The Catalytic-Dependent and -Independent Roles of Lsd1 and Lsd2 Lysine Demethylases in Heterochromatin Formation in Schizosaccharomyces pombe.

In eukaryotes, heterochromatin plays a critical role in organismal development and cell fate acquisition, through regulating gene expression. The evolutionarily conserved lysine-specific demethylases, Lsd1 and Lsd2, remove mono- and dimethylation on histone H3, serving complex roles in gene expression. In the fission yeast Schizosaccharomyces pombe, null mutations of Lsd1 and Lsd2 result in either severe growth defects or inviability, while catalytic inactivation causes minimal defects, indicating that Lsd1 and Lsd2 have essential functions beyond their known demethylase activity. Here, we show that catalytic mutants of Lsd1 or Lsd2 partially assemble functional heterochromatin at centromeres in RNAi-deficient cells, while the C-terminal truncated alleles of Lsd1 or Lsd2 exacerbate heterochromatin formation at all major heterochromatic regions, suggesting that Lsd1 and Lsd2 repress heterochromatic transcripts through mechanisms both dependent on and independent of their catalytic activities. Lsd1 and Lsd2 are also involved in the establishment and maintenance of heterochromatin. At constitutive heterochromatic regions, Lsd1 and Lsd2 regulate one another and cooperate with other histone modifiers, including the class II HDAC Clr3 and the Sirtuin family protein Sir2 for gene silencing, but not with the class I HDAC Clr6. Our findings explore the roles of lysine-specific demethylases in epigenetic gene silencing at heterochromatic regions.


Introduction
In eukaryotes, DNA is packaged with its associated proteins, forming chromatin. Nucleosomes refer to the basic units of chromatin; 147 base pairs of DNA wound around a histone core, made up of two copies each of histones H2A, H2B, H3, and H4 [1]. Histones play important roles in many cellular processes, including transcription, repair, recombination, and higher order chromatin organization [2,3]. The N-terminal tails of these histones are unstructured, radiate out from the nucleosome core, and are subject to covalent modifications, such as methylation, phosphorylation, and acetylation by chromatin-modifying enzymes [4]. These modifications adjust the density of chromatin, the dynamic associations between adjacent nucleosomes, and the recruitment of various

Heterochromatin Formation and Maintenance Assay Using an ade6 + Reporter
The original K∆::ade6 + strain was generated previously [29]. Cells that carry the K∆::ade6 + off epiallele and formed red colonies on low adenine medium (YE) were selected for genetic crosses. All strains carrying K∆::ade6 + off in different genetic backgrounds were made by genetic crossing. After genotyping, colonies were amplified on rich medium and then stored in −80 • C. For heterochromatin formation and maintenance assays, stored cells were recovered from −80 • C, grown at 30 • C for 2 days, then were used for RNA extraction and RT-PCR (Supplemental Figure  S9). Individual cells were plated on YEA plates using a dissection microscope and were grown at 30 • C until small colonies appeared (4 days for lsd1/2 C-terminal mutants or 2 days for other strains).
The YEA plates were then replica-plated on YE plates and grown for an extra 2-4 days until colony color could be observed ( Figure 5B). White (W) or red (R) colonies from YE plates were diluted in the water, and individual cells were dissected on YEA plates and grown for 2-4 days until small colonies appeared, then were plate-replicated on YE plates and grown for extra 2-4 days until the pigments of colonies could be assessed. Around 60 colonies for each genetic background were documented, and the percentage of colored and white colonies were calculated ( Figure 5C). Individual red, white, and color-mixed colonies were cultured to exponential phase growth in liquid YEA, before being collected, either for RNA extraction ( Figure 5D) or ChIP ( Figure 5E).

Growth Curve
Growth curves were generated using a plate reader, as described previously [73]. Strains with indicated genotypes ( Figure 2B and Figure 4A,B) were recovered from −80 • C, and grown on YEA plates at 30 • C for 2 days. Three replicates of each strain were inoculated in 1mL of liquid YEA, and the optical density (OD, 600nm) was measured. All strains were then diluted in liquid YEA to 0.1 OD and resuspended in wells to a total of 200 µL in an optical flat-bottom 96-well plate (Falcon). Cell growth was monitored using a microplate reader (Synergy H1 Hybrid Multi-Mode Reader, Biotek Instruments, Vermont, USA) with an incubation temperature of 30 • C, continuous fast orbital shaking, and OD (600 nm) readings every 2 min for 24 h. Growth curves for individual colonies were generated and analyzed by Gen5 microplate reader software (Biotek).

A N-Terminal Region of Lsd2 Is Required for Nuclear Localization and Viability
Since the essential lsd2 + cannot be completely deleted, we generated viable, partially functional mutant alleles for mechanistic studies. While Lsd1 and Lsd2 are structurally similar, Lsd2 has a unique N-terminal Cells 2020, 9,955 6 of 24 section (M1-P376) that contains an intron and is absent in Lsd1 ( Figure 1A). We generated multiple N-terminal truncated mutants of Lsd2 and assessed their relative viability and growth defects ( Figure 1B and Supplemental Figure S1). The removal of large sections of the N-terminus resulted in either no or very minor growth defects, as seen in lsd2-N1 (lsd2∆Q157-P376) and lsd2-N2 (lsd2∆E57-P376) (Supplemental Figure S1). The complete removal of the N-terminus resulted in lethality, as seen in lsd2-N3 (lsd2∆M1-P376) ( Figure 1B). The lethality of this deletion indicates an essential function of this N-terminal region.
Cells 2020, 9, x FOR PEER REVIEW 6 of 23 growth defects ( Figure 1B and Supplemental Figure S1). The removal of large sections of the Nterminus resulted in either no or very minor growth defects, as seen in lsd2-N1 (lsd2ΔQ157-P376) and lsd2-N2 (lsd2ΔE57-P376) (Supplemental Figure S1). The complete removal of the N-terminus resulted in lethality, as seen in lsd2-N3 (lsd2ΔM1-P376) ( Figure 1B). The lethality of this deletion indicates an essential function of this N-terminal region. Heterozygous diploids containing one copy of the mutant allele (linked with resistance to the antibiotic G418) and one copy of the wild-type (WT) allele were sporulated and tetrads of spores were arranged in columns. Red box: sample of a tetrad that shows selective growth on rich media with or without G418. (C) Visualization of GFP-tagged full-length or N-terminal truncated Lsd2 proteins in viable heterozygous diploid strains by confocal fluorescence microscopy. Indicated strains were also stained with Hoechst to show nuclear localization. Heterozygous diploids containing one copy of the mutant allele (linked with resistance to the antibiotic G418) and one copy of the wild-type (WT) allele were sporulated and tetrads of spores were arranged in columns. Red box: sample of a tetrad that shows selective growth on rich media with or without G418. (C) Visualization of GFP-tagged full-length or N-terminal truncated Lsd2 proteins in viable heterozygous diploid strains by confocal fluorescence microscopy. Indicated strains were also stained with Hoechst to show nuclear localization.  To examine cell morphology, we next observed these mutant strains via confocal fluorescence microscopy with calcofluor white staining (Supplemental Figure S5). Consistent with dilution assay results, catalytic lsd1-ao and lsd2-ao mutants appeared much like wild-type, with no obvious defects in morphology. lsd1-ΔHMG cells, however, displayed prominent defects such as elongated cells, branching cells, and cells with multiple septa, indicating defective cell division. While lsd1-ΔC appeared like wild-type in morphology, the lsd2-ΔC mutant showed similar defects as lsd1-ΔHMG, Figure 2. The C-terminal domains of Lsd1 and Lsd2 are vital for proper growth and genome-wide gene expression. (A) Schematic representations of WT Lsd1 and Lsd2 compared to catalytically inactive and C-terminal truncated mutants. (B) Growth curves generated by a plate reader show the growth rates of indicated WT, lsd1, and lsd2 mutants over a 24 h period. (C) Comparison of DEGs between lsd1-∆HMG and lsd2-∆C using linear regression analysis. (D) Loss of silencing at peri-centromeric region, mating type locus, and sub-telomeric region in lsd1-∆HMG and lsd2-∆C compared to WT. Cen 2: Centromere II. Mat: Mating-type locus. Tel2R: right telomere II. Normalized RNA-seq reads are aligned with H3K9me2 enrichment as measured by ChIP, which indicate heterochromatic regions. Data are plotted along with chromosome position. (E) qRT-PCR analysis of the silenced dg/dh repeats in the peri-centromeric region and cenH (mat locus). * p ≤ 0.05 and ** p ≤ 0.01 as determined by student's t test comparing the indicated samples with WT. Error bars represent s.e.m.
Based on the vastly different growth phenotypes observed, we generated additional mutants by removing more amino acids from the N-terminus. lsd2-N4 (lsd2∆G48-P376) was viable and showed only mild defects in growth ( Figure 1A and Supplemental Figure S1). Additional removal of the following 13 amino acids immediately upstream ( 35 SFYSMNTSENDPD 47 ) was catastrophic, as lsd2-N5 (lsd2∆S35-P376) was confirmed inviable ( Figure 1B). We were unable to generate strains in which small regions proximal to the intron near the start codon (the exon are spliced together to form the W8 codon of Lsd2) were removed, suggesting that these truncations may interrupt the splicing of lsd2 + or affect its nuclear localization. The confocal imaging of heterozygous diploid strains with GFP-tagged lsd2-N3, lsd2-N4, and lsd2-N5 reveals that only the lsd2-N4-GFP localizes to the nucleus, while lsd2-N3-GFP and lsd2-N5-GFP do not. Therefore, the 13 amino acid sequence ( 35 SFYSMNTSENDPD 47 ) near the N-terminal region contains essential residues that govern the nuclear localization of Lsd2 and are required for viability ( Figure 1C).

The HMG-Box Domains Are Essential for Lsd1 and Lsd2 Functions
The featured domain organization of Lsd1 and Lsd2 in S. pombe includes a SWIRM domain, an amine oxidase (AO) domain, and an HMG-box ( Figures 1A and 2A). The high-mobility group (HMG)-box domain proteins play essential roles in recognizing DNA or nucleosomes during various DNA-dependent processes including transcription, replication, and repair [74]. Both Lsd1 and Lsd2 contain C-terminal-HMG-box domains, but their functions in S. pombe have not been studied sufficiently. The N-terminal mutant strains of lsd2 + were either inviable or showed minimal growth defects ( Figure 1 and Supplemental Figure S1), and lsd1∆ had severe growth defects. Therefore, we endeavored to generate additional mutants that are more suitable for functional studies. Lsd1 and Lsd2 mutants with intermediate growth defects indicative of functional abnormalities were produced by deleting portions of the C-terminus. We separately truncated the C-terminal HMG-box domain (∆HMG) and just the C-terminal region immediately downstream of the HMG-box domain (∆C; retaining the HMG domain) (Figure 2A). Haploid lsd1-∆HMG (lsd1∆841-1000) and both lsd1-∆C (lsd1∆934-1000) and lsd2-∆C (lsd2∆1203-1273) alleles are viable and amenable to laboratory manipulation ( Figure 2B and Supplemental Figure S2A). However, deletion of the HMG domain was catastrophic for Lsd2, as lsd2-∆HMG (lsd2∆1113-1273) is inviable (Supplemental Figure S2A). Additionally, lsd1-∆HMG and lsd2-∆C show defects at the mating-type locus (Supplemental Figure  S2B), consistent with the previously reported observation that the HMG domain of Lsd1 is required for effective imprinting at the mating type locus [63].
While lsd1-∆HMG, lsd1-∆C, and lsd2-∆C were viable as haploid cells, they did form smaller colonies than wild-type cells under the same growth conditions, suggesting growth defects (Supplemental Figure S2A). Since the catalytic mutants of Lsd1 and Lsd2 (lsd1-ao and lsd2-ao) barely show growth defects [39,49,63], we investigated whether processes other than lysine demethylation were affected in these C-terminal mutants. We compared the doubling time of wild-type, catalytic, and C-terminal truncated mutants of lsd1 and lsd2 in liquid media ( Figure 2B). We found that lsd1-∆HMG and lsd2-∆C have intermediate growth defects, and are therefore more applicable for laboratory manipulation. Further dilution assays demonstrated that the catalytic mutants of Lsd1/2 show wild-type-like growth and resistance to heat stress, but C-terminal mutants have varying degrees of growth defects and heat sensitivities on solid rich media (Supplemental Figure S3). The analysis of homozygous and heterozygous diploids by the dilution assay revealed that the growth defects and heat sensitivities of these mutants are not dominant negative effects (Supplemental Figure S4).
To examine cell morphology, we next observed these mutant strains via confocal fluorescence microscopy with calcofluor white staining (Supplemental Figure S5). Consistent with dilution assay results, catalytic lsd1-ao and lsd2-ao mutants appeared much like wild-type, with no obvious defects in morphology. lsd1-∆HMG cells, however, displayed prominent defects such as elongated cells, branching cells, and cells with multiple septa, indicating defective cell division. While lsd1-∆C appeared like wild-type in morphology, the lsd2-∆C mutant showed similar defects as lsd1-∆HMG, although not quite to the same degree, with long or branched cells commonly observed. Consistent with our mutant phenotypes, lsd1∆ mutants were reported to display flocculation, morphological irregularities, and multi-nucleate phenotypes [34,39]. The results indicate that the HMG-box domain of Lsd1 is important for proper cell growth and morphology.

Lsd1 and Lsd2 C-Terminal Mutants Affect Genome-Wide Gene Expression and Display Defective Silencing at Constitutive Heterochromatic Regions
As Lsd2 is essential, transcriptome analysis in S. pombe has been limited to the loss of Lsd1 function. To explore the function of Lsd2 in differential gene expression (DGE), we analyzed the transcriptomes of lsd2-∆C mutants using RNA-sequencing (RNA-seq). We also performed RNA-seq in lsd1-∆HMG to compare with lsd2-∆C, as both are novel Lsd mutants. When compared with previously published expression data in lsd1∆ [39], our data agree with the expression profiles for the 68 highest-confidence target genes of Lsd1 (Supplemental Figure S6 and Supplemental Table S3). As expected, DEGs (both upand down-regulated) in lsd2-∆C mutant are highly similar to those of lsd1-∆HMG mutants (r 2 =0.904) ( Figure 2C). This result indicates widely overlapping functions between the two proteins. We observed more up-regulated than down-regulated genes (using a 2-fold cut-off) in both lsd1-∆HMG and lsd2-∆C mutants, suggesting that the dominant roles of Lsd1 and Lsd2 involve the repression of gene expression (Supplemental Figure S7). Previous studies have identified the loss of heterochromatic silencing and a moderate drop in heterochromatin-associated H3K9me2 levels in lsd1∆ [39,49], We also observed a loss of silencing at the peri-centromeric, mating-type, and sub-telomeric regions in lsd1-∆HMG and lsd2-∆C mutants ( Figure 2D). At the mating-type locus, the upregulated genes are mainly located at the L region, which has a low coverage of H3K9 methylation ( Figure 2D). We also performed qRT-PCR to compare the silencing defects between lsd C-terminal mutants with the catalytic mutants (lsd1-ao and lsd2-ao). Consistent with previous findings, the catalytic mutants only showed weak silencing defects ( Figure 2E). Notably, lsd2-∆C mutants lost repression to a similar degree as lsd1-∆HMG, suggesting that even a mild loss of function of Lsd2 has a robust effect on epigenetic silencing ( Figure 2E). Furthermore, both RNA-seq and qRT-PCR results agree that lsd1-∆HMG and lsd2-∆C have stronger defects at the outer-centromeric and sub-telomeric regions than at the mating-type locus. Altogether, the data suggests that Lsd1 and Lsd2 function to repress transcription at the constitutive heterochromatic loci, contributing to essential epigenetic silencing in fission yeast.

C-Terminal Mutants of Lsd1 and Lsd2 Retain Some Enzymatic Activities
Our data suggest that the C-terminal mutations affect the additional functions of Lsd1 and Lsd2 beyond their catalytic amine oxidase-related activities. However, it was unknown to what degree the amine oxidase function is affected by C-terminal truncation. Alternatively, this result could merely reflect the differences in the functions of Lsd1 and Lsd2. To determine whether the C-terminal truncations affect the amine oxidase functions within the same protein, C-terminal truncations were introduced in strains already bearing the catalytic mutation to generate the double mutant alleles of lsd1 and lsd2 (lsd1-ao-∆HMG, lsd1-ao-∆C, and lsd2-ao-∆C). While we were unable to isolate viable lsd1-ao-∆HMG and lsd2-ao-∆C strains following multiple transformations, we did generate viable lsd1-ao-∆C strains that showed slight synthetic growth defects and temperature sensitivity compared to either single mutant allele strains ( Figure 3A). Further investigation by qRT-PCR showed that lsd1-∆C slightly exacerbated the silencing defect of lsd1-ao at the centromere ( Figure 3B). These data indicate that the C-terminus-related activity of Lsd1 does not completely overlap with the amine oxidase activity of Lsd1. Deletion of the HMG domain in lsd1-ao (lsd1-ao-∆HMG) likely causes severe defects, and would therefore be lethal. Likewise, the loss of both the amine oxidase function and the C-terminus in Lsd2 (lsd2-ao-∆C) was catastrophic for the cell, consistent with the known lethality of lsd2∆ strains. As combining C-terminal mutation with the catalytic mutations within the same protein exacerbates the phenotype, the results indicate that the C-terminal mutants of Lsd1 and Lsd2 preserve some enzymatic activities.

Lsd1 and Lsd2 Serve Overlapping, but Divergent, Functions
While Lsd1 and Lsd2 are known to form a complex in S. pombe [34,49], the utility of Lsd1-Lsd2 complex formation is still unclear. To test whether the mutation of one Lsd protein affects the function of the other Lsd protein, we crossed each lsd1 and lsd2 single mutant strains, to generate all possible double mutant combinations. HMG-box domain-retaining mutants can form viable double mutant strains with the catalytic amine oxidase mutants, although they cannot be combined with one another ( Figure 3C). lsd1-ΔHMG mutants were found to be synthetic lethal with all lsd2 mutants, including lsd2-ao, further highlighting the importance of this domain. lsd1-ΔC lsd2-ao and lsd1-ao lsd2-ΔC mutants showed weak negative interactions compared to single mutant parental strains ( Figure 3A), indicating that each lsd ΔC mutant can tolerate the loss of amine oxidase activity, but not loss of the C-terminus, in the other Lsd protein. This result further supports our assertion that the C-terminal mutants of Lsd1 and Lsd2 retain some enzymatic activities. In addition, we examined this interaction by qRT-PCR and found that the lsd2-ΔC-associated silencing defects at centromeric region and mating type locus are exacerbated by lsd1-ao ( Figure 3D). This observation indicates that the two proteins work in parallel to suppress expression at these heterochromatic loci. Collectively, these data suggest that the C-terminal-related activities of Lsd1 and Lsd2 act in parallel with one another and do not fully overlap with their amine oxidase-related functions.

C-Terminal Mutants of Lsd1 and Lsd2 Exacerbate the Silencing Defects of ago1Δ
Since RNAi is the main pathway of assembly heterochromatin at the centromeric region [25], we next examined the relationship between Lsd1/2 mutants and RNAi. We combined lsd1 and lsd2 mutants with ago1Δ, and assessed the genetic interactions via the growth curve and dilution assays ( Figure 4A-B and Supplemental Figure S8). The C-terminal lsd mutants showed synthetic growth While Lsd1 and Lsd2 are known to form a complex in S. pombe [34,49], the utility of Lsd1-Lsd2 complex formation is still unclear. To test whether the mutation of one Lsd protein affects the function of the other Lsd protein, we crossed each lsd1 and lsd2 single mutant strains, to generate all possible double mutant combinations. HMG-box domain-retaining mutants can form viable double mutant strains with the catalytic amine oxidase mutants, although they cannot be combined with one another ( Figure 3C). lsd1-∆HMGmutants were found to be synthetic lethal with all lsd2 mutants, including lsd2-ao, further highlighting the importance of this domain. lsd1-∆C lsd2-aoand lsd1-ao lsd2-∆Cmutants showed weak negative interactions compared to single mutant parental strains ( Figure 3A), indicating that each lsd ∆C mutant can tolerate the loss of amine oxidase activity, but not loss of the C-terminus, in the other Lsd protein. This result further supports our assertion that the C-terminal mutants of Lsd1 and Lsd2 retain some enzymatic activities. In addition, we examined this interaction by qRT-PCR and found that the lsd2-∆C-associated silencing defects at centromeric region and mating type locus are exacerbated by lsd1-ao ( Figure 3D). This observation indicates that the two proteins work in parallel to suppress expression at these heterochromatic loci. Collectively, these data suggest that the C-terminal-related activities of Lsd1 and Lsd2 act in parallel with one another and do not fully overlap with their amine oxidase-related functions.

C-Terminal Mutants of Lsd1 and Lsd2 Exacerbate the Silencing Defects of ago1∆
Since RNAi is the main pathway of assembly heterochromatin at the centromeric region [25], we next examined the relationship between Lsd1/2 mutants and RNAi. We combined lsd1 and lsd2 mutants with ago1∆, and assessed the genetic interactions via the growth curve and dilution assays ( Figure 4A,B and Supplemental Figure S8). The C-terminal lsd mutants showed synthetic growth defects with ago1∆ ( Figure 4A,B), but the growth rates of lsd1-ao and lsd2-ao did not seem to be affected by ago1∆ (Supplemental Figure S8). When combined with ago1∆, both lsd1-∆HMG and lsd2-∆C show stronger losses of silencing at all heterochromatic regions ( Figure 4C). Surprisingly, we observed a noticeable decrease of H3K9me2 in lsd1-∆HMG and lsd2-∆C single mutant cells at all heterochromatic regions, and the levels of H3K9me2 in those mutants were further reduced when combined with ago1∆ ( Figure 4D). This finding suggests that the C-terminal domains of Lsd proteins play an unexpected role in epigenetic silencing through a mechanism that overlaps with RNAi.
Cells 2020, 9, x FOR PEER REVIEW 11 of 23 defects with ago1Δ ( Figure 4A-B), but the growth rates of lsd1-ao and lsd2-ao did not seem to be affected by ago1Δ (Supplemental Figure S8). When combined with ago1Δ, both lsd1-ΔHMG and lsd2-ΔC show stronger losses of silencing at all heterochromatic regions ( Figure 4C). Surprisingly, we observed a noticeable decrease of H3K9me2 in lsd1-ΔHMG and lsd2-ΔC single mutant cells at all heterochromatic regions, and the levels of H3K9me2 in those mutants were further reduced when combined with ago1Δ ( Figure 4D). This finding suggests that the C-terminal domains of Lsd proteins play an unexpected role in epigenetic silencing through a mechanism that overlaps with RNAi.

The Lysine Demethylase Activity of Lsd Proteins Antagonize RNAi
We next investigated whether the catalytic mutations of Lsd1 and Lsd2 affect RNAi. Unlike the C-terminal mutants of Lsd1 and Lsd2, lsd2-ao suppresses the silencing defect of ago1∆ at the centromeric region ( Figure 4C), suggesting that the amine oxidase activity of Lsd2 antagonizes the function of RNAi. We did not observe the same effects at the mating-type locus or the telomeric regions, because redundant pathways are present to ensure the silencing of these regions. As weak H3K9me2 demethylase activity by Lsd1 has been previously demonstrated in S. pombe [39,49,62], the catalytic mutants of Lsd1 or Lsd2 may alleviate the silencing defects of ago1∆ through reduced demethylation of the repressive heterochromatin marker, histone H3 dimethyl-K9 (H3K9me2). This is consistent with our detection of enhanced H3K9me2 in both lsd1-ao and lsd2-ao mutants, and the cumulative effects indicate that the catalytic activities of Lsd proteins limit H3K9 methylation and heterochromatin formation ( Figure 4D). Compared to ago1∆, the levels of H3K9me2 increase at the centromeric regions in lsd1-ao ago1∆ and lsd2-ao ago1∆ double mutant cells ( Figure 4D). In addition, the lsd1-ao lsd2-ao double mutants show the highest levels of H3K9me2 ( Figure 4D). These results agree with the role of Lsd proteins, as H3K9 demethylases as previously implicated [39,49,62].

Lsd1/2 Play Roles in the Maintenance and Re-Establishment of Heterochromatin at the Mat Locus via an RNAi-Independent Mechanism(s)
To further analyze the RNAi-independent function of Lsd1/2, we introduced Lsd C-terminal mutations into cells that lack part of the K region at the mating-type locus, but continued to repress a proximal ade6 + reporter gene (K∆::ade6 + off) ( Figure 5A). The silencing of the ade6 + reporter gene causes the accumulation of the red-pigmented metabolite 5-aminoimidazole ribotide (AIR), resulting in the formation of red colonies on low adenine medium [29]. Cells that partially or completely lose silencing of this reporter gene form pink sectoring colonies or white colonies, respectively [29]. Deleting the K region containing cenH (K∆) results in the loss of the RNAi nucleation center within the mating-type locus [23]. In K∆ cells, RNAi-independent mechanisms can partially form heterochromatin and mediate gene silencing, although the function of Clr4 and Swi6 are required [75]. Once the silencing is established (K∆::ade6 + off), it can be maintained through mitotic and meiotic cell division [75]. After we combined lsd1 and lsd2 mutants with K∆::ade6 + off cells through genetic crosses, we detected heterogeneous expression of the K∆::ade6 + reporter in all lsd1 and lsd2 mutants, as observed by mixtures of red and white colonies ( Figure 5B). To avoid the limitations of space and nutrition that could affect the size and color of colony formation, we placed single cells evenly on rich media plates using a dissection microscope, let the cells grow at 30 • C until colonies appear, and replica plated the colonies on low adenine medium (YE). We then assessed the expression of K∆::ade6 + using qRT-PCR, and we found that lsd2-∆C cells showed the strongest expression, which is consistent with their formation of the lighter pink color on YE medium (Supplemental Figure S9). Interestingly, although lsd1-ao lsd2-ao cells showed the weakest expression of K∆::ade6 + , these colonies show strong phenotypic variation, similar to lsd1-∆HMG and lsd2-∆C colonies ( Figure 4B), indicating that the roles of Lsd1 and Lsd2 are involved in the epigenetic maintenance of heterochromatin. Cells 2020, 9, x FOR PEER REVIEW 13 of 23 In wild-type cells, KΔ::ade6 + off (red colonies) can switch to KΔ::ade6 + on, forming white colonies on YE, indicating the loss of heterochromatin at the mat locus. However, through cell division, white colonies that have lost the silencing of KΔ::ade6 + can also re-establish silencing on this reporter gene, forming red colonies after generations of growth. To investigate how efficiently the switched white cells can re-establish heterochromatin at KΔ::ade6 + , we transferred individual white cells onto rich media, using a dissection microscope. After the colonies start to appear on rich medium plates, we replica-plated colonies on low adenine medium. After propagation, about 28.5% of initially white (A) An ade6 + reporter gene replaced the K region (K∆::ade6 + ). Epigenetic mechanism(s) can repress ade6 + expression (K∆::ade6 + off, red colonies); loss of the epigenetic silencing causes expression of ade6 + (K∆::ade6 + on, white colonies). (B) The K∆::ade6 + off allele was introduced into cells with indicated genotypes via genetic crosses. Samples of color and shape of colonies formed by individual cells on low adenine medium are shown. Colonies were replica-plated from rich medium (YEA) to low adenine medium (YE). (C) Cells that formed red or white colonies from (B) with indicated genotypes were dissected and grown on rich medium without selection, then were transferred to low adenine medium. Samples of red (R), white (W), and color-mixed (M) colonies on YE medium are shown on the right. Bar graph to the left shows the percentage of colonies that are red, white, or color-mixed. The percentages are listed in the table for each of the indicated genotypes. (D,E) ade6 + expression in red or white cells collected in (C) were investigated by qRT-PCR (D), and relative H3K9me2 enrichment at repeat regions (versus input) (E) were compared in red or white colonies with indicated genotypes. * p ≤ 0.05 and ** p ≤ 0.01 as determined by student's t test comparing the indicated samples with WT R for qPCR or qChIP. Error bars represent s.e.m. In wild-type cells, K∆::ade6 + off (red colonies) can switch to K∆::ade6 + on, forming white colonies on YE, indicating the loss of heterochromatin at the mat locus. However, through cell division, white colonies that have lost the silencing of K∆::ade6 + can also re-establish silencing on this reporter gene, forming red colonies after generations of growth. To investigate how efficiently the switched white cells can re-establish heterochromatin at K∆::ade6 + , we transferred individual white cells onto rich media, using a dissection microscope. After the colonies start to appear on rich medium plates, we replica-plated colonies on low adenine medium. After propagation, about 28.5% of initially white wild-type cells formed red colonies, indicating the efficient re-establishment of heterochromatin at the silent mat region in an RNAi-independent manner ( Figure 5C). We observed a reduction in H3K9me2 levels between red and white colonies in wild-type backgrounds by qChIP and qRT-PCR ( Figure 5D,E). In lsd1-ao lsd2-ao cells, only about 6.1% of initially white colonies switched to mixed colonies, suggesting a defect in re-establishment of heterochromatin. Interestingly, almost 100% of lsd1-ao lsd2-ao red cells gave rise to variegated colonies, indicating that the catalytic activities of Lsd1 and Lsd2 proteins are important for the maintenance of heterochromatin at the mat locus using an RNAi-independent mechanism ( Figure 5C). Similar to lsd1-ao lsd2-ao white cells, about 3.4% lsd1-∆HMG or 5.9% lsd2-∆C white cells can form pigmented colonies, suggesting a re-establishment deficiency. Our results support that these proteins play important roles in the establishment and maintenance of epigenetic silencing at the mating-type locus in an RNAi-independent manner.

Loss of Epe1 Suppresses the Silencing Defects of Lsd1 and Lsd2 Mutants
In addition to the two Lsd proteins, Epe1, a JmjC domain containing protein, is an anti-silencing factor predicted to function as a histone H3K9 demethylase based on sequence analysis [76][77][78][79][80], although the activity has not yet been formally confirmed. Deletion of Epe1 suppresses the silencing defects caused by the loss of most heterochromatic factors including HDACs and RNAi [76][77][78][79][80]. Epe1 also promotes histone turnover in heterochromatin and acts to prevent spreading of heterochromatin beyond its boundaries [37,40]. Since Lsd proteins are histone demethylases and lsd1∆ was reported to cause spreading beyond heterochromatin boundaries, [39,49] we wondered whether Lsd1 or Lsd2 is functionally redundant to Epe1. We therefore combined lsd mutant alleles with epe1∆ and assessed their genetic interactions. Dilution assays indicate that lsd1-∆HMG and lsd2-∆C show very weak positive interactions with epe1∆ in growth at the stress temperature (37 • C) ( Figure 6A). qRT-PCR analysis shows that epe1∆ partially suppresses the silencing defects of lsd1-∆HMG and lsd2-∆C ( Figure 6B). H3K9me2 levels are also increased in epe1∆ lsd1-∆HMG and epe1∆ lsd2-∆C double mutants compared to lsd1-∆HMG and lsd2-∆C single mutants ( Figure 6C). These data indicate that the anti-silencing mechanism of Epe1 antagonizes the silencing defects of Lsd1/2, similarly to its effect on the other silencing factors [37,[40][41][42][43][44][76][77][78][79][80]. and Lsd2 proteins are important for the maintenance of heterochromatin at the mat locus using an RNAi-independent mechanism ( Figure 5C). Similar to lsd1-ao lsd2-ao white cells, about 3.4% lsd1-ΔHMG or 5.9% lsd2-ΔC white cells can form pigmented colonies, suggesting a re-establishment deficiency. Our results support that these proteins play important roles in the establishment and maintenance of epigenetic silencing at the mating-type locus in an RNAi-independent manner. In addition to the two Lsd proteins, Epe1, a JmjC domain containing protein, is an anti-silencing factor predicted to function as a histone H3K9 demethylase based on sequence analysis [76][77][78][79][80], although the activity has not yet been formally confirmed. Deletion of Epe1 suppresses the silencing While human LSD1 interacts with HDAC complexes to mediate repression [53][54][55][56], S. pombe Lsd1 or Lsd2 does not appear to physically associate with HDACs [34,39,49]. Nonetheless, the phenotype of lsd1∆ mutant is similar to clr6-1, as described previously [34,62]. Both lsd1∆ and clr6-1 display decreased growth, protrusions, and multinucleate phenotypes. In addition, a significant overlap has been found between genes repressed by Lsd1 and those repressed by Clr6 [34,62]. These results suggest that Lsd1 and Clr6 may work synergistically in the same pathway. To explore whether the repressive functions of Lsd1 and Lsd2 are connected to HDAC-mediated repression, we combined lsd C-terminal mutants with mutants of each of the three HDACs: clr6-1, clr3∆, and sir2∆, and assessed their genetic interactions by dilution assay (Figure 7A-C). Interestingly, sir2∆ rescues the growth defects and heat sensitivity of lsd1-∆HMG and lsd2-∆C, indicating that Sir2, Lsd1, and Lsd2 are epistatic in cell growth and may work in the same pathway in controlling gene expression at euchromatic loci ( Figure 7C). Cells 2020, 9, x FOR PEER REVIEW 16 of 23 demethylation activities. Instead, we found that Lsd1 and Lsd2 repress heterochromatic transcripts, probably through both RNAi-dependent and -independent manners. Our data also suggest that Lsd1 and Lsd2 regulate each other through a previously undescribed mechanism.

The N-Terminus of Lsd2
We identified a region spanning 13 amino acids (SFYSMNTSENDPD) near the N-terminus of Lsd2, which is required for viability. Our results indicate that the nuclear localization of Lsd2 is At all constitutive heterochromatic domains, lsd C-terminal mutants showed strong cumulative genetic interactions with clr3∆ and sir2∆, indicating that they play overlapping functions in epigenetic silencing. In particular, lsd2-∆C sir2∆ double mutants show the strongest silencing defects at centromeric repeat regions and sub-telomeric regions, while lsd1-∆HMG clr3∆ double mutants have the most robust effect at the mating-type locus ( Figure 7D). As expected, clr6-1 shows no additive interactions with lsd C-terminal mutants, indicating that Lsd1 and Lsd2 may be involved in the same pathway with Clr6 in gene silencing at heterochromatic regions ( Figure 7A,D). Next, we investigated whether the defective silencing in these mutants is linked to changes in heterochromatin formation. Indeed, H3K9me2 ChIPs showed the greater reduction of H3K9me2 levels in clr3∆ lsd or sir2∆ lsd double mutant cells at the constitutive heterochromatic regions, compared to single mutants ( Figure 7E). Altogether, our results suggest that Lsd1 and Lsd2 operate in parallel with Clr3 and Sir2, but not with Clr6, in gene silencing at heterochromatic regions.

Discussion
As the first histone demethylases revealed in S. pombe, much attention has been focused on the roles of Lsd1 and Lsd2 in antagonizing histone methylation. However, emerging views strongly indicated that their essential functions rely on their non-enzymatic activities based on the observation that cells carrying both catalytically inactive Lsd proteins are viable and do not show significant growth defects. To provide insight into the complex roles of Lsd proteins beyond their well-known histone demethylation activities, we generated and investigated new mutants of lsd1 and lsd2 with intermediate phenotypes and examined their functional interactions with each other. We demonstrated that the N-terminal domain and the C-terminal HMG-box domain of Lsd2 are essential for viability, while the HMG domain of Lsd1 is important, but not essential, for growth and morphology, suggesting divergent roles in S. pombe for Lsd1 and Lsd2. Notably, the C-terminus of Lsd1 was shown to perform functions that do not completely overlap with the amine oxidase-related demethylation activities. Instead, we found that Lsd1 and Lsd2 repress heterochromatic transcripts, probably through both RNAi-dependent and -independent manners. Our data also suggest that Lsd1 and Lsd2 regulate each other through a previously undescribed mechanism.

The N-Terminus of Lsd2
We identified a region spanning 13 amino acids (SFYSMNTSENDPD) near the N-terminus of Lsd2, which is required for viability. Our results indicate that the nuclear localization of Lsd2 is affected by the loss of this region, suggesting that it contains critical residues that mediate the nuclear localization of Lsd2. While human LSD1 contains a predicted nuclear localization signal (NLS) 112 RRKRAK 117 [81], we could not find the similar NLS sequence in S. pombe Lsd1. We also did not find the similar 13 amino acid NLS segment in human LSD2. In addition, human LSD2 has an N-terminal zinc finger domain that seems to be required for histone demethylase activity and may mediate other molecular interactions [56], but this domain is not found in S. pombe Lsd2. The fact that S. pombe Lsd proteins share about 40% sequence similarity to that of human homologs [56], while Lsd proteins are completely missing in budding yeast, suggests that Lsd proteins are essential in organisms that have H3K9 methylation. This study provides the first evidence of the importance of the N-terminus of Lsd2 in S. pombe.

Mutual Regulation of Lsd1 and Lsd2
Our data implicate that Lsd1 and Lsd2 regulate each other through an unknown mechanism, unlikely to be related to histone demethylation. While functions of Lsd proteins beyond histone demethylation have not been well-demonstrated in S. pombe, non-histone substrates have been described for human LSD1 and LSD2. For instance, LSD1 demethylates p53 to inhibit apoptosis [82], targets IFITM3 (interferon-inducible transmembrane family protein 3) to restrict influenza A virus infection in response to IFNα [83], and has been shown to activate HIV transcription through the demethylation of the HIV Tat protein [84]. LSD2 has been found to target non-histone substrates as well, promoting the degradation of OGT (O-GlcNAc transferase) through unexpected E3 ubiquitin ligase activity, unrelated to its amine oxidase function [85]. Alternative mechanisms have also been suggested for LSD1, which promotes prostate cancer proliferation through an amine oxidase-independent mechanism [86]. In S. pombe, Lsd1 and Lsd2 may regulate one another through direct demethylation. However, up to date, no non-histone substrates have been identified for these enzymes. Alternatively, it is possible that the cross-regulation of Lsd1 and Lsd2 is not mediated by demethylation, but by demethylation-independent functions related to their C-terminal domains. Either way, the notion that Lsd1 and Lsd2 regulate the activities of one another within the same complex is intriguing and will be further investigated in our future studies.

The Catalytic Activities of Lsd Proteins Antagonize RNAi
RNAi is the primary pathway of heterochromatin assembly at the centromeric regions, but is dispensable for the maintenance of heterochromatin at the mat locus and sub-telomeres [8]. The simultaneous loss of Lsd1 and Lsd2 catalytic activities slightly increases RNA levels from the repetitive regions ( Figure 4C), suggesting that other H3K9 demethylases, such as Lid2 or Epe1, play major roles in the dynamic assembly of heterochromatin at these regions [45,76]. In the absence of RNAi, the impaired amine oxidase activities of Lsd1 and Lsd2 alleviate the silencing defects, resulting in the reduction of repeat transcript levels and the elevation of H3K9me2 levels at the main heterochromatic regions. Previous studies showed that both Lsd1 and Lsd2 are enriched at very low levels at specific heterochromatic regions, compared to their levels at euchromatic regions [34,49]. It is possible that, in the context of elevated transcription at centromeric repeat regions in cells lacking RNAi machinery, the enrichment of Lsd proteins is enhanced, which facilitates the demethylation of H3K9 and further the transcription of the repeats. It is also possible that catalytically inactivated Lsd proteins increase their association with chromatin and physically occlude the recruitment of another demethylase [87].

The Role of Lsd Proteins in Maintenance and Re-Establishment of Epigenetic Silencing
Heterochromatin assembly requires sequential steps [8,29]: it is nucleated at designated genomic loci, spread to surrounding regions [88], and its structure is maintained during DNA replication and propagated through multiple cell divisions [19,77]. Factors that are involved in heterochromatin formation often participate at specific step(s) rather than throughout the process [8]. In particular, the RNAi silencing pathway is required for the nucleation of heterochromatin formation [88]. Once silencing is established, RNAi machinery is dispensable [89]; the heterochromatic state can be maintained without RNAi. At the mating-type locus, RNAi machinery is required to nucleate heterochromatin at the cenH region [46], but is dispensable for its maintenance [23]. In addition to RNAi machinery, multiple mechanisms are present at the mating type locus to ensure the efficient establishment of heterochromatin [27][28][29]. Without the cenH region, RNAi-independent pathways can form heterochromatin, inefficiently, at the K region [75,89]. This finding is based on the studies using K∆::ura4 + off allele [75,89]. In addition to ura4 + , ade6 + gene is a widely used reporter in S. pombe for RNAi-dependent and independent heterochromatin formation at centromeric regions and the mating-type locus [29,77,90,91]. When the K region is replaced by an ade6 + reporter gene, the reporter gene can be silenced by heterochromatin (K∆::ade6 + off) [29]. Although the silenced status of this allele can be maintained through multiple generations, a low percentage of K∆::ade6 + off cells lose silencing and switch to K∆::ade6 + on status, accompanied by the loss of heterochromatin ( Figure 5C, WT, R to W). Gene silencing and heterochromatin can also be re-established during cell division ( Figure 5C, WT, W to R). Using this system, we found that pre-established K∆::ade6 + off alleles are unstable in all Lsd mutants ( Figure 5). Interestingly, although cells that lack catalytic activates of both Lsd1 and Lsd2 show fairly normal growth, they form phenotypically variable colonies, indicating the unstable inheritance of the K∆::ade6 + off allele. Once the cells lost the heterochromatin and formed white colonies, all Lsd mutants struggled to re-establish the K∆::ade6 + off state, indicating the defective assembly of heterochromatin at the mat locus without the K region. In lsd1-ao lsd2-ao double mutant cells, we consistently detected enhanced H3K9me2 on K∆::ade6 + off alleles, but still observed phenotypic variegation, suggesting that heterochromatin may not be the sole mechanism to explain this phenotype. In addition, peri-centromeric, but not mating-type or sub-telomeric regions, show great overlap between changes in H3K9me2 enrichment and gene expression in the C-terminal mutants of Lsd1 and Lsd2 ( Figure 2D). This observation further supports that the roles of Lsd1 and Lsd2 in gene silencing are not exclusively reliant on H3K9me. Overall, our results suggest that both catalytic and non-catalytic roles of Lsd proteins are involved in heterochromatin maintenance and re-establishment.

The Roles of LSD Proteins beyond Their Catalytic Activities
We consistently detected cumulative silencing defects and the reduction of H3K9me2 when combining both C-terminal mutants of Lsd proteins with ago1∆ ( Figure 4C,D). This result suggests that the functions of Lsd proteins play dominant roles in epigenetic silencing at all heterochromatic regions besides their catalytic activities. They may operate within protein complexes formed with additional factors that exhibit coregulatory or scaffolding functions. In mammals, such interactions have been shown to modulate catalytic activities, substrate specificities, and/or localizations of Lsd proteins that link to various biological processes [56]. For example, LSD1 interacts with different transcriptional repressors, including Co-REST and Co-REST like proteins, which enforces repressed chromatin status and gene silencing [50][51][52]. LSD1 also interacts with the NuRD complex, which displays nucleosome remodeling and HDAC activity, and is important for a wide variety of cellular processes [53][54][55][56], In S. pombe, in addition to H3K9 methylation, the deacetylation of histones by histone deacetylases (HDACs) is essential for gene repression and heterochromatin assembly [31,46,92]. Although direct physical interactions with HDACs were not identified [34,49], S. pombe Lsd proteins cooperate with HDACs.
At heterochromatic regions, the C-terminal mutants of Lsd proteins show strong additive silencing defects with Clr3 (Class II) and Sir2 (Sirtuin family), but not with Clr6 (Class I) HDACs. This finding suggests that Lsd proteins coordinate with Clr3 and Sir2 in heterochromatin assembly and repression of the repeat transcripts. In a genome-wide study of HDAC functions, Clr6 was found to be the main HDAC involved in promoter-localized repression, while Sir2 and Clr3 target many loci for repression, mainly in heterochromatic regions [31,33]. As significant overlap has been described between genes repressed by Lsd1 and those repressed by Clr6, our observation that the combination of clr6-1 and the C-terminal mutants of Lsd proteins have a weak or no genetic interaction in cell growth (Figure 7), is consistent with a model in which Lsd proteins act in concert with Clr6. We also found that sir2∆ partially alleviates the growth defects of lsd C-terminal mutants. This positive genetic interaction suggests that the roles of Lsd proteins may act in the same pathway with Sir2 at euchromatic loci. However, results from dilution assays provide a whole-cell summary of the genetic interaction and may not reflect genetic interactions with respect to a particular function (e.g., heterochromatic silencing; Figure 7). Considering the complicated and multifaceted regulation of gene expression by Lsd proteins, the antagonistic interaction between Lsd1/2 with Sir2 reflects the additional functions of Lsd1/2, beyond epigenetic silencing.

Conclusions
Here, we explore the catalytic-dependent and -independent roles of conserved lysine demethylases Lsd1 and Lsd2 in epigenetic silencing of constitutive heterochromatin in S. pombe. We reveal several novel findings: (1) an N-terminal peptide is essential for the nuclear localization of Lsd2; (2) Lsd1 and Lsd2 regulate each other; (3) the catalytic roles of Lsd1 and Lsd2 antagonize RNAi in heterochromatin formation; (4) Lsd1 and Lsd2 play roles in the maintenance and establishment of heterochromatin; and (5) Lsd1 and Lsd2 cooperate with Class II and Sirtuin family HDACs in constitutive heterochromatic silencing. A summary table showing all the genetic interactions between the Lsd1/2 mutants and silencing factors that were investigated in this study is available in Supplemental Figure S10.
Supplementary Materials: The following are available online at http://www.mdpi.com/2073-4409/9/4/955/s1, Supplemental Figure S1: Lsd2 N-terminal truncations retaining the nuclear localization sequence (S35-D47) display either no, or mild, growth defects. Supplemental Figure S2: lsd2-∆HMG cells are not viable and lsd1-∆C, lsd1-∆HMG, and lsd2-∆C cells have growth defects. Supplemental Figure S3: Serial dilution demonstrating the relative growth defects and heat sensitivities (37 • C) of Lsd1 and Lsd2 mutant strains (schematic of mutations in each indicated strain are displayed in Figure 2A). Supplemental Figure S4: Lsd1 and Lsd2 C-terminal mutations do not have dominant negative effects. Supplemental Figure S5: The C-terminal domains of Lsd1 and Lsd2 are vital for proper growth and morphology. Supplemental Figure S6: The similarity between RNA-seq analysis (lsd1-∆HMG, this study) and expression array data (lsd1∆, . Supplemental Figure S7: Volcano plots highlighting significant genes that are differentially expressed (DEGs) in lsd1-∆HMG or lsd2-∆C mutants using RNA-seq. Supplemental Figure S8. Serial dilutions demonstrating the genetic interactions between lsd1 and lsd2 catalytic mutant strains combined with ago1∆. Supplemental Figure S9. ade6 + expression from cells collected in Figure 5B was monitored by qRT-PCR. Supplemental Figure S10. A summary table showing all the genetic interactions between the Lsd1/2 mutants and silencing factors that were investigated in this study. Supplemental Table S1: A list of stains used in this study. Supplemental Table S2: A list of oligos used in this study. Supplemental Table S3: A list of DEG in lsd1-∆HMG and lsd2-∆C cells.