HDAC8 mutations in Cornelia de Lange Syndrome affect the cohesin acetylation cycle

Abstract

Cornelia de Lange syndrome (CdLS) is a dominantly inherited congenital malformation disorder caused by mutations in the cohesin-loading protein NIPBL^1,2 for nearly 60% of individuals with classical CdLS^3-5 and in the core cohesin components SMC1A (~5%) and SMC3 (<1%) for a smaller fraction of probands^6,7. In humans, the multi-subunit complex cohesin is comprised of SMC1, SMC3, RAD21 and a STAG protein to form a ring structure proposed to encircle sister chromatids to mediate sister chromatid cohesion (SCC)⁸ as well as play key roles in gene regulation⁹. SMC3 is acetylated during S-phase to establish cohesiveness of chromatin-loaded cohesin^10-13 and in yeast, HOS1, a class I histone deacetylase, deacetylates SMC3 during anaphase^14-16. Here we report the identification of HDAC8 as the vertebrate SMC3 deacetylase as well as loss-of-function HDAC8 mutations in six CdLS probands. Loss of HDAC8 activity results in increased SMC3 acetylation (SMC3-ac) and inefficient dissolution of the “used” cohesin complex released from chromatin in both prophase and anaphase. While SMC3 with retained acetylation is loaded onto chromatin, ChIP-Seq analysis demonstrates decreased occupancy of cohesin localization sites that results in a consistent pattern of altered transcription seen in CdLS cell lines with either NIPBL or HDAC8 mutations.

Human SMC3 is acetylated by ESCO1 and ESCO2, homologues of yeast ECO1 and has been shown to be important for the establishment of sister chromatid cohesion^{10,11,13,17,18}. Using a monoclonal antibody specific for acetylated SMC3 (SMC3-ac)¹⁸, we found that although total SMC3 levels remain stable throughout the cell cycle, SMC3-ac rapidly disappears during mitosis suggesting coordinated deacetylation (Supplementary Fig. 1).

We therefore used RNA interference-based screening of all known human HDACs and Sirtuins to identify HDAC8 as the vertebrate SMC3 deacetylase (Supplementary Fig. 2). Loss of HDAC8 activity using either HDAC8 RNAi or the HDAC8-specific inhibitor PCI-34051 (PCI; Fig. 1a,b) does not alter cell cycle progression, but clearly increases SMC3-ac in both soluble and chromatin fractions throughout the cell cycle (Fig.1c lanes 4 and 6, Fig. 1e lanes 18-22, 29-33, Supplementary Fig. 3d lanes 22-28 and 36-42). Nearly all of HDAC8 is present in the soluble fraction in both asynchronous and synchronized cultures (Fig. 1c,e). These data indicate that HDAC8 is present and active throughout the cell cycle, and that soluble SMC3-ac is its deacetylation target, similar to Hos1 in yeast^14-16. Notably, the increase of SMC3-ac in the soluble fraction in the absence of HDAC8 activity implies that SMC3-ac is able to dissociate from chromatin but fails to be deacetylated. In addition, we unexpectedly observed few sister-chromatid cohesion defects with loss of HDAC8 activity alone (Supplementary Fig. 4).

Figure 1. HDAC8 is an SMC3 deacetylase.

a, HeLa cells were transfected with HDAC8 siRNA (Lanes 1-4) or incubated with 25 μM PCI, the HDAC8-specific inhibitor (Lanes 5-9) for the indicated times. Total cell lysates were prepared and analyzed by immunoblotting using anti-SMC3-ac, SMC3 and HDAC8 antibodies. Numbers beneath SMC3-ac bands indicate quantification of SMC3-ac levels normalized to SMC3 levels and the 0 hour time point. b, Acetylated SMC3 was prepared by co-immunoprecipitation with SMC1A from HeLa cell chromatin extracts. Immunoprecipitates were incubated with recombinant purified HDAC8 or mutant HDAC8 protein at 30°C for 1 h and analyzed by immunoblotting as in (a). c, Unsynchronized HeLa cells transfected with HDAC8 siRNA for 48 h were fractionated into soluble and chromatin fractions and immunoblotted as in (a). d, HeLa cells were synchronized by double thymidine arrest and released into the presence or absence of 25 μM PCI for the indicated time and verified by FACS analysis for cell cycle progression. e, Cell extracts of the PCI-treated cells in (d) were fractionated into soluble and chromatin fractions and were analyzed by immunoblotting as in (a). Histone H3 Serine 10 phosphorylation (H3S10P) is a marker of prophase and the onset of mitosis.

To understand the role of HDAC8 in genome-wide regulation of cohesin dynamics, we performed ChIP-Seq analysis of synchronized HeLa cells transfected with control or HDAC8 RNAi (Fig. 2) and immunoprecipitated with either an anti-RAD21 antibody to detect total cohesin or with an anti-SMC3-ac antibody. Although total cellular cohesin shows no decrease (Supplementary Fig. 5a,b), and there is a high degree of overlap between SMC3-ac, cohesin and CTCF¹⁹ localization sites in treated and untreated cells, high read numbers and tight correlations between experimental replicates enabled us to identify a 17% loss of total cohesin localization peaks with reduced HDAC8 activity (Fig. 2a-c, Supplementary Fig. 5d-h). Furthermore, despite using conditions that increase total SMC3-ac more than two-fold (Fig. 1e, Supplementary Fig. 3d), we note a 16% loss of SMC3-ac localization sites with HDAC8 reduction (Fig. 2a-d, Supplementary Fig. 5f-g). Finally, we found that in both control and HDAC8-depleted cells, SMC3-ac preferentially localizes to downstream regions of genes relative to the distribution of RAD21 (Fig. 2c,d, and Supplementary Fig. 5f-g). Together, this data demonstrates decreased occupancy of cohesin localization sites with the loss of HDAC8 activity, an observation similarly noted for NIPBL haploinsufficient CdLS cells²⁰.

Figure 2. Cohesin and SMC3-ac localization sites in control- and HDAC8 RNAi-treated HeLa cells.

a, Example of ChIP-Seq data (Ensemble Gene position 1.68–2.10 Mb of human chromosome 12). ChIP-Seq data are shown in reads per million. Regions in which signals were significantly enriched (see Methods) are in red. Binding profiles for RAD21 and acetylated SMC3 in control and HDAC8 RNAi-treated HeLa cells are shown. b, Venn diagrams showing numbers and overlap of localization sites. c, Example of region-specific localization showing RAD21 and SMC3-ac binding sites in the MYC gene locus. Ensemble Gene position 128.728-128.773 Mb of human chromosome 8 is shown. d, Classification of RAD21 and SMC3-ac localization sites as a log ratio to background reads. Significance was calculated using a 2-sample test for equality of proportions; *, **, *** and -, indicate p<0.05, p<0.001, p<0.0001 and p>0.05, respectively. “Upstream” and “downstream” are defined as within 5kb from gene boundaries, respectively. Acetylated SMC3 preferentially localizes to the downstream end of genes.

With the known role of cohesin regulation in CdLS and the observations that reduction of either HDAC8 or NIPBL lead to decreased cohesin occupancy of localization sites, we hypothesized that HDAC8 mutations may cause CdLS. We screened this X-linked gene in 154 individuals with CdLS negative for mutations in NIPBL, SMC1A and SMC3, as well as RAD21, STAG2, ESCO1, ESCO2 and MAU2/SCC4. We identified four de novo missense mutations and one de novo nonsense mutation in HDAC8 (Supplementary Table 1 and Fig. 3a). In addition, one familial mutation (c.1001A>G; p.H334R) was identified in a boy, his mildly affected sister and his unaffected mother, where the mutant allele was inactivated in her blood. This mutation was also one of the de novo mutations in an unrelated girl. None of the mutations were seen in 290 ethnically matched control chromosomes or in 629 individuals of the 1000 Genomes Project²¹. Despite the small numbers and the varied clinical features in females due to random X-inactivation, these children demonstrate growth, cognitive and facial features consistent with those caused by mutations in NIPBL (“classical” CdLS). Both expression studies (Supplementary Fig. 6a) and X-inactivation studies (data not shown) demonstrate complete skewing toward the normal allele in the blood of females with HDAC8 mutations, indicating strong selection against the mutation. This limited the number of cell lines available for subsequent studies. However, immunoblotting demonstrated minimal HDAC8 protein expression in lymphoblastoid cell lines (LCLs) from a hemizygous boy with a p.G320R mutation as well as the skin fibroblasts from a female Lyonized to express a p.H180R mutant allele (Fig. 3b), indicating protein instability in each case. Consistent with this, assessment of SMC3-ac demonstrates increased levels in both the p.G320R LCLs and the p.H180R fibroblasts, while the total amount of SMC3 (Fig. 3b) and cell cycle distribution for the LCLs are unchanged (Supplementary Fig. 6b).

Figure 3. HDAC8 mutations in CdLS.

a, Facial features of individuals with HDAC8 mutations labeled with corresponding mutation and sex; ♂=male, ♀=female. b, Western blotting of protein from lymphoblastoid cell lines (LCL) and fibroblasts (Fib). Sex is indicated under Controls (1,2,3) or mutation designation. Primary antibody is indicated at the left. c, HDAC8 mutations disrupt deacetylase activity. Bar graphs demonstrate the effect of HDAC8 mutations on deacetylase specific activity (nmol substrate. μmol enzyme⁻¹. min⁻¹). All assays were performed in triplicate. Error bars indicate standard deviation. Enzymatic activity for all mutations is significantly less than wild type (p<0.05). d, Localization of HDAC8 mutations on the crystal structure (PDB accession code 3F06²³). Mutated residues are in red. e, Expression profile of HDAC8-mutant LCL is consistent with that seen in those with NIPBL mutations. The summed standard deviations of gene expression compared to NIPBL-mutant from control LCLs for 32 genes was determined for each sample. Box graphs demonstrate 25th, median and 75th percentiles of each group. Whiskers indicate minimum and maximum values. Note that for the HDAC8 male (p.G320R) the values are for biological replicate samples for a single cell line (denoted by *). Unpaired two-tailed t test significances are indicated.

We expressed the HDAC8 missense mutations in E. coli, purified and assayed each for deacetylase activity. These data (Fig. 3c) demonstrate that the p.H180R mutation severely abrogates HDAC8 deacetylase activity, and demonstrate significant activity losses for the p.G320R, p.T311M and p.H334R mutations, consistent with high conservation of each residue and position in the structure of HDAC8–substrate complexes ^22,23 (Fig. 3d, Supplementary Fig. 6c-e). Finally, purified wild type, but not mutant HDAC8, can rescue the SMC3 overacetylation seen in HDAC8 mutant LCLs (Supplementary Fig. 6g,h).

We have previously demonstrated that mutations in NIPBL, the primary cause of CdLS, result in consistent, reproducible changes of genes expressed from LCLs derived from individuals with CdLS²⁰. Using Nanostring multiplex expression analysis to avoid variation introduced by PCR or RNA amplification-based assessments and a previously validated 32-gene CdLS classifier set²⁰, we compared expression of LCLs from the male with the p.G320R mutation and from two girls (with the p.H180R and p.H334R mutations) who express only the normal allele in these cells, to 10 normal controls and 12 LCLs with loss of function NIPBL mutations. In this assay, the aggregate 32-gene LCL expression profile of the single male with an HDAC8 mutation strongly correlates with that seen of NIPBL-mutant cell lines, while the profile for the two female lines, which express the wild type allele, correlates with normal controls (Fig. 3e). These data support the hypothesis that loss of HDAC8 activity results in widespread transcriptional dysregulation as seen in NIPBL-mutated CdLS cells. The loss of cohesin binding sites in both HDAC8 RNAi-depleted HeLa cells and NIPBL-mutant LCLs²⁰ was also noted in the p.G320R HDAC8 mutant LCLs and fibroblasts derived from the girl with the p.H180R mutation, where small but consistent changes in transcriptional dysregulation were also noted (Supplementary Figs 6a,7,8). Together these data suggest that loss of HDAC8 activity leads to a common pathogenic mechanism for “classical” CdLS that converges on the reduction of bound cohesin complexes.

To further understand the endogenous effect of HDAC8 loss on cohesin, we compared each cohesin subunit in chromatin-bound and unbound fractions in the LCLs with a hemizygous p.G320R mutation versus control LCLs and unexpectedly noted accumulation of RAD21 fragments in HDAC8 mutant cells (Supplementary Fig. 9a, lanes 2,4,6). Using a RAD21 N-terminal-specific antibody, we verified accumulation of the RAD21 N-terminal separase cleavage fragment (RAD21-N) in mutant LCLs, HDAC8 RNAi treated HeLa cells and HDAC8-mutated fibroblasts expressing catalytically inactive HDAC8 versus wild type HDAC8 (Fig. 4, Supplementary Fig. 9). The accumulation of this cleavage fragment in the chromatin-bound fraction of HDAC8-mutant cells suggests that it may remain bound to the cohesin complex, so we assessed whether RAD21-N co-immunoprecipitates with other cohesin subunits. In HDAC8 mutant cells, RAD21-N co-immunoprecipitates with SMC1A, SMC3, STAG1, and STAG2 (Fig. 4b), strongly suggesting that without functional HDAC8, the cleaved N-terminus of RAD21 remains attached to the cohesin complex, presumably via acetylated SMC3²⁴. To further confirm that clearance of cleaved RAD21 is dependent on HDAC8, we performed complementary analyses in synchronized HeLa cells with or without the HDAC8 inhibitor PCI (Fig. 4c,d). These data demonstrate the appearance of the N-terminal and C-terminal RAD21 fragments upon entry into mitosis (marked by H3S10-P) and activation of separase (Fig. 4d, lanes 9-12 and 21-24) as well as their persistence in both soluble and chromatin fractions (Fig. 4d lanes 18-20) when most cells are in G1 phase (data not shown). In contrast, in untreated cells, these cleavage products are predominantly detected in the soluble fraction (Fig. 4d lanes 2-4), and largely disappear 3 h post-release.

Figure 4. Retention of RAD21-N and Sororin on cohesin in the absence of HDAC8.

a, RAD21 separase cleavage sites. Vertical lines indicate cleavage sites. Gray region indicates the RAD21-N antibody epitope. b, SMC3, STAG1 and STAG2 were immunoprecipitated from soluble extracts from normal (C) and HDAC8-mutant (H) cell lines. Full-length RAD21 (FL), the cleaved N-terminal fragment of RAD21 (N), acetylated SMC3 (SMC3-ac), and STAG1/2 were analyzed by SDS-PAGE and immunoblotting. c, Experimental schema of panel (d). HeLa cells were synchronized by double thymidine arrest (DTA) and released (R) into the presence or absence of PCI. Seven hours after release, cells were treated with RO-3306 (RO) to arrest at the G2/M boundary. After removing RO-3306, cells were cultured in the presence or absence of nocodazole (Noc) or MG-132 (MG) to arrest in metaphase. Cells were harvested at indicated time points after release from RO-3306 arrest. d, Cell extracts were fractionated into soluble and chromatin-bound fractions. Full length RAD21 (FL), cleaved C-terminus fragment of RAD21 (C), cleaved N-terminus fragment of RAD21 (N), acetylated SMC3 (SMC3-ac), SMC3, phosphorylated Histone H3 at Serine 10 (H3S10-P), and α-Tubulin were analyzed by SDS-PAGE and immunoblotting. H3S10-P is used as a metaphase marker. e, Experimental schema of panel F and G. HeLa cells were synchronized and released and treated with RO-3306 (RO) as in D. After removing RO-3306, cells were cultured in the presence or absence of PCI and harvested at the indicated times. f, FACS analysis of HeLa cells in (g). g, Cell extracts were fractionated into soluble and chromatin-bound fractions. Levels of SMC3-ac, RAD21-N, full length RAD21 (FL), Sororin, WAPAL, and Cyclin B (a G2 marker) and SMC3 were analyzed by SDS-PAGE and immunoblotting.

We also made several observations regarding HDAC8 activity in the prophase removal of cohesin^25,26. Loss of HDAC8 activity does not disrupt the prophase pathway of cohesin removal from chromatin, as demonstrated by a consistent reduction of chromatin-bound cohesin in the presence of metaphase inhibitors in either control or PCI-treated cells (Fig. 4d, lanes 13 and 17 vs. 21-24). However, HDAC8 activity is necessary following prophase removal since, in metaphase inhibitor-treated cells, soluble SMC3-ac is markedly increased in the presence of PCI (Fig. 4d, lanes 9-12). Finally, with WAPAL depletion, which diminishes prophase removal of cohesin^18,27,28, we noted that SMC3-ac is substantially increased (Supplementary Fig. 9e).

To test whether other cohesion-promoting proteins might remain associated with SMC3-ac in the absence of HDAC8 activity, we analyzed HeLa and patient-derived cells, which also demonstrated increased retention of Sororin^18,29 and RAD21-N, but not WAPAL, on soluble SMC3-ac (Fig. 4e-g, Supplementary Fig. 9d). Taken together, these data strongly suggest that HDAC8 is necessary for deacetylation of SMC3 following its removal from chromatin in both prophase and anaphase pathways to allow proper dissolution of pro-cohesive elements and allow for recycling of “refreshed” cohesin for the following cell cycle (Supplementary Fig. 10).

In summary, this work demonstrates that HDAC8 functions as a vertebrate SMC3 deacetylase to facilitate renewal of cohesin following its removal from chromatin in prophase or anaphase and that loss of HDAC8 activity results in decreased cohesin at localized sites to cause both cellular and clinical features of CdLS.

METHODS

Antibodies and reagents

Primary antibodies used in this study are as previously described for RAD21 (hSCC1)¹⁹, SMC3-ac¹⁸ and RAD21 (Scc1) N terminus³⁰. Antibodies used are as follows: SA1, SA2, SMC1A, HDAC6, HDAC10, SIRT4, SIRT5, SIRT7 were from Abcam (Cambridge, MA); α-Tubulin was from Sigma (Tokyo, Japan); Histone H3(S10P), HDAC1, HDAC2, HDAC3, HDAC4, HDAC5, SIRT1, SIRT3, SIRT6, SIRT7 and MAD2L1 were from Cell Signaling Technology (Danvers, MA); HDAC7, HDAC11 and SIRT2 were from Millipore (Billerica, MA) and SMC3 was from Bethyl Laboratories (Montgomery, TX). HDAC8 antibodies were purchased Santa Cruz Biotechnology (Santa Cruz, CA) or derived as mouse monoclonal HDAC8 antibody raised against peptide CDAYLQHLQKVSQEGDDDHPDS. PCI-34051 was purchased from Sigma Chemical (St. Louis, MO) or synthesized by Synstar Japan Co. TSA, sodium butyrate (SB) and nicotinamide (Nico) were from Wako Chemical (Osaka, Japan). Suberoyl anilide bishydroxamide (SAHA) was from Toronto Research Chemicals (North York, Ontario, Canada). Nocodazole was from Calbiochem (Darmstadt, Germany).

Cell culture and synchronization

HeLa cells were cultured in DMEM (Invitrogen, Camarillo, CA) supplemented with 0.2 mM L-glutamine, 100 units/ml penicillin, 100 μ/ml streptomycin and 10% FCS. LCLs were cultured in RPMI-1640 supplemented with 0.2 mM L-glutamine, 100 units/ml penicillin, 100 μg/ml streptomycin and 20% FCS. Fibroblasts were cultured in DMEM (Invitrogen) supplemented with 0.2 mM l-glutamine, 100 units/ml penicillin, 100 μg/ml streptomycin and 20% FCS. For cell cycle synchronization, HeLa cells were synchronized by double thymidine arrest (DTA; 14–16 h in the presence of 2 mM thymidine, 8 h release, 16 h in the presence of 2 mM thymidine) and harvested 6 h and 13 h after the second release for enrichment in the G2 phase and in the G1 phase, respectively. For enrichment in prometaphase of mitosis, nocodazole (330 nM) was added 8 h after release from the second thymidine block for 1 or 2 h, and cells were harvested by shake off.

Protein extraction and immunoprecipitations

To obtain total cell extracts, cells were lysed with lysis buffer (20mM Tris-HCl pH8.0, 100mM NaCl, 10mM KCl, 5mM MgCl2, 0.2% NP-40, 10% Glycerol, CompleteTM, phosSTOP) and DNA in the chromatin fraction was digested by the treatment of benzonase (Novagen). To obtain soluble cell extracts the cells were lysed with lysis buffer containing 200mM NaCl and centrifugated at 20,000 × g. Immunoprecipitations were performed using ProteinG magnetic beads conjugated with 5μg antibodies and incubated with the cell extracts for 5hr at 4C. After washing with Lysis buffer, the beads were suspended with SDS-PAGE sample buffer.

RNA isolation and RT-PCR

Total RNA was isolated using Trizol (Invitrogen) and Nucleospin RNA II (Macherey-Nagel, Düren, Germany) following the manufacturer’s instructions. Synthesis of cDNA was performed using SuperScript III First-Strand Synthesis System (Invitrogen). cDNA was amplified with KAPA SYBR Fast qPCR kit (Kapa Biosystems, Mowbray, South Africa) on an ABI 7500 Real Time PCR system (Applied Biosystems, Carlsbad, CA). Primer sequences were: HDAC9 Forward-CTTTGTCAGGTTCCTGCTGTTCTC, Reverse-TTCCTGTTTCCACAAGGCATTTC; ACTB Forward-TGGCACCCAGCACAATGAA, Reverse-CTAAGTCATAGTCCGCCTAGAAGCA and GAPDH Forward-GCACCGTCAAGGCTGAGAAC and Reverse-TGGTGAAGACGCCAGTGGA.

Immunofluorescence microscopy

HeLa cells grown on 18-mm coverslips were fixed with 4% PFA and permeabilized with PBS containing 0.5% Triton X-100 for 10 min. After blocking with goat serum in PBS, cells were incubated with primary antibodies for SMC3 or SMC3-ac. DNA was stained with DAPI.

Chromatin fractionation

Performed as previously described³¹.

Chromosome spreads

Cells treated with nocodazole (100 ng/ml) were harvested by mitotic shake off and hypotonically swollen in 40% PBS/ 60% tap water for 5 min at room temperature. Cells were fixed with Carnoy’s solution (methanol:acetic acid=3:1), dropped on glass slides and dried. Slides were stained with 5% Giemsa (Merck, Tokyo, Japan), and washed with water, air-dried, mounted with Entellan (Merck, Tokyo, Japan).

Purification of recombinant HDAC8 and SMC3-ac deacetylation assay

Full-length cDNA of HDAC8 and the enzymatically dead HDAC8 mutant (D101A/Y306F)²² was subcloned into pFASTBacHT (Invitrogen). The Bac-to-Bac baculovirus expression system (Invitrogen) was used to produce recombinant virus according to the manufacturer’s protocol. To express recombinant protein, Sf9 cells were infected with each baculovirus and incubated for 60 hours at 27C. Total cell lysates were prepared using Laemmle SDS buffer and boiling for 5 min. For purification of HDAC8 protein, infected cells were lysed with buffer A (20mM Tris-HCl(pH7.5), 300mM NaCl, 0.1% NP-40,10mM 2-ME). After centrifugation at 3000×g the supernatants were incubated with Ni-NTA beads for 2 h at 4C. The beads were washed with buffer A, washed with buffer B [20mM Tris-HCl (pH7.5), 100mM NaCl, 10% Glycerol 0.05% NP-40, 1mM DTT] and eluted with buffer B containing various concentrate of imidazole (50, 100, 200, 350mM). The elution fractions including HDAC8 protein were collected and concentrated using a Centricon concentrator. For detection of SMC3 deacetylation by HDAC8, solubulized total lysates treated with benzonase or SMC1A immunoprecipitants were used as substrates. Substrates were incubated with rHDAC8 or rHDAC8 mutant in reaction buffer [50mM Hepes-KOH(pH7.4), 10% Glycerol, 10mM KCl, 2.5mM MgCl2, 100mM NaCl , 0.1mM PMSF, 50ng/ml BSA, 0.02% NP-40) for 1 hour at 30C and then boiled with SDS sample buffer.

RNA interference

The siRNA oligonucleotides were annealed according to manufacturer’s instruction and used at a final concentration of 100–200 nM. The siRNA transfections were performed using oligofectamine (Invitrogen). Oligonucleotides targeting the firefly luciferase GL2 were used as controls. All siRNA sequences are listed in Supplementary Table 2.

Human Subjects

All individuals enrolled in the study were diagnosed by clinical geneticists experienced in the diagnosis of CdLS to have clinical features consistent with a diagnosis of CdLS. All patients and family members were enrolled in the study under an Institutional Review Board-approved protocol of informed consent at The Children’s Hospital of Philadelphia or the Institut für Humangenetik Lübeck.

Mutation Screening

All genes discussed were screened for mutations in the coding exons and intron-exon boundaries using PCR of genomic DNA followed by sequencing. Primers were designed using ExonPrimer ³². Primer sequences and PCR conditions are available upon request. Sequencing was performed using BigDye Terminator v3.1 cycle sequencing and analyzed on an ABI 3730 (Applied Biosystems, Carlsbad, CA). All probands were pre-screened and were negative for mutations in NIPBL, SMC1A and SMC3.

Reference sequences and HDAC8 conservation analysis

RefSeq ID numbers for mRNAs and proteins, respectively, referenced in this work include: HDAC8 (NM_018486, NP_060956), NIPBL (NM_133433, NP_597677) and SMC1A (NM_006306, NP_006297). HDAC8 protein sequences for human (AAF73428), Bos taurus (DAA12953), rat (AAI62023), mouse (CAM17598), Danio rerio (NP_998596), Xenopus laevis (NP_001085711), Xenopus tropicalis deduced from mRNA BC161282, Drosophila melanogaster (AAC61494) and for Saccharomyces cerevisiae HDAC8-like proteins, HOS2 (NP_011321) and RPD3 (AAT92832) were aligned by the ClustalW method³³ using MacVector software (Accelrys Corp, San Diego, CA).

Mapping mutations to the HDAC8 crystal structures

The identified HDAC8 mutations were mapped onto the crystal structure data of human HDAC8-substrate complex (PDB accession 3F06²³) to visualize location with respect to active domains of the protein using Cn3D ³⁴, and PyMol ³⁵ software.

Expression analysis of CdLS 32 gene classifier

Lymphocyte culture, and RNA extraction were performed as previously described²⁰. Genes and Nanostring³⁶ (Seattle, WA) probeset sequences are listed in Supplementary Table 3. For each sample 100 ng of RNA was analyzed. Hybridizations were carried out by mixing 5ul of each RNA sample (normalized to 20 ng/ul) with 20 ul of NanoString nCounter Reporter probes and 5 ul of nCounter Capture probes (30 ul total reaction volume) and incubating the hybridizations at 65°C for 18 hours. Post-hybridization purification, NanoString reporter capture, stretching and imaging was performed as described³⁶. To account for differences in assay efficiency the data was normalized to the sum of 6 positive internal control RNA spikes that ranged from 0.125-128fM. Internally normalized data was then subsequently normalized to the geometric mean of the ACTB and RPL19 spike-normalized counts. To assess reproducibility, Pearson’s r correlation coefficient was calculated for three technical replicates (average r=0.99986) and 11 biological replicates (average r=0.99675). To define a “CdLS” profile, average expression and standard deviations were calculated for 32 genes using RNA from lymphoblastoid cells lines from 10 normal controls and 12 NIPBL truncating or nonsense mutations. For each patient gene expression assay, the number of deviations in expression from normal was determined and divided by the average NIPBL expression standard deviation. These products were then summed for each gene to give a score representing a sum of deviation from normal in a “CdLS” manner for all 32 genes tested for each patient. The data and calculation results for each step are in Supplementary Data File 1.

Recombinant HDAC8 protein expression and purification from E. coli

Identified human HDAC8 mutations were introduced into a previously described HDAC8-6His-pET20b construct³⁷ using a QuickChange site-directed mutagenesis kit (Agilent Genomics, Santa Clara, CA). Oligonucleotide sequences are available upon request. HDAC8 was recombinantly expressed in BL21(DE3) E. coli cells and purified over Talon resin (Clontech, Mountainview, CA) according to published procedures ²³, with minor modifications. Briefly, 50 mL cultures (Luria-Bertani media supplemented with ampicillin (50 μg/L)) were grown overnight and used to inoculate 1 L flasks (minimal media supplemented with ampicillin at 50 μg/L). Cells were grown at 37 °C until OD₆₀₀~0.5, at which point the cells were induced by the addition of isopropyl β-D-thiogalactopyranoside (0.4 mM final concentration) and ZnCl₂ (100 μM final concentration), and grown overnight at 18 °C. The cells were pelleted by centrifugation, resuspended in 25 mL of lysis buffer (50 mM Tris (pH 8.0), 500 mM KCl, 5% glycerol, 3 mM β-mercaptoethanol (BME), 115 μM phenylmethanesulfonyl fluoride), and lysed by sonication on ice. Cell debris was pelleted by centrifugation and the cell-free extract was purified by affinity chromatography (Talon resin, Clontech Labs) using a step gradient (Buffer A: 50 mM Tris (pH 8.0), 500 mM KCl, 5% glycerol, 3 mM BME; Buffer B: 50 mM Tris (pH 8.0), 500 mM KCl, 5% glycerol, 3 mM BME, 250 mM imidazole). The final yield was approximately 8.6 mg/L culture for wild-type HDAC8, 6.7 mg/L for G320R HDAC8, 6.1 mg/L for H334R HDAC8, 5.8 mg/L for H180R HDAC8, and 3.9 mg/L for T311M HDAC8; protein concentrations were determined using the colorimetric Bradford assay³⁸. The final purity of each protein sample was better than 90% as determined by SDS-PAGE analysis.

HDAC8 activity was measured using the commercially available Fluor-de-Lys HDAC8 deacetylase substrate and Developer II (Enzo Life Sciences, Plymouth Meeting, PA). All assays were run at 25°C and contained 0.5 μM enzyme and 150 μM substrate in assay buffer [25 mM Tris (pH = 8.2), 137 mM NaCl, 2.7 mM KCl, 1 mM MgCl₂]. After 30 minutes, the reactions were quenched by addition of the HDAC8 inhibitor M344 (Sigma Aldrich, 100 μM final concentration) and Developer II. Fluorescence was measured using a Fluoroskan II plate reader (excitation = 355 nm, emission = 460 nm). Product concentration was calculated from raw data using standard curves for substrate and product (Aminomethylcoumarin, Enzo Life Sciences), with the total concentration of substrate and product kept at 150 μM (before dilution with the developer solution). All activity assays were performed in triplicate. Enzyme activities are reported as nmol product·μmol enzyme⁻¹·min⁻¹.

Assessment of X-inactivation

X chromosome inactivation (XCI) was determined by evaluating the methylation status of the CAG microsatellite locus at the 5′ end of the (AR) androgen-receptor gene as previously described with modifications ^39,40. Briefly, genomic DNA isolated from peripheral blood and for each patient, two reaction digests were performed. In one reaction, 1 μg of DNA was digested in 25 μl with the methylation-sensitive restriction enzyme HpaII (New England Biolabs), which cuts the active unmethylated allele. In the other reaction, DNA was incubated in enzyme digest buffer without enzyme. After a 16 hour incubation at 37°C, digestion was terminated by incubation at 65°C for 20 minutes. From each reaction, 2 μl was then amplified by PCR with primers flanking the polymorphic androgen receptor CAG repeat as described ³⁹; forward 5′-CTGTGAAGGTTGCTGTTCCTCAT -3′; reverse 5′-FAM-TCCAGAATCTGTTCCAGAGCGTGC -3′). ABI3730 Genetic Analyzer and GeneMapper V4.0 software (Applied Biosystems) were used for genotyping analysis. Percent X-inactivation was calculated by dividing the ratio of the allele peak volumes in the HpaII-treated by the ratio of the allele peak volumes in the untreated sample. We considered alleles separated by more than 2 CAG repeats as informative ⁴¹ and used a cutoff of >80:20% for skewed XCI and >95:5% for extremely skewed XCI.

Fibroblast immortalization and lentiviral transduction

CDL016 fibroblasts were immortalized by exogeneous expression of hTERT. HDAC8 and the enzymatically dead HDAC8 mutant (D101A/Y306F)²² were cloned into pENTR H1 Gateway vector and recombined with lentivirus vector, pCS-RfA-CG, using the LR reaction (Invitrogen). Recombinant lentiviruses were produced from 293FT cells transfected with CS-CMV-HDAC8 or CS-CMV-HDAC8mut and packaging vectors. The fibroblasts were incubated with recombinant lentiviruses in the presence of 8 μg/ml poly-brene and transduction efficiency was measured using GFP expression.

Chromatin immunoprecipitation and qPCR

Cells were crosslinked with 1% formaldehyde for 10 min, quenched with 125◻mM glycine, and prepared for ChIP as previously described¹⁹. ChIP was performed as previously described using RAD21/SCC1, SMC3-ac, and control antibodies¹⁹. Briefly, crosslinked cell lysates solubulized by sonication were incubated with protein A or protein G Dynabeads (Dynal) crosslinked with the antibodies for 14 h at 4 °C. After this, beads were washed several times and eluted with elution buffer (50 mM Tris, 10 mM EDTA, 1% SDS) for 20 min at 65 °C. The eluates were incubated at 65 °C overnight to reverse crosslinks and then treated with RNaseA and then with proteinase K. The samples were further purified by phenol-chloroform extraction and an additional purification step using a PCR purification kit (Qiagen). Real-time PCR was performed by using KAPA SYBR Fast qPCR kit (Kapa Biosystems) on an ABI 7500 PCR system (Applied Biosystems). The results were presented as a percentage of input chromatin that was precipitated. Primers used in this study were listed in Supplementary Table 4.

ChIP-seq analyses

DNA from whole cell extracts (WCE) and ChIP fractions was further sheared to an average size of approximately150bp by ultrasonication (Covaris, Woburn, Massachusetts), end-repaired, ligated to sequencing adapters and amplified according to manufacturers instructions (Applied Biosystems SOLiD Library Preparation Protocol). Gel-purified amplified DNA between 100 and 150bp was sequenced on the Applied Biosystems SOLiD platforms (SOLiD 3 and 5500) to generate single-end 50-bp reads. Sequenced reads of both ChIP and WCE were aligned to the human genome (UCSC hg19) using Bowtie⁴² allowing 3 mismatches in the first 28 bases per read (−n3 option). All duplicate reads and those without unique alignment were removed from further analysis. Sequencing, read and mapping data are summarized in Supplementary Table 5. We further analyzed only uniquely aligned reads. Each aligned read was extended to a predicted fragment length of 150bp. Reads were summed in 10bp bins along the chromosome for ChIP and WCE, respectively. To facilitate comparison of detected peaks between different ChIP experiments, we normalized the read number of each bin per million uniquely mapped reads for that chromosome for both ChIP and WCE samples (i.e. read number for each bin * 1,000,000 / total number of reads mapped onto reference genomes). This value was further smoothed with a 500bp width (50 bins). Scanning the genome with a 300 bp (30 bin) sliding window, a one-sided Wilcoxon rank-sum test was performed to estimate the enrichment p-value for each window. The fold enrichment (ChIP/WCE) for each window was also calculated. To call peaks, we employed stringent criteria to identify windows which satisfied both fold enrichment > 3.0 and p-value < 1e-4 criteria for candidate binding sites⁴³. To further eliminate uncertain sites, we removed the regions with low ChIP reads using the criteria: 1). the average number of ChIP reads in region / the average number of ChIP reads in the genome < 3.0 or 2). the maximum read intensity in ChIP bins was less than 0.2 per million reads mapped. Peak calling summary statistics are summarized in Supplementary Table 6.

RNA-seq analyses

We prepared 2 replicates for each sample for RNA-Sequencing (Solexa HiSeq2000, Illumina, San Diego, CA). Sequenced RNA reads were aligned to RefSeq RNA database (NM* accession numbers only) using Bowtie⁴² allowing three mismatches in the first 28 bases per read (−n3 option), and aligned transcript reads were merged for a single gene. To accomodate transcriptional variants, we used both uniquely- and multiply-aligned reads. Multiply aligned reads were divided equally amongst all locations (N-times matched reads were weighted as 1/N reads). Reads that aligned to reference database more than 10 times were discarded. The expression level of each gene was calculated in RPKM (reads per kilobase per million reads)⁴⁴ and normalized by TMM (trimmed mean of M values)⁴⁵. RPKM scores of the two replicates were averaged to calculate the enrichment ratio. Using a 2.5 RPKM threshold, we obtained a total of 9,763 expressed genes (Supplementary Table 7).

Data availability

ChIP-Seq and RNA-Seq data from this study is available from the Sequence Read Archive (SRA) database (http://www.ncbi.nlm.nih.gov/sra) under the accession number SRP011927.