Set2 methyltransferase facilitates cell cycle progression by maintaining transcriptional fidelity

Abstract

Methylation of histone H3 lysine 36 (H3K36me) by yeast Set2 is critical for the maintenance of chromatin structure and transcriptional fidelity. However, we do not know the full range of Set2/H3K36me functions or the scope of mechanisms that regulate Set2-dependent H3K36 methylation. Here, we show that the APC/C^CDC20 complex regulates Set2 protein abundance during the cell cycle. Significantly, absence of Set2-mediated H3K36me causes a loss of cell cycle control and pronounced defects in the transcriptional fidelity of cell cycle regulatory genes, a class of genes that are generally long, hence highly dependent on Set2/H3K36me for their transcriptional fidelity. Because APC/C also controls human SETD2, and SETD2 likewise regulates cell cycle progression, our data imply an evolutionarily conserved cell cycle function for Set2/SETD2 that may explain why recurrent mutations of SETD2 contribute to human disease.

INTRODUCTION

Histone post-translational modifications (PTMs), including acetylation, methylation, phosphorylation, and ubiquitylation, are major contributors to chromatin dynamics and the spatio-temporal regulation of DNA-dependent transactions such as transcription, replication and DNA repair (1). These modifications are deposited (written), interpreted (read), and removed (erased) by epigenetic machinery (2–4), which is often recurrently mutated or overexpressed in human diseases (5). Numerous PTMs occur across the unstructured tail and globular domains of histones (1), yet it is poorly understood how these PTMs (singly or in combination) contribute to chromatin structure and function.

Histone lysine methylation plays a significant role in gene transcription (6). For example, while methylation of histone H3 at lysine 4 (H3K4me) is linked to active transcription and marks active promoters and enhancers, methylation of H3 at lysine 36 (H3K36me) contributes to transcription elongation and marks the transcribed regions of genes (7,8). H3K36 is methylated co-transcriptionally by Set2 and functions, partly, to maintain chromatin structure and prevent inappropriate transcription from cryptic promoters during transcription elongation (9–11). This function of H3K36me is dependent on the recruitment/activation of the Rpd3S histone deacetylase complex (9–11) and inhibition of histone exchange (12,13). Thus, one consequence of disrupting Set2 function is loss of transcriptional fidelity, which leads to reduced life span in Saccharomyces cerevisiae and Caenorhabditis elegans (14,15). Cryptic transcription can occur in the sense direction, and cryptic transcripts can potentially be translated (16). As well, cryptic transcription can occur in the antisense direction, producing ‘Set2-repressed antisense transcripts’ (17). Intriguingly, H3K36me also dictates the choice between non-homologous end joining and homologous recombination pathways in DNA double-strand break repair (18–22). Consistent with this function of H3K36me in regulating genome stability, the responsible enzymes that methylate human H3K36 (e.g. NSD2, which mediates H3K36me2, and SETD2, which mediates H3K36me3) are overexpressed or recurrently mutated in a variety of cancers such as renal, breast, and hematological malignancies (5,15).

Although a requirement for Set2/H3K36me is well established in transcriptional fidelity, and, more recently, in nutrient stress and carbon starvation transcriptional programs (23,24), we sought to determine exactly why such regulation exists and whether Set2/H3K36me might also regulate transcriptional programs that are precisely timed and highly tuned, such as the cell cycle transcriptional program. In this report, we identify a function for Set2-mediated H3K36me in cell cycle control. We show that absence of Set2/H3K36me disrupts cell cycle progression and that the Anaphase Promoting Complex/Cyclosome (APC/C) complex degrades Set2 in a cell cycle-dependent manner. Moreover, deletion of SET2 causes increased antisense cryptic transcription of cell cycle-regulated genes, and this antisense transcription is correlated with mis-regulated sense transcription. Overall, our results suggest that suppression of cryptic transcription by Set2/H3K36me is a general mechanism to maintain the fidelity of highly tuned and highly regulated transcription programs. Because we found that human SETD2 is similarly cell cycle-regulated in an APC-dependent manner, our results also suggest a conserved and basic function for H3K36me in cell cycle control.

MATERIALS AND METHODS

Strains and plasmids

Unless otherwise indicated, all strains are in BY4741 background. bub1-kd (SBY11006) and its corresponding wild-type were a gift from Sue Biggins (Fred Hutchison); the GAL-CDC20 strain was provided by David Morgan (UCSF). cdc4-1 and cdc53-1 were gifts from Jennifer Benanti (UMASSMED, Worcester). An H3–H4 wild-type shuffle strain (FY2162) was a gift from Fred Winston (Harvard University) and H3–H4 (K36A) was a gift from Jerry Workman (Stowers Institute). SET2 was deleted by gene replacement using the PCR toolkit (SET2::NAT). The two D-Boxes in Set2 protein were mutated using site directed mutagenesis using Lightning multisite mutagenesis kit. D-box mutants were integrated at the endogenous SET2 locus using two-step integration method (25). All yeast strains and their genotypes are listed in Supplementary Table S2.

Immunoblots

Generally, yeast strains were grown to an A₆₀₀ of 0.6–0.8, and whole cell extracts were prepared by the SUMEB protocol (http://research.fhcrc.org/gottschling/en/protocols/yeast-protocols/protein-prep.html). Protein extracts were applied to either 8% (Set2, Clb2) or 15% (H3K36me2/3 and H3) SDS-PAGE gels. Transfers were performed at 45 mA/gel for 1.5 h using a Hoeffer semi-dry apparatus. Immunoblotting was performed at 4°C (overnight) using the following antibody dilutions: H3K36me3, 1:1000 (ab9050); H3K36me2, 1:1000 (Active Motif 39255); H3, 1:5000 (Epicypher 13-0001); G6PDH, 1:20,000 (Sigma); Clb2, 1:1000 (Santa Cruz sc-9071); HA, 1:1000 (UNC-Antibody Core HA-C5).

Yeast spotting assays

Indicated strains were grown overnight and 5-fold serial dilutions (starting A₆₀₀ 0.25) of these strains were spotted either on minimal medium lacking uracil or on medium containing benomyl (mitotic inhibitor).

Yeast cell cycle synchronization

Yeast cell cycle synchronization was performed according to Keogh et al. (9). Briefly, to arrest cells in G1, α-factor was added to WT and set2Δ cells (A₆₀₀ ∼0.2–0.3) for 3 h. Microscopy and cell cycle markers such as Clb2 confirmed cell cycle arrest. The cells were washed (twice) and re-suspended in fresh medium, and samples (either for protein or RNA extraction) were taken at indicated time points. For nocodazole arrest, cells were grown to A₆₀₀ = 0.3 in medium containing DMSO, and nocodazole was then added to 30 μg/ml for 3 h. Arrest was monitored by the appearance of large budded cells.

Flow Cytometry

Overnight wild-type (BY4741) and set2Δ cultures were inoculated at an A₆₀₀ of 0.1 and allowed to grow to no >0.25 absorbance units. α-factor was added to cultures at a concentration of 25 nM, and cells were arrested in G1 for 3 h at 30°C. The cultures were then washed three times in sterile water and released into YPD growth medium. Aliquots of cells were collected and fixed (for at least 24 h at 4°C) in 70% cold ethanol at specified time intervals before proceeding to propidium iodide (PI) staining. Fixed cells were washed (1×) in 50 mM citrate buffer, pH 7.5. After washing, cells were incubated with 250 μg/ml RNAse A at 50°C for 2 h. Proteinase K was then added to 1 mg/ml, and incubation was continued for 2 h at 50°C. Cells were washed once with citrate buffer and resuspended at a density of 2 × 10⁶ cells/ml. PI was added to a final concentration of 8 μg/ml, followed by 1–2 h incubation at room temperature, after which 10 000 events were recorded on Accuri C6 flow cytometer.

RNA extraction and RT-PCR

Total RNA was prepared from 10 A₆₀₀ units of mid-log phase cells (WT and set2Δ) using hot acid phenol-chloroform extraction followed by ethanol precipitation. 10 μg of crude RNA was treated with DNAse I (Promega) followed by purification using an RNeasy minikit (Qiagen). 500 ng–1 μg of total RNA was used to synthesize cDNA with the SuperScript II first strand synthesis system (Life Technologies); the cDNA was diluted 1:10–1:50 prior to PCR amplification. For cDNA synthesis, either random hexamer primers or gene-specific forward primers were used to detect the sense and antisense RNA, respectively. Primer sequences are available upon request.

Periodic transcription and gene length analysis

A list of Saccharomyces cerevisiae periodic genes was downloaded from Cyclebase (http://www.cyclebase.org/CyclebaseSearch), and the top 800 periodic genes were selected for subsequent analysis. We defined the periodic and non-periodic genes by selecting the top 800 and the last 1000 genes, respectively. For gene annotations for the NET-seq analysis, we relied on the annotation from Xu et al. (2009). The list of genes that give rise to cryptic transcripts in set2Δ was obtained from Lickwar et al. 2009. Statistical significance for the number of genes between different classes (Figure 1 and Supplementary Figure S1) was calculated by hyper-geometric tests, with 5460 genes serving as the total number of genes expressed in a mitotic cell (26). To identify the functionally enriched classes, the candidate list of genes was subjected to ClueGO analysis in Cytoscape (version 3.4) (27), and functional classes with P value less than 0.05 are shown in Figure 1A. For Supplementary Figure S1, DAVID (https://david.ncifcrf.gov) (28) was used to analyze the functional enrichment of either 244 genes (from (14)) or 721 genes (from (29)) that commonly give rise to cryptic transcripts and classes that show a functional enrichment with P-value <0.05 (Bonferroni corrected) is shown. Inbuilt R-codes (version 3.3.2) were used to determine the length distribution of periodic genes versus a set of random genes.

Figure 1.

Long genes prone to cryptic transcripts are associated with cell cycle related functions. (A) Network analysis, following pathway enrichment analysis, of genes that give rise to cryptic transcript from (32) showing the ontology relationships between the pathways. Genes that give rise to cryptic transcripts (429 genes from Lickwar et al. (32)) in set2Δ cells reveal an enrichment of these genes with the cell cycle, mitosis, DNA replication and cell polarity. The direction of the arrows shows the ontogeny of related functional categories. (B) A Venn diagram comparing genes that are periodic (top 800 from Cyclebase) to those that give rise to cryptic transcripts in set2Δ cells reveal a significant overlap (p = 0.00548). (C) Venn diagram showing that genes which give rise to cryptic transcription in set2Δ cells also tend to be long (P < 1 × 10⁻²⁰). (D) Plot showing the relationship between long genes (800 random genes) and periodic genes (top 800 from Cyclebase) reveal periodic genes tend to be longer than expected by chance (P< 2.531 × 10⁻⁵ Kolmogorov–Smirnov test). (E) and (F) RT-qPCR analysis of sense and antisense transcripts originating from the SWE1 (SUT201) and CDC28 genes (SUT_FS0074), demonstrating the reduced sense transcript levels of SWE1 and CDC28 in set2Δ cells.

Elongating RNAPII occupancy analysis

We obtained NET-seq data for WT and set2Δ from (30) and converted the data into genome format for efficient retrieval (31). Around each position from –200 bp to +200 bp from TSS and TTS, we used a 50 bp window and recorded the ratio of total sense reads to antisense reads. The average ratio for each position was then calculated by averaging over all periodic and non-periodic genes for WT and set2Δ datasets. The P-values for each position were then calculated by comparing the distributions of sense-antisense ratios using the Kolmogorov–Smirnov test.

Mammalian cell lines and cell cycle analysis

Human renal cell carcinoma 786-O and U2OS cells were acquired from the American Type Culture Collection (Manassas, VA, USA). Cells were cultured in Dulbeccos modified Eagle's medium (DMEM, Gibco/Life Technologies, Carlsbad, CA, USA) supplemented with 10% FBS (Gemini Bio-Products, West Sacramento, CA, USA), non-essential amino acids, l-glutamine, penicillin, and streptomycin. All cultures were maintained at 37°C in 5% CO₂. SETD2-knockout (KO) cells and a genetic complementation with a H3K36me3 active form of SETD2 was previously described (42). For Nocodazole arrest, U2OS cells at 60–70% confluency were synchronized to mitotic phase using addition of 200 ng/ml of nocodazole for 16 h. The mitotic cells were collected by shake off, washed three times with warm media and replated. Cells were trypsinized and collected at each intended time points. Cells were lysed in NETN buffer (150 mM NaCl, 0.5 mM EDTA, 50 mM Tris, pH 7.5, 0.5% NP-40) supplemented with protease and phosphatase inhibitors on ice for 20 min. Immunoblots were performed using the following antibodies at the indicated dilutions: Fzr1 and SETD2 antibodies were purchased form Abcam and used at dilutions of 1:300. CycE (1:5000) and p-Histone H3 (1:1000) antibodies were from Cell Signaling Technologies. Cyclin A (1:5000), Cyclin B (1:10 000) and Ran (1:8000) antibodies were obtained from Santa Cruz Biotechnology Inc. All antibodies were diluted in 5% milk made in phosphate buffered saline. For Hydroxyurea (HU) synchronization, 786-O and SETD2 knockout and rescue cells were incubated with 1 μM HU for 12 h and released by washing with PBS. The cells were collected at the indicated time, fixed with 70% cold ethanol and stored in –20°C for overnight. Cells were then stained with propidium iodide (PI, 50 μg/ml) for determination of total DNA content and analyzed by flow cytometry. At least 20 000 cells were acquired on Beckman Coulter CyAn ADP (Beckman Coulter, Indianapolis, IN, USA) using 488 nm excitation laser and data were analyzed with FlowJo software (TreeStar, Ashland, OR, USA).

RESULTS

Relationship between gene length, Set2-suppression of cryptic transcription, and cell cycle regulation

During transcription elongation, Set2/H3K36me maintains chromatin structure and prevents cryptic transcription in the wake of RNAPII passage (9–11,13). Prevention of cryptic transcription by Set2 has been linked to lifespan control in multiple organisms and the nutrient stress response in yeast. To determine what other cellular processes might be dependent on Set2 suppression of cryptic transcription, we examined a comprehensive dataset of cryptic transcripts, and their corresponding genes, arising from the absence of Set2 (32). A functional enrichment analysis of this dataset revealed that a significant proportion of the cryptic transcripts in SET2-deleted (set2Δ) cells (429 genes identified with high confidence) was functionally enriched for cell cycle-associated processes, most notably ‘cell cycle’, ‘cell division’, ‘mitotic cell cycle’, and ‘cytoskeletal reorganization’ (Figure 1A and Supplementary Table S1). Other significantly enriched pathways were ‘regulation of cell signaling’ and ‘response to stress’, pathways both consistent with our previous findings (24) and intimately tied to cell cycle progression.

To further understand how these cryptic transcripts are related to the cell cycle, we compared the number of genes that give rise to cryptic transcripts in set2Δ cells with the number of genes that are transcribed in a highly periodic manner (i.e. the 800 top ranked periodically expressed genes from Cyclebase (http://www.cyclebase.org/CyclebaseSearch) (33). This comparison showed a significant overlap (P = 0.00548) between genes that give rise to cryptic transcripts and genes that are periodic (Figure 1B). Recently, Sen et al. showed that loss of H3K36me in S. cerevisiae and C. elegans abrogated transcriptional fidelity and resulted in decreased lifespan. In agreement with our finding that cryptic transcription is over-represented in periodic genes, a functional enrichment analysis of 244 genes identified by Sen et al. as giving rise to cryptic transcripts during aging also showed an over-representation of cell cycle-associated genes (Supplementary Figure S1A). In addition, a significant proportion of these 244 genes was also identified as being periodically expressed, as defined by Santos et al. (33) (Supplementary Figure S1B). Finally, from a dataset of Pelechano and Steinmetz (29), we again observed a significant over-representation of cell cycle and cell division genes that showed cryptic transcription after loss of SET2 (Supplementary Figure S1C).

Consistent with a study by the Workman group (34), the analyses above also revealed that genes which give rise to cryptic transcripts are, on average, longer than expected by random chance (P = 1 × 10⁻²⁰) (Figure 1C). Because a high proportion of genes associated with cryptic transcription is represented by cell cycle-related functions, and these genes also tend to be long, we asked whether periodic genes are themselves longer than expected by chance. Indeed, we found that the top 800 periodic genes are longer than predicted by random chance (P < 2.531 × 10⁻⁵) (Figure 1D). Interestingly, mammalian genes involved in the G2-M transition also tend to be longer than expected by chance (35). Collectively, these results indicate that cell cycle-regulated genes tend to be long and, thereby, more dependent than other gene classes on the Set2/Rpd3 pathway to prevent cryptic transcription.

Set2 is required to maintain transcriptional fidelity during cell cycle progression

Because cell cycle genes appear to be highly susceptible to cryptic transcription, we next asked whether Set2/Rpd3S might contribute to regulation of periodic genes by suppressing cryptic antisense transcription. Significantly, several studies have demonstrated the potential of antisense transcripts to directly cause transcriptional interference, especially when an antisense transcript overlaps a sense transcription start site (TSS) (24,36–38). Therefore, we identified key cell cycle-regulated genes that exhibited a stable unannotated transcript (SUT) overlapping the start site of the sense promoter, and this list included key cell cycle regulatory genes such as FAR1, SWE1 and CDC28. Notably, an overlapping antisense transcript for CDC28 is critical in regulating CDC28 transcription during osmotic stress response (37). We used strand-specific real-time quantitative PCR (RT-qPCR) to measure the sense/antisense transcript abundance of CDC28 and SWE1 in asynchronously growing WT and set2Δ cells. These measurements revealed that the CDC28 and SWE1 sense transcripts were down-regulated in set2Δ cells, whereas simultaneously, their corresponding antisense transcripts were upregulated by more than two-fold (Figure 1E and F).

Given the above findings, we next extended our analysis to measure the sense and antisense transcript abundance of periodic genes across the cell cycle in WT and set2Δ cells. We selected FAR1 and SWE1 for these examinations, as these two genes are dynamically regulated across the cell cycle and possess annotated SUTs that extend to the TSS (Figure 2A and D). In agreement with previous studies, FAR1 sense transcription was down-regulated upon G1 arrest and release (Figure 2B) (39). In addition, the overlapping FAR1 SUT (SUT204) was cyclically regulated across the cell cycle, with its level being highest when the FAR1 sense transcript was at its lowest level (Figure 2C). In set2Δ cells, however, this pattern was significantly altered with FAR1 sense and antisense transcript levels being higher compared with the levels observed in WT cells (Figure 2B and C).

Figure 2.

The transcriptional fidelity of cell cycle-regulated genes is maintained by Set2. (A) Diagrammatic representation of the FAR1 locus and its corresponding antisense stable un-annotated transcript (SUT204). (B) qRT-PCR showing the absence of SET2 results in mis-regulation of the levels of both sense and antisense FAR1 transcripts. (C) FAR1 antisense transcript, SUT204, is increased in set2Δ cells at all points across the cell cycle, as assessed by qRT-PCR. (D) Schematic representation of the genomic locus of SWE1 and its corresponding antisense transcript, SUT201. (E) Loss of SET2 results in the mis-regulation in the levels of SWE1 sense transcript (compare the levels of SWE1 transcript at 30 and 45 min between WT and set2Δ). (F) qRT-PCR of SWE1 antisense transcript SUT201, shows altered expression kinetics in set2Δ cells. (G) Schematic representation of the FAR1 locus with an insertion of a 2 kb KANMX cassette to displace the SUT204 from the FAR1 3′ end (H) qRT-PCR to detect the levels of FAR1 transcript in a strain that lacks the 2 kb KANMX cassette and showing a similar down-regulation of FAR1 as in Figure 2B. (I) levels of SUT204 in a strain containing the 2 kb KANMX cassette showing partial abrogation of the transcript. The data in all the above experiments are represented as the standard deviations of three biological replicates with three technical replicates in individual experiments. The statistical significance was calculated using paired t test (* represents P value <0.01).

We next examined the cell cycle transcript profile of SWE1 and its antisense transcript SUT201 (Figure 2D). As expected, the SWE1 transcript level was up-regulated upon entry into S-phase (30 min), after which it returned to baseline level (Figure 2E) (40). Similar to FAR1, we found that the antisense SUT201 transcript abundance was also periodically expressed and anti-correlated with the SWE1 levels across the cell cycle (Figure 2F). In contrast, SWE1 levels were significantly reduced during S-phase in set2Δ cells, and this reduced level persisted after cells exited S-phase, compared with WT cells in which the SWE1 transcript was down-regulation post S-phase. However, and somewhat similar to SUT204, the levels of SUT201 showed aberrant regulation across the cell cycle in set2Δ cells (Figure 2F). Interestingly, the anti-correlation found with these sense and antisense transcripts were largely restricted to the times of peak expression of these genes. Together, these results document a sense/antisense anti-correlation of periodically expressed genes in WT cells, a finding that agrees with others (41). They also suggest that Set2-dependent suppression of antisense transcripts has an important function in fine-tuning the levels of sense transcripts across different phases of cell cycle.

We next sought determine the effect of uncoupling antisense transcription from sense transcription across the cell cycle for one of the aforementioned genes. To achieve uncoupling, we inserted a ∼2 kb KANMX cassette into the 3′-end of the FAR1 gene, thereby physically displacing SUT204 away from the FAR1 TSS (Figure 2G). Without the cassette insertion, and as observed in Figure 2B, FAR1 sense transcript level decreased upon entry into the S-phase, then became lower in the S- and G2-phase, and began peaking in the M-phase (Figure 2H). Significantly, insertion of the KANMX cassette caused a drastic increase in the expression level of the FAR1 sense transcript at the 60 min time interval; this increase was subsequently down-regulated to WT levels by 90 min. Across the cell cycle, the displaced FAR1 antisense transcript, SUT204, was significantly decreased compared with the WT control transcript, suggesting that this antisense transcript is key to regulation of the FAR1 sense transcript (Figure 2I).

Because antisense SUTs can potentially affect sense transcription of periodically expressed genes, we asked whether antisense cryptic unstable transcripts (CUTs), that are also regulated by the Set2/Rpd3 pathway (30), might also confer a similar fate upon cell cycle-regulated gene sense transcription. Significantly, Castelnuovo et al. reported that specific CUTs, when not terminated properly by Nrd1/Nab3/Sen1-dependent termination pathway, harbor a similar capability to suppress sense transcription (42). To explore if antisense transcripts are altering sense transcription, we analyzed NET-seq data generated from WT, set2Δ, and rco1Δ cells (30) and determined the occurrence and effect of antisense CUTs on transcription of periodic and non-periodic genes. Consistent with expectations, the absence of Set2 or Rco1 led to genome-wide increases in CUT formation at the 5′- and 3′-ends of genes (Supplementary Figure S2 and data not shown). However, the increases in CUT formation occurred irrespective of whether or not the genes were expressed periodically, revealing a broad role for Set2 in suppressing CUT formation.

We further examined the NET-seq data to assess how the absence of Set2 specifically affects CUT formation from periodically expressed genes, and whether changes in CUT formation are associated with decreased sense transcription. The NET-seq data revealed that transcription of many important cell cycle genes, including CDC5, CDC20 and CLB2, was significantly down-regulated in a set2Δ strain compared with the WT; however, these decreases did not correlate with increases in CUT formation (Supplementary Figure S3). Conversely, DBF2, a key periodically regulated gene required for mitotic exit, showed a striking increase in its 3′ antisense transcript (CUT613) that correlated with a decrease in sense transcription of DBF2, suggesting that this antisense transcript might be involved in regulating sense transcription (Figure 3A and B). To investigate the possibility of antisense-mediated regulation of DBF2, we measured its sense transcript and CUT across the cell cycle (Figure 3C and D). Remarkably, CUT613 was up-regulated in set2Δ cells as reported (30), and its up-regulation occurred in G2/M, which is when DBF2 is normally expressed. Significantly, increased CUT formation in set2Δ cells was associated with decreased sense production at all time points, as assessed by strand-specific qRT-PCR (Figure 3C and D). While these results suggest that the decrease in DBF2 might be mediated through transcriptional interference of increased CUT613 expression, we cannot rule out the possibility that DBF2 expression functions to suppress CUT613 expression. Nonetheless, these findings further support a function for Set2 in reinforcing the transcriptional fidelity of genes that are key to cell cycle regulation.

Figure 3.

Set2 is required to repress cryptic unstable transcripts at the cell cycle-regulated DBF2 locus. (A) Schematics of DBF2 gene organization alongside its neighboring gene (DRN1) and the cognate antisense transcripts CUT613 and CUT614, respectively. (B) Genome browser shot of the sense and antisense transcripts originating from the WT and set2Δ cells. (C) and (D) WT or set2Δ cells were arrested with α-factor, then released into fresh medium, after which samples were collected at indicated time points for strand-specific qRT-PCR. Results show that as DBF2 levels increase in G2/M, its corresponding antisense CUT613 is also detectable. Absence of Set2 results in significant increase in CUT613 production that correlates with sense decrease, suggesting that CUT613 antisense transcripts can interfere with sense transcription. The data in are represented as the standard deviations of three biological replicates with three technical replicates in individual experiments. Statistical significance was calculated using paired t test (* represents P value <0.01).

Set2 and H3K36 methylation are required for timely progression through the cell cycle

Having established that Set2 is required for proper expression of cell cycle-regulated genes, we next asked whether deletion of SET2 would lead to cell cycle progression defects. To address this question, we arrested WT and set2Δ cells with α-factor, released them into fresh medium, and collected cells at the indicated time points for flow cytometry (Figure 4A). Consistent with our results showing disruption of cell cycle regulated gene transcription upon SET2 loss, set2Δ cells displayed a substantial delay (∼15 min) in the release from G1 into S-phase (Figure 4B). In addition to this delayed entry into S-phase, our flow cytometry studies revealed that the number of cells in the S-phase was lower in set2Δ cells compared with WT, suggestive of a faster progression through S phase (Figure 4C). To confirm that absence of Set2 leads to faster S-phase progression, we arrested WT and set2Δ cells for 2 h with 200 mM hydroxyurea (HU), washed the cells and released them into fresh medium, and collected time points. In this experiment, set2Δ cells exited S-phase 15 min faster than the WT cells (Supplementary Figure S5), further confirming a function for Set2 in different phases of the cell cycle. In agreement with these findings, results from Biswas et al. also showed that Set2 inhibits DNA replication (43). Further support for this was also recently provided by Pai et al., who showed that Set2 in S. pombe is required for efficient DNA replication by regulating the MBF transcriptional program (44). However, the mechanistic basis by which Set2 controls S-phase progression and proper checkpoint activation remains unknown. Finally, although G1- and S-phase progression were perturbed, we did not observe any significant changes in the fraction of cells in the G2/M-phase of cell cycle (Figure 4D). To determine whether the set2Δ cell cycle delay was a catalysis-dependent function directed toward the H3K36 residue, we examined a strain in which H3K36 was mutated to alanine (H3K36A), hence, K36 could not be methylated. Like the absence of Set2, loss of H3K36 methylation in the H3K36A strain caused a similar delay in release from G1 (Supplementary Figure S4).

Figure 4.

Loss of Set2 and H3K36 methylation abrogates yeast cell cycle progression. (A) Representative flow cytometry profiles of WT and set2Δ cells after α-factor arrest and release, which reveal defects in the ability of set2Δ cells to properly progress through the cell cycle. Comparison of the profiles at 30, 45 and 60 min after release in G1 show set2Δ cells have a delayed entry into S-phase, which then progresses faster than that observed for WT cells. (B–D) Quantification of multiple flow cytometry profiles of WT and set2Δ cells (n > 3) in which G1, S and G2/M cells from WT and set2Δ cells are counted.

Set2/SETD2 are regulated in temporal manner during cell cycle progression

Because of the unexpected connection between Set2 and the cell cycle, we asked whether Set2 itself might be subject to cell cycle regulation. Surprisingly, as assayed by immunoblot analysis after α-factor arrest and release, we found that Set2 levels were lower in the G1- and S-phases compared with G2/M, suggesting that Set2 is specifically targeted for destruction after M-phase (Figure 5A). To test this possibility in an independent way, we performed a nocodazole arrest-release experiment and examined Set2 levels after release from metaphase arrest. Strikingly, and consistent with our α-factor arrest-release results, Set2 accumulated after nocodazole treatment, but its levels decreased upon release from nocodazole (G2/M) arrest (Figure 5B). Furthermore, Set2 levels remained low across the G1- and S-phases, as assessed by immunoblot analysis for Set2 and the control G2/M markers Clb2 and Pds1 (Figure 5B). We also observed that the level of Set2 began to increase at 90 min after release from nocodazole (Figure 5B, last lane). These data strongly suggest that Set2 turnover occurs in a cell cycle-dependent manner. Importantly, the regulation of Set2 was not due to a defect in SET2 transcription because SET2 transcript levels were not cell cycle-regulated (Supplementary Figure S6). A further analysis of H3K36me3 levels across the cell cycle agreed with our findings for Set2, with the highest methylation levels occurring in G2/M and the lowest levels in G1 and S-phase (Figure 5A). Intriguingly, loss of H3K36me3 after nocodazole release was delayed relative to the rapid loss of Set2, consistent with H3K36me being removed secondarily by H3K36 demethylases (45).

Figure 5.

Yeast and human Set2 are cell cycle regulated proteins. (A) Set2 protein is cell cycle regulated. WT and set2Δ cells were arrested in the G1 phase using α-factor and released into fresh medium. Whole cell extracts were prepared at indicated time points and probed for Set2, H3, H3K36me3, Clb2 (marker for cell cycle progression) and G6PDH (total protein loading). (B) Set2 protein levels are down-regulated during progression through mitosis and mitotic exit. Whole cell extracts were prepared at designated time points after release from nocodazole. WT cells were released into fresh medium and immunoblotted for Set2, Clb2 (a marker for cell cycle progression), Pds1 (a marker for metaphase to anaphase transition) and G6PDH (total protein loading control). Analogous to Set2, H3K36me3 levels similarly change across the cell cycle, as analyzed by immunoblotting for H3K36me2/me3, and anti-H3 (loading control). (C) Human SETD2 is a cell cycle regulated protein. U2OS cells were synchronized using nocodazole, collected at indicated time points, and immunoblotted for SETD2. CycA, CycE and H3S10P serve as markers for different cell cycle stages, whereas RAN was used as a loading control.

To ascertain whether Set2 regulation across the cell cycle is evolutionarily conserved, we performed a nocodazole arrest-release experiment with U2OS cells and immunoblotted for human SETD2 plus H3S10p, CycA and CycE (S-phase markers of cell cycle progression); RAN served as a loading control. Similar to yeast Set2, we found that human SETD2 was cell cycle-regulated, with increased accumulation during mitosis (Figure 5C).

Set2/SETD2 are targeted for destruction by the APC/C complex

To identify the cell cycle-associated degradation machinery that targets Set2, we screened temperature-sensitive alleles of members of the Skp-Cullin-F-box (SCF) containing complex and the anaphase-promoting complex/cyclosome (APC/C) to determine which, if any, is involved in Set2 degradation. Thermal inactivation of temperature sensitive SCF members (CDC4 andCDC53) in asynchronously growing strains did not alter Set2 protein abundance (Supplementary Figure S7A). In contrast, inactivation of APC/C complex members did affect Set2 abundance. Specifically, thermal inactivation of Cdc23 of the APC/C complex, followed by a cycloheximide chase, prevented the rapid turnover of Set2 observed in WT cells (Figure 6A).

Figure 6.

The Anaphase Promoting Complex (APC) promotes Set2 protein turnover across the cell cycle. (A) Compromised APC/C function stabilizes Set2 protein levels. Cycloheximide chase experiments were performed using a WT or APC mutant allele (cdc23–1) after shift to restrictive temperature (37°C), and cell extracts were immunoblotted for Set2. G6PDH was used as a loading control. (B) Set2 is not regulated by APC/C^Cdh1. Cycloheximide chase experiments were performed as described in panel (A) in WT and CHD1 deleted cells (cdh1Δ) to detect the rate of Set2 loss. (C) Set2 is regulated by the APC/C^Cdc20. A GAL-CDC20 strain was grown in dextrose for 2 h to shut off expression of CDC20, and protein extracts were immunoblotted to probe for Set2, Clb2 and G6PDH. (D) An immunoblot showing the decrease in the levels of Set2 and H3K36me3 in a strain that contains a kinase-dead allele of BUB1, a checkpoint kinase that negatively regulates CDC20. (E) Representation of Set2 protein showing the domains and two putative D-boxes in the protein sequence. For comparison, a canonical D-box sequence is shown. (F) Mutation of the conserved D-box residues in Set2 (RXXL → AXXA) enhances Set2 stability. Cycloheximide chase experiments were performed at the indicated time points on WT and the D-box mutant strains and probed for Set2 and G6PDH. (G) Set2 protein levels become uncoupled from cell cycle regulation upon D-box mutation. Lysates were prepared from WT and D-Box mutants of Set2 post G1 arrest and released at indicated time points. Lysates were probed for Set2, Clb2 and G6PDH. (H) Appropriate turnover of Set2 protein is required for proper response to mitotic poison, benomyl. Five-fold serial dilutions of indicated strains were spotted on either SC-Ura or SC-Ura with benomyl (30 μg/mL).

APC/C interacts with either Cdc20 or Cdh1 to form two distinct complexes (APC/C^Cdc20 and APC/C^Cdh1) that function during G2/M- and G1-phases of the cell cycle respectively (46). Thus, we next asked whether one or both complexes might target Set2 for degradation. Kinetics of Set2 degradation (using cycloheximide chase) were not altered upon deletion of CDH1, thus ruling out its involvement (Figure 6B). Because a CDC20 deletion is not viable, we used a GAL-inducible form of CDC20, whose expression could be turned on and off by the addition of galactose or dextrose, respectively. When CDC20 expression was off (i.e. by dextrose), we observed a significant increase in the protein levels of Set2 and Clb2, a known substrate of APC/C^Cdc20 (Figure 6C). In addition, APC/C^Cdc20 is negatively regulated by Bub1-dependent phosphorylation (47). Thus, we used a bub1Δ and a Bub1 kinase-dead mutant to increase APC/C^Cdc20 activity. In both strains, we found that Set2 and H3K36me3 were significantly reduced, supporting the conclusion that Set2 is targeted for degradation by the APC/C^Cdc20 complex (Figure 6D and Supplementary Figure S7B). Consistent with this finding, the levels of Set2 and the metaphase-to-anaphase transition marker Pds1 decreased within the first 15 min after release from nocodazole arrest (Figure 5A).

To pinpoint more precisely how Set2 is targeted for destruction, we analyzed the Set2 protein sequence and found two canonical D-box consensus motifs (RxxLxxxxN/D/E) that are associated with APC/C^Cdc20 targets (47) (Figure 6E). To determine whether these putative D-boxes control the stability of Set2, we mutated both D-boxes (changing RxxL to AxxA in each putative consensus sequence site) and replaced endogenous WT with mutant SET2 at its locus. We assessed the stability of WT and D-box mutant form of Set2 (set2D-Box) using cycloheximide chase assays as described above. Mutation of the Set2 D-boxes led to stabilization of Set2 protein (Figure 6F), confirming that the D-boxes are targets of APC/C^Cdc20. To determine whether the D-Box mutants of Set2 were defective in the cell cycle pattern of protein degradation like that depicted in Figure 5A, we arrested the WT and set2D-Box mutant strains with α-factor and assessed the dynamics of Set2 abundance at different points. We used 0, 75, 105 and 120 min time intervals because they allowed us to directly visualize the WT and D-box mutant proteins on the same immunoblot, without having to compare Set2 among different gels. After G1 arrest and release, the level of Set2 was highest at the 75 min (G2/M) interval (Figure 6G), consistent with Figure 5A. As expected, the D-box mutants showed an increase in overall Set2 protein level and Set2 was not degraded in a manner similar to WT. These observations strongly suggest that degradation of Set2 is coupled with cell cycle progression. Finally, and consistent with a function for Set2 in mitosis, we observed that deletion of SET2 rendered cells sensitive to benomyl, a microtubule poison that arrests cells in G2/M (Figure 6H). Interestingly, compared with WT cultures, under normal conditions, stabilization of Set2 (by mutating the D-boxes) conferred a slight growth advantage and an increased resistance to benomyl (Figure 6H).

To determine whether human SETD2 is also regulated by APC/C, we over-expressed the human APC/C-Cdh1 homolog (FZR1) because SETD2 did not appear to possess a putative D-box, but it did exhibit a putative KEN-box at residues 2033–2035 (47). This experiment showed that SETD2 levels were reduced after FZR1 over-expression, and SETD2 levels were restored by the proteasome inhibitor MG132 (Supplementary Figure S8A). Furthermore, down-regulation of FZR1 increased SETD2 levels (Supplementary Figure S7A). Finally, and consistent with what we observed in strains lacking Set2, SETD2 knock-out from 786-O cell lines caused a faster progression through S-phase, which was rescued by over-expression of an N-terminal truncated form of SETD2 (tSETD2) that functionally complements a SETD2 deletion (Supplementary Figure S8B) (48). These data revealed a remarkable conservation in APC/C targeting of Set2 and SETD2 for destruction, and the results imply an important and conserved function for H3K36me in cell cycle regulation.

A non-degradable form of Set2 causes transcriptional defects in cell cycle regulated genes

Because mutation of the two D-boxes in Set2 conferred stability and uncoupled the protein from cell cycle regulation, we took advantage of the D-box mutant alleles to assess the consequence of a stabilized Set2 for sense and antisense transcription of cell cycle regulator genes. Similar to our approach in Figure 2, we performed stranded qRT-PCR to measure the sense and antisense transcript levels of FAR1 and SWE1 (Supplementary Figure S9A and S9D). In agreement with Figure 2, set2Δ cells showed defects in the regulation of the FAR1 and SWE1 sense transcripts compared with their WT expression patterns (Supplementary Figure S9B and S9C). In contrast, however, the sense transcription levels of FAR1 and SWE1 were improperly regulated across the cell cycle in the set2D-Box mutant, as compared with that observed in both WT and set2Δ cells (Supplementary Figure S9B and S9C). Along with the defects observed in sense transcription, the antisense transcripts SUT204 and SUT201 were also mis-regulated in the Set2 D-Box mutant strain, as compared with their transcription in WT and set2Δ cells (Supplementary Figure S9E and S9F). These results showed that cell cycle regulation of Set2 is essential to maintain chromatin integrity and optimal sense/antisense transcript levels.

DISCUSSION

Our findings reveal a new link between Set2/SETD2 and the cell cycle. We showed that the function of Set2 in the cell cycle is catalysis-dependent, and Set2 functions, at least partly, by preventing aberrant antisense transcription that likely causes transcriptional interference within cell cycle regulator genes (see model in Figure 7). Intriguingly, cell cycle genes tend to be long, which may explain why they are more dependent for their regulation on the Set2/H3K36me pathway. Our findings suggest a model wherein Set2 methylation suppresses the generation of cryptic transcripts (i.e. CUTs and SUTs) that would otherwise interfere with the normal periodic regulation of cell cycle regulator genes. This model of transcriptional interference agrees with several recent studies showing the interference potential of neighboring antisense transcripts (24,36–38,49). Furthermore, our model also posits that other chromatin regulators that suppress antisense or cryptic transcription may be crucial regulators of the cell cycle. Consistent with this idea, a functional classification (DAVID) analysis of genes that give rise to cryptic transcripts in SPT6 and FACT mutants (spt6-1004 and spt16-197) (16) showed that the most highly represented functional categories were related to ‘cell cycle’, ‘DNA replication’, and ‘mitosis’ (data not shown). We further speculate that such regulation of (sense) mRNAs by associated (antisense) ncRNAs is a general feature of coordinated gene expression programs in other aspects of cellular biology, for example, during metabolic flux. This idea agrees with a recent report that antisense long ncRNAs regulate the induction kinetics of genes that regulate glucose to galactose switching—a switch that involves lncRNA mediated deposition of H3K36 methylation (50).

Figure 7.

Model for the function of Set2 in regulating the cell cycle by maintaining transcriptional fidelity. In WT cells, Set2-dependent H3K36 methylation is required for the recruitment/activation of the Rpd3S histone deacetylase complex, which maintains a repressed chromatin state that is refractory to pervasive cryptic transcription. Our results show that Set2 protein and H3K36me levels are cell cycle regulated, with lowest levels in the G1- and S-phases and highest in the G2/M-phase. In addition, we found that Set2 and H3K36 methylation is required for the proper expression of cell cycle regulated genes that tend to be long and highly dependent on Set2. Because Set2/H3K36me levels fluctuate across the cell cycle, we speculate that these changing levels fine-tune the levels of cryptic transcription across the cell cycle to allow for optimal expression of genes required for transition through the different cell cycle phases.

Grosso et al. reported that, in SETD2 mutant tumors, there was widespread transcriptional read-through (51), a manifestation of disrupted transcriptional fidelity that impacted neighboring gene expression in clear cell renal carcinoma (ccRCC). In addition, Park et al. reported another direct connection between SETD2 and mitosis, wherein cytoplasmic methylation of tubulin by SETD2 maintained genome stability by promoting proper chromosome segregation (52). Consistent with this observation, our results show that SETD2 levels are highest during mitosis and are down-regulated upon entry into G1, an observation that mirrors budding yeast (Figure 5). Our results also show that set2Δ cells are sensitive to the mitotic poison benomyl (Figure 6); likewise, Schizosaccharomyces pombe set2Δ cells are sensitive to the mitotic poison thiabendazole (53), suggesting that the function of Set2 in mitosis is conserved. Although it is not known whether Set2 catalyzes methylation of a non-histone substrate in budding yeast, our results using a non-methylatable form of H3K36 strongly suggest that the function of Set2 in cell cycle control is mediated (principally) by histone methylation.

Our discovery that yeast and human Set2 are targeted for destruction by the APC/C complex also implies that Set2 and H3K36me regulate the reinforcement of transcriptional fidelity differently across different phases of the cell cycle. This reinforcement of transcriptional fidelity is disrupted when Set2 protein is stabilized by mutating its D-boxes. Interestingly, mutations in the Set2 D-boxes also altered, albeit slightly, the benomyl sensitivity of cells (Figure 6G), suggesting a function of Set2 turnover in regulation of the cell cycle. Although not directly tested here, it is nonetheless important to determine why Set2 and SETD2 are targeted by the APC/C complex. We envision that some of the tumor-promoting functions of mutant SETD2 in ccRCC and other relevant tumors may be a combination of loss of transcriptional fidelity of cell cycle regulators plus mis-regulation of non-histone targets of SETD2 such as tubulin.

SUPPLEMENTARY DATA

Supplementary Data are available at NAR Online.

ACKNOWLEDGEMENTS

The authors gratefully acknowledge Sue Biggins (Fred Hutchinson Cancer Center, Seattle), David Morgan (University of California, San Francisco), Jerry Workman (Stowers Institute), and Fred Winston (Harvard Medical School, Department of Genetics) for providing yeast strains and plasmids. We also thank Howard Fried for editorial suggestions.

FUNDING

NCI [R01 CA198482 to W.K.R.], [T32 CA009156 to Y.C.C.]; V Foundation Translational Science Award; Lineberger Comprehensive Cancer Center/FACS facility supported by the CCSG [P30 CA016086]; NIH [R01GM68088 to B.D.S.]; The UNC Flow Cytometry Core Facility is supported in part by Cancer Center Core Support Grant to the UNC Lineberger Comprehensive Cancer Center [P30 CA016086]. Funding for open access charge: NIH [R01GM110058].

Conflict of interest statement. None declared.

Author notes

These authors contributed equally to this work as first authors.

Present addresses:

Deepak Kumar Jha, Program, Division of Pediatric Hematology and Oncology, Dana-Farber Cancer Institute, Boston Children’s Hospital and Dana-Farber Cancer Institute, Boston, Massachusetts 02115, USA.

W. Kimryn Rathmell, Department of Medicine, Vanderbilt University Medical Center, Vanderbilt-gram Cancer Center, Nashville, TN 37232, USA.