indicated with magenta and blue arrows, respectively.

Binding of TetR-GFP to the tetO Array Induces Strong Transcriptional and post-transcriptional Silencing.

In the absence of TetR-GFP, only very small amounts of sRNAs were expressed by both arrays (Strain A; Fig. 1 E and F and SI Appendix, Fig. S1B). Interestingly, the lacO array was strongly enriched in H3K9me3 (Fig. 1D). An AT-rich motif “ATAACAATT” was noted in the sequence of lacO1 (SI Appendix, Fig. S1A), raising a possibility that the lacO array could nucleate heterochromatin analogously to native AT-rich DNA in N. crassa (23).

In the presence of TetR-GFP, two effects concerning the tetO array were observed. First, this segment became strongly heterochromatic (Fig. 1D: “Strain B”). Second, it also started to express very large amounts of sRNAs (Fig. 1 E and F: “Strain B”). These sRNAs were predominantly 20 to 23 nt long and had a strong bias for uracil at the 5′ position (Fig. 1B). Both processes were suppressed by tetracycline (Fig. 1 D and E: “Strain B + Tc”), implicating the binding of TetR-GFP as a causative trigger of all silencing in this system.

Interestingly, the binding of TetR-GFP to the tetO array also stimulated sRNA expression at several loci as far as 100 kbp away from the array (SI Appendix, Fig. S1B). The majority of those loci were noncoding. Such a long-distance effect may involve global repositioning of the array-carrying chromosomal segment, as reported previously in budding yeast (30), or increasing concentration of some critical RNAi factor in the vicinity of the array (33).

A Set of Endogenous sRNA Loci Provides a Standard Reference for Comparative sRNA-seq Analysis.

To compare expression of array-derived sRNAs between different conditions, a set of endogenous sRNA loci was selected as a standard reference. Such loci were chosen for their strong and invariant expression patterns, mostly corresponding to tRNA and 5S RNA genes (several examples are shown in SI Appendix, Fig. S1C). For the purpose of this analysis, those loci were represented by 463 nonoverlapping 500-bp tiles. For any given condition, the reference level was calculated as the median number of sRNA-seq reads mapped to those tiles. This value was used to normalize the levels of array-derived sRNAs (calculated as the median number of reads mapped to the 500-bp tiles overlapping the arrays). This approach could also be used to compare sRNA expression between Neurospora strains of different genetic origin (SI Appendix, Fig. S1D).

The lacO–tetO Locus Undergoes RIQ.

In Neurospora, expression of sRNAs from transgenic repeats represents a hallmark of quelling (17). Thus, we asked whether tetO-derived sRNAs required quelling factors QDE-1 (an RNA-dependent RNA polymerase), QDE-2 (an Argonaute protein), and QDE-3 (a RecQ helicase). The loss of QDE-1 suppressed tetO-derived sRNAs to the background levels corresponding to the repressor-free condition (Fig. 1 C and F). The loss of QDE-3 produced a similar result, yet some residual expression of tetO-derived sRNAs above background could still be detected (Fig. 1 C and F). In both conditions, the tetO array remained occupied by TetR-GFP (SI Appendix, Fig. S2B). On the other hand, tetO-derived sRNAs did not require QDE-2, consistent with the fact that QDE-2 was dispensable for making sRNA during quelling (Fig. 1 C and F). The levels of lacO-derived sRNAs, although being lower by two orders of magnitude compared to tetO-derived sRNAs, exhibited similar regulatory patterns (Fig. 1 C and F).

We also tested the role of ERI1, a conserved exonuclease implicated in heterochromatin assembly and sRNA production associated with antisense transcription in N. crassa (34). We found that ERI1 was not required for the expression of array-derived sRNAs in our system (Fig. 1 C and F).

Earlier studies in N. crassa linked quelling with homologous recombination (21). Thus, we asked whether the uncovered quelling-like process required RAD51 and RAD52, the two critical recombination factors. Several effects were observed in the corresponding gene-deletion strains. First, the expression of tetO-derived sRNAs was not affected by the loss of RAD51 or RAD52 (Fig. 1 C and F). Second, the expression of subtelomeric sRNAs became up-regulated (Fig. 1F and SI Appendix, Fig. S3C). Third, the proportion of background sRNAs [produced at low levels throughout the genome (35)] became strongly elevated as well (Fig. 1F and SI Appendix, Fig. S3C). The loss of RAD52 had a greater impact compared to the loss of RAD51.

In N. crassa, the expression of some sRNA types requires the acetyltransferase RTT109 (36), which catalyzes acetylation of H3K56 to control nucleosome dynamics coupled to DNA replication and repair (37). We found that the loss of RTT109 had a strong impact on tetO-derived sRNAs, reducing their levels 52-fold (Fig. 1 C and F). Interestingly, in this condition, the genome-wide pattern of sRNA expression resembled those observed in the absence of RAD51 or RAD52 (Fig. 1F and SI Appendix, Fig. S3C). These results corroborated the previously reported role of RTT109 in sRNA biogenesis in N. crassa (36) and hinted at a possibility that RTT109 exerted its specific role in the production of tetO-derived sRNAs by modulating nucleosome homeostasis linked to DNA replication, rather than recombinational DNA repair.

Taken together, these results supported two conclusions. First, in Neurospora, similarly to budding yeast (31), the repressor-bound tetO array was engaged in transcriptional (heterochromatin-mediated) silencing. Second, the same locus was also engaged in post-transcriptional (RNAi-mediated) silencing, which resembled repeat-induced quelling but still occurred normally when homologous recombination was disabled. This process was named RIQ.

Type I Topoisomerase TOP3 Is Dispensable for RIQ.

Sgs1 (the yeast ortholog of QDE-3) partners with the type I topoisomerase Top3 to form a conserved complex involved in recombinational DNA repair (38). Thus, we asked whether a Neurospora ortholog of Top3 played a role in RIQ. We found that the top3Δ condition was associated with a sixfold decrease in tetO-derived sRNAs and a 10-fold increase in lacO-derived sRNAs (Fig. 1 C and F). An elevated proportion of background sRNAs was noted as well (Fig. 1F). The latter effect is exemplified by having many sRNA-seq reads mapped to the active genes near the lacO–tetO locus (SI Appendix, Fig. S3B). These results suggested that QDE-3 mediated RIQ by a mechanism that did not require TOP3.

Heterochromatin Assembly during RIQ Is Both RNAi and Recombination Independent.

Our ChIP-seq analysis yielded several insights. First, the bulk levels of histone H3 at the tetO array decreased in the presence of TetR-GFP, indicating the occurrence of stressed (perturbed) chromatin (SI Appendix, Fig. S2C). Second, the arrays remained perfectly heterochromatic when QDE-1, QDE-2, QDE-3, ERI1, RAD51, or RAD52 were removed individually (SI Appendix, Fig. S2 A and D), supporting the idea that, in Neurospora, constitutive heterochromatin and RNAi were largely independent from one another.

Interestingly, we found that the repressor-bound tetO array had its H3K9me3 levels decreased in the absence of RTT109 (SI Appendix, Fig. S2 A and D). This effect was not associated with apparent changes in the hH3 occupancy (relative to the parental strain), implying that TetR-GFP was still present normally (SI Appendix, Fig. S2 A and C). We used fluorescence microscopy as an independent approach to confirm this conclusion (SI Appendix, Fig. S4). We note that the role of RTT109 in constitutive heterochromatin assembly was reported earlier in fission yeast (39).

RIQ Is Constrained by Constitutive Heterochromatin.

We next asked whether the expression of array-derived sRNAs was constrained by constitutive heterochromatin. We found that the lacO–tetO locus lost all H3K9me3 in the absence of the histone deacetylase HDA1 (SI Appendix, Fig. S2A). This process was linked to a decrease in the hH3 levels at the lacO array (SI Appendix, Fig. S2C), and a comparable loss of hH3 also occurred at AT-rich DNA (SI Appendix, Fig. S2C). With respect to sRNAs, two effects were observed. First, in the hda1Δ condition, the expression of tetO- and lacO-derived sRNAs surged 18- and 539-fold, respectively (Fig. 1 C and F). Second, large amounts of sRNAs also became expressed from many AT-rich regions across the genome (Fig. 1F and SI Appendix, Fig. S3 A and C). A similar pattern was reported earlier in Neurospora strains lacking all heterochromatin (40).

SAD-6 Controls All Silencing of the Repressor-Bound tetO Array.

An emerging connection between perturbed chromatin and RIQ encouraged us to investigate the involvement of chromatin remodeling factors. We started by testing CHD1 (Chromodomain Helicase DNA-Binding Protein 1), which was already implicated in transcriptional and post-transcriptional silencing in N. crassa (21, 26). We found that the loss of CHD1 had no impact on the levels of hH3 or H3K9me3 at the arrays (SI Appendix, Fig. S5 A, C, and D). To a first approximation, the levels of array-derived sRNAs were not affected as well (SI Appendix, Fig. S5B).

Our results indicated that RIQ resembled MSUD in being recombination independent. Therefore, we asked whether the SWI/SNF chromatin remodeler SAD-6, which belongs to the ATRX clade and represents a critical MSUD factor (22), was involved in RIQ. Several surprising results were observed in the sad-6Δ condition. First, the expression of tetO-derived sRNAs decreased to a very low level associated with the absence of TetR-GFP and QDE-1 (Fig. 2A and SI Appendix, Fig. S6A). We confirmed (by ChIP-seq and microscopy) that TetR-GFP was still present at the tetO array in this situation (Fig. 2B and SI Appendix, Fig. S4). Second, the tetO array was no longer enriched in H3K9me3 (Fig. 2B). Third, the loss of silencing was accompanied by a strong decrease in the hH3 occupancy over DNA containing the tetO sites (Fig. 2B). Notably, hH3 was still retained at the GmR gene (in the middle of the tetO array); however, it was not associated with H3K9me3 (Fig. 2B). Fourth, the hH3 occupancy was restored in the presence of tetracycline (Fig. 2B). Fifth, similar dynamics was also exhibited by histone H2B (Fig. 2D).

Fig. 2.

Chromatin remodeler SAD-6 controls all silencing of the repressor-bound tetO array. (A) Scatter plot showing genome-wide changes in sRNA expression (analyzed and plotted as in Fig. 1F). (B) Density of ChIP-seq reads over the arrays and the neighboring genes (analyzed and plotted as in Fig. 1D). (C) Scatter plots showing the relationship between MNase sensitivity and GC content (SI Appendix, Fig. S1E). Number of reads (calculated per 500 bp, per 1 million mapped reads) was augmented by 1 to enable log10 transformation. Tiles overlapping the lacO and the tetO arrays are shown as pink and green circles, respectively. (D) Density of ChIP-seq reads over the arrays and the neighboring genes (analyzed and plotted as in Fig. 1D). (E) Scatter plots showing the relationship between histone occupancy and GC content. Number of reads (calculated per 500 bp, per 1 million mapped reads) was augmented by 1 to enable log10 transformation. SND tiles were defined as those having the log10-transformed (IP[hH3]+1)/(Input+1) ratio of −0.5 or less. Tiles overlapping the lacO and the tetO arrays are shown as pink and green circles, respectively. (F) Histogram showing the relationship between the length and the number of SND regions. The length of each SND region corresponds to the number of consecutive SND tiles included in this region. (G) Density of ChIP-seq reads over the arrays and the neighboring genes (analyzed and plotted as in Fig. 1D).

The coordinated depletion of hH2B and hH3 [both encoded by single-copy genes in N. crassa (23)] pointed to the loss of entire nucleosomes. However, such an effect could also result from changes in epitope accessibility for ChIP. We used MNase sensitivity profiling to discriminate between these two possibilities. To this end, we found that the tetO array became progressively sensitive to MNase as its histone ChIP-seq coverage decreased (Fig. 2 C and E and SI Appendix, Fig. S2C), implying that the latter served as a reliable measure of nucleosome occupancy. Overall, the unperturbed tetO array was characterized by low and largely uniform MNase sensitivity (SI Appendix, Fig. S7A: “Strain A”). Upon binding the repressor, the sensitivity of the array increased but still remained largely uniform (SI Appendix, Fig. S7A: “Strain B”). This uniformity was disrupted in the absence of SAD-6 (SI Appendix, Fig. S7A: sad-6Δ). Corresponding hH2B/hH3 ChIP-seq profiles hinted at a possibility that some strongly protected sites in the sad-6Δ condition could still feature nucleosomes in a fraction of the nuclei (SI Appendix, Fig. S7B).

Genome-wide comparison of the sad-6+/sad-6Δ conditions yielded several results. First, the strong depletion of hH2B and hH3 in the sad-6Δ condition was restricted to the tetO array and rDNA (SI Appendix, Fig. S6 B and F). The latter result was intriguing, but it was not pursued in this study. Our further analysis revealed that the tetO array was the only genomic region incorporating more than 6 consecutive 500-bp “strongly nucleosome-depleted” (SND) tiles (defined by having the log10-transformed (IP[hH3]+1)/(Input+1) ratio of −0.5 or less; Fig. 2 E and F). Second, besides the tetO array, and possibly rDNA, constitutive heterochromatin was lost at only one additional locus, designated as “r2” (SI Appendix, Fig. S6 B, F, and J). Third, only one other unrelated locus, designated as “r1,” had its sRNA levels reduced in the absence of SAD-6 (SI Appendix, Fig. S6 E and I). In the wild type, this region produced sRNAs from both strands by a process that also required QDE-1 (SI Appendix, Figs. S3C and S6I). Fourth, the tetO array was the only genomic locus characterized by such a dramatic increase in MNase sensitivity (SI Appendix, Fig. S6 C and G).

SAD-6 Is Specifically Recruited to the Repressor-Bound tetO Array.

To test whether SAD-6 regulated the tetO array directly, we replaced the promoter of the native sad-6+ gene with a construct that i) provided an N-terminal 3xFLAG tag and ii) increased the expression of the tagged SAD-6 protein to compensate for a partial loss of its activity (SI Appendix, Fig. S6D). We found that tagged SAD-6 became highly and specifically enriched at the repressor-bound tetO array, where it was partially active (as evidenced by the intermediate levels of hH3 and H3K9me3 at the array, Fig. 2G). Genome-wide, comparable enrichment was only observed at one additional locus (designated as “r3”), which overlapped the promoter of NCU01783 (SI Appendix, Fig. S6 H and K). The latter effect was unexpected yet reproducible (SI Appendix, Fig. S6K). Taken together, these results suggested that the repressor-bound tetO array represented a particularly favorable substrate for SAD-6.

NIT-2 Is Required for the De Novo Silencing of the tetO Array Induced by Nitrogen Starvation.

Our results demonstrated that aberrant binding of TetR-GFP perturbed chromatin and initiated transcriptional and post-transcriptional silencing. It was critical to know whether TetR-GFP played a general role in this process. If so, it could be replaced by an analogous unrelated protein. Here, we took advantage of the fact that each tetO site contains two GATA motifs (SI Appendix, Fig. S1A), which can recruit GATA factors in vivo (41). In Neurospora, upon nitrogen starvation, the GATA factor NIT-2 is known to activate the expression of genes involved in nitrogen catabolism (42). The DNA-binding domain of NIT-2 is very conserved (Fig. 3A), and its capacity to recognize the GATA motif was demonstrated in vitro (42). While only a few GATA motifs are found in the promoters of genes regulated by NIT-2 (43), the occurrence of many such motifs over the tetO array was expected to result in a situation where the aberrant activity of NIT-2 could perturb chromatin analogously to the strong binding of TetR-GFP.

Fig. 3.

Nitrogen starvation induces silencing of the tetO array in the absence of the repressor. (A) Comparison of the DNA-binding domains of NIT-2 and its yeast ortholog, Gat1. (B) Series of strains created by homologous transformations, as indicated (strain names and genotypes are provided in SI Appendix, Table S2). (C) Standard protocol for inducing acute nitrogen starvation. (D) Density of ChIP-seq reads over the array and the neighboring genes (analyzed and plotted as in Fig. 1D). (E) Density of sRNA-seq reads over the array and the neighboring genes (analyzed and plotted as in Fig. 1E). (F) Scatter plots showing genome-wide changes in sRNA expression (analyzed and plotted as in Fig. 1F). Tiles overlapping the tetO array and the reference loci are shown as green and light-blue circles, respectively. (G) Density of ChIP-seq reads over the array and the neighboring genes (analyzed and plotted as in Fig. 1D). (H) Scatter plots showing the relationship between histone occupancy and GC content (nitrogen starvation in the absence of SAD-6; analyzed and plotted as in Fig. 2E). (I) Histogram showing the relationship between the length and the number of SND regions (analyzed and plotted as in Fig. 2F). (J) Scatter plots showing the relationship between MNase sensitivity and GC content (analyzed and plotted as in Fig. 2C).

To test the role of NIT-2, we created another strain carrying only the tetO array, without the adjacent lacO array (Fig. 3B: “Strain E”). This strain was also used to test for the de novo initiation of heterochromatin, as opposed to its spread from the lacO array. To this end, we found that, once TetR-GFP became available, the standalone tetO array expressed sRNAs and assembled H3K9me3 at the levels similar to those in Strain B (Fig. 3 D and E: “+ tetR-gfp”), thus demonstrating the ability to initiate de novo transcriptional and post-transcriptional silencing.

Our basic nitrogen-starvation protocol included pregrowing Neurospora in the standard minimal medium for 24 h, after which mycelial cultures were washed, transferred into a new medium lacking nitrogen, grown for additional 5 h, and harvested for analyses (Fig. 3 B and C). Thus, this protocol permitted assaying relevant parameters during the early stages of silencing. Remarkably, high levels of sRNAs and H3K9me3 were found at the tetO array upon nitrogen starvation, reaching those induced in the presence of the repressor (Fig. 3 D and E). Critically, starvation-induced sRNAs and H3K9me3 were completely abrogated in the nit-2Δ condition (Fig. 3 D–F). Genome-wide, several additional loci exhibited elevated levels of sRNAs upon starvation; for some of those loci, this effect was dependent on NIT-2 (SI Appendix, Fig. S8B).

Our further analysis revealed that the starvation-induced expression of tetO-derived sRNAs required QDE-1 and QDE-3 but not RAD51 or RAD52 (tetO-derived sRNAs decreased nearly eightfold in the rad52Δ strain, yet they still exceeded background by two orders of magnitude; Fig. 3F). Thus, this process was classified as RIQ (Fig. 3F). Furthermore, starvation-induced assembly of constitutive heterochromatin occurred normally in the absence of QDE-1, QDE-3, RAD51, or RAD52 (SI Appendix, Fig. S8A), excluding the roles of RNAi and recombination in its de novo initiation.

SAD-6 Controls All Starvation-Induced Silencing.

Thus far, our results indicated that the silencing triggered by nitrogen starvation (and mediated by NIT-2) was equivalent to the silencing induced by TetR-GFP. Subsequent experiments established that the corresponding sad-6Δ conditions were also equivalent (Fig. 3 D–J). Specifically, the following effects were noted at the array upon nitrogen starvation in the sad-6Δ strain: tetO-derived sRNAs were not induced (Fig. 3 E and F), constitutive heterochromatin was not assembled (Fig. 3D), and the array became nucleosome-depleted as well as MNase-hypersensitive (Fig. 3 D and G–J). Interestingly, the array was essentially the only genomic locus featuring the aforementioned chromatin defects (SI Appendix, Fig. S8 C and D).

Comparison of high-resolution MNase-seq profiles yielded additional insights. First, MNase sensitivity of the unperturbed tetO array was not influenced by the adjacent heterochromatic lacO array (SI Appendix, Fig. S7 A and C: “Strain A” and “Strain E”). Second, when Strain E was starved for nitrogen, MNase sensitivity of the tetO array increased but remained largely uniform (SI Appendix, Fig. S7 A and C). Third, this uniformity was compromised in the sad-6Δ condition, recapitulating the state of the tetO array bound by TetR-GFP in the absence of SAD-6 (SI Appendix, Fig. S7 A and C). Notably, the two sad-6Δ conditions featured MNase-seq profiles that were markedly different, suggesting that the fine structure of the perturbed chromatin state was influenced by the nature of the perturbing agent.

DNA Replication Is Required for Starvation-Induced RIQ.

Canonical repeat-induced quelling can be suppressed by 0.1M hydroxyurea (HU) (44), a potent yet reversible inhibitor of DNA replication in N. crassa (45). We also found that strong RIQ of the tetO array was dependent on the acetyltransferase RTT109 with known roles in DNA replication (Fig. 1 C and F) (37). Therefore, we asked whether DNA replication was required for RIQ. Focusing first on the repressor-bound tetO array, we found that a 24-h incubation of pregrown mycelia in the minimal medium containing 0.1M HU down-regulated tetO-derived sRNAs by a factor of 7 (while having no effect on lacO-derived sRNAs; Fig. 1 C and F: “Strain B + HU”). By microscopy, we confirmed that the repressor was still localized properly upon the HU treatment (SI Appendix, Fig. S4). The enlarged size of the HU-treated nuclei and the increased proportion of background sRNAs were noticed (Fig. 1F and SI Appendix, Fig. S4). This result indicated a possible connection between RIQ and DNA replication, yet because HU was added to the pregrown mycelial cultures with already active RIQ, the magnitude of this effect could not be determined with certainty.

The above question was addressed using an experimental system provided by starvation-induced RIQ (Fig. 4). To block all DNA replication before starvation, HU was added during the last 5 h of normal growth (Fig. 4A). Subsequently, the HU block could be either released or maintained during the starvation phase (Fig. 4A). Using this system, we found that maintaining the HU block suppressed all starvation-induced RIQ of the tetO array (Fig. 4 B and C). The only other cosuppressed sRNAs were those expressed from AT-rich DNA (Fig. 4C).

Fig. 4.

DNA replication is required for post-transcriptional silencing induced by nitrogen starvation. (A) Standard nitrogen-starvation protocol (Fig. 3C) was modified to incorporate treatments with HU before and (optionally) during starvation. (B) Density of sRNA-seq reads over the tetO array and the neighboring genes (analyzed and plotted as in Fig. 1E). (C) Scatter plot showing the relationship between changes in sRNA expression and GC content.

The lacO Array and Native AT-Rich DNA Can Undergo RIQ.

Our results suggested that native AT-rich DNA and the lacO array shared the capacity to nucleate constitutive heterochromatin. In addition to constitutive heterochromatin, Neurospora features facultative heterochromatin marked by H3K27me3, which is deposited by the lysine methyltransferase SET-7 (23). Compromised fitness of dim-5Δ strains can be partially restored by deleting set-7+ (40, 46). Therefore, to analyze the expression of sRNAs from the lacO array and AT-rich DNA, we moved the lacO–tetO reporter to a new genetic background in which dim-5+ and set-7+ were both deleted (Fig. 5A: “Strain C”). A matching dim-5+ strain was created by replacing csr-1+ with the dim-5+ transcription unit (16) (Fig. 5A: “Strain D”). Our analysis of these otherwise isogenic strains suggested that DIM-5 promoted hH3 occupancy over AT-rich DNA and the lacO array, while also suppressing sRNA expression from these same regions (Fig. 5 B and C and SI Appendix, Fig. S9A). Affected sRNAs required QDE-1 but not RAD51, and they were also down-regulated by HU (Fig. 5D and SI Appendix, Fig. S9 B–D). The roles of QDE-3 and RAD52 were not tested, because we were unable to delete qde-3+ or rad52+ in our basic dim-5Δ strain.

Fig. 5.

SAD-6 also regulates post-transcriptional silencing of native AT-rich DNA. (A) Series of strains created by homologous transformations, starting from Strain C. The latter was produced by crossing Strain A to another strain carrying dim-5Δ and set-7Δ alleles (strain names and genotypes are provided in SI Appendix, Table S2). (B) Scatter plot showing the relationship between changes in sRNA expression and GC content (analyzed and plotted as in Fig. 4C). Tiles overlapping the lacO and the tetO arrays are shown as pink and green circles, respectively. (C) Scatter plots showing the relationship between histone occupancy and GC content (analyzed and plotted as in Fig. 2E). (D) Scatter plots showing genome-wide changes in sRNA expression (analyzed as in Fig. 1F). Populations of sRNAs affected differentially by the deletions of qde-1+ and sad-6+in Strain C are indicated (brown arrows). Reference loci are shown as light-blue circles. (E) Changes in sRNA expression across Chromosome I, obtained as ratios of sRNA-seq reads using the same array of 500-bp tiles as in Fig. 1F. Denominator values were augmented by 1. (F) Euler diagrams showing the relationships among several groups of genomic loci (represented by the same set of 500-bp tiles as in Fig. 1F). AT-rich DNA is defined by GC content ≤38%. Other thresholds correspond to the fivefold change in sRNA expression. (G) According to the model, chromatin can exist in two principal states, normal and perturbed. Perturbed state can be promoted by the aberrant activities of chromatin-associated proteins, such as TetR-GFP or NIT-2. This state can be resolved by chromatin remodelers, such as SAD-6, which are also required to initiate silencing.

In the dim-5Δ condition, two results concerning the function of SAD-6 at AT-rich DNA were observed. First, SAD-6 was required for the optimal expression of most AT-rich sRNA loci suppressed by DIM-5 (Fig. 5 D–F). Second, the loss of SAD-6 was associated with a broad increase in MNase sensitivity of AT-rich DNA (SI Appendix, Fig. S9 E and F). This effect was not evident for heterochromatinized AT-rich DNA (SI Appendix, Fig. S6C). For the majority of the AT-rich loci engaged in RIQ, the increase in MNase sensitivity was moderate and largely uniform (SI Appendix, Fig. S9F).

The lacO Array Becomes Genetically Unstable in the Absence of Heterochromatin.

We found that the lacO array could still undergo RIQ in the absence of SAD-6 (SI Appendix, Fig. S9D). This result was not surprising, as SAD-6 was also partially dispensable for RIQ of AT-rich DNA. However, in this sad-6Δ strain, approximately one-half of the lacO array was deleted (SI Appendix, Fig. S9D). Because such deletions were never observed in the other sad-6Δ strains, this result indicated that the lacO array became genetically unstable when stripped of heterochromatin. We used multiplex PCR to test the stability of the arrays in clonal populations of nuclei of several basic strains (SI Appendix, Fig. S10). Two pairs of primers were mixed for each PCR. One pair was specific for the central portion of either the lacO or the tetO array, the second pair amplified a portion of spo11 as the positive control (SI Appendix, Fig. S10B). Using this approach, frequent deletions in the lacO array were uncovered, but only in the heterochromatin-deficient background (SI Appendix, Fig. S10B: “Strain C”). Such deletions corresponded to recombination products involving the three perfect repeats present in the lacO array (SI Appendix, Fig. S1A). Thus, the lacO array differed from the tetO array in becoming genetically unstable in its perturbed state.

HDA1 and CHAP, Members of the HCHC Complex, Play Different Roles at the lacO Array.

We next analyzed the role of constitutive heterochromatin in RIQ of the lacO array. In N. crassa, constitutive heterochromatin is controlled by several protein complexes, most notably by DCDC and HCHC (23). DCDC catalyzes H3K9me3 and contains DIM-5 (the catalytic subunit), DIM-7, DIM-9, CUL4, and DDB1; whereas HCHC mediates histones deacetylation and contains HDA1 (the catalytic subunit), along with HP1, CDP-2, and CHAP (23). With respect to DCDC, the loss of DIM-7 or DIM-9 affected the lacO array similarly to the loss of DIM-5: H3K9me3 was absent, the hH3 occupancy decreased, and RIQ became strongly activated (SI Appendix, Fig. S9 G–I). These results suggested that DIM-5, DIM-7, and DIM-9 played similar roles in maintaining the state of the lacO array, consistent with them being members of one protein complex.

A more nuanced outcome was observed regarding the function of the HCHC complex at the lacO array. While the loss of HDA1 largely recapitulated the loss of DCDC (SI Appendix, Fig. S9 G–I), the lacO array still featured high levels of hH3 and H3K9me3 without CHAP (SI Appendix, Fig. S9 G–I). Surprisingly, in this strain, the lacO array was also found to undergo RIQ (SI Appendix, Fig. S9 G–I). Genome-wide, the hda1Δ condition was associated with a moderate depletion of hH3 over AT-rich DNA; however, this effect was much less dramatic compared to that observed upon deleting hda1+ in Strain B (SI Appendix, Figs. S2C and S9I). These results suggested that, while HDA1 acted as a critical regulator of heterochromatin at the lacO array, CHAP was involved in fine-tuning the balance between heterochromatin formation and RIQ activation.

At the lacO Array, RIQ Can Be Decoupled from Heterochromatin Assembly.

Our results showed that the lacO array could activate RIQ, assemble heterochromatin, and sustain high levels of genetic instability. It was interesting to know whether any of those properties could be decoupled. To this end, we identified two clonal lineages of Strain C with spontaneous lacO deletions of 5,247 and 8,263 bps (SI Appendix, Fig. S11A). The corresponding truncated arrays were named d1 and d2, respectively (SI Appendix, Fig. S11A). We found that lacO(d1) but not lacO(d2) could still induce RIQ (SI Appendix, Fig. S11B). However, once DIM-5 was provided by transformation, both lacO(d1) and lacO(d2) became strongly heterochromatinized (SI Appendix, Fig. S11C), suggesting that the ability of the lacO array to activate RIQ could be decoupled from its ability to assemble heterochromatin.

In principle, lacO(d1) could trigger RIQ because it contained some sequence motifs not present in lacO(d2). Alternatively, a threshold amount of lacO DNA could be required. To differentiate between these possibilities, we generated three dim-5Δ strains with overlapping 1.4 to 1.6 kbp fragments of lacO(d1) (SI Appendix, Fig. S11 D and E). Neither fragment triggered RIQ; however, they all formed heterochromatin equally well once DIM-5 was provided by transformation (SI Appendix, Fig. S11 F and G). These results suggested that the ability to activate RIQ represented an emergent property of the lacO array, possibly controlled by a threshold-dependent mechanism.

Discussion

The great diversity and abundance of repetitive DNA necessitated the evolution of various processes to keep this fraction of the genome in check. While some of these processes, including the KZFP-TRIM28-SETDB1 pathway in animals (1) and the RITS-CLRC relay in fission yeast (47), employ strong sequence-specific cues to target dispersed repeats, others become activated by a large number of repeats organized as tandem arrays (48). Such processes were suggested to involve transcription-based mechanisms (48–50), homologous DNA–DNA pairing (16, 51), and non-B-DNA structures (52, 53). Tandem repeats were also found to associate with diverse nonhistone proteins, including transcription factors (48, 54, 55).

We have found that in N. crassa, transcriptional and post-transcriptional silencing can be induced de novo by the aberrant activities of two unrelated DNA-binding proteins (a synthetic TetR repressor and the endogenous GATA factor NIT-2) at the tetO array. While transcriptional silencing was mediated by canonical constitutive heterochromatin, post-transcriptional silencing resembled quelling but still occurred in the absence of RAD51 or RAD52, two critical recombination factors. Therefore, this process was named RIQ. In both situations involving TetR-GFP or NIT-2, the tetO array featured perturbed chromatin, as evidenced by decreased nucleosome occupancy and high MNase sensitivity. This state was ameliorated by SAD-6, which was also needed to initiate all silencing at the tetO array. Moreover, the properties of perturbed chromatin associated with RIQ of the tetO array were also relevant for RIQ of native AT-rich DNA. Thus, the state of perturbed chromatin emerged as a general signal for silencing (Fig. 5G). Importantly, while the loss of nucleosomes was linked to all de novo silencing in this study, the exact nature of the inducing signal remains unknown. Other candidate signals include R-loops and noncanonical DNA structures, which are all expected to correlate with nucleosome occupancy.

At the tetO array, heterochromatin and RIQ are initiated concomitantly yet independently from one another. In principle, with respect to heterochromatin assembly, the role of SAD-6 may be similar to the role of ATRX in the deposition of histone H3.3, which also requires DAXX (although no fungal ortholog of DAXX has been identified) (11). On the other hand, the role of SAD-6 in RIQ appears profoundly mysterious.

RIQ Is a Distinct RNAi Process in N. crassa.

Apart from quelling and MSUD, N. crassa sports two other RNAi pathways that can dynamically designate novel sRNA-producing loci (56). The first (qiRNA) pathway is linked to DNA damage and requires RAD51 and RAD52, as well as the classical quelling factors QDE-1 and QDE-3 (56). The second (disiRNA) pathway targets nonrepetitive loci associated with sense/antisense transcription and does not require QDE-1 and QDE-3 (35), but depends on the exonuclease ERI1 (34). Our results set these processes apart from RIQ (which needs QDE-1 and QDE-3 but not RAD51, RAD52, or ERI1). Yet RIQ and canonical quelling share the requirement for DNA replication and the H3K56 acetyltransferase RTT109 (which is also linked to DNA replication) (37). In the case of RIQ, replication may be required to generate nascent chromatin (57), which may be particularly prone to perturbation by the aberrant activity of some DNA-binding proteins (such as TetR-GFP and NIT-2).

Several types of RIQ appear to exist with respect to the requirement for SAD-6. While RIQ of the tetO array induced by TetR-GFP is absolutely dependent on SAD-6, low levels of tetO-derived sRNAs are still observed without SAD-6 upon nitrogen starvation. Further, RIQ of AT-rich DNA is only partially dependent on SAD-6, and RIQ of the lacO array occurs normally without SAD-6, implying that SAD-6 can be replaced by another factor [altogether, the N. crassa genome encodes 24 ATPases with predicted chromatin-remodeling functions (58)]. Nevertheless, all types of RIQ involve perturbed chromatin, require QDE-1, and occur without RAD51, the only RecA-like recombinase in N. crassa (58). All types of RIQ must also rely on transcription to produce single-stranded RNA that can be used as a template for synthesizing double-stranded RNA. While QDE-1 was proposed to mediate this step in the qiRNA pathway (59), the nature of the DNA-dependent RNA polymerase at the basis of RIQ remains unknown.

In Neurospora, two peculiar populations of sRNAs were previously described: One population corresponds to subtelomeric sRNAs, and the other comprises sRNAs expressed at low background levels throughout the genome (35). Notably, we have found that the levels of background sRNAs were apparently increased in all conditions linked to replication and recombination defects. In a subset of these conditions (specifically, without RAD51, RAD52, or RTT109), the levels of subtelomeric sRNAs became elevated as well. Neither of these effects was associated with the loss of QDE-3, consistent with the idea that QDE-3 plays a redundant role in DNA repair (60).

Role of Chromatin Remodeling in the De Novo Initiation of Silencing.

The role of ATP-dependent remodelers in heterochromatin (re)assembly has been well established, with some archetypal examples including (in addition to ATRX-DAXX) the nucleolar remodeling complex (NoRC) (61), the nucleosome remodeling and deacetylase (NuRD) complex (62), Snf2/Hdac-containing repressor complex (SHREC) (63), as well as the helicases SMARCAD1 (64), HELLS/LSH (65) and its plant counterpart DDM1 (66, 67). These factors make nucleosomes more accessible or replaced altogether to promote the incorporation of histone variants or modifications favoring heterochromatin. Our results suggest that the aberrant activity of DNA-binding proteins can produce a state of perturbed chromatin that may constitute a particularly favorable substrate for the remodeling-dependent silencing. Curiously, the occurrence of diverse transcription factors at mammalian pericentromeres was reported (48, 54, 55). While some of these factors could induce the silencing directly, by recruiting additional enzymatic activities or driving transcription of noncoding RNAs (54), others may do so indirectly, by producing a state of perturbed chromatin. In budding yeast, the tight binding of LacI and TetR repressors to their operator arrays was shown to trigger transcriptional silencing (30, 31). According to our data, this process i) can be induced by other DNA-binding proteins, ii) requires chromatin remodeling, and iii) may also initiate post-transcriptional silencing.

How does SAD-6 recognize perturbed chromatin? Because SAD-6 controls RIQ of diverse loci (i.e., the tetO array versus AT-rich DNA), it is unlikely to rely on sequence-specific signals or secondary structures that may recruit its orthologs in other situations (68). On the other hand, SAD-6-dependent RIQ is associated with low nucleosome occupancy and high MNase sensitivity, suggesting that SAD-6 may favor nucleosome-free DNA. This propensity is shared by metazoan ATRX, which also localizes to nucleosome-depleted regions, including heavily transcribed genes (12), free proviral DNA (69), and R-loops (13). In human cells, 87% of ATRX sites are present in open chromatin (70), and ATRX was also shown to interact with free nucleic acids in vitro (71). Yet the exact nature of the SAD-6 recruitment mechanism and its relationship with the initiation of silencing all remain to be established.

In this study, the state of perturbed chromatin was induced by the aberrant activities of nonhistone proteins at the tetO array. Yet, an analogous state might also be attained by other means, for example, by recombination-independent homologous pairing (72). This idea is supported by the role of SAD-6 in MSUD, which relies on such pairing to find gaps in sequence identity between homologous chromosomes (20, 22). Thus, the proposed model (Fig. 5G), although derived from a synthetic experimental system, may also underpin other instances of the de novo initiation of transcriptional and post-transcriptional silencing in eukaryotes.

Materials and Methods Summary

A combination of ChIP-seq, MNase-seq, and sRNA-seq approaches was used to analyze the state of synthetic repetitive loci (represented by the lacO and the tetO operator arrays) and native AT-rich DNA. The following processes were assayed: i) formation of constitutive heterochromatin, ii) histone occupancy, iii) sensitivity to MNase, and iv) sRNA expression.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

All NGS data produced in this study were submitted to the Sequence Read Archive [BioProject PRJNA1109611] (73). Neurospora strains produced in this study were deposited into the Fungal Genetics Stock Center (www.fgsc.net, accessions 27332-27373). Plasmid maps, custom genome references, and 500-bp tiles corresponding to the reference sRNA-seq loci are available at Figshare (74). All other data are included in the article and/or SI Appendix.