1. Chromosomes and Gene Expression
  2. Genetics and Genomics
Download icon

H3K9me3 is required for inheritance of small RNAs that target a unique subset of newly evolved genes

  1. Itamar Lev  Is a corresponding author
  2. Hila Gingold
  3. Oded Rechavi  Is a corresponding author
  1. Tel Aviv University, Israel
Research Communication
  • Cited 1
  • Views 1,108
  • Annotations
Cite this article as: eLife 2019;8:e40448 doi: 10.7554/eLife.40448

Abstract

In Caenorhabditis elegans, RNA interference (RNAi) responses can transmit across generations via small RNAs. RNAi inheritance is associated with Histone-3-Lysine-9 tri-methylation (H3K9me3) of the targeted genes. In other organisms, maintenance of silencing requires a feed-forward loop between H3K9me3 and small RNAs. Here, we show that in C. elegans not only is H3K9me3 unnecessary for inheritance, the modification’s function depends on the identity of the RNAi-targeted gene. We found an asymmetry in the requirement for H3K9me3 and the main worm H3K9me3 methyltransferases, SET-25 and SET-32. Both methyltransferases promote heritable silencing of the foreign gene gfp, but are dispensable for silencing of the endogenous gene oma-1. Genome-wide examination of heritable endogenous small interfering RNAs (endo-siRNAs) revealed that endo-siRNAs that depend on SET-25 and SET-32 target newly acquired and highly H3K9me3 marked genes. Thus, ‘repressive’ chromatin marks could be important specifically for heritable silencing of genes which are flagged as ‘foreign’, such as gfp.

Editorial note: This article has been through an editorial process in which the authors decide how to respond to the issues raised during peer review. The Reviewing Editor's assessment is that all the issues have been addressed (see decision letter).

https://doi.org/10.7554/eLife.40448.001

Introduction

RNA interference (RNAi) responses are inherited in Caenorhabditis elegans nematodes across generations via heritable small RNAs (Alcazar et al., 2008; Buckley et al., 2012; Vastenhouw et al., 2006). In worms, exposure to a number of environmental challenges, such as viral infection (Gammon et al., 2017; Rechavi et al., 2011), starvation (Rechavi et al., 2014), heat (Klosin et al., 2017), and growth in liquid (Lev et al., 2018) induces heritable physiological responses that persist for multiple generations. Inheritance of such transmitted information was linked to inheritance of small RNAs and chromatin modifications, and hypothesized to protect and prepare the progeny for the environmental challenges that the ancestors met.

By base-pairing with complementary mRNA sequences, small RNAs in C. elegans control the expression of thousands of genes, and protect the genome from foreign elements (Luteijn and Ketting, 2013; Malone and Hannon, 2009). Via recruitment of RNA-binding proteins, small interfering RNAs (siRNAs) can induce gene silencing also by inhibiting transcription (Castel and Martienssen, 2013).

Small RNA-mediated transcription inhibition involves modification of histones, however the exact role that histone marks play in inheritance of RNAi and small RNA synthesis is still not entirely clear (Rechavi and Lev, 2017). In C. elegans small RNAs that enter the nucleus were shown to inhibit the elongation phase of Pol II (Guang et al., 2010); In addition, nuclear small RNAs are thought to recruit histone modifiers to the target’s chromatin, resulting in deposition of histone marks such as histone H3K9-tri methylation (H3K9me3) and H3K27me3 (Gu et al., 2012; Lev et al., 2017; Mao et al., 2015).

The interactions between small RNAs and repressive chromatin marks are reciprocal: in Arabidopsis thaliana (Holoch and Moazed, 2015; Molnar et al., 2010) and Schizosaccharomyces pombe (Moazed et al., 2006; Verdel et al., 2004Hall et al., 2002) small RNAs and repressive histone marks form a self-reinforcing feed-forward loop, where nuclear small RNAs induce deposition of repressive histone marks, and in turn the repressive chromatin marks recruit the small RNA machinery to synthesize additional small RNAs. Whether a similar feedback operates in worms and other organisms, is still under investigation. In Neurospora crassa, transgene-induced small RNAs work independently of H3K9me3 (Chicas et al., 2005). In C. elegans, it was previously suggested that H3K9me is required for RNAi inheritance (Shirayama et al., 2012). However, studies from different groups have shown that the situation is more complex, and that H3K9me could be dispensable, and can even suppress heritable silencing of some targets (Kalinava et al., 2017; Lev et al., 2017; Minkina and Hunter, 2017).

In C. elegans H3K9me is considered to depend mainly on the methyltransferases MET-2, SET-25, and SET-32 (Kalinava et al., 2017; Spracklin et al., 2017; Towbin et al., 2012). H3K9 methylation by MET-2 and SET-25 occurs in a step-wise fashion – after MET-2 deposits the first two methyl groups (H3K9me1/2), SET-25 can add the third methyl group (me3) (Towbin et al., 2012). In the germline, however, SET-25 is capable of tri-methylating H3K9 in a MET-2-independent manner (Bessler et al., 2010; Towbin et al., 2012). SET-32-dependent H3K9me3 is at least in part independent of the activity of SET-25 or MET-2 (Kalinava et al., 2017).

To study the roles of H3K9me3 in the maintenance of heritable small RNAs, we examined the inheritance of small RNAs in mutants defective in these histone methyltransferases. Although H3K9me3 was thought to be required for heritable RNAi (Ashe et al., 2012; Gu et al., 2012), we found the heritable RNAi-responses are greatly potentiated in met-2 mutant background (Lev et al., 2017). Our data indicated that the enhanced strength of the RNAi responses in met-2 mutants stems from a genome-wide massive loss of different endogenous small RNA (endo-siRNAs) species. In normal circumstances, these endo-siRNAs compete with exogenously derived siRNAs over shared biosynthesis components required for small RNA production or inheritance (Lev et al., 2017). In addition, we found that the accumulated sterility (or ‘Mortal Germline’, Mrt phenotype) of met-2 mutants results from dysfunctional small RNA inheritance (Lev et al., 2017).

However, our previous results regarding the role of H3K9me1/2 (deposited by MET-2) did not rule out the possibility that H3K9me3 is yet required for efficient heritable silencing of gfp transgenes: We found that RNAi responses in met-2 mutants nevertheless lead to marking of the target gene’s histones with a heritable H3K9me3 modification. Further, a comparison of the H3K9me3 signal on the gfp locus in different mutants has shown that anti-gfp RNAi responses were strongly inherited only in genetic backgrounds where some H3K9me3 trace could be detected (i.e, in wild type, met-2, and met-2;set-25 mutants). In set-25 single mutants, where no statistically significant H3K9me3 footprint could be detected, anti-gfp RNAi was only weakly inherited. Previously, set-25 mutants were reported to be deficient in heritable RNAi responses targeting different fluorescent transgenes (Ashe et al., 2012; Lev et al., 2017).

RNAi silencing of the endogenous oma-1 gene is also inherited transgenerationally. In contrast to anti-gfp heritable RNAi responses, for which H3K9me3 is important, we detected an enhancement in the inheritance potency of anti-oma-1 RNAi in set-25 mutants (Lev et al., 2017). However, in that study we did not examine whether an H3K9me3 footprint was deposited on the endogenous gene oma-1 in the set-25 background (Lev et al., 2017). The publication of a recent paper (Kalinava et al., 2017) which described strong anti-oma-1 RNAi inheritance in met-2;set-25;set-32 triple mutants, despite the absence of a detectable H3K9me3 footprint, prompted us to re-examine the inheritance of anti-gfp RNAi in this triple mutants. We hypothesized that gene-specific characteristics lead to contrasting requirements for H3K9me3 and specific methyltransferases. In this manuscript, we describe an asymmetry in the requirement for H3K9me3 and specific methyltransferases for heritable RNAi responses aimed against the endogenous gene oma-1 and the foreign gene gfp. These differences led us to perform a genome-wide analysis of H3K9 methyltransferase-dependent small RNAs, which revealed that the endo-siRNAs, which depend on H3K9me3 target newly acquired C. elegans genes that might be considered ‘foreign’, similarly to gfp.

Results

Recently Kalinava et al. examined the heritable RNAi responses against oma-1 also in a triple mutant, lacking the three main C. elegans H3K9 methyltransferases, SET-25, SET-32 and MET-2 (Kalinava et al., 2017). The authors reported that silencing of oma-1 was independent of H3K9me3, as in these mutants RNAi responses raised against the oma-1 gene were heritable despite the lack of an H3K9me3 trace (Kalinava et al., 2017).

We successfully replicated the results of Kalinava et al., and came to the same conclusion, that the met-2;set-25;set-32 triple mutant worms inherit RNAi responses against the oma-1 gene, also when we used a different assay for inheritance (Figure 1A and Figure 1B, upper panel). Unlike Kalinava et al., which used qPCR to score for downregulation of oma-1 expression, we targeted a redundant, temperature-sensitive and dominant oma-1 allele, that in the restrictive temperatures does not allow the development of embryos unless silenced (as previously described [Alcazar et al., 2008]). Upon shifting to 20 degrees, only worms that silence the oma-1 gene in a heritable manner survive.

Figure 1 with 3 supplements see all
Heritable RNAi responses against the oma-1 and gfp genes have different requirements for H3K9me3 methyltransferases.

(A) Scheme depicting the different requirements for H3K9 methyltransferases in RNAi inheritance responses aimed at different genes. In the parental generation, worms are exposed to RNAi by growing on plates seeded with dsRNA-producing bacteria. Next, worms are transferred to plates seeded with control bacteria (that do not express dsRNA) to lay the eggs the next generation. (left) Only worms that inherit small RNAs that silence the temperature-sensitive dominant allele of oma-1 can hatch. Heritable RNAi responses aimed against the endogenous oma-1 gene do not require H3K9me3 methyltransferases. (right) Inheritance of anti-gfp small RNAs lead to heritable silencing of the gfp transgene (Pmex-5::gfp::h2b transgene). Heritable RNAi responses aimed against the foreign gfp gene strongly depends on H3K9me3 methyltransferases. (B) Inheritance of anti-oma-1 RNAi response in H3K9me3 methyltransferase mutants. The percentage of fertile worms per replicate and generation is presented (N = 12, three biological replicates). (upper panel) RNAi inheritance dynamics in met-2;set-25;set-32;oma-1 mutants compared to oma-1 mutants. (lower panel) RNAi inheritance dynamics in set-32;oma-1 mutants compared to oma-1 mutants. (C) Inheritance of anti-gfp RNAi response in H3K9me3 methyltransferase mutants. In each generation the percentage of worms silencing a germline expressed GFP transgene is presented (N > 60, five replicates). (upper panel) RNAi inheritance dynamics in met-2;set-25;set-32 triple mutants. (lower panel) RNAi inheritance dynamics in set-32 single mutants. Error bars represent standard error of mean. *p-value<0.05, **p-value<0.005, ***p-value<0.001, ****p-value<0.0001, Two-way ANOVA, Sidak's multiple comparisons test.

https://doi.org/10.7554/eLife.40448.002

In parallel we discovered, surprisingly, that in contrast to anti-oma-1 inheritance, heritable silencing of a gfp transgene was defective in the same triple mutants (Figure 1C, upper panel, p=0.0014, 2-way ANOVA). In addition, we also confirmed (Spracklin et al., 2017) that while set-32 single mutants are deficient in inheriting RNAi responses raised against the gfp transgene (Figure 1C, lower panel, p=0.0026, 2-way ANOVA), they are capable (Kalinava et al., 2017) of inheriting responses raised against oma-1 (Figure 1B, lower panel, p=0.8487, 2-way ANOVA). Previously we have shown that while set-25 mutants are defective in inheritance of anti-gfp RNAi, weak inheritance responses can still be observed (Lev et al., 2017). Similarly, we were able to detect weak inheritance responses that last at least until the F3 generation also in met-2;set-25;set-32 and set-32 mutants (Figure 1—figure supplement 1, p-value < 0.0001 for met-2;set-25;set-32 and set-32 in the F3 generation, Two-way ANOVA). Together with our previous data, which showed that set-25 is required for inheriting anti-gfp RNAi, but not anti-oma-1 RNAi (Lev et al., 2017), these results suggested that heritable RNAi requires H3K9 methyltransferases in a gene-specific manner.

The levels of RNAi-induced H3K9me3 do not explain the gene-specific requirements of methyltransferases for heritable RNAi

Histone methyltransferase mutants may affect RNAi-induced H3K9me3 levels in a gene-specific manner, thus leading to different inheritance dynamics for each gene. To test this possibility, we performed anti-H3K9me3 Chromatin Immunoprecipitation (ChIP) on F1 met-2;set-25;set-32 triple mutant progeny, that were derived from parents exposed to anti-oma-1 RNAi, anti-gfp RNAi, or untreated controls. Using qPCR we found, as was discovered before (Kalinava et al., 2017) that in met-2;set-25;set-32 triple mutants the RNAi-induced H3K9me3 signal was significantly reduced (p-value=0.0007 and 0.0009, Two-way ANOVA, for gfp and oma-1, respectively). Importantly, this was true for both the oma-1 and gfp loci (Figure 2A). Interestingly, in naive wild-type animals, that were not treated with RNAi, the levels of H3K9me3 on gfp were significantly higher than on oma-1 (Figure 2B, p-value = 0.0039, Two-Way ANOVA) and an additional germline-expressed gene dpy-28 (Figure 2B, p-value = 0.0176, student's t-test). We discuss the possible contribution of this RNAi-independent H3K9me3 signal below. Regardless, as no differences can be found in the RNAi-induced fold changes in H3K9me3 levels between gfp and oma-1 (Figure 2A), the levels of RNAi-induced H3K9me3 cannot explain the gene-specific requirements of methyltransferases for heritable RNAi.

The fold change in RNAi-induced H3K9me3 on oma-1 and gfp is comparable.

(A) The RNAi-induced H3K9me3 footprint on the RNAi-targeted genes. The fold change in H3K9me3 levels in F1 progeny of animals exposed to RNAi versus untreated control animals. The H3K9me3 footprint levels were assessed using a qPCR quantification of ChIP experiments conducted with both wild type (left) and met-2;set-25;set-32 mutants (right). Filled or empty circles represent qPCR data obtained using two different primer sets that span different parts of the examined locus. (B) H3K9me3 levels on the gfp and oma-1 genes in naive untreated wild type animals. The deltaCt numbers used to obtain the fold change values were calculated using the eft-3 gene as an endogenous control. The presented data were obtained from three biological replicates. The levels of gfp and dpy-28 H3K9me3 signal in wild type animals are adapted from raw data from our previous publication (Lev et al., 2017). Two-way ANOVA, Sidak's multiple comparisons test. **p-value<0.005. Error bars represent standard deviations.

https://doi.org/10.7554/eLife.40448.006

SET-32 acts upstream to MET-2 and SET-25 to support RNAi inheritance

We previously found that in contrast to set-25 single mutants, which are deficient in RNAi-induced heritable H3K9me3 methylation (Lev et al., 2017; Mao et al., 2015), met-2;set-25 double mutants display a modest but robust H3K9me3 footprint following RNAi (Kalinava et al., 2017; Lev et al., 2017). We therefore hypothesized that in the met-2 background, an additional, perhaps otherwise inactive H3K9 methyltransferase, is expressed or activated, compensating for the absence of SET-25, to allow efficient heritable RNAi responses (see Figure 1—figure supplement 2 for summary). To test this hypothesis, we first examined whether met-2;set-32 double mutants can inherit RNAi responses raised against gfp. If SET-32 and SET-25 compensate for each other and are redundant, then met-2;set-32 double mutants are expected to strongly inherit RNAi responses, similar to met-2;set-25 double mutants (Lev et al., 2017). Our results show, that in contrast to met-2;set-25 double mutants, met-2;set-32 double mutants are defective in RNAi inheritance raised against gfp, since only a very weak response can be detected (Figure 1—figure supplement 2A). The potency of RNAi inheritance in met-2;set-32 double mutants is comparable to that of set-25 (Lev et al., 2017) and set-32 single mutants, or met-2;set-25;set-32 triple mutants (Figure 1C). These results suggest that SET-32 has a distinct role, and that it probably acts upstream to MET-2 and SET-25, in promoting RNAi inheritance. This conclusion is also consistent with the recent observation that SET-32, in contrast to MET-2 and SET-25, has an essential role in the establishment of RNAi-mediated nuclear silencing (Kalinava et al., 2018).

Unlike RNAi silencing of oma-1, silencing of sup-35 and fog-2 genes is not inherited transgenerationally

Currently, the only gene that serves to study heritable transgenerational (more than two generations) RNAi of endogenous genes is oma-1. Transgenerational RNAi inheritance requires the target gene to be expressed in the germline, and many germline genes are essential or do not have a phenotype that can be scored over many generations. The oma-1 gene can serve as a tool for studying RNAi inheritance owing to the availability of a temperature-sensitive, dominant-lethal and redundant allele that can be rescued by RNAi (Alcazar et al., 2008). In search of other endogenous target genes whose heritable silencing could be studied, we examined the inheritance of RNAi against the non-essential germline genes sup-35 and fog-2. SUP-35 is a maternally deposited toxin, expressed in the mother’s germline, suppressed by PHA-1, a zygotically expressed anti-toxin (Ben-David et al., 2017). Consequently, temperature-sensitive pha-1(e2123) mutants develop when grown at 15 degrees but arrest their development when grown in restrictive temperatures, unless exposed to anti-sup-35 RNAi. As previously described (Ben-David et al., 2017), RNAi silencing of sup-35 allowes pha-1 mutants to develop. However, we found this response was not inherited beyond the F1 generation (Figure 1—figure supplement 3A). Expression of the germline gene fog-2 is required for hermaphrodite worms to produce sperm, but is dispensable for sperm production in males (Schedl and Kimble, 1988). Silencing of fog-2 by RNAi lead to depletion of sperm (as evident by stacked oocytes), and the worms were unable to reproduce unless crossed with a male. While we found that this response was inherited to the F1 progeny, it was not inherited transgenerationally (Figure 1—figure supplement 3B). In conclusion, we could not find additional endogenous gene targets that can be transgenerationally silenced upon RNAi. Conveniently, many endo-siRNAs that target various endogenous genes are inherited transgenerationally, and such inheritance can be studied using RNA sequencing.

H3K9me3 methyltransferases are required for the biogenesis of a specific class of endo-siRNAs

Certain germline small RNAs have evolved to confer immunity against foreign genetic elements, while sparing endogenous genes (Malone and Hannon, 2009). The different requirements for particular methyltransferases and H3K9me3 for heritable silencing of gfp and oma-1 may be connected to the fact that gfp is a ‘foreign’ gene, while oma-1 is an endogenous gene. We previously found that exogenous siRNAs that target gfp are lost in set-25 mutants, and hypothesized that endo-siRNAs that target other ‘foreign’ genes would be likewise affected. Therefore, we re-analyzed our previously published small RNA sequencing data, obtained from set-25 mutants (Lev et al., 2017). However, among the targets of these differentially expressed endo-siRNAs, we could not detect striking changes (fold change >1.2) in endo-siRNAs that target transposons and repetitive elements in set-25 mutants (Figure 3A, left panel). In contrast, a subset of endo-siRNAs that target 279 different protein-coding genes was found to exhibit significant changes in set-25 mutants (adj.p <0.1, DESeq2 Figure 3A, right panel). To understand why these small RNAs are uniquely affected by SET-25, we characterized this group and the endo-siRNAs that target them. To compare the endo-siRNA pools that depend on these two H3K9 tri-methyltrasferases, we re-analyzed the recently published small RNA-seq data obtained from set-32 mutants (Kalinava et al., 2018).

Figure 3 with 2 supplements see all
A genome-wide analyses of endo-siRNAs that depend on SET-25 or SET-32.

(A) An expression analysis of endo-siRNAs targeting transposons and repetitive elements classes (left panel) or protein-coding genes (right panel). Shown are the expression values as log2 of number of Reads per Million (RPM) in set-25 mutants (y-axis) compared to wild type animals (x-axis). Gene targets of endo-siRNAs, which display significant differential expression (analyzed with Deseq2, adjusted p-value<0.1) are marked in Red (B) An analysis of H3K9me3 signals (based on published data from McMurchy et al., 2017) on differet sets of gens: highly expressed genes (top 10%, Blue), lowly expressed genes (top 10%, Yellow) and gene targets of endo-siRNAs that depend on SET-25 (based on Lev et al., 2017, Red), SET-32 (based on Kalinava et al., 2018), Dark Red) or endo-siRNAs associated with WAGOs (HRDE-1,WAGO-1, and ERGO-1, Light Red). H3K9me3 signal is aligned according to gene's Transcription Start Sites (TSS), and the regions of 1000 base pairs upstream and downstream of the TSS are shown on the x axis. The y axis shows the averaged signal of the H3K9me3 modification as a function of distance from the TSS. For statistical analysis, control data sets (shown in Gray) were created by sampling the H3K9me3 levels of randomly selected gene sets of the same size as the examined gene list. (C and D) An enrichment analysis of genes with significantly lowered levels of endo-siRNAs targeting them in set-25 and set-32 mutants compared to wild type. Fold enrichment values (log2 scale) are color coded. (C) An enrichment analysis for expression in specific tissues. (D) An enrichment analysis for different small RNA pathways. The p-values were calculated using 10,000 random gene sets identical in their size to the examined endo-siRNA-target gene list. Asterisk denotes statistically significant enrichment values (p-value<0.05).

https://doi.org/10.7554/eLife.40448.007

Since in set-25 the loss of exogenous siRNAs coincided with the loss of heritable RNAi-induced H3K9me3 (Lev et al., 2017), we first tested whether genes that were differentially targeted by endo-siRNAs in set-25 mutants were also marked by H3K9me3. By examining publicly available H3K9me3 data (McMurchy et al., 2017), we found that the 151 genes that lost the endo-siRNAs that target them in set-25 mutants were robustly marked by H3K9me3 in wild type animals (Figure 3B). We also found that in contrast, the 128 genes that had increased endo-siRNA levels that target them in set-25 and mutants were not significantly marked by H3K9me3 (Figure 3—figure supplement 1A). By analyzing an available mRNA-seq dataset (Klosin et al., 2017), we also found a significant enrichment for genes that were upregulated (at the mRNA level) in set-25 mutants amongst the list of SET-25-dependent endo-siRNA targets (1.93-fold enrichment, 18/151 genes, p-value=0.006). This suggests that endo-siRNAs that depend on SET-25 silence targeted gene. Recently, Kalinava et al. sequenced endo-siRNAs from set-32 mutants (Kalinava et al., 2018). Our analysis show that the 337 genes that had reduced levels of endo-siRNAs (fold change >2) in set-32 mutants, were also significantly marked by H3K9me3 (Figure 3B). As expected, these genes showed lower levels of H3K9me3 in set-32 mutants (Figure 3—figure supplement 1B), and genes having increased levels of endo-siRNAs were not significantly marked by H3K9me3 (Figure 3—figure supplement 1A). Together, these results support the hypothesis that H3K9me3 methyltransferases directly support the biogenesis of silencing endo-siRNAs by tri-methylating the H3K9 histones of the endo-siRNAs targeted genes.

Next we examined whether genes that display altered endo-siRNAs levels in set-25 and set-32 mutants are expressed in specific tissues. Genes that had significantly reduced levels of endo-siRNAs targeting them in set-25 or in set-32 mutants exhibited significant, but modest, enrichment for expression in the germline (Figure 3C and Figure 3—figure supplement 1C). No significant enrichment was found for other tissues (Figure 3C and Figure 3—figure supplement 1C).

To identify the small RNA pathways which are affected by set-25 and set-32, we tested whether the differentially expressed endo-siRNAs depend on particular argonautes, or associate with specific biosynthesis or functional pathways (Figure 3D). It was previously suggested that the CSR-1 argonaute carries heritable endo-siRNAs that mark endogenous genes (Claycomb et al., 2009), while the HRDE-1 argonaute carries heritable endo-siRNAs that silence foreign or aberrant elements, whose expression could be deleterious, such as transposons (Luteijn et al., 2012; Rechavi, 2014; Shirayama et al., 2012). A strong and significant enrichment (Figure 3D and Figure 3—figure supplement 1C) was found for endo-siRNAs which are carried in the germline by the argonautes WAGO-1 (Gu et al., 2009) and HRDE-1, which is required for inheritance of exogenous siRNAs (Buckley et al., 2012). Both argonautes were found to be involved in gene silencing (Buckley et al., 2012; Gu et al., 2009). Nevertheless, some of the targets of HRDE-1-bound endo-siRNAs are expressed in the germline (Figure 3—figure supplement 2A). This may explain the concurrent enrichment for both germline-expressed genes and targets of HRDE-1-bound endo-siRNAs amongst the gene targets of endo-siRNAs that depend on SET-25 or SET-32. A significant enrichment was also found for Mutator pathway small RNAs (Zhang et al., 2011), ERGO-1-dependent small RNAs, and putative piRNA targeted genes (Bagijn et al., 2012). On the contrary, a significant depletion was found for genes known to be targeted by CSR-1-carried small RNAs, a pathway that was suggested to support the expression of targeted genes (Claycomb et al., 2009; Shen et al., 2018). The helicase EMB-4 (Akay et al., 2017; Tyc et al., 2017) was shown to preferably bind introns of genes targeted by CSR-1; We could not detect a significant enrichment for genes whose introns are bound by EMB-4 (fold change = 1.07 and 0.79, p-value=0.26 and 0.002, for endo-siRNAs dependent on SET-25 or SET-32, respectively). All together, these results suggest that H3K9 methyltransferases are required for the maintenance of a specific sub-class of HRDE-1 and WAGO-1 small RNAs, that are associated with the Mutator and piRNA pathways, and that target protein-coding genes (Figure 3—figure supplement 2B).

Endo-siRNAs that depend on H3K9me3 methyltransferases target a distinctive subset of newly evolved genes

What distinguishes the target genes of endo-siRNAs that depend on SET-25 and SET-32 methyltransferases? It was recently found that periodic A/T (PATC) sequences can shield germline genes from piRNA-induced silencing and allow germline expression of genes in H3K9me3-rich genomic regions (Frøkjær-Jensen et al., 2016; Zhang et al., 2018). Fittingly, we found that genes targeted by SET-25-dependent and SET-32-dependent endo-siRNAs exhibit a moderate (~9–13% in median values) but significant reduction in PATC density compared to all protein coding genes (Figure 4A, p-value = 0.0026 and 0.0011 for SET-25 and SET-32, respectively). This feature is not general for genes targeted by WAGOs (Worm-specific Argonautes, HRDE-1, WAGO-1 and ERGO-1) associated endo-siRNAs, since these targeted genes have a higher PATC density (Figure 4A, 10% increase in average values, p-value=0.034). In addition, genes targeted by endo-siRNAs that are increased in set-25 or set-32 mutants exhibit significantly increased PATC density (Figure 4—figure supplement 1A). However, we posit that this feature is unlikely to be sufficient for distinguishing between oma-1 and gfp, since the oma-1 gene has a very low PATC density (Figure 4—figure supplement 1B).

Figure 4 with 3 supplements see all
SET-25-dependent endo-siRNAs target newly evolved genes.

(A) A PATC density analysis for SET-25 and SET-32-dependent endo-siRNAs gene targets. The PATC density values (obtained from Frøkjær-Jensen et al., 2016) are presented for all protein-coding genes, gene targets of endo-siRNAs that depend on SET-25 or SET-32 and gene targets of endo-siRNAs associated with WAGO small RNA pathways (HRDE-1,WAGO-1 or ERGO-1). **p-value<0.005, *p-value<0.05, Wilcoxon rank sum test. For clarity of display, values are shown in log2 scale (after addition of 1). The median (black line) and average levels (numbers) of PATC density levels of each plot are indicated (log2 scale). (B) An enrichment analysis of genes conserved at different levels and duplicated genes amongst gene targets of SET-25- and SET-32-dependent endo-siRNAs. The gene sets were generated based on the homology field in WormBase that details the orthologs and paralogs of each nematode gene. We defined a duplicated gene as a gene that has a paralog in C. elegans. We define genes unique to C. elegans as genes that lack an ortholog amongst the nematode species we examined (see Materials and methods). For statistical analysis, control enrichment values were obtained from 10,000 random gene sets with the same size as the examined endo-siRNA-target gene list. ****p-value<0.0001,***p-value<0.001, *p-value<0.05 (C) An analysis of endo-siRNAs fold changes in rnp-2 mutants for genes targets of endo-siRNAs downregulated or upregulated in set-25 mutants or all genes. All p-values<0.001, Wilcoxon rank sum test. (D) An analysis of intron numbers of gene targets of SET-25- and SET-32-dependent endo-siRNAs and WAGO-associated endo-siRNAs compared to all protein-coding genes. In cases of genes that have more than one transcript, the average intron value is used. The median intron number of each plot is indicated (log2 scale). ***p-value<0.001,**p-value<0.005, Wilcoxon rank sum test. (E) An analysis of splicing motif divergence score (based on Newman et al., 2018) of gene targets of SET-25 and SET-32-dependent endo-siRNAs, WAGO associated endo-siRNAs and all protein-coding genes. The median score levels of each plot are indicated. p-value>0.05, Wilcoxon rank sum test.

https://doi.org/10.7554/eLife.40448.010

The lists of genes which are targeted by SET-25- and SET-32-dependent endo-siRNAs were enriched for genes targeted by ERGO-1-dependent endo-siRNAs (Figure 3D). Many of the genes that are targeted by ERGO-1-bound endo-siRNAs are duplicated genes (Fischer et al., 2011; Vasale et al., 2010). Accordingly, we found an enrichment for duplicated genes amongst the genes that had reduced endo-siRNA levels targeting them in set-25 and set-32 mutants (Figure 4B). An additional characteristic of the set of genes targeted by ERGO-1 endo-siRNAs is an enrichment for poorly conserved genes, that have fewer introns, and possess splicing site sequences that diverge from the consensus sequence (Fischer et al., 2011; Newman et al., 2018). It was recently suggested that these poorly conserved genes are targeted for silencing because their aberrant or”non-self-like’ splicing signals are detected by the splicing machinery (Newman et al., 2018).

Therefore, we examined whether the targets of the endo-siRNAs that depend on SET-25 or SET-32 can be distinguished by their splicing signals. The changes in the endo-siRNA pool in mutants of small nuclear ribonucleoprotein-associated protein RNP-2/U1A (rnp-2) mirrored the endo-siRNA changes found in set-25 mutants (Figure 4C), but not that of set-32 mutants (Figure 4—figure supplement 2A). We also found that genes targeted by SET-25-dependent- but not SET-32-dependent endo-siRNAs bear fewer introns (Figure 4D, median of 3 and 4 compared to 4 of all protein coding genes, p-values=0.0047, and 0.42 for SET-25 and SET-32, respectively). No significant differences in the length of the coding sequences were found, hence, the difference in intron number does not simply derive from differences in gene lengths (Figure 4—figure supplement 2B, p-value = 0.8673). The lists of genes targeted by SET-25-dependent or SET-32-dependent endo-siRNAs were enriched with genes shown to be targeted by intron-targeting small RNAs (Figure 4—figure supplement 2C and D). We could not find, however, small RNAs aligning to the introns of the gfp transgene that we studied (Figure 4—figure supplement 2E, in most cases endo-siRNAs target only exons). We also did not find significant differences in the splicing motif divergence score (obtained from Newman et al., 2018). Since splicing also directly affects the RNAi machinery untangling its role in endogenous RNAi is challenging (Newman et al., 2018). In summary, splicing may be one of the factors that contribute to distinguishing genes targeted by SET-25-dependent endo-siRNAs, but not by SET-32-dependent endo-siRNAs.

In contrast, in the sets of genes targeted by either SET-25-dependent or SET-32-dependent small RNA we found a significant enrichment for newly evolved genes (Figure 4B, fold-change = 2.57 and p-value<0.0001 for both SET-25- and SET-32-dependent endo-siRNAs targets, respectively). We define newly evolved genes here as genes which had no orthologs outside C. elegans (35/151 and 78/337 of genes targeted by SET-25- or SET-32-dependent endo-siRNAs, respectively). Concordantly, in the same gene sets we also found a significant depletion for nematode-conserved genes (Figure 4B). Importantly, the sets of genes targeted by SET-25-dependent endo-siRNAs and SET-32-dependent endo-siRNAs show very small overlap (25 out of 465 genes). Thus, while SET-25 and SET-32 are required for the maintenance of endo-siRNAs that target different genes, the characteristics of these genes are very similar, that is they are distinctively newly evolved genes that have slightly lower levels of PATC sequences. Although the changes in PATC density and intron numbers that distinguish these target genes are moderate, it is possible that the cumulative effect of these small differences may result in the exposure of foreign genes that need to be silenced.

In general, we find that certain sub-classes of endo-siRNA, such as ERGO-1 and HRDE-1 bound small RNAs, target gene sets enriched for newly evolved genes (Figure 4—figure supplement 3). The significant enrichment for newly evolved genes among SET-25- and SET-32-dependent endo-siRNAs is maintained, however, even after excluding genes that are also targeted by HRDE-1, ERGO-1, WAGO-1 or Mutator endo-siRNAs (SET-25: 59/151 genes are not shared, fold enrichment = 2.97, p-value=0.0001, SET-32: 153/337 genes are not shared, fold-enrichment = 1.89, p-value=0.0012). Thus, the enrichment of newly evolved genes amongst the targets of SET-25- and SET-32-dependent endo-siRNAs is not simply due to a general preference for newly evolved genes by endo-siRNA pathways. Further, we find that newly evolved genes are marked by higher levels of H3K9me3 in comparison to the average level of H3K9me3 on protein coding genes (Figure 4—figure supplement 3). Likewise, in the absence of RNAi, in wild-type animals, gfp, the example for a foreign (non-nematode) gene that we investigated, has higher levels of H3K9me3, in comparison to the well-conserved oma-1 gene (Figure 2B). The fact that across the genome SET-25-dependent- and SET-32-dependent endo-siRNAs target newly evolved and H3K9me3 methylated genes (Figure 3B and Figure 4B), may explain why inheritance of RNAi responses raised against gfp, but not oma-1, depends on SET-25 and SET-32 (Figure 1).

In summary, our experiments reveal a specific role for histone modifications in small RNA inheritance. While in S. pombe and A. thaliana a feedback between H3K9me3 and small RNAs was suggested to be required for silencing, the worm’s RNAi inheritance machinery may use H3K9me3 as a mark that distinguishes genes identified as ‘new’. Since newly evolved genes can be disruptive, small RNAs survey these H3K9me3-flagged elements transgenerationally.

Discussion

Our study began from an investigation of a perplexing asymmetry in the requirement of specific H3K9 methyltransferases for heritable silencing of the endogenous gene oma-1 and the ‘foreign’ gene gfp. Single mutants of set-25 and set-32 and the met-2;set-25;set-32 triple mutant displayed different heritable dynamics when either the gfp or the oma-1 gene were targeted by RNAi. These results are not unique to the specific gfp transgene that was tested, since similar observations have been made with other transgenes (Klosin et al., 2017; Lev et al., 2017; Shirayama et al., 2012; Spracklin et al., 2017).

Unlike mutations in these histone methyltransferases, which negatively affect heritable silencing of gfp, but not oma-1, mutations in genes required for small RNA inheritance negatively affect heritable silencing of both oma-1 and gfp. For example, the argonaute HRDE-1 is required for inheritance of RNAi responses against both genes (Ashe et al., 2012; Buckley et al., 2012; Kalinava et al., 2017; Shirayama et al., 2012; Weiser et al., 2017). The fact that heritable RNAi responses aimed at different genes are affected by different proteins should be taken into account when studying transgenerational inheritance. Specifically, when screening for genes that affect such inheritance, one must acknowledge that heritable silencing of different targets requires different chromatin modifiers.

Future studies will hopefully reveal why some recently evolved genes, but not others, display high levels of H3K9me3 (in the absence of RNAi), and are targeted by endo-siRNAs. Recent studies examined why transgenes are sensitive to silencing by synthetic piRNAs, while endogenous germline expressed genes, including oma-1, are not. This protection was suggested to be conferred at least in part by PATC sequences, and to be independent of the genomic location of the gene (Zhang et al., 2018). PATC sequences were previously shown to allow expression of transgenes in the germline in heterochromatic areas (Frøkjær-Jensen et al., 2016). Similarly, our analysis revealed that the gene targets of SET-25-dependent and SET-32-dependent endo-siRNAs have lower levels of PATC density (Figure 4A). However, the oma-1 gene does not possess many PATC sequences (Figure 4—figure supplement 1B). An additional theory suggested that an intrinsic unknown coding-sequence feature confers resistance to silencing by piRNAs. Seth et al. have studied why a fusion between oma-1 and gfp can trans-activate silenced gfp transgenes (an effect known as ‘RNAa’,(Seth et al., 2013)). While unique ‘protective’ sequence features were not described in that work, the authors showed that an unknown coding-sequence feature, not related to the codon usage or the translation of the protein, grants the oma-1 gene with its ability to activate silenced transgenes (Seth et al., 2018). It is possible that the gene targets of SET-25- dependent and SET-32-dependent small RNAs that we describe here have unique intrinsic sequences that distinguish them as well. The different requirement of methyltransferases for heritable silencing of some genes but not others may be related to such intrinsic sequence features. Alternatively, it is possible, as was suggested in the past, that new genes are silenced because they are not licensed transgenerationally by heritable small RNAs for expression (Claycomb et al., 2009; Shen et al., 2018). If this is the case, future studies will hopefully reveal how such license is granted (See Figure 5 for Scheme).

Scheme characterizing H3K9me3 methyltransferase-dependent endo-siRNAs and their targets.

SET-25-dependent and SET-32-dependent endo-siRNAs are enriched with small RNAs known to be carried by the argonautes HRDE-1, WAGO-1 and ERGO-1 but not CSR-1. SET-25-dependent and SET-32-dependent endo-siRNAs targets are enriched with newly evolved genes, have fewer PATC sequences, and are marked with higher levels of H3K9me3. Targets of SET-25-dependent but not SET-32-dependent endo-siRNAs bear fewer introns.

https://doi.org/10.7554/eLife.40448.014

Materials and methods

Key resources table
Reagent typeDesignationSource of referenceIdentifiersAdditional information
Strain
(E. coli)
OP50(Brenner, 1974)Op50
RRID:WB-STRAIN:OP50
Strain
(E. coli)
anti-gfp RNAI bacteriaThis study
Strain
(E. coli)
anti-oma-1 RNAi bacteriaAhringer RNAi library
(Kamath and Ahringer, 2003)
4-4-3-C01
Strain
(E. coli)
anti-sup-35 RNAi bacteriaVidal RNAi library
(Rual et al., 2004)
11006-E2
Strain
(E. coli)
anti-fog-2 RNAi bacteriaVidal RNAi library
(Rual et al., 2004)
10011-C3
Strain
(C. elegans)
N2CGCN2
RRID:WB-STRAIN:N2_(ancestral)
Strain
(C. elegans)
set-32(ok1457) ; oma-1(zu405)CGCBFF25
Strain
(C. elegans)
mjIs134[pmex-5::gfp::h2b::tbb-2]Erik Miska's lab (Univeristy of Cambridge)SX1263
Strain
(C. elegans)
oma-1(zu405)CGCTX20
Strain
(C. elegans)
pha-1(e2123)CGCGE24
Strain
(C. elegans)
set-32(ok1457);mjIs134[pmex-5::gfp::h2b::tbb-2]This studyBFF24
Strain
(C. elegans)
met-2(n4256);set-25(n5021);set-32(ok1457);mjIs134[pmex-5::gfp::h2b::tbb-2]This studyBFF26
Strain
(C. elegans)
met-2(n4256);set-25(n5021);set-32(ok1457);oma-1(zu405)This studyBFF27
Strain
(C. elegans)
met-2(n4256);set-32(ok1457);mjIs134[pmex-5::gfp::h2b::tbb-2]This studyBFF28
AntibodyH3K9me3AbcamRRID:AB_306848
Software,
algorithm
GraphPad Prismhttps://www.graphpad.com/scientific-software/prism/RRID:SCR_002798
Software,
Algorithm
Image JOpensource: https://imagej.nih.gov/ij/RRID:SCR_003070
Software,
Algorithm
MATLAB MathWorkshttps://www.mathworks.com/RRID:SCR_016651

Cultivation of the worms

Standard culture techniques were used to maintain the nematodes on nematode growth medium (NGM) plates seeded with OP50 bacteria. Extreme care was taken to avoid contamination or starvation, and contaminated plates were discarded from the analysis.

RNAi bacteria

HT115 Escherichia coli strains expressing dsRNAs were used: anti-oma-1 RNAi bacteria were obtained from the Ahringer RNAi library (Kamath and Ahringer, 2003). Anti-fog-2 and anti-sup-35 RNAi were obtained from the Vidal RNAi library (Rual et al., 2004). For the sequence of the anti-gfp RNAi see supplemental data.

RNAi experiments

RNAi HT115 E.coli bacteria were incubated in lysogeny broth (LB) containing Carbenicillin (25 μg/mL) at 37°C overnight with shaking. Bacterial cultures were seeded onto NGM plates containing isopropyl β-D-1-thiogalactopyranoside (IPTG; 1 mM) and Carbenicillin (25 μg/mL) and grown overnight in the dark at room temperature. Five L4 animals were placed on RNAi bacteria plates and control empty-vector bearing HT115 bacteria plates and maintained at 20°C for 2 days and then removed. The progeny hatching on these plates was termed the P0 generation. In the next generations the worms were grown on E.coli OP50 bacteria. For anti-gfp RNAi experiments, four L4 animals were placed on plates for two days to lay the next generation. In every generation approximately 60 one day adult worms were collected and photographed per condition (see below). For anti-oma-1 experiments, in each generation twelve individual L4 staged worms were placed in individual wells of a twelve well plate. Four days later the number of fertile worms was assessed (at least one progeny) and 12 individual L4 progeny worms were chosen from the most fertile well to continue to the next generation. For and anti-sup-35 RNAi experiments, in each generation 12 individual L4 staged worms were placed in individual wells of a 12-well plate. Two days later the adult worms were removed. Two days later the number of developing worms was counted and twelve individual L4 progeny worms were chosen from the well with the highest number of developing progeny to continue to the next generation. For and anti-fog-2 RNAi experiments, five L4 worms were crossed on RNAi bacteria for 24 hr. The crossed worms were transferred to fresh RNAi bacteria plates. In each generation, five resulting L4 progeny from the cross were crossed on control bacteria plates and ~40 L4 worms were picked to control bacteria plates and photographed a day later. The number of sterile worms with stacked oocytes was assayed.

Germline GFP expression analysis

Percentage silencing analysis: for each condition, around 60 animals were mounted on 2% agarose slides and paralyzed in a drop of M9 with 0.01% levamisole/0.1% tricaine. The worms were photographed with 10x objective using a BX63 Olympus microscope (Exposure time of 200 ms, and gain of 2). The images were analyzed with ImageJ2 software, and the percentage of worms lacking any observable germline GFP signal was calculated.

GFP expression level analysis: for each condition, the GFP fluorescence level of the background and of oocyte nuclei of at least 30 worms was calculated using ImageJ2.

CTCF value was calculated as follows: CTCF = Integrated density of selected object X – (area of selected object X * mean fluorescence of background readings). The obtained CTCF value was normalized to the average CTCF value obtained from photographs of control animals of the same genotype, generation and age which were fed on control plates.

Chromatin immunoprecipitation

Chromatin immunoprecipitation experiments were conducted as described in Lev et al. (2017). For anti-H3K9me3 ChIP experiments the abcam, ab8898 antibodies were used.

qPCR reactions

All Real time PCR reactions were performed using the KAPA SYBR Fast qPCR and run in the Applied Biosystems 7300 Real Time PCR System.

The primer sequences used in qRT-PCR:

gfp set #1 FOR: ACACAACATTGAAGATGGAAGC

gfp set #1 REV: GACAGGTAATGGTTGTCTGG

gfp set #2 FOR: GTGAGAGTAGTGACAAGTGTTG

gfp set #2 REV: CTGGAAAACTACCTGTTCCATG

oma-1 set#1 FOR: AACTTTGCCCGTTTCACC

oma-1 set#1 REV: TCAAGTTAGCAGTTTGAGTAACC

oma-1 set#2 FOR: TTGTTAAGCATTCCCTGCAC

oma-1 set#2 REV: TCGATCTTCTCGTTGTTTTCA

(The above primer set was adapted from Spracklin et al., 2017)

dpy-28 FOR: CTGATGGATCCAGAGTTGG

dpy-28 REV: CTGCTATACGCATCCTGTTC

eft-3 FOR: CCAACATGATTAGTCAGATGACC

eft-3 REV: CTAGGAGTTAGATGTGCAGG.

Information on the sequencing libraries analyzed in this paper

All the studied publicly available sequencing libraries were prepared from synchronized young adult worms grown at 20 degrees (Kalinava et al., 2018; Klosin et al., 2017; Lev et al., 2017; McMurchy et al., 2017). For more information see the original publications and GEO information: A. set-25 small RNAs (Rechavi and Lev, 2017; GEO accession: GSE94798), B. set-25 mRNA (Klosin et al., 2017; GEO accession: GSE83528), C. set-32 small RNAs (Kalinava et al., 2018; GEO accession: GSE117662, the set-32 (red11) allele data were used)., D. set-32 and wild type H3K9me3 ChIP-seq (Kalinava et al., 2018; GEO accession: GSE117662). E. wild type H3K9me3 ChIP-seq (McMurchy et al., 2017; GEO accession: GSE87524).

Bioinformatic genome-wide endo-siRNAs analysis

Small RNA analysis was conducted as previously described (Lev et al., 2017). Briefly, adapters were cut from the reads using Cutadapt (Martin, 2011). Reads that were not cut or were less than 19 bp long, were removed. The quality of the libraries was assessed by FastQC (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Reads were mapped to the C. elegans genome (WS235) using Bowtie2 (Langmead and Salzberg, 2012). In total 31,053,062, 21,913,420 and 18,372,739 reads were mapped in the three wild type biological repeats and 21,258,241, 19,925,004 and 21,391,091 reads were mapped in the three set-25 biological repeats. The mapped reads were then counted using the python script HTseq_count (Langmead and Salzberg, 2012) using. gff feature file from wormbase.org (version WBcel235). Differential expression was analyzed using DESeq2 (Love et al., 2014). p-adjusted value <0.1 was regarded as statistically significant. The set-32 data from Kalinava et al (Kalinava et al., 2018) GEO was analyzed in a similar fashion. The reads the 5' barcode and 3' linker were trimmed using Cutadapt (Martin, 2011), in accordance the information supplied by Kalinava et al in GEO (accession number: GSE117662). Next, reads were filtered to lengths of 20–23 bp and aligned (not allowing mismatches) to the C. elegans genome (ce10) by Shortstack (Axtell, 2013). In total 1,342,884 and 1,006,568 reads were mapped in the wild type and set-32(red11) small RNA samples, respectively. The reads mapping to each genomic were counted by HTseq_count (Langmead and Salzberg, 2012). Since one biological sample was available, significantly altered small RNAs were defined as genes having fold change of larger than 2 (up-regulated) or smaller than 0.5 (downregulated).

Bioinformatic genome-wide analysis of H3K9me3 signal

For analysis of H3K9me3 signal on different genes in wild type worms, the processed H3K9me3 data (aligned and normalized) from the McMurchy et al. study was used (McMurchy et al., 2017; GEO accession GSE87524). The shown H3K9me3 signal represents the averaged H3K9me3 signal in two replicates of young adults. For analysis of the H3K9me3 levels in wild type and set-32 mutants the raw data from the Kalinava et al. study was used (Kalinava et al., 2018; GEO accession: GSE117662). The raw data were analyzed in a similar fashion to the analysis conducted by McMurchy et al. Briefly, adaptors were trimmed using Cutadapt (Martin, 2011) and aligned using Bowtie2 (Langmead and Salzberg, 2012). H3K9me3-enriched regions were identified using MACS2 (Lupien et al., 2008) and the H3K9me3 signal was corrected for biases using BEADS (Cheung et al., 2011).

Bioinformatic mRNA expression analysis

Processed files with raw counts of reads mapping to each gene were downloaded from GEO (Klosin et al., 2017; GEO accession: GSE83528). Differential expressed genes were detected using DESeq2 (adjusted p-value<0.1).

Bioinformatic gene enrichment analysis

The enrichment values denote the ratio between (A) the observed representations of a specific gene set within a defined differentially expressed genes group, to (B) the expected one, that is the representation of the examined gene set among all protein-coding genes in C. elegans. The analysis was done for 15 gene sets: (1) 7727 genes enriched in oocytes gonads (Ortiz et al., 2014) and 9012 genes enriched in spermatogenic gonads (Ortiz et al., 2014); we excluded genes with expression lower than 1 RPKM(2) 11427 genes expressed in isolated neurons (Kaletsky et al., 2016). (3) 7176 genes expressed in intestine (Gerstein et al., 2010) (4) 2957 genes expressed in pharynx (Gerstein et al., 2010) (5) 2526 genes expressed in body muscle (Gerstein et al., 2010) (6) 4146 targets of CSR-1 (Claycomb et al., 2009) (7) 1478 targets of HRDE-1 (Buckley et al., 2012) (8) 87 targets of WAGO-1 (Gu et al., 2009) (9) 399 targets of ALG-3/4 class small RNAs (Conine et al., 2010) (10) 1823 targets of mutator class small RNAs (11) 721 EGO-1 dependent small RNA gene targets (Maniar and Fire, 2011), (12) 23 gene targets of small RNAs up-regulated in ego-1 mutants (Maniar and Fire, 2011), (13) 49 genes targeted by 26G-RNAs enriched in ERGO-IP (Vasale et al., 2010) (14) 77 genes depleted of 22G-RNAs in ergo-1 mutants (Vasale et al., 2010), and (15) 348 putative piRNA gene targets (Bagijn et al., 2012). The putative piRNA gene targets were defined as genes for which, in at least one transcript, the ratio of the # 22G-RNA reads at piRNA target sites between wild type to prg-1 is at least 2 (linear scale). Note that the indicated number above achieved after intersection between the various published data sources and the records appears in the *.gff file used by us.

The enrichment value of a given gene set i in differentially expressed gene targeting small RNAs was calculated using the following formula:

Enrichment=ObservedExpected=fractionofgenesbelongtotheithsetamongdifferentiallyexpressedSTGsfractionofgenesbelongtotheithsetamongallthegenes

Obtaining the observed-to-expected ratios, we then calculated the corresponding p-values using 10,000 random gene groups identical in size to that of the examined group of differentially expressed genes. Next, the enrichment values of the random sets are ranked and the p-value is determined by the ranking of the examined gene set amongst the ranking of all enrichment values of the random sets.

Gene sets by conservation

The classification of gene sets by conservation was done by mining the ‘Homology’ field of all the C. elegans protein-coding genes in WormBase (www.wormbase.com). We defined the following three gene sets (Figure 4B):

  1. Unique to C. elegans – C. elegans genes which have no orthologues gene in any of the following species: B. malayi, C. brenneri, C. briggsae, C. japonica, C. remanei, O. volvulus, P. pacificus and S. ratti.

  2. Caenorhabditis only - C. elegans genes which have at least one orthologues gene in one of the C. brenneri, C. briggsae, C. remanei and C. japonica species, and have no orthologues gene in any of the B. malayi, O. volvulus, P. pacificus and S. ratti species.

  3. Conserved among nematodes - C. elegans genes which have at least one orthologues gene in one of the C. brenneri, C. briggsae, C. remanei and C. japonica species, and in addition have at least one orthologues gene in one of the B. malayi, O. volvulus, P. pacificus and S. ratti species.

Statistical analysis

For RNAi experiments, Two-way ANOVA tests were used to compare the percentages of the RNAi-affected worms (GFP silencing or fertility for the oma-1 assay) between the tested genotypes. In cases of multiple comparisons between genotypes and across generations, Sidak multiple comparison tests were applied. For GFP fluorescence experiments, Two-way ANOVA tests were used to compare the normalized GFP expression levels between the genotypes and across the biological repeats. For H3K9me3 qPCR-ChIP experiments Two-way ANOVA tests were used to compare the delta-delta-Ct (or delta-Ct) values between the gfp and the oma-1 loci obtained using two different primer sets. In cases of comparisons between genotypes and loci the Sidak multiple comparison tests were applied. Biological replicates were performed using separate populations of animals. Statistical tests were performed using GraphPad Prism software (Graphpad Prism) version 6. The statistical analysis used for each of the bioinformatics analyses is listed under the corresponding bioinformatics methods.

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
    The genetics of Caenorhabditis Elegans.
    1. S Brenner
    (1974)
    Genetics 77:71–94.
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
    Model-based analysis of ChIP-Seq (MACS)
    1. M Lupien
    2. J Eeckhoute
    3. CA Meyer
    4. Q Wang
    5. Y Zhang
    6. W Li
    7. JS Carroll
    8. XS Liu
    9. M Brown
    10. W Li
    11. XS Liu
    12. M Lupien
    13. J Eeckhoute
    14. C Meyer
    15. Q Wang
    16. Y Zhang
    17. W Li
    18. J Carroll
    19. X Liu
    20. M Brown
    21. T Kim
    22. B Ren
    23. V Iyer
    24. C Horak
    25. C Scafe
    26. D Botstein
    27. M Snyder
    28. P Brown
    29. B Ren
    30. F Robert
    31. J Wyrick
    32. O Aparicio
    33. E Jennings
    34. I Simon
    35. J Zeitlinger
    36. J Schreiber
    37. N Hannett
    38. E Kanin
    39. T Volkert
    40. C Wilson
    41. S Bell
    42. R Young
    43. A Barski
    44. S Cuddapah
    45. K Cui
    46. T Roh
    47. D Schones
    48. Z Wang
    49. G Wei
    50. I Chepelev
    51. K Zhao
    52. G Robertson
    53. M Hirst
    54. M Bainbridge
    55. M Bilenky
    56. Y Zhao
    57. T Zeng
    58. G Euskirchen
    59. B Bernier
    60. R Varhol
    61. A Delaney
    62. N Thiessen
    63. O Griffith
    64. A He
    65. M Marra
    66. M Snyder
    67. S Jones
    68. T Mikkelsen
    69. M Ku
    70. D Jaffe
    71. B Issac
    72. E Lieberman
    73. G Giannoukos
    74. P Alvarez
    75. W Brockman
    76. T Kim
    77. R Koche
    78. W Lee
    79. E Mendenhall
    80. A O’Donovan
    81. A Presser
    82. C Russ
    83. X Xie
    84. A Meissner
    85. M Wernig
    86. R Jaenisch
    87. C Nusbaum
    88. E Lander
    89. B Bernstein
    90. D Johnson
    91. A Mortazavi
    92. R Myers
    93. B Wold
    94. C Wei
    95. Q Wu
    96. V Vega
    97. K Chiu
    98. P Ng
    99. T Zhang
    100. A Shahab
    101. H Yong
    102. Y Fu
    103. Z Weng
    104. J Liu
    105. X Zhao
    106. J Chew
    107. Y Lee
    108. V Kuznetsov
    109. W Sung
    110. L Miller
    111. B Lim
    112. E Liu
    113. Q Yu
    114. H Ng
    115. Y Ruan
    116. R Redon
    117. S Ishikawa
    118. K Fitch
    119. L Feuk
    120. G Perry
    121. T Andrews
    122. H Fiegler
    123. M Shapero
    124. A Carson
    125. W Chen
    126. E Cho
    127. S Dallaire
    128. J Freeman
    129. J Gonzalez
    130. M Gratacos
    131. J Huang
    132. D Kalaitzopoulos
    133. D Komura
    134. J MacDonald
    135. C Marshall
    136. R Mei
    137. L Montgomery
    138. K Nishimura
    139. K Okamura
    140. F Shen
    141. M Somerville
    142. J Tchinda
    143. A Valsesia
    144. C Woodwark
    145. F Yang
    146. W Johnson
    147. W Li
    148. C Meyer
    149. R Gottardo
    150. J Carroll
    151. M Brown
    152. X Liu
    153. J Song
    154. W Johnson
    155. X Zhu
    156. X Zhang
    157. W Li
    158. A Manrai
    159. J Liu
    160. R Chen
    161. X Liu
    162. J Carroll
    163. C Meyer
    164. J Song
    165. W Li
    166. T Geistlinger
    167. J Eeckhoute
    168. A Brodsky
    169. E Keeton
    170. K Fertuck
    171. G Hall
    172. Q Wang
    173. S Bekiranov
    174. V Sementchenko
    175. E Fox
    176. P Silver
    177. T Gingeras
    178. X Liu
    179. M Brown
    (2008)
    Cell 132:958–970.
    https://doi.org/10.1186/gb-2008-9-9-r137
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
    fog-2, a germ-line-specific sex determination gene required for hermaphrodite spermatogenesis in Caenorhabditis Elegans
    1. T Schedl
    2. J Kimble
    (1988)
    Genetics 119:43–61.
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
  64. 64
  65. 65

Decision letter

  1. Patricia J Wittkopp
    Senior and Reviewing Editor; University of Michigan, United States

In the interests of transparency, eLife includes the editorial decision letter, peer reviews, and accompanying author responses.

[Editorial note: This article has been through an editorial process in which the authors decide how to respond to the issues raised during peer review. The Reviewing Editor's assessment is that all the issues have been addressed.]

Thank you for submitting your article "H3K9me3 is Required for Inheritance of Small RNAs that Target a Unique Subset of Newly Evolved Genes" for consideration by eLife. Your article has been reviewed by four peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Patricia Wittkopp as the Senior Editor. The reviewers have opted to remain anonymous.

The Reviewing Editor has highlighted the concerns that require revision and/or responses, and we have included the separate reviews below for your consideration. If you have any questions, please do not hesitate to contact us.

As you will see in the reviews below, the reviewers agreed on the importance of this topic and the potential impact of the work. However, they also identified concerns with the statistical analyses and strength of the evidence supporting conclusions drawn. During the post-review discussion, after reading the other reviews, at least one of the reviewers stated that s/he became more concerned about the statistics and the reviewer who raised the statistical concerns feels the paper should not be published without quite significant changes. I hope the reviewers' detailed comments below will help you decide how to address these concerns and best proceed with this work. At this point, given the concerns expressed below, it may be best to withdraw this manuscript and return it at a later date when you have had a chance to address the quite significant issues raised by the reviewers.

Separate reviews (please respond to each point):

Reviewer #1:

This paper explores interesting observations, that some particular genes respond differently to RNAi from each other. Oma-1 is a germ line gene that for some reason ignores histone methylation pathways whereas GFP does not ignore these pathways. The authors make the interesting connection that GFP is a foreign gene and therefore may be surveilled more than the endogenous oma-1. And when they compare siRNAs that are loaded into the germline inheritance pathway, they find that genes that novel to C. elegans, more likely to be foreign, are enriched. This is definitely worth publishing. I am not informaticist so I cannot judge the scatterplots in Figures 4,5 so I only weigh in on the RNAi assays for GFP fluorescence or fertility. Those results in Figures 1,2,3 are convincing. I am a little ashamed of myself for not being able to judge the statistics which is why I choose anonymity. Paper is good to me. The discussion is excellent too.

One suggestion: an often overlooked feature of the GFP fusion genes that everyone in C. elegans uses is that the introns are artificial and constructed by Andy Fire back when dinosaurs roamed the Earth. If any intron is likely to be viewed as foreign it would be these introns. Deep sequencing of strains carrying GFP fusion genes should show if these introns are highly subject to siRNA production. Need to filter based on the intron exon junctions of these artificial genes. This is not a key suggestion. Optional for the authors.

Reviewer #2:

In this paper, the authors attempt to parse the differential roles of the three known C. elegans H3K9 histone methyltransferases (HMTs), set-32, set-25, and met-2. One of the first observations they note is the differential role set-32 plays in transgenerational epigenetic silencing of the oma-1 locus and a gfp transgene. More specifically, they find that set-32 is not required to silence oma-1 but is required to silence gfp. This suggests that there may be differential roles between silencing of genes derived endogenously and exogenously. They use ChIP-qPCR to show that this finding is not a result of differential H3K9me3 deposition at the indicated targets. The authors use the minimal data they collected regarding differences in silencing oma-1 and gfp in set-32 mutants coupled with data from previous studies (predominantly, Kalinava 2017, Towbin 2012, Lev 2017) to draw conclusions about the epistasis of the three HMTs. The focus of their paper seems to pivot from epistasis analysis of set-32 to the function and targets of set-25 where they implicate this HMT in specifically targeting newly evolved genes. They identify 151 protein-coding genes that are the specific targets of SET-25 silencing. These genes are enriched for HRDE-1 and WAGO-1 targets, H3K9me3 presence, germline expression, and most interestingly, newly evolved genes. The computational finding that set-25 specifically regulates newly evolved genes is novel and intriguing, yet, experimentally unvalidated.

The implications of this work would be significant as it represents an important and valuable synthesis of data from multiple labs that have struggled to clearly address to this point, frequently arriving at apparently contradictory results. Though the question is interesting, the authors fail to arrive at a convincing conclusion from their analysis. The experiments and lack rigor and the computational analysis is opaque. The very weak first half of the paper, in which the authors piece together datasets from multiple authors, coupled with the second half of the paper that lacks validated experimental evidence to support their model makes this paper unsuited for publication in eLife.

Major Critiques:

1) Figures 1, 2, 3: The first three figures could be combined into one figure. The data are incomplete and, as presented, offer little novel insight. To build a cohesive story, including both novel data and condensation/replication of the numerous published studies referenced in text and outlined below, RNAi inheritance assays must be replicated in all single, double, and triple mutants, minimally including both an exogenous and endogenous target, and results correlated with H3K9me3 deposition at each target in each genetic mutant. This is essential to support the major claims of this paper.

a) Figure 1B (upper and lower) – redundant with Kalinava et al., 2017. (different experimental scheme, same data and interpretation)

b) Figure 1C (lower) – Redundant with Spracklin et al., 2017.

c) Figure 1D – Redundant with Figures 1B-C.

d) Figure 1A-B – Zeller et al., 2016 demonstrate that met-2;set-25 worms have reduced fertility at 20 and 25 degrees C due to accumulation of DNA damage in the germline. This will impact the total% fertile hermaphrodites used as a readout for anti-oma-1 RNAi inheritance when analyzing inheritance defects across all HMT mutants.

e) Figure 2A – Loss of H3K9me3 at the oma-1 locus in treated and untreated worms in the met-2;set-25;set-32 mutant is redundant with Kalinava et al., 2017. Accumulation of WT levels of H3K9me3 marks across the genome is known to rely on MET-2, SET-25, and SET-32 as shown in Towbin et al., 2012, Mao et al., 2015, and Zeller et al., 2016.

f) Figure 2B – Demonstrates that H3K9me3 signal is higher on the gfp locus than oma-1 locus in untreated worms, yet maintains that since there is no observable fold change between H3K9me3 in treated/untreated (2A) this observation does not account for the difference in RNAi inheritance phenotypes observed in figure 1. This assumption is discrediting a valid interpretation that increased base level H3K9me3 deposition at the gfp loci compared to oma-1 is important for the reported differences in RNAi inheritance.

g) Figure 3 – this figure in insufficient to support the epistasis argument presented in text. It should be condensed into Figure 1. Additionally, the figure contains a typo that renders it uninterpretable: either the data are switched or the y-axis is incorrect.

2) Figure 4: This figure requires more rigorously controlled analysis and should focus on novel insight.

a) Figure 4A – The authors fail to do due diligence in showing that the set-25 targets they identify are solely set-25 targets, and that they do not overlap with other HMT targets. Performing set-32 RNA-seq is a necessary control to ensure that these are not a set of common targets simply hypersensitive to loss of a single HMT.

b) Figure 4B-C – It is unsurprising that targets of an H3K9me3 methyltransferase are enriched for H3K9me3. Additionally, it logically follows that these targets are downstream of a pathway known to deposit H3K9me3 (HRDE/WAGO) and not a licensing pathway (CSR). However, the authors should discuss why these targets appear oocyte specific (shown in 4C), while simultaneously HRDE-1 target-enriched. HRDE targets germline non-expressed genes, pseudogenes, transposons, etc. The authors fail to explain or comment on this paradox.

3) Figure 5: Experimental validation is necessary to uphold the claims of this figure, and the ultimate conclusions of this paper. Failure to experimentally and convincingly show that SET-25 is specifically required to target newly evolved genes renders this paper as nothing more than a prediction founded on unsubstantiated computational outputs lacking critical controls. These controls are especially important in datasets containing small sample sizes (151 identified SET-25 genes) which are more prone to bias.

a) Figure 5A-B – the authors show that SET-25 targets are enriched for newly evolved genes. However, they did not perform the important converse experiment: what percentage of newly evolved genes are SET-25 targets? This analysis is necessary to isolate common identifiers between the subset of newly evolved genes targeted by SET-25 as compared to all newly evolved genes. Without this analysis, their claims into the function of SET-25 are baseless.

b) The computational analysis lacks validation by experimental evidence. A computational model is most useful when it is predictive, and the authors have not demonstrated that here. There are multiple ways in which the authors can test their model, a few of which are listed below:

i) Test multiple transgenes for silencing +/- set-25

ii) Validate the transgenerational silencing difference of some of the 151 identified "foreign" SET-25 targets as compared to additional endogenous loci.

iii) Add "foreign" introns or interfere with splice junctions of known (licensed) genes and ChIP-qPCR for H3K9me3 +/- set-25

iv) Mutate gfp transgene to include introns of highly conserved gene or introduce better splice donors/acceptors

4) Figure 6: The model figure must include the full scope of the work. Currently, it only incorporates data collected in Figures 4-5, but does not help the reader understand the genetic interplay of the HMT pathway, and the relationship between H3K9me3 deposition and RNAi inheritance elucidated throughout the first half of the paper.

Minor Critiques:

1) Figure 1D – these data are redundant to panels (1A-1C). Although the data do illustrate the weak RNAi inheritance responses in later generations, this can still be seen in the previous panels, and the space would be better utilized expanding Figure 1 as outlined in point 1 above. This data could instead be included as a supplemental figure. Furthermore, it is unclear exactly what is being graphed. The Materials and methods section implies they did ImageJ quantification of GFP signal and that they analyzed ~1350 – 1450 germlines, however, there do not appear to be that many points on the graph. Better graph description or clarification of data collection is necessary.

2) Grammar and punctuation throughout – including incorrect possessives (third paragraph of Introduction: "Worms small RNAs…"), missing commas, incomplete sentences (final sentence of Results section titled "Set-25 is required for the maintenance of a specific class of endo-siRNAs"), etc.

3) Figure 3 results typo – "…met-25;set-25;set-32 triple…" is incorrect (no met-25)

4) Citing primary literature – especially in Introduction. (e.g. First intro sentence should include primary citations in addition to "reviewed in Rechavi and Lev, 2017")

5) Figure 3 requires a schematic of the genetic pathway of the three HMTs. A figure here, or incorporated into the final model, would be beneficial in helping the reader understand the complex genetics at play here.

6) Discussing enrichment of SET-25-dependent endo-siRNA targets – GFP is not a newly-evolved gene. It is, understandably, a proxy for the purposes of this paper, but to define it as such significantly changes the meaning, especially with regards to the definition used throughout Figure 5.

Reviewer #3:

Rechavi and colleagues propose that, in C. elegans, H3K9me3 is important for siRNA-based silencing of newly evolved genes based on the depletion of siRNAs from a very narrow set of genes, that tend to be C. elegans-specific, in an H3K9me3-defective mutant, set-25. The data is consistent with a correlation existing between genes that are non-conserved and dependence on set-25 for siRNA production. It is not clear, however, if this applies specifically to the subset of genes for which siRNAs are dependent on set-25 or if it is a general theme for all genes targeted by the WAGO/Mutator branch of the siRNA pathway. A revised approach for analyzing the data, described below, would help to clarify this point.

It would be helpful if the Results section included a description of the gfp transgene including what endogenous elements it includes and if the endogenous elements are expressed in the germline normally.

Figure 1D. The results and figure legend sections describing Figure 1D would be clearer if there was a brief description of what is being measured.

Figure 2A. Do the open and closed circles represent two different primer pairs as in 2B?

Figure 2B. As this is a qPCR assay, results from two distinct primer sets shouldn't be directly compared without absolute quantification, which is not described as being done in the Materials and methods. It's also a bit concerning that qPCR results from completely distinct experiments – oma-1 from one experiment and gfp from a different experiment done at a different time – are compared side by side.

How many biological replicates were included int the small RNA sequencing experiment in Figure 4A? Even though the sequencing data was described previously, as this is the major data in the manuscript, additional details should be provided – strains, developmental stages, replicates, synchronization, total mapped reads in each library, library preparation method, etc.

What was the fold change cutoff for identifying the 279 genes – 1.2 fold, as was used in describing transposons?

It appears that few genes have more than a very modest reduction or increase in siRNA levels. If the downstream analysis was done with only genes that were depleted of siRNAs by 3x it might lead to more pronounced effects.

Figure 5. A general concern with the data in Figure 5, is that while the results may be significant, and when you have Ns as large as several thousand it doesn't take much difference to be significant, the effects are quite modest and as such the biological relevance is a bit questionable. For example, the difference in PATC density appears to be less than 10% – is there biological relevance to a motif being present 3.5 times compared to 3.8 times? Similarly, do the authors believe that there is biological relevance in a set of genes having a median of 3.6 vs 4 introns?

There must be thousands of C. elegans-specific genes, but only a tiny fraction are dependent on set-25 for optimal siRNA production. Thus, set-25/H3K9me3 is presumably a minor player in regulating these so-called newly evolved genes. What is the role of non-set-25 dependent siRNAs in regulating these genes?

Are the C. elegans-specific genes more likely to be targeted by the subset of set-25 dependent siRNAs than non-set-25 dependent WAGO/Mutator siRNAs?

In each of the plots in Figure 5, it would be more appropriate if the comparisons of set-25 targets was to WAGO targets as opposed to all coding genes, as the observed results could potentially apply to any random set of WAGO targets and may not distinguish the subset of genes affected by set-25. This is sort of eluded to in the manuscript but not tested.

Reviewer #4:

In this manuscript, Lev et al., demonstrate that H3K9me3 is required for inheritance of RNA silencing at some genes but not others. Specifically, H3K9me3 seems to be required for RNA silencing at foreign and newly evolved genes. This is a very intriguing idea, especially because there has been some disagreement as to when and how H3K9me3 is required for RNAi and inheritance.

Overall, this paper makes some interesting observations, especially between oma-1 and GFP, but this analysis could really be taken to the next level by more broadly by addressing how consistent this observation is genome-wide (and beyond just the set-25 mutant small RNA data). Also, the discovery of SET-25 targeting recently evolved genes is not surprising because of the significant overlap with ERGO-1, WAGO-1, and HRDE-1 target genes. Multiple publications have shown that these ERGO/WAGO/HRDE targets are overall less conserved, have fewer introns, and are spliced less efficiently, so SET-25-dependent siRNA target genes also having some of these features seems expected.

Specific suggestions

Need more/better background about the three methyltransferases. What about tissue-specific expression? Presumably all three are expressed in the germline? A chart or table summarizing the inheritance phenotypes (previously published and new) associated with each methyltransferase alone and in combination with each other would also be useful, either in Figure 1 or the supplement.

Figure 1.

While Figure 1 does a nice job summarizing the differences in inheritance between GFP and oma-1, much of this data is redundant with Kalinava, 2017, and Spraklin, 2017. To make this point more conclusively, inheritance assays of several more target genes (endogenous vs. exogenous) would be necessary to demonstrate that this is really a difference between these two types of genes, rather than something specific to oma-1 or GFP uniquely.

Figure 2.

In B, H3K9me3 levels should be measured across more genes. Do SET-25 dependent siRNA target genes have H3K9me3 at levels similar to GFP? What about siRNA target genes that are not SET-25 dependent? Is there an overall pattern? This could also be analyzed more thoroughly with the McMurchy data set.

Figure 4. In B, how are random set controls generated? Should be indicated briefly in figure legend and in Materials and methods

Figure 5A. While PATC, rnp-2, and # of introns are all technically statistically significant, the graphs are not especially convincing. The differences are quite subtle and the majority of the SET-25 genes overlap with the bulk of the protein coding genes. Indicating the mean numerically in each panel (for PATC density, fold enrichment, or # introns) could help.

How is the bootstrap control performed for panel B? No indication in the Materials and methods.

Bigger picture questions/suggestions –

Do endogenous set-25-dependent siRNA target genes also lose siRNAs in a heritable way? Could be addressed through crosses between set-25 and wild-type, followed by sequencing or rt-qPCR.

There is somewhat of a disconnect between the first few figures, looking at the relationship between the three methyltransferases, and the later figures looking only at SET-25-dependent siRNA targets. How do the SET-25-dependent siRNA target genes compare to the small RNAs dependent on the other to H3K9 methyltransferases (MET-2 and SET-32, alone or in combination)? Is it possible that other HRDE/WAGO/ERGO target genes (or specifically other newly evolved genes) are methylated by other methyltransferases?

It would be relatively easy to look at H3K9me3 across all targets of the HRDE/WAGO/ERGO pathways similarly to Figure 4B. How does the H3K9me3 signal change in the methyltransferase single, double and triple mutants?

Does loss of H3K9me3 at endogenous targets change the mRNA expression of these genes?

Minor Comments:

Figure 1. The diagram in panel A could be more clear. I generally assume a lightening bolt is indicated mutagenesis, not RNAi. Need to make it clear that RNAi is only occurring in generation 1 and subsequent generations are moved to OP50 plates.

Need to indicate in the results and figure legend what GFP transgene is being used for this assay.

How is GFP + or – defined. Materials and methods describe a calculation for fluorescence intensity but figure legend indicates% worms with GFP fluorescence. Is there a cutoff above which GFP is "on" and below which GFP is "off"?

As for the aesthetics of Figure 1, in panel D, wild-type should either be above the mutants, or ideally, they should in the same graph for easier comparison. It is a little bit confusing as to which labels go with which graphs and which wild-type data goes with which mutant data.

Labeling in Figure 2 is also confusing. Open vs. closed circles should be indicated on graph. Colors should be consistent between A and B.

Figure 3 should be moved to be part of Figure 1.

Figure 4. Are tissue enrichment boxes in C supposed to be colored? Text says fold enrichment for oocytes is 1.24 – shouldn't this be colored pale pink?

Figure 5. Is the number of stars for P value cutoffs consistent between different panels of this figure? Panel B is also made difficult to interpret because one's eye is drawn to blue dots, rather than the red bars. D and E, could be zoomed in on the lower part of the graph, to potentially allow for observation of a difference in the means (especially in D, which is supposedly different from the control).

Typos in text:

Subsection “SET-25-dependent endo-siRNAs target a unique subset of newly evolved genes”, first paragraph – "SET-25-dependnt endo-siRNAs"

Last sentence of results ends in comma.

Double period at the end of Figure 5 legend.

Figure 3—figure supplement 1A "Radom set control"

https://doi.org/10.7554/eLife.40448.028

Author response

Reviewer #1:

This paper explores interesting observations, that some particular genes respond differently to RNAi from each other. Oma-1 is a germ line gene that for some reason ignores histone methylation pathways whereas GFP does not ignore these pathways. The authors make the interesting connection that GFP is a foreign gene and therefore may be surveilled more than the endogenous oma-1. And when they compare siRNAs that are loaded into the germline inhertance pathway, they find that genes that novel to C. elegans, more likely to be foreign, are enriched. This is definitely worth publishing. I am not informaticist so I cannot judge the scatterplots in Figures 4,5 so I only weigh in on the RNAi assays for GFP fluorescence or fertility. Those results in Figures 1,2,3 are convincing. I am a little ashamed of myself for not being able to judge the statistics which is why I choose anonymity. Paper is good to me. The discussion is excellent too.

We thank the reviewer for the kind words.

One suggestion: an often overlooked feature of the GFP fusion genes that everyone in C. elegans uses is that the introns are artificial and constructed by Andy Fire back when dinosaurs roamed the Earth. If any intron is likely to be viewed as foreign it would be these introns. Deep sequencing of strains carrying GFP fusion genes should show if these introns are highly subject to siRNA production. Need to filter based on the intron exon junctions of these artificial genes. This is not a key suggestion. Optional for the authors.

That is a great suggestion, but endo-siRNAs target exons (almost exclusively), as RdRP amplification it thought to occur on the spliced mRNA (Sapetschnig et al., 2015). Nevertheless, we examined the siRNAs that are aligned to the gfp construct that we used in our study, and our analysis (Figure 4—figure supplement 2) shows no targeting of the synthetic introns of the GFP construct. Therefore this does not explain how gfp is different from oma-1.

Reviewer #2:

In this paper, the authors attempt to parse the differential roles of the three known C. elegans H3K9 histone methyltransferases (HMTs), set-32, set-25, and met-2. One of the first observations they note is the differential role set-32 plays in transgenerational epigenetic silencing of the oma-1 locus and a gfp transgene. More specifically, they find that set-32 is not required to silence oma-1 but is required to silence gfp. This suggests that there may be differential roles between silencing of genes derived endogenously and exogenously. They use ChIP-qPCR to show that this finding is not a result of differential H3K9me3 deposition at the indicated targets. The authors use the minimal data they collected regarding differences in silencing oma-1 and gfp in set-32 mutants coupled with data from previous studies (predominantly, Kalinava 2017, Towbin 2012, Lev 2017) to draw conclusions about the epistasis of the three HMTs. The focus of their paper seems to pivot from epistasis analysis of set-32 to the function and targets of set-25 where they implicate this HMT in specifically targeting newly evolved genes. They identify 151 protein-coding genes that are the specific targets of SET-25 silencing. These genes are enriched for HRDE-1 and WAGO-1 targets, H3K9me3 presence, germline expression, and most interestingly, newly evolved genes. The computational finding that set-25 specifically regulates newly evolved genes is novel and intriguing, yet, experimentally unvalidated.

The implications of this work would be significant as it represents an important and valuable synthesis of data from multiple labs that have struggled to clearly address to this point, frequently arriving at apparently contradictory results.

We thank the reviewer and agree that lots of data were collected over the years, and that it is still a challenge to connect all the dots.

Though the question is interesting, the authors fail to arrive at a convincing conclusion from their analysis. The experiments and lack rigor and the computational analysis is opaque. The very weak first half of the paper, in which the authors piece together datasets from multiple authors, coupled with the second half of the paper that lacks validated experimental evidence to support their model makes this paper unsuited for publication in eLife.

We think, and the other 3 reviewers appear to share our view, that our conclusions stand, and that this critique is not justified. We cannot refute this statement (since no specific problems are mentioned, it’s just general negativity), and therefore we choose instead to address all the specific comments, one-by-one.

Major Critiques:

1) Figures 1, 2, 3: The first three figures could be combined into one figure. The data are incomplete and, as presented, offer little novel insight.

The three figures show different things, we think lumping all these data together would be too overwhelming. Therefore, for the sake of clarity, we leave the figures separated. However, in line with this suggestion, reviewers #3 and #4 suggested that we move Figure 3 to the supplementary material, and we did so in the revised manuscript.

To build a cohesive story, including both novel data and condensation/replication of the numerous published studies referenced in text and outlined below, RNAi inheritance assays must be replicated in all single, double, and triple mutants, minimally including both an exogenous and endogenous target, and results correlated with H3K9me3 deposition at each target in each genetic mutant. This is essential to support the major claims of this paper.

Indeed it’s always important to replicate previous results, and we made sure to do it. We explicitly say when we replicate previous findings, for example see Results section “We successfully replicated the results of Kalinava et al., and came to the same conclusion ….”, and subsection “The levels of RNAi-induced H3K9me3 do not explain the gene-specific requirements of methyltransferases for heritable RNAi”: “we found, as expected (Kalinava et al., 2017)”. Now rephrased following this comment: " Using qPCR we found, as was discovered before "

We do not think every single experiment from the past must be repeated (specifically not all the experiments that we did ourselves in the past), this would not add new information and no genetic backgrounds are missing from our current analysis, that could change any conclusion. Please see the new Figure 1—figure supplement 2, where we summarize all the conclusions that were made in the past and all the new conclusions, with regard to the H3K9 methylation mutants and their effect on RNAi inheritance.

a) Figure 1B (upper and lower) – redundant with Kalinava et al., 2017. (different experimental scheme, same data and interpretation)

As the reviewer says: It’s a different experimental scheme, it’s not the same experiments as conducted by Kalinava et al. (they examined mRNA expression by real time PCR, and we examine the phenotype). Additionally, Kalinava et al. did not analyze anti-GFP RNAi, and obviously the comparison with GFP is the main point here (in the figure). Therefore this figure is not redundant.

In the previous comment the reviewer asked that we replicate all previous findings, and here he/she says we shouldn’t, so the different concerns contradict.

b) Figure 1C (lower) – Redundant with Spracklin et al., 2017.

Figure 1C is not redundant since Spracklin et al., 2017 did not analyze heritable silencing of oma-1. In contrast to heritable silencing of gfp (which was studied by Sparcklin et al), and we show that heritable silencing of oma-1 does not require set-32.

c) Figure 1D – Redundant with Figures 1B-C.

Figures 1D (Now Figure 1—figure supplement 1) shows the absolute quantification of GFP in contrast to the binary quantification (expressing / not expressing) presented in Figures 1B and C. We think this is helpful information and therefore do not find this figure redundant. The figure shows, moreover, that there’s weak inheritance in the methyltransferase mutants, which is different from what was previously suggested in the literature (Sparkling et al., 2017).

d) Figure 1A-B – Zeller et al., 2016 demonstrate that met-2;set-25 worms have reduced fertility at 20 and 25 degrees C due to accumulation of DNA damage in the germline. This will impact the total% fertile hermaphrodites used as a readout for anti-oma-1 RNAi inheritance when analyzing inheritance defects across all HMT mutants.

This difference can’t explain our conclusion since in the met-2, set-25 and set-32 mutants we do not see a reduction, on the contrary, we see an increase (due to silencing of oma-1) in the number of live (hatching) progeny. Therefore, if anything, when we say RNAi inheritance is still functional in these mutants we underestimate the effect.

e) Figure 2A – Loss of H3K9me3 at the oma-1 locus in treated and untreated worms in the met-2;set-25;set-32 mutant is redundant with Kalinava et al., 2017.

We explicitly say in the text that we replicated Kalinava et al’s results. As the reviewer noted himself/herself above, replication is valuable. In the text, we wrote: “we found, as expected (Kalinava et al., 2017)”. Now rephrased following this comment: "Using qPCR we found, as was discovered before."

Moreover, this result was necessary for the comparisons of RNAi-induced H3K9me3 levels deposited on the gfp and oma-1 genes.

Accumulation of WT levels of H3K9me3 marks across the genome is known to rely on MET-2, SET-25, and SET-32 as shown in Towbin et al., 2012, Mao et al., 2015, and Zeller et al., 2016.

We use the reduction in RNAi-induced H3K9me3 levels as a control. We did not claim that this is one of the novelties of our paper. (And in any case Towbin et al. and Zeller et al. did not examine RNAi-induced H3K9me3, and we have previously shown that this is different than non RNAi-induced K9me.).

f) Figure 2B – Demonstrates that H3K9me3 signal is higher on the gfp locus than oma-1 locus in untreated worms, yet maintains that since there is no observable fold change between H3K9me3 in treated/untreated (2A) this observation does not account for the difference in RNAi inheritance phenotypes observed in Figure 1. This assumption is discrediting a valid interpretation that increased base level H3K9me3 deposition at the gfp loci compared to oma-1 is important for the reported differences in RNAi inheritance.

The reviewer misunderstood, we did say that the higher base levels of H3K9me3 on gfp loci compared to oma-1 could be important for the reported differences in RNAi inheritance, please see subsection “Endo-siRNAs that depend on H3K9me3 methyltransferases target a distinctive subset of newly evolved genes”: “Further, we find that newly evolved genes have higher levels of H3K9me3.., Likewise, in the absence of RNAi, in wild-type animals, gfp, the newly evolved gene that we investigated, has higher levels of H3K9me3, in comparison to the well-conserved oma-1 gene“. To further clarify this in the revised manuscript we now refer to this conclusion again earlier in the paper, in subsection “The levels of RNAi-induced H3K9me3 do not explain the gene-specific requirements of methyltransferases for heritable RNAi”.

g) Figure 3 – this figure in insufficient to support the epistasis argument presented in text.

We do not agree, and moreover this observation is in accordance with new studies that show that SET-32 plays a role in the initial steps of RNAi inheritance (Kalinava et al., 2018). We added this statement to the revised manuscript in subsection “SET-32 acts upstream to MET-2 and SET-25 to support RNAi inheritance”.

It should be condensed into Figure 1.

This figure was moved to the supplementary information, now Figure 1—figure supplement 2.

Additionally, the figure contains a typo that renders it uninterpretable: either the data are switched or the y-axis is incorrect.

Thank you, the typo is now corrected.

2) Figure 4: This figure requires more rigorously controlled analysis and should focus on novel insight.

We do not agree and do not know how this critique could be used to improve the paper, as it’s not clear what would satisfy the reviewer’s definition of “novel insight” (and he/she does not specify which controls are missing). In contrast, we think this is a very useful and novel figure. To the revised figure we also added new data on set-32 (see below) and WAGO targets that further strengthen our conclusions.

a) Figure 4A – The authors fail to do due diligence in showing that the set-25 targets they identify are solely set-25 targets, and that they do not overlap with other HMT targets.

We never claimed that we saw anything that is true in general to all known HMTs. The Gu lab recently sequenced small RNAs from set-32 mutants, the data was released (December 2019) 5 months after our manuscript was submitted (so obviously we couldn’t analyze it before, and therefore there was no “due diligence” to make). In the revised manuscript we include an analysis of these newly released data (Figures 3 and 4 and their figure supplements), which further support our conclusions and show that SET-25-dependent endo-siRNAs targets are unique yet their characteristics are shared with SET-32-dependent endo-siRNAs targets (for example high H3K9me3, newly evolved genes, HRDE-1 targets..).

Performing set-32 RNA-seq is a necessary control to ensure that these are not a set of common targets simply hypersensitive to loss of a single HMT.

An analysis of set-32 RNA-seq is added to the revised manuscript, see Figures 3 and 4. We found that endo-siRNAs that depend on the two HMTs (SET-32 and SET-25) target different genes (very small overlap of 25 out of 465 genes). Importantly, however, the characteristics of these gene targets are very similar – H3K9me3 methylated, known targets of HRDE-1 and ERGO-1 argonauts, and newly evolved.

b) Figure 4B-C – It is unsurprising that targets of an H3K9me3 methyltransferase are enriched for H3K9me3.

We do not agree. It is certainly surprising that endogenous small RNAs are directly affected by H3K9 methylation. The effects of the methylations could have been indirect (as is the case sometimes, for example as we demonstrated in Lev et al. Current Biology 2017). Further, in general it’s not a valid critique to speculate if something is surprising or not.

Additionally, it logically follows that these targets are downstream of a pathway known to deposit H3K9me3 (HRDE/WAGO) and not a licensing pathway (CSR). However, the authors should discuss why these targets appear oocyte specific (shown in 4C), while simultaneously HRDE-1 target-enriched. HRDE targets germline non-expressed genes, pseudogenes, transposons, etc. The authors fail to explain or comment on this paradox.

We don’t think it’s a paradox. HRDE-1 is expressed specifically in the germline. Gene targets of HRDE-1 are not completely silenced in the germline. Indeed, we find a large overlap between HRDE-1 targets (Buckley et al., 2012) and germline expressed genes ((Buckley et al., 2012), (1000/1048 HRDE-1 targets!), See Figure 3—figure supplement 2. We elaborate on this now in subsection “H3K9me3 methyltransferases are required for the biogenesis of a specific class of endo-siRNAs”.

3) Figure 5: Experimental validation is necessary to uphold the claims of this figure, and the ultimate conclusions of this paper. Failure to experimentally and convincingly show that SET-25 is specifically required to target newly evolved genes renders this paper as nothing more than a prediction founded on unsubstantiated computational outputs lacking critical controls. These controls are especially important in datasets containing small sample sizes (151 identified SET-25 genes) which are more prone to bias.

It's not clear how to test this (that newly evolved genes are subjected to SET-25-dependent heritable RNAi). The data generated an interesting hypothesis that maybe future studies could further study. In addition, the experiments with the two reporter genes (oma-1 and gfp) support these ideas.

a) Figure 5A-B – the authors show that SET-25 targets are enriched for newly evolved genes. However, they did not perform the important converse experiment: what percentage of newly evolved genes are SET-25 targets? This analysis is necessary to isolate common identifiers between the subset of newly evolved genes targeted by SET-25 as compared to all newly evolved genes. Without this analysis, their claims into the function of SET-25 are baseless.

The converse fold change of enrichment of SET-25 targets amongst newly evolved genes is 2.57 (the same as the direct fold change enrichment value presented in the paper) and the is p-value < 0.0001.

The fold enrichment is calculated (described in the Materials and methods section) as follows:

Enrichment=ObservedExpected=fractionofgenesbelongtothei-thsetamongdifferentialexpressedSTGsfractionofgenesbelongtothei-thsetamongallthegenes

Therefore, converse enrichment calculations do not give values different from the enrichment values. The P value might change due to the randomization and changes in the size of the examined gene set.

b) The computational analysis lacks validation by experimental evidence. A computational model is most useful when it is predictive, and the authors have not demonstrated that here. There are multiple ways in which the authors can test their model, a few of which are listed below:

i) Test multiple transgenes for silencing +/- set-25

We previously tested an additional transgene for silencing with set-25 mutants and got the same results (see Sup Figure 2B in Lev et al., 2017). Others have tested different gfp transgenes and reached similar conclusions (Ashe et al., 2012; Spracklin et al., 2017).

ii) Validate the transgenerational silencing difference of some of the 151 identified "foreign" SET-25 targets as compared to additional endogenous loci.

We add new mRNA analyses that show that some of these genes are re-expressed in set-25 mutants (subsection “H3K9me3 methyltransferases are required for the biogenesis of a specific class of endo-siRNAs”).

iii) Add "foreign" introns or interfere with splice junctions of known (licensed) genes and ChIP-qPCR for H3K9me3 +/- set-25

iv) Mutate gfp transgene to include introns of highly conserved gene or introduce better splice donors/acceptors

These experiments are extremely time consuming and very technically challenging, and therefore are out of the scope of the paper (this could be a stand-alone paper). We generated multiple interesting hypotheses that would generate future work.

4) Figure 6: The model figure must include the full scope of the work. Currently, it only incorporates data collected in Figures 4-5, but does not help the reader understand the genetic interplay of the HMT pathway, and the relationship between H3K9me3 deposition and RNAi inheritance elucidated throughout the first half of the paper.

We thank the reviewer for this helpful suggestion. Since we added to the revised manuscript genome wide analyses of set-32 mutants, we incorporate the conclusion of these experiments to the Model Figure. In addition, we added a supplementary figure that explains the interplay between the HMT mutants (Figure 1—figure supplement 2).

Minor Critiques:

1) Figure 1D – these data are redundant to panels (1A-1C). Although the data do illustrate the weak RNAi inheritance responses in later generations, this can still be seen in the previous panels, and the space would be better utilized expanding figure 1 as outlined in point 1 above. This data could instead be included as a supplemental figure.

As we wrote above (this was raised in a previous comment of the same reviewer) Figure 1D (Figure 1—figure supplement 1 in the revised version) shows the absolute quantification of GFP, and therefore we do not think it is redundant. Moreover, it shows that there’s weak inheritance in the mutants (which is different from what was previously suggested in the literature). Nevertheless, we have no objection to moving it to the supplemental information, and we did so in the revised manuscript (now shown in Figure 1—figure supplement 1).

Furthermore, it is unclear exactly what is being graphed. The Materials and methods section implies they did ImageJ quantification of GFP signal and that they analyzed ~1350 – 1450 germlines, however, there do not appear to be that many points on the graph. Better graph description or clarification of data collection is necessary.

Since it’s a scatter column graph, points with the similar values “fall” on each other. We clarify this better in the legend of the revised version.

2) Grammar and punctuation throughout – including incorrect possessives (third paragraph of Introduction: "Worms small RNAs…"), missing commas, incomplete sentences (final sentence of Results section titled "Set-25 is required for the maintenance of a specific class of endo-siRNAs"), etc.

We corrected every typo that we could detect.

3) Figure 3 results typo – "…met-25;set-25;set-32 triple…" is incorrect (no met-25)

Fixed.

4) Citing primary literature – especially in Introduction. (e.g. First intro sentence should include primary citations in addition to "reviewed in Rechavi and Lev, 2017")

We don’t think this is a pattern in the paper, and believe in certain cases (where too much work must be referred to) it’s appropriate to reference reviews, but in this specific case we added additional references to primary papers.

5) Figure 3 requires a schematic of the genetic pathway of the three HMTs. A figure here, or incorporated into the final model, would be beneficial in helping the reader understand the complex genetics at play here.

We now added a summarizing supplemental figure that gives more background on the three different methyltransferase. See Figure 1—figure supplement 2B.

6) Discussing enrichment of SET-25-dependent endo-siRNA targets – GFP is not a newly-evolved gene. It is, understandably, a proxy for the purposes of this paper, but to define it as such significantly changes the meaning, especially with regards to the definition used throughout Figure 5.

We agree, indeed, we made sure that we did not say it’s a newly evolved gene anywhere in the paper. As the reviewer writes, we say that it could perhaps be seen as a proxy. We toned down these statements in the revised text.

Reviewer #3:

Rechavi and colleagues propose that, in C. elegans, H3K9me3 is important for siRNA-based silencing of newly evolved genes based on the depletion of siRNAs from a very narrow set of genes, that tend to be C. elegans-specific, in an H3K9me3-defective mutant, set-25. The data is consistent with a correlation existing between genes that are non-conserved and dependence on set-25 for siRNA production. It is not clear, however, if this applies specifically to the subset of genes for which siRNAs are dependent on set-25 or if it is a general theme for all genes targeted by the WAGO/Mutator branch of the siRNA pathway. A revised approach for analyzing the data, described below, would help to clarify this point.

We thank the reviewer for his/her excellent comments, we added the suggested new analyses to the revised manuscript.

It would be helpful if the Results section included a description of the gfp transgene including what endogenous elements it includes and if the endogenous elements are expressed in the germline normally.

Added (Figure 4—figure supplement 2E).

Figure 1D. The results and figure legend sections describing Figure 1D would be clearer if there was a brief description of what is being measured.

Added to the figure legend. (Now Figure 1—figure supplement 1).

Figure 2A. Do the open and closed circles represent two different primer pairs as in 2B?

Yes, we wrote it in the legend but further clarified in the revised text since it obviously wasn’t clear enough.

Figure 2B. As this is a qPCR assay, results from two distinct primer sets shouldn't be directly compared without absolute quantification, which is not described as being done in the Materials and methods. It's also a bit concerning that qPCR results from completely distinct experiments – oma-1 from one experiment and gfp from a different experiment done at a different time – are compared side by side.

The qPCR results shown in Figure 2B are normalized to a common control gene eft-3, as noted in the figure legends. In the revised manuscript we show data from the gfp experiment of an additional germline-expressed gene, dpy-28, that shows that this gene has low levels of H3K9me3, similarly to the oma-1 gene, and lower than gfp (Figure 2B). It is also worth noting that the gfp and oma-1 H3K9me3 ChIP experiments were done using the same strain (SX1263), grown on the same control bacteria, using the same ChIP protocol and reagents.

How many biological replicates were included int the small RNA sequencing experiment in Figure 4A? Even though the sequencing data was described previously, as this is the major data in the manuscript, additional details should be provided – strains, developmental stages, replicates, synchronization, total mapped reads in each library, library preparation method, etc.

We had 3 biological replicates (explained in the statistical analysis part of the Materials and methods). All the other details that the reviewer asked for are also added to the revised manuscript. See Information on the sequencing libraries analyzed in this paper section in the Materials and methods.

What was the fold change cutoff for identifying the 279 genes – 1.2 fold, as was used in describing transposons?

It appears that few genes have more than a very modest reduction or increase in siRNA levels. If the downstream analysis was done with only genes that were depleted of siRNAs by 3x it might lead to more pronounced effects.

In this analysis we did not use a fold change cutoff, but an adjusted p-value cutoff of the DESeq2 algorithm. Adding a fold change cutoff did not change our results (Supplementary file 3).

Figure 5. A general concern with the data in Figure 5, is that while the results may be significant, and when you have Ns as large as several thousand it doesn't take much difference to be significant, the effects are quite modest and as such the biological relevance is a bit questionable. For example, the difference in PATC density appears to be less than 10% – is there biological relevance to a motif being present 3.5 times compared to 3.8 times? Similarly, do the authors believe that there is biological relevance in a set of genes having a median of 3.6 vs 4 introns?

That’s a good point, we agree with the reviewer, and therefore we acknowledge this in the revised manuscript and detail the exact differences in means. It also worth noting that the PATC density distributions were graphed in log2 scale such that the actual differences are larger than seen (~10% in median, and ~25% in average levels). In the revised version we note in the figure that log2 scale is shown in addition to the explanation in the figure legends. We also added analysis of PATC density levels of targets of WAGO associated endo-siRNAs (now Figure 4A) and endo-siRNAs upreulgated in set-25 and set-32 mutants (Figure 4—figure supplement 1) that shows that the reduced PATC density levels are specific to endo-siRNAs that depend on SET-25 or SET-32. Moreover, in the revised paper we further explain how many complementary signals, even if some are relatively small (in average intron number or PATC levels) could together expose genes that need to be silenced (See subsection “Endo-siRNAs that depend on H3K9me3 methyltransferases target a distinctive subset of newly evolved genes” paragraph three).

There must be thousands of C. elegans-specific genes, but only a tiny fraction are dependent on set-25 for optimal siRNA production. Thus, set-25/H3K9me3 is presumably a minor player in regulating these so-called newly evolved genes. What is the role of non-set-25 dependent siRNAs in regulating these genes?

Are the C. elegans-specific genes more likely to be targeted by the subset of set-25 dependent siRNAs than non-set-25 dependent WAGO/Mutator siRNAs?

Another good point. We analyzed this in Figure 4—figure supplement 3 in the revised manuscript. We found that the ERGO-1, WAGO-1, and HRDE-1 and Mutator class small RNAs show significant enrichment for targeting newly evolved genes. However, while the fold enrichment for newly evolved genes is comparable between endo-siRNAs that depend on SET-25 or SET-32, and those that depend on other WAGO/Mutator factors, we found that the SET-25/32 dependent endo-siRNAs that target newly evolved genes do not fully overlap with the endo-siRNAs that depend on the WAGO pathways. Further, the enrichment for newly evolved genes is maintained for SET-25/32 dependent endo-siRNAs, even when we removed the targets shared with ERGO-1, WAGO-1, and HRDE-1 and Mutator from the analysis.

In each of the plots in Figure 5, it would be more appropriate if the comparisons of set-25 targets was to WAGO targets as opposed to all coding genes, as the observed results could potentially apply to any random set of WAGO targets and may not distinguish the subset of genes affected by set-25. This is sort of eluded to in the manuscript but not tested.

These comparisons are added to the revised manuscript. See Figures 3 and 4.

Reviewer #4:

[…] Specific suggestions –

Need more/better background about the three methyltransferases. What about tissue-specific expression? Presumably all three are expressed in the germline? A chart or table summarizing the inheritance phenotypes (previously published and new) associated with each methyltransferase alone and in combination with each other would also be useful, either in Figure 1 or the supplement.

Excellent idea, we now added a summarizing supplement figure that gives more background on the three different methyltransferase. See Figure 1—figure supplement 2.

Figure 1.

While Figure 1 does a nice job summarizing the differences in inheritance between GFP and oma-1, much of this data is redundant with Kalinava, 2017, and Spraklin, 2017. To make this point more conclusively, inheritance assays of several more target genes (endogenous vs. exogenous) would be necessary to demonstrate that this is really a difference between these two types of genes, rather than something specific to oma-1 or GFP uniquely.

Unfortunately there are no good RNAi inheritance assays for other endogenous germline-expressed genes (a problem the field deals with for a long while). The oma-1 assay that Alcazar et al. developed is special, perhaps because they found a redundant dominant lethal and temp sensitive germline allele that is perfect for RNAi inheritance assays.

In response to this comment we add new data (Figure 1—figure supplement 3) showing that we tried to develop new assays for examination of heritable silencing of other endogenous germline genes, fog-2 and sup-35 (since these have phenotypes that can be followed across generations). Unfortunately, we find that silencing of these genes is not inherited transgenerationally, and therefore these genes cannot be used as tools.

Figure 2. In B, H3K9me3 levels should be measured across more genes. Do SET-25 dependent siRNA target genes have H3K9me3 at levels similar to GFP? What about siRNA target genes that are not SET-25 dependent? Is there an overall pattern? This could also be analyzed more thoroughly with the McMurchy data set.

We added this analysis to the revised manuscript, see Figure 4. Our analysis of H3K9me3 levels on WAGO-associated endo-siRNA targets, which is based on the McMurchy data, shows that these targets are methylated to a lesser degree in comparison with genes targets of SET-25-dependent- and SET-32- dependent endo-siRNAs.

Figure 4. In B, how are random set controls generated? Should be indicated briefly in figure legend and in Materials and methods

This was written in the Materials and methods, we added this now to the legends as well.

Figure 5A. While PATC, rnp-2, and # of introns are all technically statistically significant, the graphs are not especially convincing. The differences are quite subtle and the majority of the SET-25 genes overlap with the bulk of the protein coding genes. Indicating the mean numerically in each panel (for PATC density, fold enrichment, or # introns) could help.

How is the bootstrap control performed for panel B? No indication in the Materials and methods.

We added the mean to the graph panels and text. It is also important to note that the PATC data is presented in log2 scale such that the actual differences are larger than what they may seem in the graph (~10% in median and ~25% on average), and that such reduction in PATC density was not found in the new controls we added (Figure 4A and Figure 4—figure supplement 1). We also explain now in the text how the cumulative effect of small differences in intron numbers and PATC might allow recognition of genes that need to be silenced (See subsection “Endo-siRNAs that depend on H3K9me3 methyltransferases target a distinctive subset of newly evolved genes” paragraph four). We also explain how the bootstrap control is performed (previously it was in the Materials and methods, now we add it also to the legends).

Bigger picture questions/suggestions –

Do endogenous set-25-dependent siRNA target genes also lose siRNAs in a heritable way? Could be addressed through crosses between set-25 and wild-type, followed by sequencing or rt-qPCR.

We think these small RNAs are heritable since they are bound by Argonauts involved in small RNA inheritance such as HRDE-1 and WAGO-1. This could be interesting to further study in the future.

There is somewhat of a disconnect between the first few figures, looking at the relationship between the three methyltransferases, and the later figures looking only at SET-25-dependent siRNA targets. How do the SET-25-dependent siRNA target genes compare to the small RNAs dependent on the other to H3K9 methyltransferases (MET-2 and SET-32, alone or in combination)? Is it possible that other HRDE/WAGO/ERGO target genes (or specifically other newly evolved genes) are methylated by other methyltransferases?

We added to the revised manuscript an analysis of small RNAs from set-32 mutants, and we compare it to set-25 mutants (Figures 3 and 4, and their figure supplements, subsection “H3K9me3 methyltransferases are required for the biogenesis of a specific class of endo-siRNAs”). We described the changes in small RNAs in met-2 mutants in Lev et al. Current Biology 2017, and therefore did not go into it again, also because we think (as we described in that paper), that the effects of MET-2 are often indirect.

It would be relatively easy to look at H3K9me3 across all targets of the HRDE/WAGO/ERGO pathways similarly to Figure 4B.

We added these analyses to the revised text, see Figure 4. It shows that these WAGO targets are marked by H3K9me3, but targets of SET-25 or SET-32 dependent endo-siRNAs have higher levels of H3K9me3.

How does the H3K9me3 signal change in the methyltransferase single, double and triple mutants?

By analyzing the recent Kalinava data set (Kalinava et al., 2018), we tested how the H3K9me3 levels on SET-32-dependent endo-siRNA target genes change in set-32 mutants and found, as might be expected, that gene targets of SET-32-dependent endo-siRNAs have reduced levels of H3K9me3 in set-32 mutants (Figure 3—figure supplement 1, discussed in paragraph two of subsection “H3K9me3 methyltransferases are required for the biogenesis of a specific class of endo-siRNAs”).

Does loss of H3K9me3 at endogenous targets change the mRNA expression of these genes?

We added this analysis to the revise text. We find that targets of SET-25-dependent endo-siRNAs are significantly enriched with genes that are re-expressed in set-25 mutants (Klosin et al., 2017) see paragraph two of subsection “H3K9me3 methyltransferases are required for the biogenesis of a specific class of endo-siRNAs”.

Minor Comments:

Figure 1. The diagram in panel A could be more clear. I generally assume a lightening bolt is indicated mutagenesis, not RNAi. Need to make it clear that RNAi is only occurring in generation 1 and subsequent generations are moved to OP50 plates.

Need to indicate in the results and figure legend what GFP transgene is being used for this assay.

How is GFP + or – defined. Materials and methods describe a calculation for fluorescence intensity but figure legend indicates% worms with GFP fluorescence. Is there a cutoff above which GFP is "on" and below which GFP is "off"?

Thank you, all these clarifications were made in the Figure, figure legends and Materials and methods.

As for the aesthetics of Figure 1, in panel D, wild-type should either be above the mutants, or ideally, they should in the same graph for easier comparison. It is a little bit confusing as to which labels go with which graphs and which wild-type data goes with which mutant data.

We now place them one above the other in Figure 1—figure supplement 1.

Labeling in Figure 2 is also confusing. Open vs. closed circles should be indicated on graph. Colors should be consistent between A and B.

Fixed.

Figure 3 should be moved to be part of Figure 1.

We moved this figure to the Supporting materials, we think it was too overwhelming. Now Figure 1—figure supplement 2.

Figure 4. Are tissue enrichment boxes in C supposed to be colored? Text says fold enrichment for oocytes is 1.24 – shouldn't this be colored pale pink?

Fixed.

Figure 5. Is the number of stars for P value cutoffs consistent between different panels of this figure?

Definitely, yes.

Panel B is also made difficult to interpret because one's eye is drawn to blue dots, rather than the red bars. D and E, could be zoomed in on the lower part of the graph, to potentially allow for observation of a difference in the means (especially in D, which is supposedly different from the control).

Fixed. We removed the dots, and with regards to Figure D, we show the median values in the figures.

Typos in text:

Subsection “SET-25-dependent endo-siRNAs target a unique subset of newly evolved genes”, first paragraph – "SET-25-dependnt endo-siRNAs"

Last sentence of results ends in comma.

Double period at the end of Figure 5 legend.

Figure 3—figure supplement 1A "Radom set control"

Thank you. All fixed.

https://doi.org/10.7554/eLife.40448.029

Article and author information

Author details

  1. Itamar Lev

    Department of Neurobiology, Wise Faculty of Life Sciences & Sagol School of Neuroscience, Tel Aviv University, Tel Aviv, Israel
    Contribution
    Conceptualization, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Project administration, Writing—review and editing
    For correspondence
    itamai.et@gmail.com
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-9100-5100
  2. Hila Gingold

    Department of Neurobiology, Wise Faculty of Life Sciences & Sagol School of Neuroscience, Tel Aviv University, Tel Aviv, Israel
    Contribution
    Conceptualization, Software, Formal analysis, Investigation, Methodology, Writing—original draft, Project administration, Writing—review and editing
    Competing interests
    No competing interests declared
  3. Oded Rechavi

    Department of Neurobiology, Wise Faculty of Life Sciences & Sagol School of Neuroscience, Tel Aviv University, Tel Aviv, Israel
    Contribution
    Conceptualization, Supervision, Funding acquisition, Writing—original draft, Project administration, Writing—review and editing
    For correspondence
    odedrechavi@gmail.com
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-6172-3024

Funding

Israel Science Foundation (1339/17)

  • Itamar Lev
  • Hila Gingold
  • Oded Rechavi

European Research Council (335624)

  • Itamar Lev
  • Hila Gingold
  • Oded Rechavi

Adelis Foundation (01430001000)

  • Oded Rechavi

Paul G. Allen Family Foundation

  • Oded Rechavi

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

Acknowledgements

We thank all the Rechavi lab members for the helpful comments and fruitful discussions. We thank Yael Mor for the fruitful discussions and asistance with formulating the newly evolved genes hypothesis. Some strains were provided by the CGC, which is funded by NIH Office of Research Infrastructure Programs (P40 OD010440). We thank Yosef Shiloh, Yael Ziv, for their assistance and advice. Special thanks to Dror Cohen for the illustrations that he contributed. This work was supported by the ERC (grant #335624) and the Israel Science Foundation (grant #1339/17) and OR gratefully acknowledges the support of the Allen Discovery Center of the Paul G Allen Frontiers Group and the support of the Adelis foundation (no. 01430001000).

Senior and Reviewing Editor

  1. Patricia J Wittkopp, University of Michigan, United States

Publication history

  1. Received: July 26, 2018
  2. Accepted: February 26, 2019
  3. Version of Record published: March 14, 2019 (version 1)

Copyright

© 2019, Lev 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.

Metrics

  • 1,108
    Page views
  • 153
    Downloads
  • 1
    Citations

Article citation count generated by polling the highest count across the following sources: PubMed Central, Crossref, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Chromosomes and Gene Expression
    2. Developmental Biology
    Albert Tsai et al.
    Research Advance
    1. Cell Biology
    2. Chromosomes and Gene Expression
    Konstadinos Moissoglu et al.
    Research Article