Site-speci c gains and losses of heterochromatin accelerate the age-related neurodegeneration through the cascading destruction of KDM3B- centered epigenomic network

Mi-Jin An Chung-Ang University Ji-Young Kim Jinhong Park Chung-Ang University Jinho Kim Chung-Ang University Dae-Hyun Kim Chung-Ang University Geun-Seup Shin Chung-Ang University Hyun-Min Lee Chung-Ang University Ah-Ra Jo Chung-Ang University Chul-Hong Kim Chung-Ang University Mi Jin Kim Chung-Ang University Jeongkyu Kim Chung-Ang University Sangmyung Rhee Chung-Ang University Sang-Beom Seo Chung-Ang University Jung-Woong Kim (  jungkim@cau.ac.kr ) Chung-Ang University https://orcid.org/0000-0003-4458-7213


87
Characterization of heterochromatin organization and transcriptomic profile involved in retinal 88 aging. The retina is composed of six neuronal cell types, including rod and cone photoreceptors (Fig. 1a). 89 To investigate the organization pattern of chromatin during retinal aging, we observed the distribution of 90 heterochromatin in the nuclei of photoreceptors using DAPI staining. Heterochromatin was detected at the 91 nuclear periphery and within chromocenters in cone photoreceptors of 2-month-old mice, but less 92 condensed heterochromatin was observed in the center of the nucleus in 18-month-old mice (Fig. 1b). 93 However, the distribution of heterochromatin did not change in the rods of both 2-and 18-month-old mice. 94 The number and mean area of chromocenters in cone nuclei decreased in the 18-month-old mice compared 95 to that in the 2-month-old mice ( Fig. 1c and Extended Data Fig. 1a). Several studies have shown that 96 repressive histone modifications, such as H3K9me2/3, are hallmarks of pericentromeric heterochromatin 97 that play a pivotal role in heterochromatin formation 15,16 . We also observed the distribution of silencing 98 histone modification (H3K9me1/2/3) and active modification (H3K9ac and H3K4me3) in cones and rods 99 during aging. H3K9me2/3 was abundant in the heterochromatin at the nuclear periphery in 2-month-old 100 mice with conventional nuclear architecture, whereas H3K9me2/3 showed unusual distribution in both 101 euchromatin and heterochromatin (but not in chromocenters) in 18-month-old mice (Extended Data Fig.   102 1b). We did not observe any differences in the distributions of H3K9me1 and active modification between 103 2-month-old mice and 18-month-old mice (Extended Data Fig. 1b). Furthermore, we confirmed the global 104 histone modification of whole retina using immunoblotting (Extended Data Fig. 1c). H3K9me2 level was 105 slightly decreased in 18-month-old mice, whereas H3K9me3 level was increased in whole retina of 18-106 month-old mice. These data indicate that genome-wide organization of chromatin is altered during retina 107 aging. 108 To investigate whether epigenetic regulators are involved in "loss of heterochromatin features" 109 during neuronal aging, we performed gene expression profiling using RNA-sequencing analysis of 1-, 6-, 110 functions in cone photoreceptor cells (Extended Data Fig. 3g). 136 To demonstrate the functional role of Kdm3b in the eye, we observed the morphology and the 137 structure of the enucleated eyeball and retinal tissue. We clearly detected smaller eyes and abnormal corneal 138 phenotypes in 12-month-old Kdm3b +/mice compared to those in 1-month-old mice (Extended Data Fig.   139 3h). To investigate the structural and morphological differences in the retina, we measured the thickness of 140 the entire retinas of 1-, 6-, and 12-month-old Kdm3b +/+ and Kdm3b +/mice. The retinal outer nuclear layer 141 (ONL) of the Kdm3b +/mice was thinner than that of the Kdm3b +/+ mice (Fig. 2a, and Supplementary Fig.   142 1a,b). The mean ONL thickness of Kdm3b +/mice was significantly reduced in 12-month-old mice (1,600 143 m inferior: 97 ± 10 m versus 57 ± 2.95 m, 1,600 m superior: 84.6 ± 9.35 m versus 54.7 ± 2 m) ( Fig.   144 2b). However, there was no difference between the mean ONL thickness of 1-month-old Kdm3b +/+ and 145 Kdm3b +/mice (Fig. 2a, b). Consistent with these results, we observed that the number of nuclei in the ONL 146 was markedly reduced in Kdm3b +/mouse retinas among both 6-month-old (266 ± 3.06 versus 239 ± 2.34 147 nuclei/100 m 2 , P = 1.93E-5) and 12-month-old mice (239 ± 3.38 versus 205.1 ± 3.27 nuclei/100 m 2 , P = 148 9.0E-6) (Fig. 2c). We also observed that the mean thickness of the inner nuclear layer (INL) and the number 149 of nuclei in the INL and ganglion cell layer (GCL) were decreased in 1-and 6-month-old mice (Extended 150 Data Fig. 4a, b). The decrease in ONL thickness in the Kdm3b +/-retina could be because of reduced number 151 of photoreceptor cell nuclei, including those of the rods and cones. To determine whether the rod and cone 152 populations decreased in the ONL of the Kdm3b +/-retina, we performed IHC with cone-cell specific (anti-153 CAR; cone arrestin) and rod-cell specific (anti-RHO; rhodopsin) antibodies. We found that Kdm3b 154 haploinsufficiency induced different morphologies and reduced the number of cones (14.6 ± 0.41 versus 155 9.8 ± 0.26 cones/100 m 2 P = 4.93E-34) in 12-month-old mice, but it was not associated with differences 156 in the rods (Fig. 2d, and Extended Data Fig. 4c).

157
Vertebrate rod and cone photoreceptors include outer segments (OS) and cell bodies that contribute 158 to phototransduction, are required for cell viability, and a synaptic terminal able to signal to second-order 159 neurons 18 . To determine the progression of cone loss in aged Kdm3b +/-retinas, we examined the density of cones and the morphological changes of cone OS by immunostaining analysis of whole mouse retinas.

161
Consistent with that of retina sections, the mean number of cones was significantly decreased in 12-month-162 old Kdm3b +/-mouse retinas, but the mean diameter of the cone-cell bodies was not significantly different 163 between Kdm3b +/+ and Kdm3b +/-mice (Fig. 2e, and Supplementary Fig. 1c,d). The mean length of the inner 164 segment of cones was significantly decreased in 12-month-old Kdm3b +/-mouse retinas (Fig. 2f). However, 165 the mean circumference of the cone OS was similar between Kdm3b +/+ and Kdm3b +/-mouse retinas (Fig.   166 2f). Even though the overall mean cone lengths were similar, the mean lengths of the basal side and cone-167 cell body in the Kdm3b +/mouse retinas were longer than those of the Kdm3b +/+ mouse retinas (Extended 168 Data Fig. 4d). Interestingly, we observed that the mean presynaptic area of cones (pedicle) was smaller in  Transcriptomic profiling of Kdm3b +/-mouse retinas. After determining the morphological changes in the 174 Kdm3b +/mouse retinas, we analyzed differential transcript expression in Kdm3b +/-retina. We performed 175 transcriptome profiling of 12-month-old Kdm3b +/+ and Kdm3b +/-mouse retinas by RNA-sequencing 176 analysis. Principal component analysis (PCA) plot showed that Kdm3b +/-retinas accounted for the largest 177 variance, and biological replicates showed great reproducibility (Fig. 3a). Volcano plots present the 178 statistical significance of differential transcript expressions with respective fold-changes (P < 0.05, absolute 179 log2 fold-change (log2 FC) > 0.5) compared to the expression observed in the control group (Fig. 3b).  in Kdm3b +/-retinas enabled their classification into several categories that were associated with mRNA 187 splicing and DNA damage stimuli, and downregulated DEGs were involved in synapse assembly and 188 vesicle transport (Fig. 3d). Gene set enrichment analysis (GSEA), which shows the biological pathways 189 and processes of the DEGs, also revealed that genes associated with cellular responses to DNA damage 190 stimuli and apoptotic processes were upregulated, while the repressed genes in Kdm3b +/-mouse retinas were 191 involved in synapse assembly and vesicle-mediated transport (Fig. 3e). The selected genes in these groups   The mean Müller cell counts were significantly increased in 6-and 12-month-old Kdm3b +/-retinas compared to that in age-matched Kdm3b +/+ retinas (Fig. 4a, and Extended Data Fig. 6a). To further confirm 211 whether retinal cell damage induces photoreceptor cell death through apoptosis, we stained 1-and 12-212 month-old Kdm3b +/+ and Kdm3b +/retinas with the executive apoptotic marker, cytochrome c. The intensity 213 of cytochrome c was significantly increased in 12-month-old Kdm3b +/-retinas compared to that in 1-month-214 old Kdm3b +/-retinas. However, this was not observed in 12-and 1-month-old Kdm3b +/+ retinas (Fig. 4b). 215 Interestingly, we observed an increased release of cytochrome c in the cones of 12-month-old Kdm3b +/-216 retinas (Fig. 4c, and Supplementary Fig. 2a). Furthermore, we used TUNEL staining to determine whether  old Kdm3b +/retinas (3.8 ± 0.35 versus 2.9 ± 0.27, P = 0.0285) (Fig. 5c). These findings show that KDM3B 241 is required for the synaptic components and the visual acuity associated with cone photoreceptors.  Site-specific regulation of H3K9 methylation patterns by KDM3B in mouse retinas. Because KDM3B is required for the physiological and functional roles of cones, we performed ChIP-seq analysis to determine 261 whether KDM3B was directly involved in gene expression via histone demethylase activity. Over 32 262 million reads were obtained for KDM3B ChIP-seq; these were aligned with the mouse reference genome 263 (GRCm 38.99), and peak calling was performed (Supplementary Table 5). The majority of KDM3B peaks 264 were globally found within 100 Kb from the transcription start site (TSS) (Fig. 6a). The distribution of 265 KDM3B-binding peaks signified that the majority of KDM3B was localized in intergenic regions (50%) 266 and introns (42%) (Fig. 6b). To further test the functional importance of KDM3B in gene expression 267 regulation, we performed the GO analysis of the KDM3B-binding peaks. Interestingly, the KDM3B ChIP-268 seq signals were highly enriched for genes involved in nervous system development, such as synapse 269 assembly and organization, neuron generation, and differentiation ( Fig. 6c and Supplementary Table 6).

270
Previous studies have determined that KDM3B has demethylase activity for H3K9me1 and manner. Also, the enrichment of H3K9me2 in cluster 3 proceeded as a demethylase-independent function 284 of KDM3B (Fig 6e). The scatter-density plot of the H3K9me2 level at the TSS between Kdm3b +/+ and Kdm3b +/mouse retinas revealed a markedly negative correlation (Fig. 6f). Furthermore, using GO analysis, 286 we found that the H3K9me2 enriched target genes in Kdm3b +/retinas (cluster 2), indicating that repressed 287 genes are involved in synaptic signaling, synapse organization, and chemical synaptic transmission (Fig.   288 6g). On the other hand, the H3K9me2 peaks in Kdm3b +/+ (cluster 3) were enriched for negative regulation 289 of apoptosis and signal transduction (Fig. 6g). We obtained similar results for the global pattern of 290 H3K9me1 peaks and biological processes by GO analysis (Fig. 6h-j). Together, these ChIP-seq analyses 291 demonstrated that KDM3B is associated with H3K9me1 and H3K9me2, suggesting that KDM3B 292 exclusively regulates gene sets between neuronal homeostatic genes (synaptic transmission, vesicle 293 transport) and cell-death related genes.  Table 7). We next focused on the 803 downregulated genes in Kdm3b +/mice to determine 305 whether gene expression was modulated by the histone demethylation activity of KDM3B. H3K9me1 and 306 H3K9me2 levels were significantly increased in association with downregulated genes in Kdm3b +/mice, 307 revealing the deposition of histone methylation due to the ablation of KDM3B enzymatic functions (Fig.   308 7g). Additionally, H3K9me3 level was slightly increased in some sets of downregulated genes (Fig. 7g). 309 downregulated in Kdm3b +/mice (Fig. 7h). Based on ChIP-seq data, we selected ten target genes, such as 311 synapse assembly genes and apoptosis-induced genes for validation by ChIP-qPCR. KDM3B was enriched 312 in the target gene promoters (within ~2 Kb from TSS) representing the normal range of ChIP-seq peaks for 313 H3K9me1 and H3K9me2 in Kdm3b +/+ , and the signal was significantly elevated in Kdm3b +/-mouse retinas 314 (Fig. 7i, and Extended Data Fig. 9, 10). These results suggest that KDM3B modulated the transcriptional 315 dynamics of synaptic and apoptotic genes by balancing the H3K9 methylation status.  Since the AP-1 family regulates the various cellular processes, including cell proliferation, 330 differentiation, and apoptosis 23 , it is reasonable to speculate that KDM3B modulates neuronal homeostatic 331 genes and cell death-related genes by interacting with AP-1 transcription factors. To test this hypothesis, 332 we first confirmed that AP-1 family (c-Fos, c-JUN, and JUNB) mRNA and protein levels were stably 333 expressed up to 12 months in mouse retina (Extended Data Fig. 11b, c). Transcriptome profiling showed 334 that c-Fos and JUNB mRNA levels increased in Kdm3b +/-mice retina (Extended Data Fig. 11d). In addition, public mRNA-sequencing data (GSE 74660) from mice retina tissue showed that the expression of c-Fos, 336 c-JUN and JUNB genes was higher in cone than in rod photoreceptors (Extended Data Fig. 11e). Next, we 337 analyzed whether AP-1 transcription factors were involved in KDM3B recruitment in vivo. Finally, co-338 immunoprecipitation experiments showed that KDM3B was associated with c-Fos and JUNB in mice retina 339 and human lung cancer cells (Fig. 8b and Extended Data Fig. 11f). Additionally, ChIP-qPCR with c-Fos 340 and JUNB antibodies confirmed that both transcription factors were associated with KDM3B target genes 341 (Fig. 8c, and Extended Data Fig. 11g). These results suggest that KDM3B, c-Fos, and JUNB can physically 342 interact with and could regulate a set of KDM3B target genes. ages of Kdm3b +/+ mouse retinas (1.5, 3, 6, and 12 months) compared to those of Kdm3b +/mouse retinas.

357
Network robustness is a central question in systems biology and medicine, helping us understand 358 how subtle changes or failures lead to the development of disease 26 . Therefore, we measured the network 359 robustness between the Kdm3b +/+ and Kdm3b +/genotypes by random link removal. We simulated random 360 link removal with a fraction and measured the fraction of the largest cluster size ( ( ) ⁄ (0)) as an 361 order parameter (Fig. 8f). Since all epigenomic networks of KDM3B are based on centralized KDM3B 362 networks and finite-sized networks, we can predict that the critical threshold goes to 1 27 . Therefore, we 363 focused on the difference of ( ) ⁄ (0) for various values of . The order parameters of all four 364 Kdm3b +/+ network stages showed more robustness for link failure than those of Kdm3b +/- (Fig. 8f). These reduced in Kdm3b +/mouse retina by regulating apoptotic genes ( Fig. 2 and 4). Furthermore, KDM3B plays 415 an important role in synapse assembly and vesicle transport in cone ribbon synapses through multi-416 dimensional epigenomic network analysis (Fig. 8d-f). We suggest that the cone-cell type specificity and KDM3B functional network can cause age-related diseases, such as AMD. Future studies will demonstrate 418 whether KDM3B is associated with cone-cell type specificity in the human retina. Kdm3b +/mouse retinas showed that the synapse assembly genes including Rims2, Nrxn2, CtBP2, Nedd4, 434 and vesicle-transport genes, including Tmed10, Vamp4 were downregulated by increasing H3K9 435 methylation (Fig. 7). This suggests that the presynaptic terminal abnormalities in cones proceed from 436 epigenetic modification complications followed by incorrect assembly of ribbon synapses. GRNs affect all biological phenomena associated with maturation (development), rapid 458 destruction (disease), and slow destruction (aging) (Fig. 8g). In terms of network topology, conventional 459 GRNs cannot differentiate between subtle changes, such as environmental adverse effects and aging. We properties are not significantly different between Kdm3b +/+ and Kdm3b +/mice (Fig. 8d), we found interior 480 topological differences by measuring network dissimilarity and robustness (Fig. 8f). Eventually, Kdm3b    Here, represents Jensen-Shannon divergence. The first term represents dissimilarity in average node 657 connectivity is the graphs averaged node-distance distribution. The second term shows dissimilarity in 658 a node dispersion metric ( ( ) ). And the last term denotes the difference for distribution of - Image analysis. The number of nuclei in ONL was counted on the middle of the Kdm3b +/+ and Kdm3b +/-664 mice retina. The thickness of ONL was measured using the 'length measurement' plugin of ImageJ software.

665
The number of cone photoreceptors was counted in the middle of Kdm3b +/mouse inferior retina. Size of 666 cone photoreceptor synapse was measured by using the 'area measurement' plugin of ImageJ software.

667
Intensities of GFAP and cytochrome c were measured using the 'histogram' plugin of ImageJ software. The