SS18-SSX forms novel condensates and re-organizes the 3D genome

In synovial sarcoma samples, there is a chromosome translocation that results in the replacement of C-terminal 8 amino acids of wild type SS18 by 78 amino acids of SSX genes (SSX1, SSX2 or SSX4) to form onco-protein SS18-SSX (Fig. 1a). To better understand this Onco-transformation event, we show that SS18-SSX1 can induce extensive activation of embryonic developmental genes (Supplementary Fig. 1a, b), like PAX6, NKX3-2 and SOX2, and the down regulation of genes are involved in extracellular matrix/structure organization (Supplementary Fig. 1c). To understand the molecular basis of transforming SS18 into oncogenic SS18-SSX, we analyzed their condensates and show that SS18-SSX is at least 300% more efficient than SS18 (Fig. 1b, c), with less protein expressed on western blot (Fig. 1d). We next performed half-FRAP assay [13] to probe the nature of SS18/SS18-SSX condensates and show that the unbleached half condensates exhibit distinct molecular dynamics (Fig. 1e), suggesting that the mechanism of SS18-SSX condensate formation might be closer to low-valency interactions with spatially clustered binding sites (ICBS) than to the liquid–liquid phase separation (LLPS). It has been reported that SS18-SSX could recognize H2AK119ub histone modification [18], which can provide the scaffold for low-valency interactions. To test it, we mutated all the acidic amino acids which has been reported to recognize H2AK119ub [18], by alanine in the C terminus of SSX1. We show that the resulting mutant fails to form condensates (Supplementary Fig. 2a-c), suggesting that acid residue mediated low-valency interaction is responsible for the emerging property of SS18-SSX.

We then tested the role of tyrosine residues in SS18 IDR previously shown to be critical for condensate formation of SS18 protein [21], and show that once the tyrosine enriched in the SS18-IDR mutated totally, SS18-SSX also fails to form condensates in comparison with other enriched amino acids Gln, Gly and Pro (Supplementary Fig. 3a, b). Furthermore, although SS18(YA)-SSX mutant loses the capability to form condensates, it remains competent in recognizing and binding to chromatin regions with H2AK119ub histone modification (Supplementary Fig. 3c, d). However, SS18(YA)-SSX mutant fails to activate the downstream genes (Supplementary Fig. 3e) and loses tumorigenicity based on EdU cell proliferation assay (Supplementary Fig. 3f, g), colony formation assay (Supplementary Fig. 3h, i) and tumor-bearing mouse models (Supplementary Fig. 3j-l), suggesting that the tyrosine-based condensation also plays a critical role in its tumorigenic potential.

Given the more pronounced condensate formation phenomenon of pathological SS18-SSX condensates than wild type SS18 condensates (Fig. 1b, c), we hypothesize that it may be able to transform the 3D genome architecture. To test this, we performed Hi-C experiments to probe the effect of SS18-SSX onco-fusion on 3D chromatin structure. Compared to wild type SS18 and condensate-deficient SS18(YA)-SSX, SS18-SSX nuclei appear to be smoother, consistent with a more ordered structure (Fig. 1f), characteristic of enhanced long-range chromatin interaction (Supplementary Fig. 4a, b). Consistently, this enhancement occurs more frequently in regions with H2AK119ub modification than those without (Supplementary Fig. 4c-f, Fig. 1h). Additionally, compared with wild type SS18 or condensate-deficient mutant SS18(YA)-SSX, enhancement of long-range (more than 1 M) chromatin interaction by SS18-SSX is positively correlated with the intensity of H2AK119ub modification (Fig. 1g). Similar structural changes could also be observed when tested in IMR-90 cells (Supplementary Fig. 4g, h). These results suggest that SS18-SSX possess a unique ability to promote long-range chromatin interactions and a more compact architecture by at least enriching H2AK119ub modified chromatin.

SS18-SSX condensates exclude HDACs

The new properties emerged from SS18-SSX fusion suggest that it may elicit new function inside nuclei. To test this hypothesis, we analyzed several candidate biomacromolecular complexes like BRG1/BAF we previously reported [21]. For example, we show that SS18 innate or native condensate could enrich acetyltransferase CBP/p300 while allowing free access of HDAC1/2 (the most abundant HDACs in BJ cells) and corresponding repressor complexes containing HDAC1/2, like SIN3 complex (SIN3A), NuRD complex (RBBP4) and CoREST (RCOR1), into the interior of the condensates (Fig. 2a). Interestingly, while SS18-SSX onco-condensates remain competent in enriching CBP/p300 (Fig. 2b) and BRG1 (Supplementary Fig. 5a), it readily excludes HDAC1/2 and corresponding repressor complexes (Fig. 2b), which could also be observed by HDAC1/2 CUT&Tag experiment (Supplementary Fig. 5b, c). We also show that the exclusion behavior of HDACs also occurs in endogenous SS18-SSX condensates engineered by gene editing in human embryonic stem cells (hESCs) (Supplementary Fig. 6a-c), synovial sarcoma cell line HS-SY-II (Supplementary Fig. 6i) and in primary tumor cells from synovial sarcoma patients (Supplementary Fig. 6j). Interestingly, the knock-in of C-terminal SSX at the SS18 locus led to the activation of genes for neural differentiation (Supplementary Fig. 6d, e) and neural rosette-like structures in hESCs (Supplementary Fig. 6g, h) without intervening pluripotent genes expression (Supplementary Fig. 6f). These results suggest that SS18-SSX acquire the ability to exclude repressors through its oncogenic translocation.

Fig. 2

SS18-SSX1 excludes HDAC1/2 complexes from condensates. Representative immunofluorescent images of endogenous p300/CBP, HDAC1/2 or HDAC1/2 associated transcription repressive complexes SIN3/NuRD/CoREST and lentiviral expression of SS18-EGFP (a) or SS18-SSX-EGFP (b) on the left panels. Scale bars, 5 μm. The violin plots on the right panels show the quantitative analysis of co-localization. Outside and inside groups indicate the distribution of random pixels’ fluorescence intensity normalized by Z score from outside and inside of SS18 (a) or SS18-SSX (b) condensates. Two-sided Wilcoxon test adjusted for multiple comparisons. n = 30 pixels, from 3 nuclei. *p < 0.05, ****p < 0.0001. ns, not significant. SS, SS18-SSX1

Forced re-entry of HDAC1 into SS18-SSX condensates neutralizes its oncogenic potential

The exclusion of HDACS further intrigues us to probe the underlying mechanisms. First, we wish to test if endogenous HDACs can be forced to return to the condensates. To this end, we took advantage of the well-known N12(Sall4)-NuRD interaction module [22, 23] and fuse it or its variants to SS18-SSX (SS) and show that N12, not its variant N12(R3A) nor N12(K5A), can pull endogenous HDAC1 into the condensates (Fig. 3a, b), presumably as part of NuRD complex [22, 23]. Restoration of HDAC complexes in the condensates severely dampens the activation of SS18-SSX downstream genes such as PAX6 and NKX3-2 (Fig. 3c) and almost abolishes the tumorigenicity of SS18-SSX tested by EdU cell proliferation assay (Fig. 3d, e), colony formation assay (Fig. 3f, g) and tumor-bearing mouse models (Fig. 3h, Supplementary Fig. 7a, b). These results suggest that the active exclusion of HDAC repressor complexes by SS18-SSX onco-condensates may play a critical role in synovial sarcomagenesis.

Fig. 3

Interference HDACs complexes exclusion alleviates oncogenicity of SS18-SSX condensates. a Schematic illustration shows the SS18-SSX fused transcription factor SALLs derived N12 domain in the N terminus could recruit HDAC1/2 involved NuRD complex. N12 with R3A or K5A mutation loses that capability. SS, SS18-SSX. b Left panels are representative immunofluorescent images of endogenous HDAC1 in the BJ cells with lentiviral expression of SS18-SSX, N12 fused SS18-SSX and N12 with R3A or K5A mutation fused SS18-SSX, respectively. Scale bars, 5 μm. The violin plots on the right panels show the quantitative analysis of co-localization. Outside and inside groups indicate the distribution of random pixels’ fluorescence intensity normalized by Z score from outside and inside of SS18-SSX, N12 fused SS18-SSX and N12 with R3A or K5A mutation fused SS18-SSX condensates, respectively. Two-sided Wilcoxon test adjusted for multiple comparisons. n = 30 pixels, from 3 nuclei. ****p < 0.0001. ns, not significant. SS, SS18-SSX1. c The expression of representative downstream genes of SS18-SSX in BJ fibroblasts with lentiviral expression of EGFP, SS18-SSX, N12 fused SS18-SSX and N12 with R3A or K5A mutation fused SS18-SSX, respectively. Data are mean ± s.d., two-sided, unpaired t-test, **p < 0.01, ***p < 0.001, ****p < 0.0001. n = 3 independent experiments. SS, SS18-SSX1. d Representative image of EdU assay in synovial sarcoma cell line SW982 with lentiviral expression of (b). Scale bars, 200 μm. SS, SS18-SSX1. e Histogram shows the ratio of EdU positive cells of (d). Data are mean ± s.d., two-sided, unpaired t-test, **p < 0.01, ns, not significant. n = 3 independent experiments. SS, SS18-SSX1. f Representative images of colony formation assay in synovial sarcoma cell line SW982 overexpressing mCherry, SS18-SSX, N12 fused SS18-SSX and N12 with R3A or K5A mutation fused SS18-SSX, respectively. One well of a 6-well plate was seeded with 1,000 cells and cultured for 10 days. Colonies were stained with 0.1% crystal violet. SS, SS18-SSX1. g Histogram shows the number of colonies in one well from (f). Data are mean ± s.d., two-sided, unpaired t-test, ****p < 0.0001, n = 6, from 3 independent experiments. SS, SS18-SSX1. h Tumor growth curve of SW982 cell overexpressing mCherry, SS18-SSX and N12 fused SS18-SSX xenografts established in NCG mice for 21 days. Data are mean ± s.d., two-sided, unpaired t test of n = 4 mice per group from two biological replicates, ****p < 0.0001

Re-entry of HDAC1 to SS18-SSX condensates via N12 further encouraged us to probe the exclusion mechanism. To this end, we focused on HDAC1, which has N-terminal histone deacetylase and C-terminal disordered region (Supplementary Fig. 8a). By deletion analysis, we show that its C-terminal disordered region is both required and sufficient for exclusion by SS18-SSX condensates (Supplementary Fig. 8b). We then analyzed the amino acid distribution of HDAC1 and found that its C terminus is enriched with acidic amino acids (Glu and Asp), basic amino acids (Lys) and serine (Supplementary Fig. 9a). We performed mutation analysis and show that mutating the charged residues render HDAC1 accessible again into SS18-SSX1 condensates, especially the one on acid residues, i.e., HDAC1(A) (Supplementary Fig. 9b). These results provide preliminary insight into the inclusion and exclusion of protein complexes into condensate.

SS18-SSX condensates accumulate H3K27ac that can be targeted pharmacologically

One of the consequences of HDAC exclusion may be the accumulation of acetylated histones. To test this hypothesis, we focused on H3K27ac and show that H3K27ac distribution is dramatically different between SS18 to SS18-SSX (Fig. 4a, b), with significant enrichments in the onco-condensates, which has not been observed from the other common active epigenetic modifications, such as H3K9ac, H3K4me3, and H3K36me2/3 (Supplementary Fig. 10a), and indeed, the interference of HDACs exclusion via N12 would significantly reduce the abnormal deposition of H3K27ac and the downstream gene expression (Supplementary Fig. 10b). Dramatically, H3K27ac and H2AK119ub have been transformed from mutually exclusive to inclusive as a result of SS18-SSX1 expression (Fig. 4c). Consequently, SS18-SSX1 is able to endow de novo H3K27ac in the entire SS18-SSX binding sites as shown by CUT&Tag experiments (Fig. 4d), although the intensity of global H3K27ac modification has been attenuated, especially at the intrinsic SS18 binding sites (Supplementary Fig. 11a-c). Additionally, the H3K27ac sites are more localized at 5’UTR and distal intergenic regions upon SS18-SSX expression (Supplementary Fig. 11d).

Fig. 4

Onco-condensate remodels H3K27ac. a Representative image of immunofluorescence for H3K27ac in the BJ fibroblasts with lentiviral expression of SS18-EGFP or SS18-SSX1-EGFP. Scale bars, 5 μm. b Violin plot shows the pixel intensity distribution of immunofluorescence for H3K27ac in (a). Outside and inside groups indicate the distribution of random pixels’ fluorescence intensity of H3K27ac normalized by Z score from outside and inside of SS18 or SS18-SSX1 condensates, respectively. Two-sided Wilcoxon test adjusted for multiple comparisons. n = 30 pixels, from 3 nuclei. ****p < 0.0001. c Representative image of immunofluorescence for H3K27ac and H2AK119ub in the BJ fibroblasts with lentiviral expression of SS18-SSX1. Scale bars, 5 μm. d Heatmap shows the intensity changes of H3K27ac occupancy at SS18-SSX1 binding sites upon expression of the onco-fusion protein. e Jointly analysis of Hi-C, CUT&Tag of histone modification, SS18-SSX1 and cohesin-CTCF at three representative loci in BJ fibroblasts with the expression of SS18-SSX or condensate-deficient SS18(YA)-SSX as control. The green boxes indicate the newly formed long-range interaction. The blue and black dotted lines indicate the intrinsic chromatin domains. f Venn plot shows the intersection of cohesin component RAD21 binding sites in BJ cells with the overexpression of SS18-SSX1 and SS18(YA)-SSX as control. SS, SS18-SSX1. g Venn plot shows the intersection of cohesin component SMC3 binding sites in BJ cells with the overexpression of SS18-SSX1 and SS18(YA)-SSX as control. SS, SS18-SSX1. h Venn plot shows the intersection of CTCF binding sites in BJ cells with the overexpression of SS18-SSX1 and SS18(YA)-SSX as control. SS, SS18-SSX1. i H3K27ac histone modification sites were divided into three groups by Venn plot, i.e., control specific sites, SS18-SSX1 specific sites and the overlapped sites. j The colocalization analysis between the overlapped H3K27ac sites in (i) with cohesin component RAD21 binding sites in BJ cells with the overexpression of SS18-SSX1 and SS18(YA)-SSX as control, respectively. SS, SS18-SSX1. k The colocalization analysis between the specific H3K27ac sites in (i) with cohesin component RAD21 binding sites in BJ cells with the overexpression of SS18-SSX1 and SS18(YA)-SSX as control, respectively. SS, SS18-SSX1

The dramatic co-localization of H3K27ac, H2AK119ub (a gene silencing marker by PRC1 [18]), and SS18-SSX, but not SS18, highlights extensive remodeling of chromatin when HDACs are excluded. We found the occupancy of H2AK119ub modification and SUZ12/PRC2 decreased and BMI1/PRC1 increased at the sites with dynamic changes of H3K27ac/me3, except for just a few genes, like NKX3-2 (Supplementary Fig. 11e-g).

We next analyzed the histone modification and Hi-C data combined with cohesin-CTCF, we show that the newly formed long-range interaction between H2AK119ub modified chromatin upon SS18-SSX expression could cross the boundary of the original chromatin domain established by cohesin-CTCF and endows aberrant H3K27ac deposition at these sites that were supposed to be silent (Fig. 4e). In comparison to condensate-deficient mutant SS18(YA)-SSX control, the expression of SS18-SSX in BJ cells attenuates the occupancy of cohesin but enhances the CTCF binding moderately (Supplementary Fig. 12a). Additionally, cohesin-CTCF shares most of the binding sites in the control and SS18-SSX overexpressed BJ cells (Fig. 4f-h). Intriguingly, when the total H3K27ac sites are divided into three groups, i.e., control specific sites, SS18-SSX1 specific sites and the overlapped sites (Fig. 4i, Supplementary Fig. 12b), we show that both the control specific and the overlapped sites have a large overlap with cohesin-CTCF binding sites, but the newly formed SS18-SSX1 specific sites are independent with cohesin-CTCF (Fig. 4j, k, Supplementary Fig. 12c-f). These data indicate that the oncogenic condensates formed by SS18-SSX may have created special microenvironment to enrich chromatin with H2AK119ub and CBP/p300, and simultaneously exclude HDAC1/2 complexes, which makes it favorable to deposit H3K27ac at these regions and activate the expression of downstream genes. Indeed, downstream genes activated by SS18-SSX indeed contain a stronger H2AK119ub modification (Supplementary Fig. 12g).

The imbalance of H3K27ac modification created by HDAC exclusion may be corrected by CBP/p300 inhibition. We tested this hypothesis with the compound A-485, a highly selective and drug-like CBP/p300 catalytic inhibitor [24], to decrease the over-deposited H3K27ac modification in SS18-SSX condensates (Supplementary Fig. 13a). We show that A-485 dramatically impede the activation of SS18-SSX downstream genes (Fig. 5a). A-485 treatment also reduces the expression of over 60% genes up-regulated by SS18-SSX1 (Fig. 5b) but has minimal impact on down-regulated and none-regulated genes (Supplementary Fig. 13b). Consistently, A-485 blocks tumorigenicity of onco-condensate validated by EdU cell proliferation assay (Fig. 5c, d) and tumor-bearing mouse models constructed by SW982 with overexpression of SS18-SSX (Fig. 5e, Supplementary Fig. 13c, d) and synovial sarcoma cell line HS-SY-II (Supplementary Fig. 13e-g), which suggested that targeting H3K27ac histone modification may be efficacious for treating synovial sarcoma carrying SS18-SSX.

Fig. 5

H3K27ac deposition can be targeted pharmacologically. a The expression of representative genes in BJ fibroblasts expressing control or SS18-SSX1 treated with DMSO or 1 μM A485 inhibitor. Data are mean ± s.d., two-sided, unpaired t-test; n = 3 independent experiments. **p < 0.01, ***p < 0.001, ****p < 0.0001. b Venn plots showing the intersection between upregulated genes and A485 treatment downregulated genes in BJ fibroblasts expressing SS18-SSX1. c Representative image of EdU assay in synovial sarcoma cell line SW982 with lentiviral expression of EGFP or SS18-SSX1 treated by DMSO or 1 μM A485. Scale bars, 200 μm. d Histogram shows the ratio of EdU positive cells in (c). Data are mean ± s.d., two-sided, unpaired t-test, n = 4, from 2 independent experiments. e Tumor growth curve of SW982 cell overexpressing SS18-SSX1-EGFP xenografts established in NCG mice. Mice were treated intraperitoneally with vehicle control or A-485 at 100 mg/kg/dose, twice daily for 12 days (BID × 12). Data are mean ± s.d., two-sided, unpaired t test of n = 7 mice per group from two biological replicates, **p < 0.01. f A Model for condensate remodeling as a driver for synovial sarcoma. SS18 WT protein forms innate phase separation for its normal physiological function (left panel). In contrast, SS18-SSX1 has enhanced condensates formation tendency mediated by ICBS mechanism via H2AK119ub histone modification and remodels 3D genome structure (right panel). The enhanced onco-fusion condensates remodel H3K27ac by excluding HDACs complexes, which resulting in aberrant activation of embryonic developmental genes