H2A.Z-dependent and -independent recruitment of metabolic enzymes to chromatin required for histone modifications

H2A.Z plays a fundamental role in the regulation of transcription and epigenetics, however, the mechanisms that underlie its functions are not fully understood. Using rapid chromatin immunoprecipitation-mass spectrometry, we uncovered the association of H2A.Z-bound chromatin with an array of tricarboxylic acid cycle and beta-oxidation enzymes in the mouse heart. Recombinant green florescence fusion proteins combined with mutations of putative nuclear localization signals of select enzymes, including acetyl-CoA acyltransferase 2 (ACAA2), oxoglutarate dehydrogenase (OGDH), and isocitrate dehydrogenase 2 confirmed their nuclear localization and chromatin binding in both rodent and human cells. Conclusively, chromatin immunoprecipitation-deep sequencing, confirmed the selective association of ACAA2 and OGDH with H2A.Z-occupied transcription start sites. Finally, human H2A.Z-deficient HAP1 cells exhibited reduced chromatin-bound metabolic enzymes, with the exception of pyruvate dehydrogenase, accompanied with reduced posttranslational histone modifications. Thus, the data show that metabolic enzymes are recruited to active promoters for potential site-directed epigenetic modifications.


26
The highly conserved, histone variant, H2A.Z gene is unique in many ways when contrasted with those of core 27 histones. Mainly, it is a single copy gene that does not exist within the histone clusters known in human and 28 mouse genomes, it includes introns, and is polyadenylated 1 , all of which underscore its specialized nature. We 29 currently know that it selectively associates with transcriptionally active, as well as, inactive genes. For example, 30 in yeast, Htz1 has been shown to suppress the spread of heterochromatin into transcriptionally active genes 31 near the telomeres 2 , while, in contrast, its abundance was shown to negatively correlate with transcriptional 32 rates 3 . Furthermore, changes in growth conditions induced translocation of Htz1 from transcriptionally active to 33 inactive genes 4 . More precisely, Htz1 is found at the transcription start site (TSS) of nearly all genes in 34 euchromatin, in the -1 and +1 nucleosomes flanking a nucleosome free region in the active genes, while present 35 mainly in the -1 nucleosome in inactive genes 5 . Likewise, in Drosophila, H2Av is present at thousands of both 36 transcriptionally active and inactive genes in euchromatin, as well as, in heterochromatic chromocenter of 37 polytene chromosomes 6 , whereas its density negatively correlates with that of RNA polymerase II (pol II).

38
Conversely, other studies have shown that H2A.Z vs. H2A at the +1 nucleosome facilitates pol II progression 7 .

39
In murine embryonic stem cells, we also see this bifunctionality, where in the undifferentiated state, H2A.Z, in 40 conjunction with the polycomb subunit Suz12, is present at silenced homeodomain genes involved in 41 differentiation, whereas in committed neuronal progenitor cells, its associates with highly expressed genes 8 .

42
Other than destabilizing nucleosomes at the +1 position, we have little understanding of how H2A.Z selectively 43 regulates transcriptional activation v. deactivation, or if it has any role in metabolism-induced transcriptional 44 remodeling.
that while tGFP was predominantly located in the cytosol and to a minimal extent in the mitochondrial/membrane 135 (mito/mem) fraction, its fusion with ACAA2, OGDH, or IDH2, resulted in its redistribution, to include the nucleus 136 and chromatin-bound protein fractions upper 2 panels). Likewise, the endogenous enzymes were 137 detected in the mito/mem fraction, which contains the mitochondria, as confirmed by VDAC1, in addition to the 138 nuclear and chromatin fractions, as confirmed by TFIIB and histone H3 (Fig. 3a-c, second panels). Ultimately, 139 substitutions of key lysine residues with glutamine in the putative NLS of ACCA2 (mtACAA2) or OGDH 140 (mtOGDH) significantly reduced their nuclear import and chromatin associations, proving that this localization is 141 specific and requires an NLS . Note, the chromatin bound proteins were not subjected to 142 crosslinking, thus, only the directly-or tightly-associated proteins are retained in this fraction, such as the 143 histones and RNA polymerase II. The ratios of ACAA2-tGFP in the different fractions show that % of total wild 144 vs. NLS-mutant protein is lower in the cytosol and mitochondria, and higher in the nuclear and chromatin fractions 145 as the translocation of the mutant to the nucleus is reduced in the latter (Fig. 3a, 3d-f). On the other hand, the 146 mtOGDH-tGFP protein was predominantly in the mitochondrial, while minimally detected in the nucleus vs. the 147 wild type OGDH-tGFP (Fig. 3b,. Similar results were observed in human colon cancer cells (supplementary 148 Fig. 4S). Interestingly, the levels of the ACAA2-tGFP protein were increased when the cells were incubated with 149 palmitate vs. glucose, particularly in the nucleus and chromatin-bound fractions (supplementary Fig. 4S).

150
Additionally, the metabolic enzymes were also detected in the nucleus of the mouse heart tissue and isolated 151 myocytes, although there was no significant difference when growth-induced with pressure overload or 152 endothelin-1 (supplementary Fig. 5S). These data confirm that mitochondrial enzymes reside in the nucleus in 153 significant concentrations and that at least ACAA2 and OGDH harbor NLSs that mediate their nuclear import.

154
The data also suggest that nuclear and chromatin enzymes are limiting and subject to regulation by metabolic 155 substrates.

156
OGDH and ACAA2 associate with H2A.Z-bound transcription start sites. The co-precipitation of Expression Omnibus (GEO) datasets (accession pending). The results show that ACAA2 and OGDH, similar to 163 H2A.Z 17 , predominantly associate with TSSs, as observed with heatmaps of the sequence tags and the curves 164 of the average signals aligned to a region encompassing -2000 to +2000 bp from the TSS, with no substantial 165 differences observed between their total levels in the normal v. growth-induced hearts ( Fig. 4a-b). Additionally, 166 the binding of the ACAA2 and OGDH coincided with that of H2A.Z (when analyzed over the length of the gene 167 and including -2000 bp upstream of the TSS) at values higher than expected for a random event (r=0.813, 0.803, 168 0.888, 0.894, in normal or growth-induced hearts for each gene, respectively ( Fig. 4c and supplementary Fig.   169 6S). Additionally, we found that ACAA2's and OGDH's chromatin binding sites extensively overlap (r= 0.86, Fig.   170 4c). This supports our conclusion that H2A.Z is a major recruiter of metabolic enzymes.

171
ACAA2 selectively and exclusively associates with H2A.Z-bound TSS. The Avg Val of the sequence tags 172 from the ACAA2 ChIP-Seq analysis were sorted into ACAA2-positve and -negative TSSs (-1000 to +1000 bp 173 from TSS) for transcriptionally active genes (determined by RNA pol II binding), in parallel with those of H2A.Z,

174
H3K9ac, Cdk9, and RNA pol II ChIP-Seq data. The data were graphed as violin plots representing the median, 175 quartiles, and distribution and probability density of the tags. This revealed that ACAA2 associates with the TSS 176 of 4204 genes (36.5% of genes expressed in the heart) that are also all H2A.Z-bound (approx. 90% of expressed 177 genes are H2A.Z-bound, see Fig. 5a-c). Conversely, not all H2A.Z-bound genes were associated with ACAA2, 178 suggesting selectivity and the involvement of other regulators (Fig. 5a-c). Notably, ACAA2-positive TSSs exhibit 179 a higher median for bound H2A.Z (8% and 6% higher for normal and growth-induced hearts, respectively) and 180 the H3K9ac mark (13.6% and 12% higher for normal and growth-induced hearts, respectively) relative to those 181 negative for ACAA2, as reflected in the violin plots . Interestingly, both H2A.Z and H3K9ac show similar 182 patterns of tag density distribution, as it is altered in the absence vs. presence of ACAA2, demonstrating a 183 positive correlation between the two marks. ACAA2-positive TSSs also exhibit significantly higher levels of pol II 184 and Cdk9 peaks, denoting higher transcriptional activity (supplementary Fig. 7S). On the other hand, there is no 185 correlation between changes in ACAA2 abundance during cardiac growth with the upregulation or 186 downregulation of Cdk9, H3K9ac, or pol II (Fig. 5d-e). Thus, although ACAA2 preferentially associates with ACAA2, in growth-induced v. normal hearts. Broadly, functional pathway analysis shows that these three 191 categories of ACAA2-bound genes encompass pathways involved in endoplasmic protein processing and 192 proteolysis, metabolism, and RNA transport and protein synthesis, respectively (supplementary Tables 2-4S).

193
The sequence tags of the ACAA2 ChIP-Seq were also aligned with those of H3K9ac, a histone mark that is 194 associated with active promoters; TFIIB, which demarcates the TSS; RNA pol II, which reflects transcriptional 195 activity; CDK9, which reflect transcriptional elongation; and ANP32E, which is a known H2A.Z-interacting protein.
196 Figure 5a shows the changes in peak densities of these molecules in the normal v. growth-induced hearts, across 197 the chromosomal coordinates of the TSS of Rrbp1, a ribosome binding protein; Eif4g1, which is involved in 198 translation initiation; Gtf2b, required for initiation of transcription; and Ubc, a substrate for protein ubiquitination.

199
These represent genes that exhibited an increase or a decrease in ACAA2 abundance in growth-induced v.

200
normal hearts (e.g. Rrbp1 and Eif4g2, respectively) or remain unchanged (e.g. Gtf2b and Ubc). Notably, these 201 genes contrasted with all cardiac-specific genes (shown are Actc1 and Actn2), which have no detectable ACAA2, 202 coinciding with the lack of, or undetectable, H2A.Z 17 (Fig. 5a). Therefore, these data reveal, for the first time, 203 the nuclear localization and chromatin binding of a beta-oxidation enzyme and validate our RIME analysis. Also, 204 consistent with our H2A.Z-RIME, ACAA2 associated exclusively with H2A.Z-bound TSSs, providing support of 205 specificity for this association.

206
OGDH exclusively associates with all H2A.Z-bound TSS. The Avg Val of the sequence tags from the OGDH 207 ChIP-Seq analysis were sorted into OGDH-positive and -negative TSSs (-1000 to +1000 bp from TSS) of 208 transcriptionally active genes (determined by RNA pol II binding), in parallel with those of H2A.Z, H3K9ac, Cdk9, 209 and RNA pol II ChIP-Seq data. The data were graphed as violin plots representing the median, quartiles, and 210 distribution and probability density of the tags. This analysis revealed that OGDH preferentially associates with 211 H2A.Z-bound TSSs with substantially higher H2A.Z densities (4.3-and 4.5-fold higher medians in sham and 212 TAC hearts, respectively, vs. OGDH-negative genes), which includes 89.9% (10,362) of expressed genes ( Fig.   213 6a-c). This also coincides with substantially higher levels of H3K9ac (5.3-and 6.5-fold higher medians for sham 214 and TAC hearts, respectively, vs. OGDH-negative genes) and Cdk9 (2.25-and 2.5-fold higher medians for sham induced vs. normal hearts, which is characteristic of a pause-release in transcription, was uniquely observed in 218 OGDH-positive TSSs ( Fig. 6b and supplementary 7S-c).

219
Of the 10,362 OGDH-positive genes, 992 exhibited ≥1.25-fold increase of OGDH, while 993 exhibited ≤ 0.75-220 fold downregulation in OGDH at the TSSs, in growth-induced v. normal hearts (Fig. 6d-e). Similar to ACCA2, 221 there is no correlation of the changes in OGDH abundance with the those observed in Cdk9, H3K9ac, or pol II 222 in normal v. growth-induced hearts, and, thus, transcriptional activity ( Fig. 6d-e). The sequence tags of the OGDH 223 ChIP-Seq were also aligned with those of H3K9ac, TFIIB, RNA pol II, Cdk9, ANP32E, and ACAA2 across the 224 genome (Fig. 6a). Figure 6a shows the changes in peak densities in the normal v. growth-induced hearts, across 225 the chromosomal coordinates of TSS regions of Ndufb10, which exhibits an increase, Prkab2, a decrease, and  Tables 5-6). In contrast to OGDH-positive genes, a uniform increase in TSS-and 230 gene body-pol II was observed in the OGDH-negative genes, in growth-induced v. normal hearts ( Fig. 6c and   231 supplementary Fig. 7S-d). Notably, gene ontology analysis of these genes included the terms sarcomere, Z disc, 232 myofibril,..etc. that characterize cardiac muscle-specific genes (supplementary Table 7). These have relatively 233 very low or no detectable H2A.Z, as seen in figures 5a (Actc1 and Actn2) and 6a (Tnnt2), figures 6c and 234 supplementary 7S-d, and as we have previously reported 17 .

235
While the vast majority of OGDH peaks overlapped with H2A.Z, 53 peaks appeared to exhibit H2A.Z-236 independent chromatin binding. Specifically, these peaks were identified in the terminal exon of 53 zinc finger 237 proteins (Zfp, Fig. 6a and supplementary Fig. 8S). Ultimately, the binding of OGDH to the TSS of H2A.Z-bound 238 housekeeping and ZFP genes appears to be conserved in humans, as we determined in a human colon cancer 239 cell line (supplementary Fig. 9S-a-c). Thus, the data suggest that OGDH is dependent on H2A.Z for its 240 recruitment to TSSs, however, other factors maybe required for its recruitment to select intragenic sites within

241
Zfp genes and that these findings are highly conserved between mouse and human cells.

249
Although TFIIB directly binds to DNA elements near the TATA-box, this interaction is not preserved in our protein 250 fractionation method, which does not involve protein crosslinking, thus, resulting in its localization to the 251 nucleoplasm. This contrasts with the histones, which are wrapped with chromatin, and, accordingly, strictly 252 localize to the chromatin-bound fraction of proteins. Thus, detection of chromatin-bound PDHA1 and ACAA2, 253 and its disruption by knockdown of H2A.Z, confirms their relatively tight association with chromatin in an H2A.Z-254 dependent fashion. Moreover, the data show that H2A.Z is required for H3 acetylation and methylation, plausibly 255 as a result of recruitment of metabolic enzymes. This is further supported by a reduction in H3ac after knockdown 256 of ACAA2 (supplementary Fig. 12S).

257
Knockdown of OGDH inhibits H4 succinylation. OGDH has been reported to bind to chromatin in U251 258 glioblastoma cells where it mediates succinylation of H3K79 14 . To test the impact of OGDH on histone 259 succinylation in normal cardiac myocytes, we knocked it down using shRNA (sh-OGDH). This treatment induced 260 a significant reduction of chromatin-bound OGDH (79 ± 7 %), and H4K12suc (62 ± 7 %) v. control levels, but not 261 of H3K27me2/3, H3K9me1/2/3, H3, or H4. TFIIB was used as a nuclear marker. In addition, knockdown of OGDH 262 was associated with a reduction in chromatin-bound ACAA2, indicative of the codependence of ACCA2 on 263 OGDH for its recruitment to chromatin. We conclude that OGDH is required for histone succinylation, plausibly 264 through conversion of alpha-ketoglutarate into succinyl-CoA at TSSs. On the other hand, we predict the reduction 265 in H3 acetylation maybe secondary to the reduction in chromatin-bound ACAA2, since knockdown of ACAA2, 266 resulted in 88% reduction in H3 acetylation (supplementary Fig. 11S).

H2A.Z knockout in human HAP1 cells inhibits chromatin association of metabolic enzymes and
268 posttranslational histone modifications. To validate the above data and investigate its relevance in human 269 cells, we analyzed human near-haploid HAP1 cells with a 2 bp deletion in exon 3 of H2A.Z (∆H2A.Z). These deficient in the ∆H2A.Z cells (Fig. 8c). This loss is associated with more than 90% reduction in chromatin 273 bound mitochondrial enzymes, including OGDH, ACAA2, HADHA, IDH2, SDHA and SDHB, and to a lesser 274 extent their nucleoplasmic levels ( Fig. 8a-b). Except for IDH2, the mitochondrial content of these enzymes was 275 also reduced, whereas OGDH was undetectable. We predict that this reduction in total enzyme content is a 276 result of direct, or indirect, H2A.Z-dependent transcription of their genes. As for OGDH, it is unclear why it is 277 completely lost from the mitochondria, in particular, in the absence of H2A.Z.

278
In contrast to the above tested enzymes, while the results show that PDHA1 exhibited strong localization to the 279 nuclear and chromatin fractions, it was the only enzyme for which neither its expression nor chromatin binding 280 were impacted by the knockout of H2A.Z. This proved its H2A.Z-independent chromatin association and, 281 thereby, the selectivity of H2A.Z-dependent recruitment of metabolic enzymes. Other noted differences between 282 the enzymes, include the finding that only OGDH and PDHA1 were detected in the cytosol, the unexpected 283 complete loss of OGDH in the mitochondrial fraction in the ∆H2A.Z cells, and the equivalent reductions of SDHB 284 in all fractions in the ∆H2A.Z cells that suggests its independence of H2A.Z for chromatin binding. Also, notable, 285 is the fact that the nuclear localization of mitochondrial enzymes was selective, since the mitochondrial proteins 286 VDAC1 and PDK1 were not detected in the nucleus, and were, thus, used as mitochondrial markers and internal 287 controls in our blots. Thus,

289
In this study, we have identified a plethora of metabolic enzymes that bind to the TSSs of transcriptionally active 290 genes in a H2A.Z-dependent and, less frequently, -independent fashion. These were discovered by an unbiased 291 screen using anti-H2A.Z chromatin immunoprecipitation-mass spectrometry. This approach provides the unique 292 advantage of identifying proteins that associate with H2A.Z in its native conformation within the nucleosome.

293
One of the disadvantages, though, as with other immunoprecipitation approaches, is the likelihood of non-294 specific bindings. To eliminate those from our analysis, we applied the following measures; each sample 295 analyzed consisted of a pool of 20 independent heart, each sample was analyzed twice by mass spec, the H2A.Z 296 pulldown was analyzed in 2 independent samples (the normal heart and the growth-induced), we only considered 297 the co-immunoprecipitated proteins that exhibited ≥ 2-fold enrichment with H2A.Z vs. IgG control, and finally, we 298 validated this finding for 7 of 29 enzymes identified using a combination of various methods that included immunocytochemistry, tGFP fusion proteins, NLS mutagenesis, ChIP-Seq, and H2A.Z knockout. So far, all the 300 metabolic enzymes that we have tested, including ACAA2, OGDH, IDH2, PDHA1, HADHA, SDHA, and SDHB, 301 were confirmed for their nuclear localization and chromatin binding by two or more of the methods listed above, 302 providing confidence in our RIME findings.

303
We preformed the RIME assay in total heart tissue for the purpose of preserving the 3D milieu of the cells, which 304 is critical for their transcriptional integrity, as we ascertained that the signals obtained by this approach are 305 predominantly derived from cardiac myocytes. This is supported by the fact that smooth muscle actin (Acta2), 306 which is expressed in smooth muscle cells and myofibroblasts in the heart, and ATPase plasma membrane Ca 2+ 307 transporting 4 (Atp2b4), which is ubiquitously expressed, including in epithelial cells, have no detectable RNA 308 pol II binding compared to its high abundance in the corresponding cardiac genes, cardiac actin (Actc1) and

309
ATPase sarcoplasmic/endoplasmic reticulum Ca 2+ transporting 2 (Atp2a2, supplementary Fig. 14S). We then 310 confirmed the findings by the various methods listed above, in both rodent and human cell lines, including 311 isolated mouse and rat myocytes, human iPSC-derived cardiac myocytes, human colon cancer cells, and human 312 near-haploid cells with or without a H2A.Z deletion. We find that the nuclear localization of the metabolic enzymes 313 is conserved between species and largely H2A.Z-dependent.

314
H2A.Z is a highly conserved histone variant that plays an essential role in sensing and responding to metabolic 315 and environmental cues, however, the underlying mechanisms remain elusive. It has, though, been shown to 316 interact with 93 proteins, mostly identified by unbiased screens using co-fractionation, affinity capture-MS, and  immunoprecipitation-mass spectrometry of endogenous proteins assay approach 18 with anti-H2A.Z we identified chain amino acid metabolism. The fact that these enzymes are associated with chromatin at the TSS explains 327 how metabolites could be directly delivered to target genes where they are used as substrates for histone 328 modifications (e.g. acetyl-CoA, succinyl-CoA, among other short acyl-CoA metabolites) or as co-factors (aKG) 329 for histone modifying enzymes, thereby, allowing promoters to immediately sense and respond to metabolic 330 cues. Consistent with an H2A.Z-dependent recruitment, genes that are devoid of H2A.Z, also lack metabolic 331 enzymes (ACAA2 and OGDH) at their TSS and, conclusively, knockdown or knockout of H2A.Z abrogates 332 chromatin association of multiple enzymes including ACAA2, OGDH, IDH2, HADHA, SDHA, and SDHB.

333
Interestingly, however, PDHA1 was an exception, as it retained its chromatin association in the absence of H2A.Z 334 in HAP1 cells. Therefore, we conclude that while H2A.Z may be required for the recruitment of multiple metabolic 335 enzymes to chromatin, there are some that are H2A.Z-independent, for which the recruiting protein remains to 336 be determined. On the other hand, we speculate that the absence of H2A.Z and metabolic enzymes at the 337 promoters of constitutively expressed, tissue-restricted genes (e.g. sarcomeric proteins), which distinguish an 338 organ's unique functionality, ensures these are not impacted by metabolic fluctuations, as one would expect.

339
This contrasts with housekeeping and inducible genes that would be immediately modulated in response to 340 changes in oxygen and/or metabolic substrate availabilities, as a mechanism of cellular adaptation. Our results

341
are also consistent with a proteomics study that identified all TCA cycle enzymes in the nucleus of normal and 342 cancer cells 16 .

343
H2A.Z is required for transcriptional memory in yeast where it is incorporated in newly deactivated INO1 at the 344 nuclear periphery 27, 28 , and in hippocampal memory in mice, where it negatively controls fear memory through 345 suppressing the expression of specific memory-activating genes 29 , and is necessary for neurogenesis and 346 normal behavioral traits in mice 30 . As well established, memory is a function of epigenetics, wherein histone 347 acetylation is an essential regulator, demonstrated by Mews et al 31 . In that study, the authors reported that 348 ACSS2 (a cytosolic and nuclear enzyme that converts acetate into acetyl-CoA) binds near the TSS of 349 hippocampal neuronal genes, where its knockdown diminishes long-term spatial memory and an acetylation-350 dependent cognitive process. This suggests a potential link between H2A.Z and the intricate and precise 351 regulation of core histone modifications. Concordantly, we show that knockdown or knockout of H2A.Z reduces The association of metabolic enzymes with the TSS of genes could potentially explain the mechanism of targeted 355 histone modifications, and, therefore, the direct regulation of transcription via glucose and fatty acid metabolism.

356
Furthermore, the data expands the range of locally-delivered modifying substrates and regulatory co-factor to 357 include not only acetyl-CoA or succinyl-CoA, but also citrate, aKG, succinate, fumarate, and other short chain 358 acyl-CoAs, which considering the complexity of gene regulation and memory, is not surprising.

359
While acetylation and methylation of histones is a key player in transcriptional regulation and memory, 360 succinylation is another modification that has received less attention. 367 peaks, covering 89.9% of TSSs that co-localized and co-immunoprecipitated with H2A.Z, with the exemption of 368 tissue-restricted genes, which lack any substantial amount of H2A.Z. We predict that the difference in the number 369 of peaks is likely due to differences in the cell types or the antibodies used for ChIP-Seq. Other support of nuclear 370 localization of mitochondrial enzymes was reported by Jiang et al, showing that phosphorylated fumarase 371 interacts with H2A.Z in response to ionizing radiation-induced activation of DNA-dependent protein kinase in 372 U2OS cells, a function that regulates DNA repair 32 . Our study extends these findings to include nuclear 373 occupancy of all the enzymes in the TCA cycle and beta-oxidation spiral, where they are recruited to chromatin 374 via H2A.Z. Thus, our data support the concept that local production of the TCA cycle intermediates is necessary 375 for transcriptional regulation. Other than fumarate and succinyl-CoA, production of citrate, aKG, succinate, and 376 acetyl-CoA, among others, have the capacity of either generating the substrates that are directly required for 377 histone modification, or alternatively, generating metabolites that regulate histone modifying enzymes.

378
H2A.Z's function does not always correlate with transcriptional activity, as noted by others and us. Not only do suppressed/unexpressed genes (e.g. Wnt1, Noggin, Tbx1) have substantial amounts of H2A.Z at the TSS and 382 in gene body (supplementary data, Fig. 15S and Table 8). On the other hand, moderately-expressed 383 housekeeping genes that are amenable to incremental modulation by external stimuli have the highest levels of 384 H2A.Z at their TSS, whereas, inducible genes have high levels of H2A.Z at the TSS that extends into the gene 385 body. The latter pattern enhances the responsiveness of inducible genes to stimuli, as previously reported in 386 Arabidopsis thaliana 33 . While these results support the role for H2A.Z in transcriptional regulation, they suggest 387 that it is not required for constitutive transcription but rather for strictly regulated transcription. Our findings show

394
One of the unresolved issues that needs to be thoroughly investigated, is the exact stoichiometry of the enzymes 395 and the interdependence of their bindings and functions, which is apt to determine the concentrations of the 396 metabolites that are produced/consumed at the TSSs. Accordingly, we expect that the composition of these 397 enzymatic complexes to define the underlying histone modifications and the responsiveness of the genes' 398 expression to oxidative or metabolic cues, while any perturbation may result in pathogenesis. This is underscored 399 by the fact that while ACAA2 fully overlaps with ODGH at TSSs, the precise pattern of binding and their 400 responsiveness to growth stimuli in the heart are distinct. For instance, in the Ubc gene, the two start sites exhibit 401 differential binding to ACAA2 and OGDH, particularly, during growth, when OGDH shows a decrease in 402 abundance at the first TSS accompanied by an increase at the second TSS, whereas, the ACAA2 assembled at 403 the first TSS remains unchanged (Fig. 5). Meanwhile, H2A.Z abundance does not vary significantly. Another 404 incompletely resolved matter, is how these enzymes are imported into the nucleus. We were able to confirm that, 405 at least, ACAA2 and OGDH harbor NLSs that are required for their nuclear import ( Fig. 3

410
Animal care -All animal procedures used in this study are in accordance with US National Institute of Health

411
Guidelines for the Care and Use of Laboratory Animals . All protocols were approved by the 412 Institutional Animal Care and Use Committee at the Rutgers-New Jersey Medical School.

413
H2A.Z rapid immunoprecipitation mass spectrometry of endogenous proteins (RIME) -Male C57/Bl, 12 414 wk-old mice, 10 each, were subjected to a sham or transverse aortic constriction (TAC) procedure. After 1 wk, 415 the hearts were isolated, pooled for each condition, and sent to Active Motif for RIME analysis by anti-H2A.Z  false discovery rates (FDR) algorithm. Peptide identifications were also required to exceed specific database 429 search engine thresholds. X! Tandem identifications required at least. Protein identifications were accepted if 430 they could be established at greater than 5.0% probability to achieve an FDR less than 5.0% and contained at 431 least 1 identified peptide. Protein probabilities were assigned by the Protein Prophet algorithm 34 . Proteins that 432 contained similar peptides and could not be differentiated based on MS/MS analysis alone were grouped to 433 satisfy the principles of parsimony. Proteins sharing significant peptide evidence were grouped into clusters.

434
ANALYSIS -The total spectra counts for all genes in the three samples, were normalized to the corresponding each sample. Fold enrichment of total spectra for sham : IgG and TAC : IgG, for each gene, was calculated.

437
Seventy-three of these had a ≥ 2-fold enrichment, those are shown in figure 1 and supplementary figure 1S.

438
Culturing rat neonatal cardiac myocytes -Cardiac myocytes were cultured as described in our previous 439 reports 35 . Briefly, hearts were isolated from 1 day old of Sprague-Dawley rats. After dissociation with 440 collagenase, cells were subjected to Percoll gradient centrifugation followed by differential pre-plating for 30 min 441 to enrich for cardiac myocytes and deplete non-myocyte cells. Myocytes were cultured in Dulbecco's Modified

442
Eagle's medium supplemented with 10% fetal bovine serum (FBS). All experiments were initiated after a 24 h 443 culturing period.

444
Culturing mouse adult cardiac myocytes -Adult cardiac myocytes were isolated and cultured from C57/Bl 445 mice (8-9 wks old), according to the protocol described by Ackers-Johnson et al. 36 . Briefly, mice were anesthesia 446 with Ketamine/Xylazine/Acepromazine (65/13/2 mg/kg) by intraperitoneal injection. The mouse chest cage was 447 then opened, the ascending aorta clamped, and both descending aorta and inferior vena cava cut. First, 448 ethylenediaminetetraacetic acid (EDTA) is injected into the base of the right ventricle, the heart is then transferred 449 into a petri dish, and a second EDTA injection is administered into the left ventricular wall above the 450 apex. Following this, the cells are dissociated using collagenase, and the rod-shaped myocytes are differentially 451 separated by gravity, where calcium is re-introduced. The cells are plated on laminin-coated dishes or glass 452 slides in M199 medium with 5 % FBS for 1 hour, after which the FBS is replaced with 0.1% bovine serum albumin 453 for longer culturing periods.

454
Human iPSC-derived cardiac myocyte cultures -Cardiac myocytes derived from human iPSCs were 455 purchased from Cellular Dynamics International and cultured as recommended by the manufacturer.

456
Colon cancer cell culture -SW620 were purchased from the American Type Culture Collection (ATCC). Cells 457 were cultured in Leibovitz's (Gibco) medium with 10% FBS and maintained in a CO2 free incubator.

537
Signaling Technology, cat # 26865), followed by next generation sequencing (Active Motif). We have previously 538 reported the results of our ChIP-Seq for RNA pol II 42 , H3K9ac 42 , TFIIB 35 , H2A.Z 17 , and ANP32E 17 , and, thus, 539 are not further described here. Briefly, ChIP libraries were sequenced using NextSeq 500, generating 75-nt 540 sequence reads that are mapped to the genome using BWA algorithms. The reads/tags were extended in silico 541 by 150-250 bp at their 3'end (fragments), the density of which is determined along the genome, divided in 32 nt 542 bins, and the results saved in bigWig and BAM (Binary Alignment/Map) files. Fragment peaks were identified 543 using MACS, which identifies local areas of enrichment of Tags, defined as 'intervals', while overlapping intervals are grouped into 'Merged Regions'. The locations and proximities to gene annotations of intervals and active 545 regions are defined and compiled in Excel spreadsheets, which include average and peak fragment densities.

546
Regarding tag normalization and input control, the sample with the lowest number of tags is used for 547 normalization of all samples, while the input is used to identify false positive peaks. The statistics for ACAA2 and 548 OGDH ChIP-Seq results, including total number of reads, peaks, empirical FDR, and peak calling parameters 549 are listed in supplementary Table 1.

550
In addition, we separately analyzed the fragment densities by gene region, where the average value (Avg Val) 551 of fragment densities at the TSS (-1000 to +1000) and in-gene/gene body (+1000 to 3' end) regions for all genes 552 were calculated separately. Subsequently, we used these values to sort genes according to TSS-pol II, -ACCA2, 553 or -OGDH, occupancy.

554
ChIP-Seq analysis software -The heatmaps, curves, and histograms shown in figure 4a-c, were generated 555 using EaSeq 43 . Images of sequence alignments of fragments across chromosomal coordinates were generated 556 using the Integrated Genome Browser (Fig. 6-7

562
The RNA polymerase II AND H3K9ac ChIP-Seq data (accession: GSE50637), the TFIIB ChIP-Seq data 563 (accession: GSE56813) and H2A.Z and ANP32E ChIP-Seq data (accession: GSE104702) are available in the 564 Gene Expression Omnibus Datasets. The ChIP-Seq data for ACAA2, OGDH, and Cdk9 will be deposited in GEO 565 with a private link for the reviewers, which will be made public upon acceptance.

767
The sequence tags for OGDH, H2A.Z, H3K9ac, ACAA2, and pol II, at the TSS (-1000 to +1000), from these 768 groups, were plotted as violin plots, in which the horizontal solid line represent the median, and the dashed lines 769 the quartiles, whereas, the shape of the violin reflect the tags' density distribution. repeats), was quantitated and plotted as the mean ± SEM relative to the control, adjusted to 1. Error bars 778 represent standard error of the mean, and *p = 0.03 v. each's corresponding control. c. shRNA targeting OGDH 779 or a control construct, was delivered to isolated rat neonatal cardiac myocytes using adenoviral vectors (moi 30).

780
After 48 h, organelles were isolated, fractionated, and analyzed as described in (a). The Western blot signals 781 (n=4, each, from 4 repeats), was quantitated and plotted as the mean ± SEM relative to the control, adjusted to 782 1. Error bars represent standard error of the mean, and *p = 0.035 v. each's corresponding control.  c.   F igu re 2S. Imag es from t he h uman prot ein at las proj ect (www.proteinatlas.org). The Human Protein Atlas is a Swedish-based project that includes antibody-based imaging of human tissue and cell lines, and is open access for scientists allowing free use of the data, given that it is properly cited [Ref. 18]. Shown are images from that project that include normal human heart sections immuno-stained with a. anti-OGDH and b. anti-PDHB, c. human breast cancer sections immuno-stained with anti-HADBH, d. A431 cells immuno-stained with anti-OGDH, and e. A-431 cells immuno-stained with anti-MRSP36. Direct links to the web pages are listed beneath each image. Note, different antibodies had differential affinities to the nuclear v. mitochondrial form of a given protein, as demonstrated in our data f. We also stained the human colon cancer cell line SW620 with the a second OGDH antibody that targets the N-terminus v. the C-terminus, used in the main figure 2, for validation of the data.

Figure 4
Antibody: Sigma, cat # HPA019514, targeting the N-terminus. (same as used for human heart and A-431 in a. and d., respectively ) iab.keio.ac.jp/cgi-bin/NLS_Mapper_form.cgi. The predicted NLS is indicated by red lettering. The scoring system is such that a protein with a score of 8, 9, or 10 is exclusively nuclear; 7 or 8 is partially nuclear; a score of 3, 4, or 5 is both nuclear and cytoplasmic; and a score of 1 or 2 is cytoplasmic. The mutations generated in the OGDH and ACAA2 predicted NLS are shown at the bottom, where the substituted amino acids are indicated in red.

OGDH (TCA cycle enzyme) ACAA2 (b-oxidation enzyme)
Mutations generated in the predicted NLS sequences: OGDH a.a. 531-RKQKPVLQK mutated to RQQQPVLQQ ACAA2 a.a. 207-EVKTKKGKQ mutated to EVQTQQGKQ + + + + + + + + + + + + + + + + + + + + + + + + + + Figure 4S. Nuclear localization of metabolic enzymes confirmed by tGFP-fusion proteins and NLS mutation. SW480 human colon cancer cell were infected with a 10-30 moi of recombinant adenoviruses harboring turbo-GFP (tGFP) or a. wt ACAA-tGFP or an NLS mutant (mtACAA2-tGFP), b. wt OGDH. In a., the cells were incubated in either glucose-containing (fatty acid and serumfree) or in palmitate-BSA (glucose-free and serum-free) medium, as indicated at the top of the lanes. After 18 h, the cellular protein/organelles were fractionated into cytosol (cyto), mitochondrial and membrane (mito), nuclear (nuc), and chromatin-bound (chrom) protein fractions that were then analyzed by Western blotting for the proteins listed on the left of each panel. The fusion proteins were detected by anti-GFP (upper panels, a-b) and anti-ACAA2 or anti-OGDH (second panels, a-b), which also detect the endogenous proteins. AKT1, VDAC1, Pol II, were immunodetected for their use as internal controls for the corresponding cell fractions: cytosol, mitochondria and, nuclear and chromatin, respectively.  Figure 5S. Nuclear localization of metabolic enzymes in mouse heart and isolated cardiac myocytes, confirmed by Western blotting. Cellular organelles were extracted from a. mouse sham or TAC hearts, b. isolated mouse adult cardiac myocytes (mACM), and c. rat neonatal cardiac myocytes. These were fractionated into membrane, including mitochondria (Mem), nuclear (Nuc), and chromatin-bound (Chrom, no crosslinking applied), using a combination of differential lysis and sequential centrifugation. The protein extracted from each of these fractions was analyzed by Western blotting for the genes indicated on the right of each panel (n=3, each). Figure 6S -The association of ACAA2 and OGDH with chromatin overlaps with H2A.Z at transcription start sites. Histograms showing the distribution of fragments calculated from their overall frequencies in the ChIP-Seq of H2A.Z (X-axis) v. ACAA2 or OGDH (Y-axis), and of ACAA2 (X-axis) v. OGDH (Y-axis), over the length of the gene and including -2000 bp upstream of the TSS, as labeled. The X and Y-axes were segmented into 75 bins, and the number of fragments within each bin was counted, color coded, and plotted. The bar to the right of the plot illustrates the relationship between count and coloring. The plots represent pseudo-colored 2D matrices showing observed, expected, and observed/expected distribution, calculated from the overall frequencies of fragments on each of the axes. These show the relation between a.-b. H2A.Z and ACAA2, c.-d. H2A.Z and OGDH, e.-f. ACAA2 and OGDH, g.-h. H2A.Z and H3K9ac, all in the sham and TAC hearts. The pseudo-color corresponds to the Obs/Exp ratio, and the color intensity is proportional to the log2 of the number of observed fragments within each bin. These plots suggest that there is a positive correlation between the levels of H2A.Z and ACAA2 or OGDH, where the red indicates that this occurs more frequently than expected by chance, as denoted by the correlation coefficient listed above each plot. i. A histogram showing the relation between H3K9ac and the input, as an example of unrelated binding events, and j. as an example of a perfect correlation. This figure was generated by EaSeq software.  Figure 8S. OGDH binds to the terminal exons of zinc finger proteins, in a H2A.Z-independent manner. The alignment of the ChIP-Seq sequence tags for H3K9ac, TFIIB, pol II, Cdk9, H2A.Z, ANP32E, ACAA2, and OGDH (y-axis) across the genomic coordinates (x-axis) of Zfp90, Zfp2, Zfp146, Zfp157, Zfp180, and Zfp110 genes. The arrow shows the start and direction of transcription. The results show a substantial peak of OGDH in the terminal exons of these genes that is subject to differential regulation during cardiac hypertrophy. In addition, all genes show OGDH at their TSSs, albeit at a lower density.  Fig. 9S-a-b. Conserved, selective, binding of OGDH to H2A.Z-bound TSSs. Both mouse heart tissue and human SW480 colon cancer cell line, were subjected to H2A.Z and OGDH ChIP-Seq us ing the same antibodies and c h r o m a t i n c o n c e n t r a t i o n . T h e r e s u l t i n g s equenc e Tags form both reac tions were aligned across the coordinates for the same genes in the mouse and human genomes, as indicated. Two regions are shown, a. the first showing the PRKAB2, FMO4, and CHD1L genes in the human cells and mouse tissue, where OGDH co-localizes with H2A.Z at the TSS in all three genes in the latter, however, OGDH is absent in from the FMO5 in the human genome, b. the second region encompasses PKN2, GTF2B, CCBL2, GBP1-5,7 genes that show conserved co-localization of OGDH and H2A.Z at the TSS of the former 3 genes in the mouse and human, but differs between species for the GBP genes, which have no OGDH in of the human cells, with a relatively small peak of H2A.Z at the TSS of GBP1-4. These data reveal that the co-localization of H2A.Z and OGDH at TSSs of key specific genes are conserved between species.