Reorganization of inter‐chromosomal interactions in the 2q37‐deletion syndrome

Abstract

Chromosomes occupy distinct interphase territories in the three‐dimensional nucleus. However, how these chromosome territories are arranged relative to one another is poorly understood. Here, we investigated the inter‐chromosomal interactions between chromosomes 2q, 12, and 17 in human mesenchymal stem cells (MSCs) and MSC‐derived cell types by DNA‐FISH. We compared our findings in normal karyotypes with a three‐generation family harboring a 2q37‐deletion syndrome, featuring a heterozygous partial deletion of histone deacetylase 4 (HDAC4) on chr2q37. In normal karyotypes, we detected stable, recurring arrangements and interactions between the three chromosomal territories with a tissue‐specific interaction bias at certain loci. These inter‐chromosomal interactions were confirmed by Hi‐C. Interestingly, the disease‐related HDAC4 deletion resulted in displaced inter‐chromosomal arrangements and altered interactions between the deletion‐affected chromosome 2 and chromosome 12 and/or 17 in 2q37‐deletion syndrome patients. Our findings provide evidence for a direct link between a structural chromosomal aberration and altered interphase architecture that results in a nuclear configuration, supporting a possible molecular pathogenesis.

Introduction

Elucidating the spatial organization and the interactions of interphase chromosomes within the nucleus is pivotal for understanding gene regulation and the maintenance of cellular identity (Cremer & Cremer, 2010; Pederson, 2011; Gorkin et al, 2014; Dixon et al, 2015). Chromosomes themselves are organized into evolutionarily conserved topologically associated domains (TADs) that harbor extensive intra‐chromosomal interactions to regulate gene expression (Dixon et al, 2012). Structural chromosomal aberrations may result in TAD alterations, leading to re‐wiring of enhancer–promoter interactions that are associated with human disease (Lupianez et al, 2015). Moreover, these structural chromosomal rearrangements, especially duplications, can support the formation of new regulatory chromatin domains (neo‐TADs) leading to ectopic intra‐chromosomal contacts and differential gene expression. Thus, several studies have pointed to a potential mechanistic link between nuclear architecture and pathogenesis (Franke et al, 2016).

Normally, the positioning of genes during interphase is constrained, and only mitosis‐mediated changes or major gene expression changes can lead to gene repositioning within the nucleus (Chubb et al, 2002; Kumaran & Spector, 2008; Ferrai et al, 2010). Gene positions are dependent on the chromatin neighborhood and are not randomly located within the nucleus (Jost et al, 2015). Generally, transcriptionally active loci tend to be in the nuclear center, whereas silenced genes are located closer to the nuclear periphery, which is associated with heterochromatin (Peric‐Hupkes et al, 2010). Protein recruitment and transcriptional activity, rather than histone composition, may be responsible for directed nuclear repositioning of a genomic region (Wijchers et al, 2016).

Despite extensive investigations into intra‐chromosomal communication or cis interactions, the phenomena of inter‐chromosomal interactions (non‐homologous chromosomal contacts: NHCCs) or “chromosome kissing” remain elusive (Kioussis, 2005; Spilianakis et al, 2005). Others showed earlier that non‐randomly organized chromosomal territories of interphase chromosomes intermingle to form NHCCs (Bolzer et al, 2005; Branco & Pombo, 2006; Manvelyan et al, 2008). However, how these spatial intermingling chromosomes are arranged to one another and how specific loci are involved has received little attention.

More recently, long non‐coding RNA (lncRNA) loci have been shown to facilitate specific NHCCs (Maass et al, 2012; Hacisuleyman et al, 2014; Rinn & Guttman, 2014). For example, we determined earlier that the lncRNA locus CISTR‐ACT on chromosome 12 contacts PTHLH on chromosome 12, and SOX9 on chromosome 17 by FISH in crosslinked cells and by CRISPR live‐cell imaging (CLING) in living cells (Maass et al, 2012, 2018). Moreover, 3D‐FISH showed adjacent chromosome 12 and 17 territories (Bolzer et al, 2005). Misplacement of CISTR‐ACT caused disrupted NHCCs with dysregulation of chondrogenesis genes on chromosome 17 (PTHLH, HOXB, NOG, ETV4, SOX9, etc.), in different translocation families with autosomal‐dominant brachydactyly Type E (BDE2, OMIM 613382; Maass et al, 2012). Thus, intact nuclear locus positioning in chromatin is important to ensure gene regulation either by intra‐chromosomal, inter‐chromosomal regulation, or by associated proteins (Maass et al, 2012; Hacisuleyman et al, 2014; Wijchers et al, 2016). BDE was also linked to various heterozygous chromosome 2q37‐deletions, mutations, and partial or total deletions of HDAC4 within the autosomal‐dominant brachydactyly mental retardation syndrome (BDMR; OMIM 600430, 605314). BDMR features also short stature, obesity, craniofacial abnormalities, developmental delay, intellectual impairment, behavioral problems, neurological and cardiac alterations, and BDE (Phelan et al, 1995; Wilson et al, 1995; Wolff et al, 2002; Aldred et al, 2004; Casas et al, 2004; Chaabouni et al, 2006; Falk & Casas, 2007; Galasso et al, 2008; Williams et al, 2010a; Morris et al, 2012; Leroy et al, 2013; Villavicencio‐Lorini et al, 2013; Jean‐Marcais et al, 2015).

Isolated BDE in translocation families is similar to the form of BDE that was observed in a family with BDMR, which we have studied here (Maass et al, 2010, 2012). Thus, we aimed to investigate chromosome 2, 12, and 17 chromosomal arrangements in healthy donors and in the BDMR family by multi‐color DNA‐FISH and the genome‐wide chromosomal conformation capture technique Hi‐C. In normal state, we found that NHCCs are stable and not tissue‐specific, but single loci showed tissue‐specific interactions, while in BDMR patients with 2q37‐deletion disrupted NHCC formation, increased co‐localization of chromosomal territories, and differential gene expression were found. Our study provides new insights into the fundamental properties of NHCCs and how their misarrangement can contribute to human disease.

Results

Arrangements of chromosomes 12 and 17

To understand the normal relationship between chromosomes 12 and 17, we first investigated their spatial arrangements in mesenchymal stem cells (MSCs) from three healthy donors, unrelated to the BDMR family (Materials and Methods). To validate that these are indeed MSCs, we first performed flow cytometric analysis of MSCs to demonstrate expression of MSC surface markers CD105⁺, CD90⁺, CD73⁺, HLA‐ABC⁺, CD31⁻, CD34⁻, CD45⁻, and HLA‐DR⁻ (Appendix Fig S1A). Second, we were able to differentiate MSCs into proliferating chondrocytes, osteocytes, and adipocytes to confirm the multi‐lineage potential, fulfilling the minimal criteria for defining MSCs in vitro (Appendix Fig S1B; Dominici et al, 2006; Salem & Thiemermann, 2010).

To study NHCCs of chromosomes 12 and 17, we labeled chromosome 12 (green) and chromosome 17 (red) by whole chromosome painting (WCP) FISH (Fig 1A and B). The visual analysis of the microscopic images indicated that the chromosome 12 and 17 territories intermingled more often than separated (non‐interacting) chromosomes occurred (Fig 1B and C). To quantitatively determine the real overlap of the intermingling chromosomal territories and to distinguish between co‐localized and separated chromosomes, we analyzed the true degree of co‐localization of chromosomes 12 and 17 by calculating Manders split coefficients m1 and m2 (see Materials and Methods, Fig 1C, Appendix Fig S1C) (Manders et al, 1993; Zinchuk, Zinchuk et al, 2007). M1 describes co‐localized chromosome 17 regions with chromosome 12, while m2 represents the degree of chromosome 12 co‐localizing with chromosome 17. We analyzed the MSC nuclei and detected that both chromosomes 12 and 17 co‐localized with one another at similar levels (Fig 1D, median m1: 60%, m2: 68%, 100 comparisons of chr. 12 and 17). Low degrees of co‐localization indicated spatially separated chromosomes (Fig 1D). Next, we sought to test whether the co‐localization of chromosomes 12 and 17 is significant. By using Costes's P‐value (Costes et al, 2004), we calculated the correlation between the observed experimental FISH images and randomized test images. Specifically, to assess whether the randomized images displayed a better correlation than the original observed chromosomal arrangements, the positioning of WCP‐FISH signals was randomized 100 times per imaged nucleus. A Costes's P‐value of 1 indicated the best correlation that none of the randomized images has better correlation than the original data (Costes et al, 2004). Using these calculations, we detected that 88% (median) of the regions of interest (ROIs) were co‐localized and that intermingling chromosomal territories were more frequent than separated chromosomes (Fig 1D). This result suggests a common and significant interaction of chromosomes 12 and 17 across a cell population and when compared to randomized controls.

Figure 1. NHCC arrangements of chromosomes 12 and 17.

1. Multi‐color FISH approach to label chromosome 12 (green); PTHLH (cyan); CISTR‐ACT (magenta); chromosome 17 (red); and ETV4, HOXB, NOG, and SOX9 (green).
2. Example of multi‐color FISH. Contrast‐increased signal intensities were used to depict maximal distributions of chromosomal territories in Fig 1E (dashed lines Fig 1E). For quantitative analysis, non‐processed images were used.
3. Representative analysis of (left) raw, non‐processed WCP signals to determine the degree of co‐localization (chromosomal contact) degrees by Manders split coefficients. Middle: gray scale images of single channels of each WCP signal were used as input to (right) quantify true co‐localization, shown as white regions (ROI = region of interest).
4. Left: quantification of co‐localized (intermingling) ROIs of chromosomal territories by Manders coefficients m1 and m2. The distribution showed different degrees of co‐localization: spatially separated to strongly co‐localized regions of the chromosomal territories 12 and 17. Right: Costes P‐value (best value = 1) determined that 88% of the original data were correlated more robustly than randomized test images (100×) of chromosomal pattern.
5. Top and middle rows: schemes and representative non‐processed images of chromosomal pattern (I–V, raw WCP signals) for chromosome 12 interacting with chromosome 17 in MSCs. Dashed lines depict the maximal distribution of the chromosomal territories (Fig 1B). Either mono‐allelic or bi‐allelic intermingling was detected (pattern I–IV). In pattern V, the chromosomal territories were separated. Bottom: In the same nuclei, images of single loci were acquired. Co‐localizations of CISTR‐ACT (magenta) with either PTHLH (cyan) or one of the chondrogenesis genes (ETV4, HOXB, NOG, SOX9) on chromosome 17 (green) were identified by overlapping or merged signals (white arrowheads = BAC signals, white arrows = co‐localizations, see also Fig 1A and B). Scale bars = 5 μm.
6. Distribution of chromosomal pattern in MSCs. Patterns III (**P < 0.001) and IV (*P < 0.05) were the most frequent arrangements in 100 nuclei of each MSC donor, Student's t‐test, medians ± min‐to‐max are shown.
7. Randomizations of experimental data from three MSC donors were tested by FDR‐corrected ANOVA, and used in Monte Carlo simulations (1,000×).
8. Simulating the data of MSC1‐3 (Fig 1F) in a power analysis indicated that > 83% of the experimental results were true (P < 0.05).
9. MSCs of donor 1 were differentiated into adipocytes, chondrocytes, osteocytes, and vascular smooth muscle cells (VSMCs). In the MSC‐derived cells, patterns III and IV were the commonest (FISH, each cell type 100 nuclei).
10. Monte Carlo simulations (1,000×) determined that > 98.9% of the chromosomal pattern in MSC‐derived cells were statistical true hits (P < 0.05).
11. Manders split coefficients m1 and m2 determined significant differences of the degrees of co‐localization between patterns I–IV and V (ANOVA ***P < 0.0001, 220 comparisons of 4 independent replicates).
12. Costes P‐value analysis randomized data of original images (100×). A maximal Costes P‐value of 1 means that none of the randomized images had a better correlation than the original image (ANOVA ***P < 0.0001).
13. With regard to the chromosomal pattern of chromosomes 12 and 17, separate or joined chromosome X or Y in male MSCs and MSC‐derived cells of donor 1 were quantified in at least 100 nuclei of each tissue. In the most frequent arrangements of patterns III and IV, separated chromosomes X and Y were also highly frequent and more common than joined X and Y (FISH, each cell type 100 nuclei, medians, 25^th and 75^th percentiles ± min‐to‐max are shown).

Next, we explored how chromosomes 12 and 17 were arranged relative to one another, and if this pattern differ between individual MSC donors. Specifically, we distinguished intermingled from separated chromosomes due to the degree of co‐localization, and interestingly, we noted recurring chromosomal pattern among the intermingled chromosomes 12 with 17. Based on the chromosomal arrangements to one another, we classified different NHCC pattern (Fig 1E, Appendix Fig S2A). In pattern I or II, at least one allele of chromosome 12 or 17 was proximal to the other. Pattern III showed either a chromosome 12–17 mono‐allelic or bi‐allelic neighboring. Pattern IV demonstrated bi‐allelic proximity resulting in tetramer‐like complexes, and pattern V represented distinct separated chromosomal territories without any crosstalk (Fig 1E). Next, we counted the occurrence of this chromosomal pattern in 100 nuclei from each of the three MSC donors. Independent of the donor, we found that pattern III (~31%) and IV (~35%) were the most common spatial arrangements (Fig 1F, Appendix Fig S2B, P < 0.05, P < 0.01). To test how meaningful our results from 100 nuclei of each of the three different donors were over randomized controls, we randomized the experimental data, compared the ratios of chromosomal patterning by FDR‐corrected ANOVA testing, and ran Monte Carlo simulations (1,000×, Fig 1G). The simulations determined that more than 83% of the analyzed chromosomal arrangements were true hits with significant P‐values of < 0.05 (confidence interval: 82.2–86.8%, Fig 1H). Collectively, these findings suggested that while the positioning of the entire territories of chromosomes 12 and 17 was variable between individual cells, they favored specific NHCC pattern that recur significantly in different individuals.

Next, we asked whether the detected chromosomal pattern I–V exhibited tissue specificity. For this purpose, we differentiated MSCs (MSC1 donor) into adipocytes, chondrocytes, osteocytes, and vascular smooth muscle cells (VSMCs). In these MSC‐derived cell types, we observed only minor changes in the distributions of pattern compared to non‐differentiated native MSCs. Similar to MSCs, pattern III (mono‐allelic: 25–35%) and pattern IV (bi‐allelic: 25–44%) were most often detected across multiple cell lineages (Fig 1I). As described above, we also tested for statistical significance by randomizing and by simulating the experimental data (Monte Carlo: 1,000×) from MSC‐derived cell lineages. We detected that > 98.9% of the data were statistically significant (P < 0.05, confidence interval 98–99.4%, Fig 1J). Thus, the detected chromosomal pattern is stable and not tissue‐specific.

Since we detected that chromosomes 12 and 17 occurred in five recurrent pattern across five different cell types (MSCs and MSC‐derived cells), we asked whether there were any differences in the co‐localized chromosomal regions between each pattern or cell type. Thus, we quantified the direct overlap (co‐localization) between chromosome 17 with 12 (m1), or 12 with 17 (m2) by Manders coefficients (Fig 1K, 220 chromosomes, 12/17 comparisons). M1 (62.6 ± 24.1%) and m2 (59.6 ± 20.9%) of pattern I–IV showed significantly more co‐localization, than m1 (28.5 ± 24.1%) and m2 (33.6 ± 15.6%) in pattern V (ANOVA ***P < 0.0001, Fig 1K). Also, Costes‐generated P‐values were significantly different between I–IV and V, without significant differences within pattern I–IV (ANOVA ***P < 0.0001, median of pattern I–IV: 0.92, pattern V: 0.12, Fig 1L). As a result, the majority of the observed datasets displayed higher correlations than randomized test images. These findings indicate that the chromosomal interactions are stable across chromosomal pattern I–IV, independent of the arrangement to one another. Interestingly, even when all four chromosomes were positioned in the tetramer‐like complex of pattern IV, the co‐localized regions interacted to a similar extent that was detected for pattern I–III. Collectively, we detected that similar levels of interaction occur between chromosome 12 and 17, independent of the investigated cell type.

To determine whether the identified chromosomal pattern occurs relative to other chromosomes of similar size, we selected the single chromosomes X and Y, which are similar in size to chromosome 12 and 17 (± 20 Mb). We performed four‐color WCP‐FISH in parallel for chromosomes X, Y, 12, and 17 in MSCs and MSC‐derived cells and analyzed 100 nuclei for each cell type. Separated X and Y chromosomes were observed more frequently (~26–28%) than intermingling X and Y chromosomes (~12–13%, Fig 1M, Appendix Fig S2C–H). Interestingly, these WCP experiments demonstrate similar spatial arrangements and distributions among chromosomes X, Y, 12, and 17. NHCC pattern III and IV were again the commonest (Fig 1M, compare with Fig 1F and I). Altogether, we observed that NHCCs are frequent, stable, and specific with respect to chromosomes 12 and 17 relative to X and Y, and not tissue‐specific.

Locus‐specific NHCCs in MSC‐derived cells

Having analyzed the positioning of chromosomes 12 and 17, relative to one another, we next sought to investigate NHCCs at the level of individual gene loci which we had also labeled by BACs in our WCP‐FISH approach described above (Fig 1A and B). To accomplish this aim, we focused on co‐localizations of the loci: CISTR‐ACT (magenta, chr. 12), PTHLH (cyan, chr. 12), the HOXB cluster (chr. 17), and the chondrogenesis genes ETV4, NOG, and SOX9 (green, chr. 17) with fluorescence‐labeled bacterial artificial chromosomes (BACs, Figs 1A and EV1A and C). With this approach, we were able to study whether or not any locus‐specific co‐localizations belonged to specific NHCC pattern (Fig 1E).

Figure EV1. WCP probes and BACs for DNA‐FISH .

• A–D
  FISH examples of metaphase chromosomes showing whole chromosome painting by WCP probes and single BACs labeling (A) PTHLH (WI2‐405L10) or CISTR‐ACT (WI2‐1544O11) on chromosome 12, or (B) randomly chosen regions as controls on chromosome 15, or (C) SOX9 (WI2‐934C8), HOXB (WI2‐1875I9), NOG (WI2‐675E8), ETV4 (WI2‐1441J11) on chromosome 17, or (D) a CNV for pod‐FISH on chromosome 2 with WCP probes for chromosome 2q, either in Texas Red, or DEAC, or SpectrumOrange, or SpectrumGreen or Bio/Cy5.
• E
  pod‐FISH on metaphase chromosomes of whole blood with BAC RP11‐80D14 and BAC RP11‐696D23 (ΔHDAC4) to track non‐affected chromosome 2 (red signal) and the HDAC4 deletion chromosome 2 (no signal) in the BDMR family.
• F
  Examples of WCP‐FISH signals in combination with pod‐FISH and HDAC4 deletion‐tracking on interphase chromosomes in a nucleus of one non‐affected and one affected individual (scale bar = 5 μm, arrowheads exemplarily depict BAC signals).

By quantifying co‐localizations in MSCs of three donors, we found that mono‐allelic co‐localizations between CISTR‐ACT and the loci on chromosome 17 (SOX9, or HOXB, or ETV4, or NOG; 46–57%) were more frequent than bi‐allelic tetramer‐like interactions (4–11%, Fig 2A). The intra‐chromosomal CISTR‐ACT–PTHLH interaction occurred in ~72% of the nuclei (Appendix Fig S2I),

Figure 2. CISTR‐ACT's inter‐chromosomal interactions.

1. FISH co‐localization frequencies of CISTR‐ACT–chromosome 17 interactions (ETV4, HOXB, NOG, or SOX9, each 100 nuclei) in MSCs and T‐L = T lymphocytes. Mono‐allelic inter‐chromosomal contacts were more often detected than bi‐allelic contacts. Loci on chromosome 15 were controls. Co‐localizations of loci on chromosomes 12–17 in MSCs were more often observed than co‐localizations between chromosomes 12 and 15 in T‐L (***P < 0.001, χ² test, each 100 nuclei)
2. Tissue‐specific co‐localization frequencies of CISTR‐ACT with either of the chondrogenesis genes on chromosome 17 revealed more mono‐allelic contacts in chondrocytes and osteocytes, than in adipocytes, or VSMCs, or MSCs (100 nuclei each cell type; *P = 0.02; **P = 0.003, ***P < 0.0001, χ² test).
3. Four‐color 3D‐FISH in MSCs determined that the territories of chromosomes 12 (red) and 17 (yellow) were interacting. Exemplifying 3D reconstructions of z‐stacks show that CISTR‐ACT (cyan) pinpointed the chondrogenesis genes on chromosome 17 (green).

To determine whether NHCCs were specific to adherent cells, we also analyzed primary human T lymphocytes grown in suspension and used control loci on chromosome 15 to compare with the chromosome 12–17 combinations (Fig EV1B). Overall, we observed relatively few inter‐individual differences. Co‐localizations were present in MSCs and T lymphocytes, but less frequent between CISTR‐ACT and the loci on chromosome 17 when compared to chromosomes 12–15 in T lymphocytes (P < 0.001, Fig 2A).

Above, we demonstrated that NHCCs are frequent and not tissue‐specific and that co‐localizations of individual loci are common in MSCs. Next, we wanted to determine co‐localization frequencies in other MSC1‐derived cell types (adipocytes, chondrocytes, osteocytes, and VSMCs), that could be disease‐relevant, since CISTR‐ACT is associated with BDE, which phenotypically manifests during chondrogenesis (Maass et al, 2012). For this aim, we quantified CISTR‐ACT co‐localizations with chondrogenesis genes on chromosome 17 and assessed whether the chromosomal pattern of these co‐localizations occurred in similar proportions to those identified above (Fig 2B, 100 nuclei each). Notably, the majority of mono‐allelic CISTR‐ACT co‐localizations with chromosome 17 chondrogenesis genes were within pattern III or IV in chondrocytes (P < 0.01, P < 0.001) and osteocytes (P < 0.05, P < 0.001). However, far fewer co‐localizations were observed in MSCs, or adipocytes, or VSMCs (Fig 2B). As a result, we observed significant differences between the different cell types, with more co‐localizations being detected in BDE‐relevant cell types: chondrocytes and osteocytes (Fig 2B). A similar, although non‐significant, trend was observed for CISTR‐ACT–PTHLH intra‐chromosomal co‐localizations, with more being present in chondrocytes and osteocytes (Appendix Fig S2I). Thus, NHCC pattern is stable and non‐tissue specific on a chromosomal scale, but more cell‐specific and dynamic at more refined scale of individual loci.

Finally, we confirmed the direct contact between chromosome 12 and 17, and investigated CISTR‐ACT's involvement, using 3D‐FISH (Fig 2C). Computational 3D reconstructions of nuclei showed that chromosomes 12 and 17 interacted and that the CISTR‐ACT locus was in spatial proximity to the chondrogenesis genes on chromosome 17 (Fig 2C). Any interaction of the chromosomal territories or specific loci corresponded to pattern V (Appendix Fig S2J). Collectively, these data indicate that the NHCC arrangements of the whole chromosomes 12 and 17 are similar across different cell types; in contrast, locus‐specific contacts do exist, and appear to be tissue‐specific, such as the CISTR‐ACT locus. Overall, chromosomes 12 and 17 were specifically organized in recurring arrangements across different cell types and were stable in proximity.

Clinical and molecular characterization of the BDMR family

Because we observed frequent chromosomal pattern between chromosomes 12 and 17 in normal karyotypes, we then sought to compare the results to a disease state, specifically the chromosome 2q37‐deletion syndrome or BDMR. We recruited a three‐generation family with autosomal‐dominant BDMR, who had typical BDE features, comprising shortening of the metacarpals and metatarsals, short stature, and cone‐shaped epiphyses (Figs 3A and B, and EV2A–C; Bell, 1951; Wilson et al, 1995). Notably, developmental delay, speech impairment, mental retardation (MR), and behavioral abnormalities such as autism or autism spectrum disorders were not clinically apparent in our patients (Williams et al, 2010a; Wheeler et al, 2014; Jean‐Marcais et al, 2015). The findings suggest that our family has BDMR, but without MR (Fig EV2A).

Figure 3. Molecular characterization of the HDAC4 deletion within the BDMR family.

1. Pedigree of the three‐generation BDMR family with autosomal‐dominant inheritance.
2. Hand‐roentgenograms of affected individual II‐1 displayed shortened metacarpals, especially of digit III and IV and cone‐shaped epiphysis. X‐rays of the feet showed variability in the shortening of metatarsals and distal phalanges. Arrows (black and white) indicate some BDE‐affected bones.
3. Genomic localization and intron–exon structure of the HDAC4 gene on chromosome 2q37. Positions refer to the human genome assembly hg19. The deletion breakpoints (BP) were assigned between intron 2 and exon 27 to position 240,196,438 and 239,974,129 bp (hg19), spanning 222,309 bp. The regulatory features (Ensembl or ENCODE) of the breakpoint‐neighboring region revealed that distinct ChIP‐seq peaks, Fantom 5 enhancers, TFBS, CTCF sites, and multiple DNAseI HS are located in the HDAC4 gene body, and they flank the re‐ligated deletion breakpoints.
4. Immunoblotting detected HDAC4 haploinsufficiency in the affected persons. Quantification showed ˜50% less expression compared to two non‐affected probands (n = 3, ± SD, Student's t‐test P = 0.0131, NON = controls, AFF = affected individuals).

Figure EV2. Molecular characterization of the BDMR family and HDAC4 depletion in chondrocytes.

• A
  Clinical features of the BDMR family without intellectual disability. The patients' sole behavioral peculiarity is a “very friendly” disposition in all affected members and III‐1 recently developed self‐injuring behavior (skin picking). Genotype–phenotype comparisons and phenotype—2q37‐breakpoint correlations showed that clinical features are variable in ˜50% of patients (Aldred et al, 2004; Casas et al, 2004; Falk & Casas, 2007; Villavicencio‐Lorini et al, 2013).
• B, C
  Hand (B) and feet (C) photographs of the affected individual II‐1 with obvious BDE of fingers and toes.
• D
  qPCR was used to reduce the critical deletion region on chromosome 2q37; genomic positions refer to human genome assembly hg19. Amplification values were normalized to the non‐affected proband I‐2. The HDAC4 haploinsufficiency was determined between position 239,871,764 and 240,198,505 bp (hg19, chr2q37) by a reduction of approximately 50% of the qPCR amplicons (means ± SD).
• E
  Electropherogram of the deletion‐breakpoint spanning sequence determined by PCR amplification with oligonucleotides flanking the deletion breakpoints. The black bar represents the re‐ligated breakpoints.
• F
  PTHLH, its downstream target ADAMTS7 (Bai et al, 2009), and CISTR‐ACT were downregulated in MSC‐derived chondrocytes of affected persons (AFF), when compared to non‐affected controls (NON). RUNX2 was not significantly differentially regulated (**P < 0.001, Student's t‐test, means ± SD).
• G
  Normal intellect was already reported in BDMR (Wheeler et al, 2014; Jean‐Marcais et al, 2015). However, intellectual disability in BDMR can phenotypically overlap with the Smith‐Magenis syndrome (SMS), causally associated with mutations in RAI1 or interstitial 17p11.2 deletions (Juyal et al, 1996; Slager et al, 2003; Williams et al, 2010b). Williams and colleagues reported retinoic acid‐induced 1 (RAI1) gene downregulation in BDMR patients with haplo‐insufficient HDAC4 (Williams et al, 2010a). In RAI1 Xenopus morphants, defective neural crest formation and abnormal brain development were observed, confirming the SMS clinical features (Tahir et al, 2014). Due to the observed RAI1 downregulation in lymphoblastoid cells (LCLs) of BDMR patients with developmental delay (Williams et al, 2010a), we tested RAI1 expression in MSC‐derived chondrocytes and in MSC‐derived neuron‐like cells in our samples. In both tissues, we observed upregulated RAI1 levels in AFF compared to controls (*P < 0.05, two‐tailed Student's t‐test, means ± SD).
• H
  Consistent with these results, we observed the same in siRNA‐mediated depletion experiments of HDAC4 in the human C28/I2 chondrocyte cell line (Goldring et al, 1994). Upon HDAC4‐depletion, PTHLH and CISTR‐ACT were downregulated, RAI1 was significantly upregulated, and Runx2 expression was not significantly elevated, as detected in patient‐derived chondrocytes or Hdac4 ^−/− mice (Vega et al, 2004; Appendix Table S1, *P < 0.05, **P < 0.001, two‐tailed Student's t‐test). In contrast to the results reported by Williams et al, we found that RAI1 was upregulated in patient material and in siRNA‐depleted HDAC4 chondrocytes, a finding that could be a clue to mental retardation in BDMR. This contradiction could be due to the use of chondrocytes and neuron‐like cells, instead of LCLs (Williams et al, 2010a).
• I
  Examples of magnified WCP‐FISH images of chromosomal patterns II and III to show individual loci within the chromosomal territories (arrowheads = BAC signals of loci, * = chromosome 2q with HDAC4 deletion, blue = chr. 2q, green = chr.12, red = chr.17, light green = RP11‐80D14, pink = RP11‐696D23, gray = DAPI).

To elucidate molecular mechanisms responsible for this phenotype, we first excluded mutations in GNAS1 or PTHLH that are known to cause BDE or BDMR by Sanger sequencing (Rickard & Wilson, 2003; Klopocki et al, 2010; Maass et al, 2010). Next, we focused on HDAC4, which is often deleted in 2q37‐deletion syndrome (Aldred et al, 2004). We identified a partial heterozygous HDAC4 deletion, spanning 222.3 kb from intron 2 to the 3′ UTR of HDAC4, which would represent the smallest deletion interval identified in BDMR thus far (Figs 3C and EV2D–E). Using publicly available ChIP‐seq data (H3K27ac, H3K27me, H3K4me1, H3K9me3, CTCF), Fantom5 enhancer data (Andersson et al, 2014), and conserved transcription factor binding site data (TFBS), we analyzed regulatory features at the HDAC4 locus in MSCs, human embryonic stem cells (H1‐ESCs), and osteoblasts (Phillips & Corces, 2009; Maher, 2012; Flicek et al, 2014).

Our analysis revealed that major regulatory elements were deleted in our BDMR patients and the reorganization of elements flanking the deletion breakpoints could underpin the pathogenesis of BDMR (Fig 3C). For example, the 3′ UTR of HDAC4 contains a TFBS cluster with strong transcriptional activity (RNA‐seq data ENCODE, Fig 3C), while intron 2 harbors regulatory elements especially in osteoblasts. Of note, we did not detect a truncated transcript that could have hypothetically been transcribed from the remaining exons on the deletion allele (NM_006037, exon 2 → 3′ UTR, oligonucleotides Appendix Table S1). Immunoblotting revealed the expected HDAC4 haploinsufficiency with a ~50% reduction in protein expression in MSCs from affected family members (Fig 3A and D, P < 0.05, Appendix Fig S3A and B; Dominici et al, 2006; Salem & Thiemermann, 2010). Additionally, we identified differential expression of PTHLH, ADAMTS7, CISTR‐ACT, and RAI in MSC‐derived cells from affected individuals (Fig EV2F–G, Appendix Fig S3C, Appendix Table S1). Importantly, depleting HDAC4 by siRNA recapitulated these results (Fig EV2H). Collectively, these data indicate that HDAC4 haploinsufficiency affects chondrogenic and osteogenic signaling, and thereby supports BDE pathogenesis in the context of the 2q37‐deletion syndrome BDMR.

NHCCs and their interactions in the BDMR family

Having molecularly characterized the genetic aberration in the BDMR family, we next addressed the role of NHCCs in the molecular mechanisms responsible for BDMR. The phenotypic BDE features in the BDMR family were similar to two previously reported BDE families (Maass et al, 2012). Therefore, we investigated the NHCCs between chromosome 2q (HDAC4), chromosome 12 (PTHLH, 12p11.2; CISTR‐ACT, 12q13.13), and/or chromosome 17 (SOX9, 17q24.3) by WCP‐PCP‐FISH, and we assessed how these might be altered in affected individuals of the BDMR family.

Within our WCP‐FISH approach, we also discriminated between parental chromosomes 2 by performing parental origin determination FISH (pod‐FISH) with a copy number variant (CNV) on chromosome 2 (BAC RP11‐80D14; Weise et al, 2010, 2015). To distinguish the NHCCs in non‐affected and affected state (HDAC4 deletion), we used BAC RP11‐696D23 hybridizing exactly within the HDAC4 deletion (Figs 4A and EV1D–F) and performed WCP‐FISH as described above. Next, we discriminated between co‐localized and separate states, as described previously and classified the chromosomal pattern again: in patterns I and II, both alleles of chromosomes 2q, or 12, or 17, respectively, were positioned next to at least one allele of the other chromosomes (Fig 4B). Moreover, our strategy enabled us to distinguish between interactions of the deletion‐carrying chromosome 2 (pattern III) and the non‐affected chromosome 2 (pattern IV) with either chromosome 12 and/or 17 (Fig EV2I). Since both copies of chromosome 2 were unaffected in healthy individual I‐2, a discrimination of signals was not possible in pattern IV. In pattern V, both alleles intermingled, and pattern VI represented separated chromosomes without any contacts to each other (Fig 4B, Appendix Fig S2J). By pod‐FISH, we detected that the maternal or paternal chromosome 2 alleles were equally positioned with chromosome 12 or 17 (45–49% and 51–55%, respectively, Fig EV3A and B). Aside from the aberrant karyotype in the BDMR family, the results of the chromosomal pattern were highly similar to those observed in the healthy karyotypes (Figs 1E–F and 4B), further demonstrating that predominant chromosomal pattern occur regardless of the MSC donor.

Figure 4. NHCCs in the BDMR family.

1. Scheme of the multi‐color FISH and pod‐FISH approaches. BAC RP11‐80D14 (light green) hybridized to a copy number variation and was selected to distinguish between the parental chromosome 2 alleles. RP11‐696D23 (magenta) hybridized within the HDAC4 deletion and tracked only the non‐affected allele in the affected persons II‐1 and III‐2 (AFF). The q arm of chromosome 2 (2q, blue), and the chromosomes 12 (green) and 17 (red) were labeled.
2. Top: schemes of six different chromosomal pattern found in MSCs. Bottom: representative images of raw, non‐processed WCP signals of chromosomes 2q, 12, and 17 and BAC signals in MSCs of one affected BDMR patient. Arrows indicate BAC signals within the territory of non‐affected; arrows with an asterisk indicate HDAC4 deletion‐affected (* del HDAC4) chromosome 2q. Pattern I or II: both alleles of the chromosomes 2q, 12, or 17 were proximal to one of the other chromosomes. Pattern III shows an allelic pairing of the affected chromosome 2q with 12 or 17. Pattern IV: nuclei without any intermingling of the deletion allele with chromosome 12 or 17. Pattern V: bi‐allelic intermingling, or pattern VI, no chromosomal crosstalk of chromosome 2q with 12 and/or 17. Scale bars = 5 μm.
3. Left: Chromosomal pattern of chromosomes 2q and 12 in 100 nuclei of non‐affected control I‐2 (NON) and BDMR‐affected persons (AFF) II‐1 and III‐2. Patterns III and IV were the commonest arrangements. Compared to I‐2, pattern III occurred significantly less in II‐1 and III‐2. Right: pattern III of chromosome 2q with 17 (100 nuclei) occurred significantly less in AFF II‐1 and III‐2, compared to I‐2 (2 Way ANOVA, ***P < 0.0001, n.d. = pattern not determined).
4. Power analysis to validate experimental data: Monte Carlo simulations (10,000×) of randomized data comparing chromosomal pattern I–VI between each of the two groups (NON, AFF), in different populations of nuclei (5–100). Simulating the used standard of analyzing 100 nuclei by FISH showed that > 99.9% of the experimental results are true (P < 0.05).
5. Distribution of co‐localization (chromosomal contact) degrees shown with Manders coefficients m1 and m2 across the chromosomal pattern I‐VI in NON and AFF (data of 4 independent replicates, medians, 25^th and 75^th percentiles ± min‐to‐max are shown). Alleles in chromosomal pattern IV in NON were not distinguishable.
6. The degrees of chromosomal co‐localization (overlaps between chromosomes 2q–12, 2q–17, 12–17, pattern I–V) were different between NON and AFF (two‐way ANOVA, ***P < 0.0001, means ± SD are shown).
7. Randomizations of the data shown in panel E determined that the chromosomal pattern of chromosomes 2q–12, and 2q–17 correlated less in the original data of affected BDMR patients, than in the control I‐2 (*P < 0.05, Mann–Whitney rank sum test).

Figure EV3. Nuclear measurements in the BDMR family.

1. Measurements of lengths and widths of the studied chromosomal territories in interphase normalized to the corresponding nucleus size (each 100 chromosomes, means ± SD).
2. Quantification of the pod‐FISH approach. Comparing the interaction frequencies of the smaller signal with the bigger RP11‐80D14 signal showed a nearly fifty‐fifty distribution, indicating random interaction between chromosomes 2 and 12 or 17, respectively.
3. The interactions between chromosomes 12 and 17 in affected individuals of the BDMR family (AFF) were not different to the interactions found in the non‐affected control (NON).
4. Representative, contrast‐enriched image of WCP‐FISH signals to measure the x‐y dimensions of the nucleus and the maximum WCP‐FISH distributions (dotted lines), and pod‐FISH‐signal distances (arrows) in the non‐affected control and affected patients of the BDMR family. The maximal width and height of the nucleus was determined to normalize the closest distances of the chromosome 2q territory, and the distances of the BAC signals RP11‐80D14 or RP11‐696D23 to the nucleus boundary.
5. Nuclei sizes, as ratios of nuclei widths to heights in probands I‐2 (NON), II‐2, and III‐2 (AFF; each 50 nuclei, means ± SD).
6. The ratio of the closest chromosome 2q distances to the nucleus boundary, normalized to every nucleus size, indicated slightly reorganized alleles in affected individuals of the BDMR family (each proband = 100 chromosomes, means ± SD).
7. Normalized distances of BAC RP11‐80D14 to the nucleus border tended to be also slightly different in affected persons, compared to I‐2 (each proband = 100 BAC signals, means ± SD).
8. Ratio of the normalized BAC RP11‐696D23 distance to the nucleus border. Due to the HDAC4 deletion, the BAC signal was not present on the second allele in affected individuals (II‐2, III‐2; each proband = 100 BAC signals, means ± SD).
9. Deletion intervals of earlier studies show that either lncRNAs (LINC001107, FLJ43879) or coding genes (TWIST2) at the HDAC4‐locus on chromosome 2q37 were deleted, implying that additional clinical phenotypes could derive from the deletion of multiple genes (Villavicencio‐Lorini et al, 2013; Jean‐Marcais et al, 2015).

We next focused on if the HDAC4 deletion could have influenced the chromosomal pattern of chromosomes 2q, 12, and 17 to one another (each MSC donor 100 nuclei). Remarkably, we detected that pattern III (deletion‐affected chromosome 2q with chromosome 12) occurred significantly less frequently in BDMR patients than in a healthy control (Fig 4C, P < 0.0001).

Interestingly, we obtained the same results when analyzing the chromosome 2q and 17 arrangements (Fig 4C). Specifically, pattern III was present to a similar extent to that described above in non‐affected control I‐2, yet markedly reduced in the BDMR patients (Fig 4C, P < 0.0001). We also analyzed chromosomes 12 and 17 (~55%), but did not find any pattern differences between patients and healthy control (Fig EV3C). By using Monte Carlo simulations (10,000×), we evaluated whether our conclusions drawn from 100 nuclei of each of the MSC donors could be true. Specifically, we randomized the experimental data of the chromosomal pattern grouped into two states (simulating non‐affected vs. affected state) and used a power analysis to determine how valid significant differences between these two groups are, if we selected 100 or less nuclei. Simulations with different fractions of nuclei determined that the differences of chromosomal arrangements in 100 nuclei, between healthy donor and affected patient, are true to >99.9% (P‐values < 0.05, Fig 4D). These findings indicate that if significant differences within the chromosomal pattern between two given groups occur, the analysis of 100 nuclei is sufficient to detect them.

Since chromosomal pattern III occurred less frequently in BDMR patients, we next addressed whether this coincided with an alteration of the co‐localized regions between chromosomes. By Manders coefficients (see Materials and Methods; 241 comparisons in NON, and 324 in AFF), we quantified the direct overlap (co‐localization) between chromosomes 2q–12, 2q–17, and 12–17. We detected that independent of the m1 or m2 coefficients, predominant co‐localized regions occurred within patterns III, IV, and V (Fig 4E). Moreover, the degree of co‐localization seemed stronger in affected patients than in the healthy control (Fig 4E). Therefore, we merged the data of the co‐localization analysis of the chromosomal patterns I–V (not VI), and Manders coefficients m1 and m2, since we were interested in the general co‐localization between two given chromosomes. A two‐way ANOVA comparing the degrees of chromosomal co‐localization within patterns I–V, and healthy versus affected state, determined that either of the studied chromosomes showed more co‐localization in the BDMR‐affected patients than in the control (Fig 4F, P < 0.0001). Also, chromosome 2q was less co‐localized with chromosomes 12 and/or 17 than the co‐localization between chromosomes 12 and 17 (Fig 4F). To validate our results, we used Costes's analysis. Interestingly, the randomized test images of chromosome 2q overlapping with 12 or 17 correlated better in the original data of the control than in BDMR patients (Fig 4G). Consistent with our previous observation that pattern III between chromosome 2q–12 and 2q–17 occur less frequently in nuclei of BDMR patients, the randomizations between chromosomes 2q–12, and 2q–17, but not 12–17, were significantly different between BDMR patients and control (Fig 4G, P < 0.05, P < 0.01). Collectively, our findings show that the occurrence of chromosomal pattern and the degree of co‐localization between chromosomes 2q, 12, and 17 are different and they support the hypothesis that these conditions may primarily contribute to BDMR pathogenesis.

While analyzing our FISH images, we noted that the affected HDAC4 locus tended to be more frequently located at the nuclear periphery compared to both HDAC4 loci on the non‐affected alleles of chromosome 2 from healthy controls (Fig 4B). Therefore, we measured distances from the chromosome 2q territories and from the pod‐FISH BAC (RP11‐80D14), normalized to nucleus sizes, to their closest nucleus boundary. We found that the HDAC4 localization in the BDMR patients was different to its position in non‐affected subjects (Fig EV3D–H). Together, we observed no major inter‐individual differences, and overall, the chromosomal pattern was similar to those found in healthy donors, indicating determined spatiality within the nucleus. However, the frequency of chromosomal pattern III was altered and the co‐localization of chromosomal regions was increased in patients. Our data therefore provide evidence that displaced inter‐chromosomal pattern and altered co‐localization of chromosomal regions were supported by repositioning the deleted HDAC4 region.

HDAC4 interactions in Hi‐C datasets

To assess our findings using a different measure of chromosome contacts, we reanalyzed previously published Hi‐C data from human MSCs (H1 cell‐derived) to determine whether NHCCs can be independently validated (Dekker, 2014; Barutcu et al, 2016). For this task, we used 40 kb bins for intra‐chromosomal interactions and 1 Mb bins for inter‐chromosomal interactions (Dixon et al, 2015). By assigning TAD boundaries to the chromosome 2q37 region, we found that HDAC4 is located within a 400 kb TAD, of which 56% are deleted in the BDMR patients (Fig 5A). Then, we analyzed all intra‐chromosomal interactions within ~4 megabases upstream and downstream of the HDAC4 locus and plotted the expected and observed interaction frequencies of intra‐chromosomal 2q37 interactions (Fig 5A, Materials and Methods). We found evidence that HDAC4 interacts, via intra‐chromosomal looping, with other loci on chromosome 2. As such, these interactions would be expected to be disrupted by the HDAC4 deletion in the studied BDMR family, or the larger 2q37‐deletions reported in other BDMR families (Fig 5A). For example, we identified an interacting region 3.5 Mb upstream of the HDAC4 locus that harbors the AGAP1 gene, which has been previously linked to the 2q37‐deletion syndrome (Fig 5A; Leroy et al, 2013).

Figure 5. Hi‐C reveals NHCCs between chromosomes 2 and 12, or 17.

• A
  Hi‐C interaction heatmap of the HDAC4 locus (anchor) at 40 kb resolution from human MSCs (top panel, hg19). The gray shaded bars highlight the intra‐chromosomal interactions of the HDAC4 locus. The insulation plot shows the TAD boundaries (middle panel). The interaction plot describes the normalized interaction frequencies of the HDAC4 deletion (anchor = orange bar) with each 40 kb bin (red dots) of the heatmap. The solid black line represents the expected interaction level (Lowess line); dashed black lines are the Lowess value ± 1 standard deviation. The loci interacting with HDAC4 were one standard deviation (SD) above the mean background (Appendix Table S2). HDAC4 displays an increased highly likely interaction frequency with a downstream‐located region (red arrow) outside of its own TAD.
• B, C
  Anchored HDAC4 locus showed similar (B) intra‐chromosomal interactions and (C) inter‐chromosomal interactions to all bins on either chromosome 12 or 17 in four different tissues. Z‐scores describe standard deviation above the mean interaction frequency.
• D
  Inter‐chromosomal interactions between chromosome 2 with either chromosome 12 or 17. In the most frequent top five interactions of HDAC4, adjacent regions to PTHLH and CISTR‐ACT on chromosome 12 and the chondrogenesis gene NOG on chromosome 17 were detected (medians, 25^th and 75^th percentiles ± min‐to‐max are shown).
• E
  Scheme of HDAC4 interacting with PTHLH, CISTR‐ACT, and NOG.

The Hi‐C analysis identified one of the strongest HDAC4 interacting regions ~660 kb downstream of HDAC4 (Appendix Table S2). This region contained several genes that, through intra‐chromosomal looping, are brought into close proximity with the HDAC4 locus, including NDUFA10, OR6B2, PRR21, and OR6B3 (Fig 5A). Although these genes have not been associated with craniofacial abnormalities, intellectual disability, chondrogenesis, or osteogenesis, to date, alterations in the long‐range interaction between HDAC4 and these genes, or long‐range regulatory regions in these regions, may be involved in human disease. Moreover, when we expanded the Hi‐C analysis to include different cell types (MSCs, H1 embryonic stem cells, mesoendoderm, trophectoderm), that can be important to manifest the other phenotypes commonly associated with BDMR, we found that intra‐chromosomal interactions of the chromosome 2q37 region were consistently detected (Fig 5B).

To refocus on the inter‐chromosomal associations of HDAC4, we performed a quantitative genome‐wide analysis of Hi‐C interactions between the HDAC4 locus and any other chromosomal region across the four different cell types (1 Mb bins, Appendix Fig S4). To quantify the results, we used z‐score values (Materials and Methods), which is a measure of how many standard deviations a particular interaction is above the mean inter‐chromosomal interaction frequency between two chromosomes, implying that stronger interactions have a higher z‐score. First, we focused on the HDAC4 locus and chromosomes 12 or 17 and found that the interaction profiles were highly similar between different tissues (Fig 5C), consistent with the idea that similar chromosomal arrangements occur across different cell types, and that chromosomes 2, 12, and 17 are in spatial proximity—as was observed in our FISH data. Interestingly, the genome‐wide comparison indicated that HDAC4 either interacted in a very defined manner with other sub‐telomeric regions, or showed no interaction at all. In particular, HDAC4 had no interactions with the p arms of several chromosomes (13, 14, 15, 21, 22), or only minor interactions (chromosomes 4, 8, 18, X, Y, Fig S4), indicating that HDAC4 preferentially interacts with a limited number of inter‐chromosomal targets.

To investigate the inter‐chromosomal interactions of chromosomes 2, 12, and 17, in a more unbiased approach without anchoring to HDAC4, we compared all bins of chromosome 2 with all bins of chromosomes 12 or 17 in MSCs. Strikingly, three out of the top five inter‐chromosomal interactions of the HDAC4 locus with chromosome 12 or 17 were regions adjacent to CISTR‐ACT or PTHLH on chromosome 12, and Noggin (NOG) on chromosome 17 (Figs 5D–E and EV4A and B). NOG antagonizes the binding of bone‐morphogenetic proteins to their receptors and is a key participant in initial cartilage condensations and cartilage morphogenesis (Brunet et al, 1998). In a reverse approach, we investigated the interactions between the anchored CISTR‐ACT, PTHLH, or NOG loci with other chromosomes, and again detected highly similar interacting regions across the different cell types, suggesting that these loci and HDAC4 are in spatial proximity and share common interaction sites (Fig EV5A–C). Altogether, our Hi‐C analysis identified interacting genomic regions that contain candidate genes that are differentially expressed (CISTR‐ACT, PTHLH) in the BDMR patients (Fig EV2F–H). Moreover, these data further validate the NHCCs that were identified by our microscopic analyses. Collectively, therefore, our data shed new light on complex nuclear signaling events that link chromosomal communication, gene expression, and molecular pathogenesis.

Figure EV4. HDAC4 interactions with chromosome 12 and 17 in Hi‐C.

• A, B
  IGV‐snapshots of the top five inter‐chromosomal interactions of HDAC4 with (A) chromosome 12, or (B) chromosome 17 (Robinson et al, 2011) in MSCs. DNase I hypersensitivity (DHS), CTCF, and RAD21 ChIP‐seq data (ENCODE) are shown for the human embryonic stem cell line H1‐ESC. Identical data for MSCs were not available. PTHLH and CISTR‐ACT were adjacent to two of the identified HDAC4 interactions on chromosome 12. NOG, a chondrogenesis gene, was differentially regulated upon CISTR‐ACT's dysregulation and was found next to one interacting bin on chromosome 17 (Brunet et al, 1998; Maass et al, 2012).

Figure EV5. Anchored interactions of PTHLH, CISTR‐ACT, and NOG with the other chromosomes in Hi‐C.

• A–C
  (A) Anchored interactions of PTHLH with all z‐scores on chromosomes 17 and 2, or (B) interactions of CISTR‐ACT with chromosomes 17 and 2, or (C) interactions of NOG with chromosomes 12 and 2. Blue lines indicate the studied, locus‐specific NHCCs on different chromosomes. Across different cell types, strong similarity in interacting regions was observed. Slight differences in NHCC contacts could indicate tissue‐specific contacts that could contribute to tissue‐specific gene regulation and variability in phenotypic peculiarities. The data show that PTHLH, CISTR‐ACT, and NOG share common interaction sites/regions.

Discussion

We performed a comprehensive characterization of the chromosomal territories 2q, 12, and 17, their arrangements to one another and their co‐localized regions, across several primary cell types in normal and in disease states to provide unprecedented in vivo relevance to a disease mechanism with potential molecular implications for the clinic. We conclude that the studied chromosomes frequently interact in similar arrangements and distributions across various primary cell types and in different individuals, supporting the idea that chromosomes are non‐randomly organized and positioned in the nucleus (Misteli, 2004; Cremer & Cremer, 2010). Chromatin interactions can be cell type specific, but the chromatin organization at TAD levels seems to be evolutionary conserved (Dixon et al, 2012, 2015; Rao et al, 2014; Vietri Rudan et al, 2015). We found that in MSCs and MSC‐derived cell types, the global chromosomal contacts occurred in similar frequencies, pattern, and co‐localization levels; however, locus‐specific NHCCs were observed between MSCs and certain MSC‐derived cell types. For the CISTR‐ACT NHCCs, we found a higher tissue‐specific interaction rate in BDE‐relevant chondrocytes and osteocytes, and previously by CLING, that these NHCCs are stable across time (Maass et al, 2018). Although live‐cell imaging approaches allow studying spatiotemporal dynamics of genomic loci, the combined multi‐color labeling of individual loci and entire chromosomes is not possible so far. Here, by using multi‐color FISH, we were able to address how individual loci are organized within their chromosomal territory and how this is altered in primary cells of a disease state. Our results of chromosomal pattern indicate that local cell‐type‐specific changes may re‐organize chromatin sub‐domains that result in a nuclear configuration that supports molecular pathogenesis. Moreover, our multi‐color FISH approach revealed a novel link between HDAC4 on chromosome 2 and loci on chromosomes 12/17.

We have also characterized the smallest observed HDAC4 deletion (~222 kb) that has been reported as responsible for the BDE phenotype in the chromosome 2q37‐deletion BDMR syndrome (Fig EV3I). Previously, the reported 2q37‐deletions ranged from 490 kb to 9.8 Mb, and therefore included more genes, and were more often associated with intellectual disability and BDE (Morris et al, 2012; Jean‐Marcais et al, 2015). For example, TWIST2 on 2q37 regulates early osteoblastogenesis (Bialek et al, 2004) and was involved in other reported 2q37‐deletions causing BDMR (Villavicencio‐Lorini et al, 2013; Jean‐Marcais et al, 2015). Investigating non‐coding loci, especially lncRNAs within the 2q37‐deletions and their functions, could help to delineate the variable BDMR outcome (Fig EV3I; Cabili et al, 2011). Others described truncated HDAC4 caused by a frame‐shift mutation with a gain‐of‐function transcriptional repression in neurons compared to the wild‐type HDAC4 allele (Williams et al, 2010a; Sando et al, 2012). We suggest that the lack of intellectual disability in our family could be due to the absence of truncated HDAC4, further deletion‐involved coding or non‐coding genes, and tissue‐specific NHCCs of HDAC4 with other loci.

Preserved NHCCs and their nuclear positioning indicate that interactions between specific loci can be key features for gene regulation. We addressed the possibility that the HDAC4 deletion impaired nuclear architecture to an extent that local repositioning of chromosomal sub‐domains occurred in individuals affected with BDMR. The situation of displaced NHCCs and the HDAC4 haploinsufficiency with reduced HDAC4 protein levels leading to differential gene expression could have facilitated BDMR pathogenesis.

Non‐homologous inter‐chromosomal contacts could happen in distinct sub‐nuclear compartments. Thus, local chromatin remodeling, in particular chromatin decondensation, can lead to nuclear reorganization of loci repositioning in development (Therizols et al, 2014). Approximately half of the TAD domain harboring the HDAC4 locus and further regulatory sequences was lost in the BDMR family. Wijchers and colleagues described that repositioning of sub‐TADs (which are only a few hundred kilobases in size) to different nuclear sub‐compartments was not due to changes in chromatin marks or transcription, but rather due to recruited protein factors (Wijchers et al, 2016).

We conclude that synergistic effects of the HDAC4 haploinsufficiency together with structural changes of the ~222 kb HDAC4 deletion caused repositioning of the deletion‐affected chromosome 2 sub‐domain and led to displaced NHCCs and altered co‐localization of the chromosomal territories that facilitated the dysregulation of chondrogenic and osteogenic signaling. Our findings further support the idea that specific euchromatic transcriptional neighborhoods exist, where loci are in spatial proximity and cluster with shared transcriptional regulators more often than with other regulators (Shopland et al, 2003; Edelman & Fraser, 2012; Beagrie et al, 2017).

By analyzing human Hi‐C data (Dixon et al, 2015), we found evidence that the deleted intra‐chromosomal interactions could support BDMR pathogenesis. Our conclusions are further supported by a recent computational polymer model that showed that local chromatin reorganization can happen in scales < 10⁵ bp (Florescu et al, 2016).

Another consequence of the deletion could cause repositioning of the chromosomal sub‐domain that in turn leads to changes in inter‐chromosomal interactions. Indeed, although we did not clearly observe this state‐of‐affairs by microscopic distance measurements, probably because of the low number of patients. Nonetheless, we detected altered chromosomal pattern and co‐localization in BDMR patients. Inter‐chromosomal Hi‐C analysis also revealed conservation of the interactions between HDAC4, CISTR‐ACT, PTHLH, and NOG and chromosomes 2, 12, and 17 across four different cell types. This finding is consistent with our FISH results and highlights the preserved organization of chromosomal territories in different cell types and from different individuals (Figs 1, 2 and 4). Given the fact that only some aspects of BDMR were present in the family we studied, slight differences in the locus‐specific NHCCs across different Hi‐C datasets could indicate that tissue‐specific contacts contribute to tissue‐specific 3D gene regulation and to the diverse phenotypic peculiarities of BDMR. However, further approaches will be necessary to better characterize tissue‐specifically the more detailed inter‐chromosomal interactions of HDAC4 and other loci.

Our findings provide several insights into the determined, non‐random interphase architecture, and how this may be disrupted in a disease state. We found that preferential chromosomal interactions and arrangements are cell type independent, while locus‐specific interactions occur across different cell types. Moreover, our data provide evidence for the impact of genomic rearrangements and their consequence on repositioning loci locally within the genomic architecture. Further studies on the positioning and interacting of chromosomal domains and their loci should specifically elucidate their impact on spatial nuclear signaling and gene regulation in health and disease.

Materials and Methods

Human material

After approval by the ethics committee (Charité Medical Faculty Berlin) and written, informed consent, we obtained human tissues and photographs to publish. Mesenchymal stromal (stem) cells (MSCs) were obtained and differentiated as previously described (MSC‐1 ♂ 18 years, MSC‐2 ♀ 59 years, MSC‐3 ♀ 61 years, I‐2 ♀ 63 years, II‐1 ♀ 40 years, III‐2 ♀ 18 years; Vermette et al, 2007; Maass et al, 2015). For pod‐FISH validation experiments on metaphase chromosomes, we used whole blood from the whole BDMR family, including non‐affected proband I‐1. A further non‐affected and non‐related proband (♂ 36 years) served as second negative MSC control in quantification experiments together with MSCs of the BDMR family.

RNA preparation and qRT–PCR analysis

RNA was prepared using Trizol^® Reagent (Ambion) and phenol/chloroform precipitation. The RNA was reverse transcribed using RevertAid first‐strand cDNA synthesis kit (Fermentas). CYBR‐green quantification (Roche) was performed according to standard protocols on ABI 7500. Oligonucleotides were designed in Primer3 (v. 0.4.0, Appendix Table S1). Human GAPDH served for normalization (Qiagen). Expression was quantified applying the ΔΔC _t method. qRT–PCR products were analyzed for amplicon size, specificity, and integrity on 3% agarose gels.

Deletion‐breakpoint mapping and identification

Identical DNA amounts were used in CYBR‐green qPCR as described above. Proband I‐2 served as control. Specific amplicons covered regions of the HDAC4 locus. In comparison with the two amplified alleles in I‐2, the heterozygous deletion was detected between amplicon HDAC4_240198 and HDAC4_239871 (Appendix Table S1). The breakpoint spanning PCR amplicon was sequenced using the Big Dye^® Terminator Cycle Sequencing Kit v1.1 (Applied Biosystems). The analysis was performed on a 3130xl Genetic Analyzer (Applied Biosystems) using Gene Mapper^® Software Version 4.0. SeqMan software (Lasergene Version 10.0; DNAStar) was used to evaluate the traces.

Immunoblotting

Immunoblotting was done according to standard procedures. The antibodies anti‐β‐tubulin (sc‐9104, Santa Cruz Biotechnology) and anti‐HDAC4 [Y416] (ab32534, Abcam) were used. Blot signals were determined by HRP‐mediated chemiluminescence (SuperSignal West Pico, ThermoScientific) and visualized in a PeqLab, Chemi‐Smart 5000. Signal quantification (Adobe Photoshop) was done on three independent blots, normalizing the HDAC4 signal against β‐tubulin in relation to two non‐affected persons.

Neurogenic differentiation of MSCs

Subconfluent MSCs were neurogenically differentiated with Promocell neurogenic differentiation medium (C‐28015) for 7 days. Every 2 days, medium was exchanged. Microscopic documentation and qRT–PCR quantification of neurogenic markers validated the successful differentiation in neuron‐like cells.

siRNA transfection

C28/I2 were cultured in DMEM/F12 (1:1), supplemented with 10% fetal calf serum (FCS), 100 U/ml penicillin, and 100 μg/ml streptomycin (Goldring et al, 1994). 5 × 10⁴ C28/I2 cells were seeded 12–24 h prior to transfection. 100 nM siRNA were transfected with Dharmafect (Dharmacon) for 48 h in C28/I2. Scrambled, HDAC4, and CISTR‐ACT siRNAs were validated (Ambion). Two different siRNAs were used. For visual simplicity, the results of one siRNA are shown. HDAC4 siRNA1‐sense: GCAAGAUCCUCAUCGUGGAtt, antisense: UCCACGAUGAGGAUCUUGCtc, HDAC4 siRNA2‐sense: GAAAAGGUUUUACAGCAAAtt, antisense: UUUGCUGUAAAACCUUUUCtg.

FISH and pod‐FISH

Molecular cytogenetic studies were performed on PHA‐stimulated peripheral blood lymphocytes and on MSCs and MSC‐derived cells of I‐2, II‐2, and III‐2 according to standard protocols (R.S. Verma, 1989). For locus‐specific labeling, BACs from BACPAC Chori were used by DNA labeling with nick translation (Roche), either directly with fluorochromes, or indirect by haptenes and detected with fluorochrome‐coupled antibodies in a posthybridization washing step. Probes for whole and partial chromosome paints (WCP/PCP) were generated by microdissection, amplified, and labeled by DOP‐PCR (standard protocols). Zeiss Axioplan 2 and Axio Imager.Z2 fluorescence microscopes (Zeiss, Jena, Germany) equipped with appropriate filter sets to discriminate between a maximum of five fluorochromes (SpectrumGreen, SpectrumOrange, TexasRed, Cy5, DEAC, see Fig EV1) and DAPI counterstaining were used for image acquisition. Digital images were captured using an IMAC S30 CCD camera and MetaSystems (Isis) software (Altlussheim, Germany). We used the same acquisition setting to image at least 100 nuclei from the same experiment, on the same slide, under the same microscopic conditions. To provide the best signal‐to‐noise ratios of every image, we used the Isis‐standardized background control algorithm to allow quantitative analysis of the WCP signals. A comparison of chromosomal interactions of images with or without background control can be found in Appendix Fig S1C. To visually assess the maximal distribution of chromosomal territories to measure dimensions of WCP‐FISH signals (Fig EV3), images were postprocessed by increasing the contrast of each acquired channel. To quantitatively measure the true degree of co‐localization (Figs 1 and 4), we selected regions of neighbored/interacting raw WCP‐FISH signals (ROI = regions of interest, not contrast‐enriched) to distinguish interacting (co‐localizing) from separated chromosomal territories and to analyze differences between chromosomal pattern or cell types, or between affected and non‐affected BDMR family members. Quantitative image analysis of single channels in gray scales was done with the Coloc 2 analysis package in Fiji. Manders split co‐localization coefficients (m1 and m2) and Costes P‐values were determined and quantified in cells of different donors and cell types in Fiji, according to the user guidelines (Manders et al, 1993; Costes et al, 2004; Zinchuk et al, 2007). The arrangements of the WCP signals to one another were analyzed in a cell‐by‐cell manner for identifying the relative chromosomal positioning within the same cell according to the described schemes shown in Fig 1 and 4. Basically, the pattern analysis was based on the arrangements of single or both alleles and their positioning to one another. Co‐localized chromosomal territories were classified into pattern I–IV, while separated, non‐interacting chromosomal territories were grouped in pattern V (Fig 1D, E, K and L). Only the images with a distinct signal of all fluorescent channels were analyzed. Images that did not show integer signals for all acquired channels were excluded. BAC clone RP11‐80D14 was used in a pod‐FISH assay on metaphase spreads and interphase nuclei to distinguish homologous chromosomes 2 (Weise et al, 2008). At least 100 nuclei were analyzed per experiment. For 3D‐reconstructions, Cell‐P software (Olympus) was used.

Hi‐C analysis of the HDAC4‐locus

The MSC Hi‐C data were derived from Dixon et al (2015). All biological replicates, which showed high correlation, were pooled. The Hi‐C mapping, the iterative correction, the binning, and the matrix generation were performed using the Hi‐C Pro software with default parameters according the manual (https://github.com/nservant/HiC-Pro; Servant et al, 2015). For the z‐score analysis of intra‐chromosomal interactions, we employed a Lowess smoothing algorithm to find the weighted average (expected interaction frequency) and the weighted standard deviation of all Hi‐C signals around the HDAC4‐locus to make the “4C‐like” plot (Sanyal et al, 2012), by using the publicly available “matrix2anchorPlot.pl” script (https://github.com/dekkerlab/cworld-dekker). The z‐scores for the inter‐chromosomal interactions in Figs 5C and EV4 were calculated by taking the mean signal and the standard deviation of the interactions between pairwise chromosomes (e.g., 2–12, 2–17, and 12–17), and the z‐scores of the interactions between HDAC4, PTHLH, CISTR‐ACT, NOG, and all of the other corresponding chromosomes were plotted. The TAD boundaries were detected using the “Insulation Plot” method with the options “–is 480000 –ids 200000 –im mean and –nt 0.15” (https://github.com/dekkerlab/crane-nature-2015) (Barutcu et al, 2015; Crane et al, 2015). To calculate the insulation score of each 40 kb bin in the chromosome 2q37 genomic region, a square of 480 kb × 480 kb (12 bins × 12 bins) was slid along the Hi‐C matrix and the mean interaction frequency was plotted. For the bins that were within ± 500 kb of the ends of the interaction matrix, an insulation score was not assigned. The window side of the insulation delta span of 200 kb was used. Boundaries with a score < 0.15 were removed.

Statistics

Functional experiments were done at least three times. Numbers (n) of experiments and numbers of analyzed nuclei are mentioned in the figure legends. Randomization of data, ANOVA (FDR‐corrected), and Monte Carlo simulations (1,000× or 10,000× with 5 SD or maximal Gaussian random error of tested samples) were done in GraphPad Prism v.7.0b. Costes's p‐value analysis with 100 randomized test images was done in Fiji. Significance was determined by two‐tailed Student's t‐test, two‐tailed Mann–Whitney U‐test, ANOVA, or χ² test (***P < 0.001, **P < 0.01, *P < 0.05). CDF and box plots (medians, 25^th–75^th percentiles, and min‐to‐max) were generated in GraphPad Prism v.7.0b.

Author contributions

LP and ST recruited the family members and clinically examined the probands. PGM conceived the study and performed the functional studies. AW, TL, and JL analyzed the chromosomal territories, and AW and KR did the pod‐FISH analysis. ARB analyzed the Hi‐C data. AA narrowed down the deletion interval and YW‐N cultivated MSCs and performed immunoblotting. PGM, AW, JLR, FCL, and SB supervised the project and wrote the manuscript.

Conflict of interest

The authors declare that they have no conflict of interest.

Supporting information

Appendix

Expanded View Figures PDF

Review Process File

The EMBO Journal (2018) 37: e96257