SIRT6-deficient brains show impaired expression of neuronal activity genes.
To characterize the transcriptional profile of the brS6KO brain tissue, we performed whole-brain RNA sequencing on brS6KO and WT mice (Fig. S1A). Gene enrichment against several datasets of the differentially overexpressed genes showed enrichment of categories in Gene Ontology (GO) related to neural activity and development (Fig. 1A, Table S1). These results suggest that SIRT6 is involved in regulating brain function and neuronal activity. To better understand the changes in cortical neurons specifically, we performed ATAC sequencing (ATAC-seq) on WT and brS6KO cortical neurons isolated from 10 month old mice, allowing us to recognize genomic regions accessible to transcription mechanisms (Fig. S1B, S1C). Enrichment of regions differentially accessed in brS6KO neurons reveals that GO categories associated with neural activity are significantly enriched (Fig. 1B, Table S2). Genes represented in both RNA-seq and ATAC-seq were mainly related to neural activity, morphology, and development (Fig. 1A, 1B).
SIRT6-KO differentially expressed genes are enriched for REST regulation.
Next, we asked whether these genes might be regulated by a common transcription factor (TF). To find potential TF candidates, we performed an enrichment analysis of the brS6KO cortical neuron ATAC-seq and the brS6KO RNA-seq against the ENCODE, and ChEA consensus TFs from a ChIPX database. We then performed a hypergeometric meta-analysis encompassing the enriched TFs to find the candidates overlapping both neuron chromatin accessibility and brain expression. Our results filtered three potential TFs: SUZ12, SMAD4, and REST (Fig. 2A). To narrow our focus, we tested their relevance to SIRT6 expression in the human brain by comparing the transcription profile of SIRT6 from six human brains (Allen Brain Atlas) to the profile of each TF13. Interestingly, the RE-1 Silencing Transcription factor (REST) correlates negatively with SIRT6 and, most significantly, in the brains of older individuals. The transcription factors SMAD4 and SUZ12 showed significant, but less prominent correlation with SIRT6 in most individuals. (Fig. 2B, Fig. S1A). REST is considered a repressive factor crucial for mediating the expression of neural development and activity genes containing the RE-1 element in their promoter14. Therefore, these results suggest that upregulated genes in brS6KO might be mediated through a lack of REST repression.
Next, we examined the proportion of REST targets among the upregulated genes in brS6KO. To do this, we overlapped the SIRT6 RNA-seq results with publicly available REST ChIP-seq data of SHSY-5Y cells5, and H1 cell line ChIP seq (GSM803365) (Fig. 2C, Fig. 2E). Over 100 genes upregulated in brS6KO are REST targets in the SHSY-5Y cell line, enriched in GO categories related to neural activity and development (Fig. 2D, Table S3). In H1 cells, more than 500 REST targets are represented in brS6KO upregulated genes and are associated with neural morphology and activity (Fig. 2F, Table S4). Altogether, our results suggest that the transcriptional alteration in brS6KO, specifically the upregulated neural genes, is mediated through REST.
REST is overexpressed in SIRT6-KO brains and cells.
REST is known as a neural gene repressor, and since its targets were upregulated, we expected to observe reduced levels of REST in brS6KO. Surprisingly, REST levels were higher in the brS6KO RNA-seq data (Fig. 3A). To validate these results in cells, we measured mRNA and protein levels of REST in the SHSY-5Y cell line with CRISPR deleted SIRT6 (SIRT6-KO). In these cells, REST was overexpressed in both mRNA and protein levels of chromatin extractions (Fig. 3B, 3C). In brS6KO mice, REST protein levels were increased in chromatin extracts (Fig. 3D).
To confirm that this overexpression is a product of SIRT6 absence, we introduced recombinant SIRT6 to the SIRT6-KO cells. mRNA measurement by qPCR shows rescue of REST levels with the introduction of SIRT6 (Fig. 3E), suggesting that the expression of this factor is SIRT6 dependent. To test if this is a conserved phenomenon and if we could use tissue culture to understand REST-related changes, we selected REST target genes from the RNA-seq results to measure in the SIRT6-KO cells (Fig. S3A). We measured mRNA levels for two representative REST target genes, SYN1 and GRIK2, and found that in both genes, mRNA levels were higher in SIRT6-KO cells, and reintroduction of SIRT6 to these cells rescued their expression (Fig. 3F). In the context of AD, SIRT6 expression declines with Braak stage progression10. To assess the levels of REST expression in AD, we used microarray data from AD and non-AD patients from GSE48350 database. By dividing the samples according to their Braak stage, we concluded that REST expression is higher in later Braak stages than earlier ones, suggesting loss of expression regulation (Fig S3B).
REST has several isoforms, some of which lack repressive capabilities and thus quench the function of the main isoform15,16. To test for the possibility of alternative splicing, we constructed probes that recognize the exon 2 and exon 3 (E2 + E3) exon junction (present in most of the transcripts) and the exon 3 and exon 4 (E3 + E4) exon junction (present in the full isoform), and we measured the abundance of the inactive splice variant relative to the whole isoform population (Fig. S3C). According to our measurements, there was no significant difference in the main REST isoform population between WT and SIRT6-KO cells (Fig. S3D), indicating that REST loss of function does not correlate with REST alternative splicing, and the lack of repressive capacity is not from the generation of different alternative spliced variants.
REST sub-cellular localization.
To address the hypothesis of REST loss of function in SIRT6 KO cells, we first tested changes in cellular localization. Previous studies showed that REST could be nuclear and cytoplasmic17. We hypothesized that changes in REST localization might hint at the causality of its impairment. To observe nuclear morphology and different nuclear regions, we performed immunofluorescence (IF) against the endogenous REST in SIRT6-KO cells (Fig. 3G). When measuring total cell intensity, REST intensity significantly increased in KO cells (Fig. 3G-H). Dividing the cellular area to its nuclear and cytoplasmic regions, REST intensity was higher in both compartments (Fig. 3H and S3E). Additionally, in the SIRT6-KO cells, REST was observed as condensates in the cytoplasm and closer to the cellular membrane, in addition to its nuclear location (Fig. 3G). This phenotype was not observed in WT cells. Quantification of these bodies shows that REST condensates accumulate in the cytoplasm of the SIRT6-KO cells (Fig. 3I). To test whether this phenomenon is reversible, we used an inducible shSIRT6 system that silences SIRT6 for 21 days using doxycycline, followed by 9 more days of recovery of SIRT6 levels (Fig. 3J and S3F). After the tenth day of silencing, REST cellular intensity increased significantly, in agreement with our previous results from SIRT6-KO cells. Interestingly, the recovery period did not reduce the mean nuclear intensity of REST, in addition to whole cell and cytoplasmic region measurements (Fig. 3J-M). Our previous results show rescue of mRNA levels of REST after 48 hours of SIRT6 reintroduction (Fig. 3E). However, in the inducible experiment, the intensity of REST does not decrease after SIRT6 recovery, suggesting that the phenotype is irreversible at the protein level, at least not within 10 days after SIRT6 re-expression.
To test whether this was due to high levels of REST leading to its aggregation, we overexpressed REST with Flag-REST plasmid under the CMV promoter, followed by immunofluorescence using Flag antibodies, which recognizes only the exogenous protein. In this experiment, we failed to see cytoplasmic REST bodies. To understand the discrepancy between exogenous REST and endogenous staining, we cloned a Flag-REST-GFP plasmid (Fig. S4A-B), since the antibodies are targeted for the C-terminus and Flag was in the N-terminus. Overexpression of this construct showed cytoplasmic aggregates (Fig S3G). It was recently published that REST accumulated in cytoplasmic autophagosomes5; therefore, we measured co-localization of REST with LC3. A fraction of REST co-localized with LC3 in both WT and SIRT6-KO cells, but there was a significant increase in SIRT6-KO cells with both Flag and GFP (Fig. 3N-O).
Last, we collected brains from mice at different ages, and separated proteins into chromatin bound and cytoplasmic fractions. With age, REST tends to localize more to the cytoplasmic fractions almost two-fold, while being depleted from the chromatin bound fraction. This suggests that the overexpression of REST first leads to its increase in all fractions, while slowly accumulating in the cytoplasm with age (Fig. 3SH-I). Note that in this method we are missing the nuclear fraction, unlike that for IF. SIRT6 has a declining trend with age but presents some variability (Fig. 3SJ).
REST is present at the nuclear Lamina.
Interestingly, we detected REST accumulating closer to the nuclear lamina, forming a ring almost co-localizing with Lamin B in WT cells (Fig. 4A-B). This was not previously seen for REST. In contrast, SIRT6-KO cells exhibited less colocalization of REST with the nuclear lamina (70% less) (Fig. 4C). When a ring was detectable, it was more homogenous and thicker inside the nucleus, measured by the co-localization of REST to the nuclear lamina and the thickness of the REST peak near the lamina (Fig. 4B and D-F), suggesting a loss of heterochromatin binding, or loss of chromatin condensation capability. Moreover, Flag and GFP staining co-localize in this ring, but not in all the cells, with GFP having a more homogenous nuclear distribution, and higher appearance in the cytoplasmic fraction (Fig. S4B). These results suggested that REST could be cleaved, separating the C and N terminus. To test that, we transfected Flag-REST-GFP and measured total protein extracts as well as cytoplasmic and chromatin bound fraction with Flag, GFP and REST antibodies. We can clearly see REST fragmentation overall (also with a higher band, only in total extraction). While Flag shows mainly a chromatin bound presence, GFP could be seen in the cytoplasm, as well as REST (which combined both exogenous and endogenous forms). This strengthens the point at which N-terminal REST is mainly nuclear and chromatin bound, while the C-terminus is cleaved and appears more in the nucleus as cytoplasm (Fig. S4F-H).
Next, we performed ChIP-seq to better understand the changes in REST targets in SIRT6-WT and KO cells. We confirmed a general enrichment for neuronal related genes (Fig. 4G). SIRT6 deficient cells have almost double the amount of peaks, probably due to higher levels in the cells (Fig. 4H), but with no difference in the distribution to promoter, intron or exon distribution (Fig. 4I). Importantly, in both WT and KO cells, a significant part of REST localizes to Lamin Associated Domains (LADs) confirming our IF results. However, the KO cells show a slightly higher proportion in the LADs (Fig. 4J). A small but significant fraction of REST co-localized with heterochromatic mark H3K27me3 in both cells (Fig. 4K),but had zero overlap with H3K9me3 mark. REST gained categories such as neurotransmitter transport and other ion activities, but lost monoatomic ion channel activity, possibly affecting neuronal activity (Fig. 4L).
REST interaction with EZH2 is abrogated in SIRT6-KO cells.
Since REST binding was not diminished in the SIRT6 deficient cells, we hypothesized that REST function as a scaffold protein15, recruiting various chromatin remodelers to its RE-1 element in neural promoters could be impaired, failing to repress its target genes. EZH2 is an H3K27 methyl transferase, part of the PRC2 silencing complex18. EZH2 is a known REST interactor and an important chromatin silencer in the brain18,19. Previous works showed that the interaction of REST and EZH2 is important for repressing REST target genes20. Interestingly, 190 genes upregulated in brS6KO are EZH2 targets (Fig. 5A) (Using publicly available hESC ChIP-seq data). To better understand REST and EZH2 gene co-regulation in SIRT6 KO cells, we analyzed the EZH2 targets together with the REST consensus peaks from our ChIP-seq data (Fig. S5A). Functional enrichment analysis of REST and EZH2 targets revealed that overlapping pathways are neuron related (Fig. S5B-C). We then checked if upregulated REST targets in brS6KO are also targeted by EZH2. A permutation test showed that REST and EZH2 regulate 16 genes in WT cells that are upregulated in brS6KO (Fig. 5B), and 21 genes in SIRT6 KO cells that are upregulated in brS6KO (Fig. 5C). Functional similarity between these genes allows significant enrichment in GO categories associated with neural pathways (Fig. 5E). Due to the small coverage of our ChIP-seq, we analyzed data from hESC for EZH2 and REST, and found that they overlap in about 500 genes, with 51 being also overexpressed in SIRT6-KO brains (Fig. 5D). Genes were enriched for synaptic and brain functioning, similarly to the ATAC-seq and RNA-seq results (Fig. 1A-B). To confirm these bioinformatic results, we tested the levels of EZH2 in SIRT6 deficient cells. Although total levels of EZH2 were not altered in the total cell extract (Fig. 5F and S5G), the chromatin bound fractions were reduced (Fig. 5G and S5E).
To determine if the interaction between REST and EZH2 is affected in the absence of SIRT6, we performed Co-Immunoprecipitation (Co-IP) of the recombinant Flag-REST expressed in WT vs. SIRT6 KO SHSY cells. We found a significant reduction in the interaction between REST and EZH2 in SIRT6-KO cells (Fig. 5H). Although REST did not show differential binding between WT and SIRT6 KO cells in our ChIP-seq (Fig. 4K-I), the impaired interaction with EZH2 led us to hypothesize that there are less H3K27me3 in regions bound by REST. To test this, we performed Co-IP against H3K27me3 from chromatin preps in WT and SIRT6 KO SHSY-5Y cells, and found that REST is less enriched in H3K27me3 in SIRT6 KO cells (Fig. 5I). These results suggest that, although REST levels are increased and its DNA binding capability not impaired, interaction with key chromatin remodelers and epigenetic markers such as EZH2 and H3K27me3 are disrupted, leading to less chromatin condensation in REST bound loci.
To assess the loss and gain of REST interaction depending on SIRT6 genotype, we overexpressed recombinant REST in WT and SIRT6KO HEK293T cells and sent the triplicate samples to mass spectrometry. We obtained 141 proteins that pass the filter (excluding non-specific protein reads that were present in the empty vector sample threshold and in 3 independent IPs). Our results show one unique interaction for REST in WT and nine novel interactions in SIRT6-KO cells. The unique interactions were defined as proteins present specifically in the WT or KO samples (Fig. S6A). In addition, we performed a differential interaction analysis. We discovered a small but significant number of proteins (9) that are gained in SIRT6-KO cells, mainly related to translation, while the lost interactions (5) were related to DNA repair and remodeling, potentially affecting chromatin condensation function (Fig. S6B-C). STRING analysis revealed that REST interacts with proteins involved in histone and chromatin remodeling (Fig. S6D-E). For our lost and gained interactions, we performed STRING analysis and allowed the addition of a second layer of interactors to better define the functionality of these proteins (Fig S6F-G). Enrichment analysis of the cellular components of these networks revealed that, in SIRT6 KO cells, REST loses interaction with proteins involved in DNA repair, while gaining interaction with translation related complexes - which are chiefly cytoplasmatic (Fig S6H). These results support the fact that, in SIRT6 KO, REST exists more prominently in the cytoplasm and loses its function in the nucleus by losing key interactors. However, we cannot discard the possibility that these changes could be the result of different protein content in a SIRT6-deficient background.
Acetylation of REST on Lysine 494 results in loss of association with the heterochromatin.
It was shown that EZH2 methylates REST on Lysine 494 (K494), increasing the interaction between these two proteins, and this modification is critical for the repression of REST target genes20. Since the interaction with EZH2 is diminished, we hypothesized that methylation could be impaired by the presence of different PTMs, such as acetylation. Since there are no commercially available antibodies for K494Ac, we performed immunoprecipitation (IP) of Flag-REST in WT and KO cells, and measured pan-acetyl-Lysine antibody (Fig. 6A). We discovered that, not only is REST acetylated, but is even more so in the absence of SIRT6. The reduced interaction with EZH2 and the existence of acetylated REST led us to hypothesize that the K494 residue might be responsible for these alterations. To check this, we generated a K494A REST mutant. Overexpression of REST variants shows that WT-REST is hyperacetylated in KO cells, while the non-modifiable K494A is generally less acetylated in WT cells and shows no change in KO cells, indicating that this is the main change occurring in SIRT6-KO cells. A similar trend is observed in the mutated K494R (Fig. 6B). These results suggest that SIRT6 mediates the balance between acetylated or methylated REST. To test the proportion of acetylated or methylated REST in SIRT6 KO cells, we performed immunoprecipitation using pan-methyl lysine (MeK) and pan-acetyl lysine (AcK) antibodies.
SIRT6 KO cells transfected with WT REST showed more REST enrichment when using AcK antibody (Fig. 6C), and less when using MeK (Fig. 6D), suggesting that REST is more likely to get acetylated than methylated in SIRT6 KO cells. In addition, cells transfected with the non-modifiable REST K494A mutant using MeK antibody showed less enrichment of REST, regardless of SIRT6 genotype - while in AcK, REST enriched similarly to the control. These results suggest that, in addition to the differential proportion of Me-REST and Ac-REST in SIRT6 KO cells, lysine 494 has a critical role in influencing REST PTM.
To understand if REST can become toxic under the different PTMs, we transfected the cells with the different REST variants and Apo-Alert Active caspase 3 detector. Interestingly, WT-REST was no different between SIRT6-WT and KO cells, but the different REST variants, particularly 494M and a lesser extent 494Q could rescue the WT cells but became more toxic in the KO cells (Fig. 6E). Suggesting that in the presence of SIRT6, REST can be neuroprotective, but in its absence, as it happens during AD, they become toxic.
REST K494 PTMs affect its nuclear location.
To assess if K494 affects REST nuclear localization, we mutated WT REST in K494 to alanine (K494A), methionine (K494M) and glutamine (K494Q). K494M and K494Q serve as PTM mimics for methylation and acetylation in this lysine residue, respectively, while alanine is a non-modifiable mutation. REST WT and the point mutants were transfected to WT and SIRT6 KO SHSY-5Y cells (Fig. 7A-D, S7A). Both WT, K494A, K494M - and K494Q REST - appear in the nucleus 48 hours after transfection. As previously shown, WT REST co-localizes closely to the nuclear lamina in WT cells, while a more diffuse pattern is seen in SIRT6-KO cells, losing the ring feature (Fig. 7A). Interestingly, K494A REST fails to localize to the nuclear lamina in both WT and SIRTKO cells (Fig. 7B). The methyl mimic K494M showed higher association with the nuclear periphery even in KO cells (Fig. 7C), while the acetyl mimic showed lower association even in SIRT6-WT cells (Fig. 7D). This indicates that K494 methylation induces its Lamin-associated location, while non modifiable or acetylated REST shows impaired recruitment to the nuclear envelope. In addition, we measured the width of the peaks that the REST mutants generate. SIRT6 genotype did not affect the width of peaks of the K494A and K494Q mutants, probably because of the decreased accumulation of these mutants to the nuclear lamina (Fig. S7C-D). In the K494M mutant, the width of the REST peak was wider in KO than in WT cells (Fig. 7B), although the laminar colocalization value improved in the KO cells. This might suggest that although REST is more localized to the nuclear lamina when methylated, it still depends on the presence of SIRT6 to have a sharper localization, probably through other effects of SIRT6 on nuclear lamina function.
To understand the changes in binding activity, we performed ChIP-seq using anti-Flag antibodies expressed REST WT, and the K494A, M and Q mutants in SHSY cells. Enrichment analysis of consensus peaks between all the mutants showed high enrichment of processes and pathways associated with neural activity (Fig. 7E-F). Once again, the different REST mutants did not affect the distribution of REST to the different regions such as promoter, introns, exon etc., similarly to what we saw with endogenous REST in SIRT6 WT and KO cells (Fig. 7E). We performed an analysis with 2 out of 3 ChIPs presenting the peak and a more stringent test with the 3 peaks present. Both analyses show the same trend (Fig. S7D-F). Interestingly, the K494M mutation bound ~ 4 times more unique targets than K494Q, and ~ 22 times more than K494A. Additionally, K494M shares more targets with REST WT than with K494Q (~ 3 times) and K494A (~ 8 times) (Fig S7D). These results suggest that under normal conditions REST is methylated in the cells, allowing improved binding, while non modifiable or acetyl mimic shown an impairment.
The difference between the point mutants is mainly in the amount of sites they can bind, but enrichment the for different categories is also affected. For example, WT binds the monoatomic channels, neurotransmitter and synaptic activity. Interestingly, only WT and acetyl mimic bind glutamate receptor activity, one of the enriched categories in the DE expressed genes in brS6KO - particularly the alanine mutant loss and monoatomic ion gated channel activity. Methyl mimic lost some categories related to glutamate receptor, symporter and general transmembrane transport, but gained guanyl nucleotide exchange factor activities. One unique acetyl mimic category was syntaxin binding, also related to SNARE binding enriched in the Alanine mutant, suggesting an overall important role for REST in membrane constitution - from ion channels to membrane composition and endocytosis (Fig. 7G).
Moreover, we see that the proportion of REST in LADs is conserved among the mutants, strengthening the notion that the plasmids are indeed binding the correct sequences as the endogenous REST, and that it is not an artifact of the Flag (Fig. 7H).
Rockowitz et. al21 showed that REST associates mainly with the transient heterochromatin mark H3K27me3 in neurons and less with the more permanent H3K9me3 mark. We see the same results here; although with smaller numbers, it is significantly enriched (Fig. 7I).
The association of REST K494M with the nuclear periphery led us to hypothesize that this mutant might help to repress REST target genes, even in the absence of SIRT6. To check this, we introduced WT or K494M REST to SIRT6 normal or deficient cells and measured the mRNA levels of REST targets SYN1 and GRIK2 (Fig. S7G). As shown previously, SIRT6 KO influences the expression of REST targets even when overexpressed in its WT form. However, REST K494M rescues the level of these target genes, even in the context of SIRT6 deficiency. These results not only suggest that the methylation of REST is important for its repressive function, but also that SIRT6 regulates REST function through the K494 residue acetylation/methylation switch.