Generation of amino acid substitution and truncated variants of KDM6A based on the mutational landscape of KDM6A in tumor tissues and cell lines. Based on 64668 tested samples from 44 tissue types, 2496 unique mutations are found in COSMIC v92 (GRCh 38, November 2020) for KDM6A. Among the cancer tissues, meninges and the urinary tract exhibit the highest mutation frequency in each more than 30% of the tested samples (411/1336 cases for UC). The three most frequent point mutations across all tissues are the synonymous mutation Q1037= (c.3111G > A) in the JmjC domain, the missense mutation T726K (c.2177C > A) and the truncating Q555* (c.1663C > T) both in the intrinsically disordered region. We selected T726K as a hotspot substitution variant and substitution mutations, typically found in urothelial cancer cell lines and tissues: E315Q (located in TPR6) and D336G (TPR7), P966R (loop in a beta sheet cavity containing the active center), V1338F (zinc-binding domain) and C1361Y (zinc-binding). To compare the impact of these substitution mutations, we additionally selected the variants Q1133A (H3-tail interaction) and H1329A (hydrophobic patch near active center) that have been shown to be catalytically inactive (14). Positions P966, Q1133, V1338 and C1361 are conserved in the KDM6A zebrafish orthologue and its closest functional homologs KDM6B and KDM6C. We created all substitution variants by site-directed mutagenesis using eGFP-KDM6Aorigin as a template (Fig. 1A). As nonsense mutations make up nearly one quarter of all point mutations and generate truncated variants with partial losses of the IDR and JmjC, we established a set of truncated variants with an eGFP-tag fused to their N-terminus: ΔTPR, ΔIDR, ΔJmjC, TPR (= ΔIDR/ΔJmjC) and JmjC (= ΔTPR/ΔIDR) (Fig. 1C). These variants were transiently transfected into urothelial cancer cell lines with wildtype (SW-1710) or mutated (T-24) KDM6A/KMT2C/D proteins. All KDM6A substitution and deletion variant proteins were detectable at the expected sizes (Fig. 1B/D), except ΔIDR, which was predicted to have the lowest solubility of all truncated variants (Table S2). In a second step, we analysed all variants for their demethylation activity. KDM6A JmjC and the flanking Zinc-binding domain are known to recognize and bind several amino acids between H3R17-H3T32 of the H3K27 di- and tri-methylated N-terminal histone tail to ensure substrate specificity (14). Consequently, mutations in the JmjC and flanking domains have a high potential to impede or even abolish the catalytic activity. However, it is unknown to what extent mutations in the TPR and IDR might contribute to KDM6A demethylase activity.
KDM6A demethylase activity is strongly affected by substitutions and deletions within the JmjC domain. To assess the demethylase activity of KDM6A variants, we established an ELISA-based H3K27me3 demethylation assay to screen for the catalytic activity of all truncated and substitution variants. The assay was carried out with the purified protein attached to GFP-dynabeads. We used fluorescence emission of the eGFP tag to determine and normalize the amount of protein before and after binding to dynabeads (Figure S2A). The activity of the variants was measured by colorimetric readout of the demethylated products and subsequent fitting via 4PL-regression (Figure S3A/B). The eGFP-KDM6A WT species served as a reference point for demethylase activity (Figure S3C). The activity of substitution variants was assessed in a quantitative manner, yielding specific activities summarized in Table 1. As shown in Fig. 1A, eGFP-KDM6A variant T726V featured a demethylase activity in the range of WT. Note that TPR substitutions E315Q and D336G had a slightly higher activity. Interestingly, activity in T726K decreased time-dependently within two days after transfection (Figure S4). H1329A and V1338F had strongly reduced activity, while P966R, Q1133A and C1361Y were considered as non-active.
Table 1
Specific activity of substitution variants.
No. | KDM6A variant, affected domain | Specific activity [10− 3 µmol min− 1 mg− 1] |
1 | WT | 2.15 ± 0.24 |
2 | E315Q, TPR | 3.96 ± 0.58 |
3 | D336G, TPR | 3.76 ± 0.89 |
4 | T726K, IDR | 0.69 ± 0.10 |
5 | T726V, IDR | 1.54 ± 0.11 |
6 | P966R, JmjC | < 0.07 ± 0.03 |
7 | Q1133A, JmjC | < 0.09 ± 0.02 |
8 | H1329A, JmjC | 0.28 ± 0.05 |
9 | V1338F, JmjC | 0.29 ± 0.03 |
10 | C1361Y, JmjC | < 0.12 ± 0.03 |
Specific activity of KDM6A variants highly depends on the mutation site. Mutations affecting JmjC domain impair catalytic activity, whereas TPR substitution mutation enhanced activity. IDR mutation T726K showed time-dependent reduction of activity. Specific activity was calculated as described in additional method 1.
The demethylase activity of the truncated KDM6A variants was assessed in a qualitative manner since changes in the fluorescence emission spectra of the truncated variants did not allow proper quantification of the amount of protein used in the assay (Figure S2C). As expected, experiments with truncated (Δ- and ΔΔ-) variants exclusively yielded activity when the JmjC domain was preserved (Figure S2D).
KDM6A single substitutions do not alter nucleoplasmic localization. In SW-1710 cells, wildtype eGFP-KDM6A was located in the nucleoplasm and in the cytoplasm, sometimes with a tendency to cytoplasmic speckle formation (dependent on the protein dose of the transient overexpression). All substitution variants showed cytoplasmic and predominantly nucleoplasmic localization with different degrees of cytoplasmic speckle formation, as well as weak accumulation around the nucleoli (Fig. 2A). KDM6A transport to the nucleus depends largely on the KMT2C/D COMPASS complex (17). Therefore, the observation that KDM6A WT and all substitution variants localized in the nucleus suggests functional nuclear import, as well as interaction with the COMPASS complex. Notably, a recent study [17] found that KDM6A TPR mutations, among them the D336G variant, predominantly localized in the cytoplasm in stably transfected HeLa cells (which have no detectable endogenous KDM6A expression). We performed a pull-down experiment of KDM6A WT, D336G and T726K variants with RBBP5 to test for impaired association with the KMT2C/D-complex. RBBP5 is a core component of KMT2 complexes that directly interacts with ASH2L, WDR5 and DPY30 within the WRAD complex to bind specific chromatin sequences, stabilize and activate the methyltransferase activity of KMT2 proteins. RBBP5 was pulled down with all three KDM6A variants to similar extents, but not with the eGFP control (Fig. 2B), suggesting that they are present in the same complex. Furthermore, perinuclear KDM6A speckles also stained DAPI positive, indicated newly formed micronuclei (Figure S5). Micronucleus formation, an indicator of genomic instability, is common in urothelial cancer cell lines and was slightly enhanced in KDM6A transfected cells. An increased tendency for DNA release was identified with some KDM6A substitution variants, which colocalized at the released DNA sequences (Fig. 2A, catalytically impaired variants). This phenomenon was much stronger in KDM6A truncation variants (see below).
KDM6A truncated variants cause severe nuclear damage. Compared to the correctly localized KDM6A control variant, deletion of any functional domain resulted in either one of two phenotypes: (1) a “mild” phenotype with weak cytoplasmic and nucleoplasmic KDM6A-Δ-variant expression as shown in Figure S7 and (2) a “severe” phenotype with a speckled, perinuclear distribution of the transfected protein. JmjC and ΔJmjC exhibited a more wildtype-like localization pattern with normally shaped nuclei, whereas following transfection of TPR and ΔIDR, nuclei were malformed or cells displayed multiple non-segregated nuclei and nuclear defects (Fig. 3, Figure S7). Nuclear DNA, which was excessively released into the cytoplasm, colocalized with KDM6A truncation variants. These results indicate the intrinsic ability of all variants to bind to chromatin, either directly through the JmjC or indirectly via other protein-protein interactions by TPR-containing variants. The presence or absence of the intrinsically disordered region (IDR) strongly affects stability and solubility of KDM6A, but not its enzymatic activity. All variants principally localized to the nucleus, but (to various extents) showed anomalies like partial nuclear redistribution to nucleoli or accumulation in perinuclear DNA-associated speckles. These observations raise two further questions, namely (1) how deletions of one or two KDM6A domains impair functional interactions with known proteins such as RBBP5, the KMT2C/D (COMPASS) complex and (2) whether Nucleophosmin (NPM1), a prominent shuttling and chaperone protein found in nucleoli, may be involved in KDM6A interactions with these organelles. To address these questions, we performed protein immunoprecipitation (Co-IP) and co-staining of transiently transfected KDM6A variants with NPM1 and RBBP5.
Full-length KDM6A is needed for maximal binding of the COMPASS-complex core component RBBP5. First, we tested for RBBP5 interaction with KDM6A by Co-IP experiments 48 h post transfection of selected KDM6A truncation variants (Fig. 4A). We had previously shown that RBBP5 was enriched in KDM6A-tagGFP2 Co-IPs in different urothelial cancer cell lines (18). Interestingly, we observed that for maximum interaction with RBBP5 all KDM6A domains are needed (Fig. 4B). Deletion of any domain impaired RBBP5 binding. Specifically, TPR-containing variants (ΔJmjC and TPR, Figure S8) and to some extent IDR-containing variants (ΔTPR) bound RBBP5 to some degree, but JmjC alone did not at all.
Intriguingly, MS data analysis from two KDM6A-tagGFP2 stably transduced urothelial cancer cell lines showed a significant enrichment of nucleolar proteins, such as Nucleolin, NPM1 and ribosomal subunits, in addition to histone variants (Figure S9, Table S4). However, in Co-IP experiments NPM1 was only significantly enriched with the KDM6A T726K mutant (Fig. 4C) whereas other selected KDM6A variants failed or gave very weak bands (ΔTPR and ΔJmjC, Fig. 4C white stars). Therefore, we searched for a complementary approach to detect associations of KDM6A with NPM1.
Scatterplots are widely used to detect colocalization. Typically, the intensities of two channels are plotted in a xy-diagram and the resulting populations elucidate on colocalization, which is displayed in Pearson (P between − 1 and 1) or channel-wise Manders values (M1, M2 between 0–1). However, one drawback of simple P, M1, or M2 values is that the connection between scatterplot populations and the spatial cellular information is lost. For this reason, we generated scatterplots to identify populations of interest, mapped them back to the respective cellular compartment using the freely available Fiji plugin ScatterJ (19) and compared the outcome with line profiles through the corresponding cellular compartments. By this approach, we identified localization and interaction changes among the KDM6A variants.
KDM6A WT and NPM1 form nucleoplasmic populations. As shown in the scatterplot for KDM6A WT and NPM1 (Fig. 5A, left panel), two unique fractions appeared at the y- or x-axis, representing signals in only one of the two channels. The corresponding eGFP-KDM6A WT-only fraction is shown in green and the NPM1-only fraction in red. A third fraction with green and red signals of different intensities is shown in orange and named from here on “intermediate” fraction. Cellular back mapping of the selected populations (Fig. 5A, middle panel), clearly showed KDM6A WT-only signal predominantly in the cytoplasm and little in the nucleoplasm. As expected, the NPM1-only signal was present in nucleoli and in nucleoplasm. The intermediate fraction was always associated with NPM1-only fraction at the nucleoli rim and within the nucleoplasm, indicating a dynamic exchange of NPM1-only, mixed complexes and KDM6A WT-only fractions at specific sites within the nucleus. The line profile through the nucleus with DAPI as a DNA indicator (Fig. 5A right panel) corroborates these findings: High NPM1 (red) signal intensities were exclusively found in nucleoli, whereas green-red overlapping signals represent the KDM6A WT-NPM1 complexes.
The T726K hotspot mutant is absent from nucleolamina and additionally forms cytoplasmic complexes. By comparing the scatterplot presentations from T726K or WT KDM6A and NPM1, we observed a much broader intermediate fraction with similar KDM6A-T726K signal intensities but higher NPM1 intensities (Fig. 5B), which was also visible in the line profile. By cellular back mapping, we identified the intermediate population in the nucleoplasm and in the cytoplasm, where KDM6A T726K (green population) was also detectable. In contrast, the nucleoplasmic NPM1-only fraction was completely absent with this mutant. In the original image (Figure S10A), DAPI staining clearly indicated DNA release into the cytoplasm. Scatterplot analysis of KDM6A T726K with DAPI (Figure S10B) showed KDM6A T726K associated to DNA in the nucleus and at the released DNA (population 3). Furthermore, we identified a DAPI-only fraction (population 4), which was also negative for NPM1 at the nuclear lamina.
Truncated KDM6A variants likewise form complexes with NPM1 at extranuclear DNA segments. As described above, truncated KDM6A variants elicited more severe nuclear DNA release and nuclear damage (Fig. 3). We exemplary selected KDM6A JmjC (Fig. 5C) and KDM6A TPR (Figure S10C) for further analysis. Scatterplot analysis and line profiling of the truncated variants with NPM1 clearly indicated an enriched, colocalizing intermediate fraction at extranuclear DNA segments. As observed before, KDM6A JmjC and TPR were also localized to the cytoplasm. In contrast to KDM6A WT, these variants were always associated with DNA, as shown by DAPI-KDM6A scatterplot analysis and cellular back mapping approaches (Fig. 5, Figure S10 B/C). By performing scatterplot analysis combined with the cellular back mapping approach, we identified KDM6A-NPM1 compositions of variable stoichiometry and different localization, which were hardly detectable by Co-IP/WB analysis (Fig. 4). Importantly, scatterplot analysis and the cellular back mapping approach highlighted differences among the wildtype, substitution and truncated KDM6A variants. The T726K variant represents a non-wildtypic phenotype with reduced enzymatic activity in a time-dependent manner and sometimes even DNA release. The severe phenotype of the truncated variants is characterized by a massive DNA release and forms nearly equal complexes with NPM1, pointing to a role of NPM1 different or in addition to the wildtype.
The severe phenotype of KDM6A truncation variants is characterized by mitotic defects and DNA release. As DNA release and micronuclei formation have been observed, we analysed whether the typical indicators for a DNA damage response are activated by KDM6A truncation variants, namely phospho-γH2A.X and RAD51. As shown in Figure S11, we detected elevated levels of phospho-γH2A.X with truncated KDM6A variants, whereas overall phospho-γH2A.X levels were unchanged in KDM6A WT transfected cells, even in cells with cytoplasmic aggregates. Visual inspection of phospho-γH2A.X in cells with truncated variants revealed defects in mitosis (Fig. 6A). These defects occurred (1) in anaphase as lagging chromosomes and multiple fragmentation events and (2) in telophase and cytokinesis by persisting chromosome bridges and accumulation of DNA damage sites at chromosome bridges (see also Figure S7C). Line profiles through the chromatin bridges with stained endogenous DNA damage response markers RAD51 and phospho-γH2A.X together with KDM6A JmjC or ΔTPR and additionally NPM1 indicated colocalization of all these proteins at DNA (Fig. 6C). Cell counting by features of DNA release, mitotic defects and micronuclei formation in untransfected cells or cells transfected with eGFP, KDM6A WT and truncated variants showed no statistically significant changes in the extent of micronuclei formation, except for ΔJmjC, but strongly increased mitotic defects and DNA release (Fig. 6B).