SETD1B controls cognitive function via cell type specific regulation of neuronal identity genes

Histone-3-lysine-4-methylation (H3K4me) is mediated by six different lysine methyltransferases (KMTs). Amongst these enzymes SET domain containing 1b (SETD1B) has been linked to intellectual disability but its role in the adult brain has not been studied yet. Here we show that mice lacking Setd1b from excitatory neurons of the adult forebrain exhibit severe memory impairment. By combining neuron-specific ChIP-seq, RNA-seq and single cell RNA-seq approaches we show that Setd1b controls the expression of neuronal-identity genes with a broad H3K4me3 peak linked to learning and memory processes. Our data furthermore suggest that basal neuronal gene-expression is ensured by other H3K4 KMTs such as Kmt2a and Kmt2b while the additional presence of Setd1b at the single cell level provides transcriptional consistency to the expression of genes important for learning & memory.


74
To study the role of Setd1b in the adult brain, we crossed mice in which exon 5 of the Setd1b 75 gene is flanked by loxP sites to mice that express CRE-recombinase under control of the 76 CamKII promoter. This approach ensures deletion of Setd1b from excitatory forebrain 77 neurons of the adult brain (cKO mice). Quantitative PCR (qPCR) analysis confirmed 78 decreased expression of Setd1b from the hippocampal Cornu Ammonis (CA) area, the dentate 79 gyrus (DG) and the cortex when compared to corresponding control littermates that carry 80 loxP sites but do not express CRE recombinase (control group). Expression in the cerebellum 81 was not affected confirming the specificity of the approach (Fig 1A). Residual expression of 82 Setd1b is most likely due to the fact that deletion is restricted to excitatory neurons while 83 other cell types are unaffected. In line with the qPCR data, SETD1B protein levels were 84 reduced in the hippocampal CA region of Setd1b cKO mice (Fig 1B). Setd1b cKO mice did 85 not show any gross abnormalities in brain anatomy as evidenced by immunohistological 86 analysis of DAPI staining, staining of marker-proteins for neuronal integrity Neuronal N 87 (NEUN), microtubule-associated protein 2 (MAP2) as well as ionized calcium-binding 88 adapter molecule 1 (IBA1) as a marker for microglia and glial fibrillary acidic protein 89 (GFAP) as a marker for astrocytes (Fig. 1C). Next, we subjected Setd1b cKO and control 90 mice to behavior testing. Notably, it was previously shown that heterozygous mice expressing 91 CRE under control of the CamKII promoter do not differ from wild type littermates (Kuczera 92 et al, 2010) (Stilling et al, 2014) and we have confirmed this in the context of the present 93 study also for behavior testing (Fig. S1). There was no difference amongst groups in the open 94 field test, suggesting that explorative behavior is normal in Setd1b cKO mice (Fig 1D). Short 95 term memory was assayed via the T-maze and was also similar amongst groups (Fig 1E).

96
Next, we subjected mice to the Morris Water Maze test to study hippocampus-dependent 97 spatial memory. While control mice were able to learn the task as indicated by a reduced 98 escape latency throughout the 10 days of training, Setd1b cKO mice were severely impaired 99 ( Fig 1F). We also performed a more sensitive analysis using a modified version of the 100 MUST-C algorithm to measure the different spatial strategies that represent either 101 hippocampus-dependent or independent abilities (Illouz et al, 2016). Our results indicate that 102 Setd1b cKO mice fail to adapt hippocampus-dependent search strategies such as "direct", 103 "corrected" and "short-chaining" (Fig 1G). Consistently, the cumulative learning score 104 calculated on the basis of these search strategies was severely impaired in Setd1b cKO mice 105 ( Fig 1H). To assess memory retrieval, a probe test was performed. Set1b cKO mice were 106 severely impaired during the probe test performed at the end of the training (Fig 1I). These 107 data show that deletion of Setd1b from excitatory neurons of the adult forebrain leads to 108 severe impairment of hippocampus-dependent learning and memory abilities. To elucidate the molecular mechanisms by which Setd1b contributes to memory 112 formation we decided to test its impact on epigenetic gene-expression in hippocampal 113 neurons. To this end we isolated the hippocampal CA region from Setd1b cKO and control 114 mice and prepared nuclei using modified fixation protocols that allowed us to perform 115 neuron-specific chromatin-immunoprecipitation (ChIP) to study histone-modifications and 116 RNA-sequencing to assay gene-expression from the same samples (Fig 2A, Fig S2). Since 117 SETD1B is a histone 3 lysine 4 (H3K4) methyltransferase we decided to analyze tri-118 methylation (H3K4me3) of histone 3 lysine 4 that is enriched at the transcription start site 119 (TSS) of active genes and is associated with euchromatin and active gene-expression. H3K4 120 methylation is believed to be a stepwise process and recent data suggest that the different 121 methylation states (from mono-to tri-methylation) at the TSS of a gene form a gradient we also analyzed mono-methylation of histone 3 at lysine 4 (H3K4me1). In addition, we 124 analyzed histone 3 lysine 9 acetylation (H3K9ac), an eu-chromatin mark that was shown to 125 partially depend on H3K4 methylation (Kerimoglu et al., 2013;Kerimoglu et al., 2017b).

126
Finally, we also performed Chip-seq for histone 3 lysine 27 acetylation (H3K27ac), another 127 euchromatin mark that is linked to active gene-expression and marks promoter elements 128 around the TSS but also enhancer regions and has not be directly linked to H3K4me3 in brain 129 tissue. We observed that loss of Setd1b leads to a substantial decrease in neuronal H3K4me3 130 across the genome while the majority of significant changes are localized to regions in close 131 proximity to the transcriptional start site (TSS) (Fig 2B, C). Similar changes were observed 132 for neuronal H3K9ac and H3K27ac, although less regions were affected when compared to 133 H3K4me3 (Fig. S3). We also observed significantly altered H3K4me1 in neurons of Setd1b 134 cKO mice (Fig 2B). These changes were also almost exclusively detected in vicinity to the 135 TSS ( Fig 2B) but in contrast to the other investigated histone-modifications, many of the 136 significantly altered genomic regions exhibited increased H3K4me1 levels in Setd1b cKO 137 mice (Fig. 2B, C). To further analyze these data, we first asked if the observed changes in showing decreased H3K27ac were mainly localized to different genes (Fig. S3). These data 143 support previous findings, showing that H3K4me3 is functionally linked to H3K9ac 144 (Kerimoglu et al., 2013;Kerimoglu et al., 2017b)and suggest that the observed changes in 145 H3K27ac are mainly due to secondary effects. Interestingly, decreased H3K4me3 in Setd1b 146 cKO manifested exclusively downstream of the TSS, indicating that loss of Sedt1b may affect 147 peak width (Fig 2D). We decided to further explore this observation and noticed that there 148 was an obvious difference amongst the genes that exhibit decreased H3K4me3 and increased 149 H3K4me1 (Fig. 2E) when compared to genes that show exclusively decreased H3K4 150 methylation around the TSS (Fig 2F). Namely, the change in H3K4me3 was most significant 151 in genes with decreased H3K4me3 and increased H3K4me1 and was characterized by a 152 substantially reduced H3K4me3 peak width (Fig 2E), when compared to genes with 153 decreased H3K4me3 and H3K4me1 (Fig 2F). Findings from other cell types suggest a 154 gradient of H3K4 methylation states in which the proximity of the mark to the TSS is 155 correlated to the level of gene-expression. Thus, genes with broader H3K4me3 peaks at the 156 TSS exhibit the highest and most consistent expression levels and represent genes of 157 particular importance for cellular identity (Benayoun et al, 2015; Soares et al., 2017). Indeed, 158 our data revealed that the genes which are characterized by decreased H3K4me3 and 159 increased H3K4me1 in Setd1b cKO mice, already exhibit significantly broader H3K4me3 160 peaks under basal conditions, when compared to genes characterized by decreased H3K4me3 161 but either decreased or unchanged H3K4me1 levels (Fig 2 G). Interestingly, these genes were 162 also expressed at significantly higher levels under baseline conditions (Fig 2 H). Taken 163 together, our findings suggest that Setd1b may be of particular importance for the expression 164 of genes linked to the specific function of hippocampal neurons. In line with this, functional 165 pathway analysis revealed that the genes with decreased H3K4me3 and increased H3K4me1 166 and thus having the broadest H3K4me3 peak under basal conditions, represent pathways 167 intimately linked to the function of excitatory hippocampal neurons (Fig 2I). Most 168 importantly, this was not the case for the genes of the other two categories (Fig 2I). In

175
To test the impact of Setd1b on gene-expression directly, we analyzed the RNA-sequencing 176 data obtained from neuronal nuclei of the same hippocampi used to generate ChIP-seq data 177 (See Fig 2A, Fig S2). In line with the established role of H3K4me3 in active gene-expression, 178 we mainly detected down-regulated genes when comparing control to Setd1b cKO mice ( Fig   179   3A). In fact, the comparatively few up-regulated genes were all lowly expressed at baseline 180 conditions suggesting rather unspecific effects (RPKM down-regulated genes =18.77 +/-1.45 181 vs. up-regulated genes RPKM = 3.08 +/-0.26; P < 0.0001). Further analysis revealed that the 182 TSS of genes down-regulated in Setd1b cKO mice is characterized by significantly reduced 183 H3K4me3 peak-width and increased H3K4me1 (Fig 3B, C). This observation was specific to 184 the genes down-regulated in Setd1b cKO mice, since random sets of genes that were not de-185 regulated in Setd1b cKO mice show normal H3K4me3 and H3K4me1 levels at the TSS ( Fig   186   3D). We also observed that the genes down-regulated as a result of Setd1b deletion were 187 characterized by a significantly broader H3K4me3 peak and higher expression under basal 188 conditions (Fig 3C, E). A functional pathway analysis revealed that the genes down-regulated 189 in Setd1b cKO mice are intimately linked to synaptic plasticity and learning and memory 190 related processes (Fig 3F). Taken together, these data further suggest that Setd1b controls a 191 specific set of genes that are characterized by a broad H3K4me3 peak at the TSS, are highly

215
To allow optimal comparison of these RNA-seq data to the gene-expression changes in 216 Setd1b mutant mice, we also performed bulk RNA-seq from the hippocampal CA1 region of 217 Setd1b cKO mice and control littermates. We observed 485 genes that were significantly 218 down-regulated when comparing control to Setd1b cKO mice (Fig 4D). These genes largely 219 overlapped with the down-regulated genes detected via neuronal specific RNA-seq in Setd1b 220 cKO mice (Fig S4). While the total number of genes differentially expressed in the 221 hippocampal CA1 region of Kmt2a, Kmt2b and Setd1b cKO mice was comparable, there was 222 little overlap amongst them (Fig 4E). Recently, RNA-sequencing data was reported for mice 223 that were heterozygous for Setd1a. Although these mutants were heterozygous constitutive 224 knock out mice and furthermore cortical tissue was analyzed instead of the hippocampus 225 (Mukai et al., 2019), it is interesting to note that there was virtually no overlap regarding the 226 genes down-regulated in Setd1a knock out mice, when compared to the data obtained from 227 our Setd1b cKO mice (Fig. S5). Further support for a specific role of Setd1b in neuronal 228 genes expression was revealed by the finding that the genes down-regulated in Setd1b cKO 229 mice exhibited a significant enrichment for neuronal identity genes, while this was not the 230 case for genes down-regulated in Kmt2a or Kmt2b cKO mice (Fig 4F). In line with these data 231 we observed that genes down-regulated in Kmt2a or Kmt2B cKO mice display decreased 232 H3K4me3 at the TSS, while the levels of H3K4me1 were unaffected (Fig 4G). In striking 233 contrast, only the genes down-regulated in Setd1b cKO mice were characterized by reduced 234 H3K4me3 and also increased H3K4me1 (Fig 4G). Consequently, the genes that exhibit

256
In our effort to further elucidate the specific role of Setd1b on neuronal gene-expression and 257 memory function we noticed that the levels of the H3K4 methyltransferases differ 258 substantially in hippocampal neurons of the adult mouse brain. Surprisingly, our RNA-seq 259 data from neuronal nuclei revealed Setd1b as the least expressed H3K4 methyltransferases 260 when compared to Kmt2a or Kmt2b (Fig 5A). These data might indicate that Sedt1b is 261 generally expressed at very low levels or that alternatively only few cells may express Setd1b, 262 a question that cannot be addressed on the basis our neuron-specific bulk RNA-seq. Thus, we 263 decided to perform single nuclei sequencing. We isolated the hippocampus from 3-month old 264 wild type mice and sorted NeuN + nuclei using our established protocol (Fig 5B). These 265 nuclei were then subjected to sequencing. As expected, we detected excitatory neurons of the 266 cornu ammonis (CA) and dentate gyrus region as well as inhibitory neurons (Fig 5C). Since 267 our analysis so far was focused on the hippocampal CA region of mice that lack Setd1b from 268 excitatory neurons, we selected the CA excitatory neurons and plotted the expression of 269 Kmt2a, Kmt2b and Setd1b. In line with the data obtained from bulk sequencing of 270 hippocampal neuronal nuclei, we observed that Kmt2a expression was most prominent when 271 compared to Kmt2b or Setd1b (Fig 5D). However, this difference was not due to the absolute 272 expression value per cell, which was comparable for all 3 KMT's ( Fig 5E). Rather, we   that these genes are also detectable in neurons that lack Setd1b and express Kmt2a or 390 Kmt2b, but to a lesser extent (See Fig 5I). This allows for some interesting 391 All of these studies found that only a small fraction of the originally activated cells 397 became reactivated during memory retrieval. It is thus tempting to speculate that the 398 activity of genes such as Setd1b might help to shape the neuronal ensemble that will 399 indeed be reactivated during memory retrieval, a hypothesis that would need to be 400 tested in further studies. Taking into account that decreased neuronal H3K4me3 levels 401 have been observed in cognitive and neurodegenerative diseases therapeutic strategies 402 that reinstate specifically the expression of neuronal plasticity genes controlled by 403 Setd1b might be particularly helpful. We suggest that the various epigenetic drugs 404 currently tested in pre-clinical and clinical settings for cognitive diseases should 405 especially be analyzed for their potential to reinstate the H3K4me3 peak width at 406 neuronal identity genes. 407 In conclusion, we show that Setd1b is essential for memory consolidation and ensures the            Fig S2).

799
Only for Setd1b cKO mice a highly significant odds ratio (Fisher's exact test) was observed,

860
Nuclei from the hippocampal CA region were subjected to FACS as depicted in Fig 2A. A.

861
Representative images showing nuclei that were sorted using the neuronal marker NeuN.