Here we present compelling evidence for the pathogenicity of non-truncating or NMD-escaping variants in the EHD2 and ARID functional domains of the ARID1B gene, thereby providing novel insights into the understanding of ARID1B-associated CSS with implications for genetic diagnosis.
We initially employed gene expression profiling and subsequent methylation analysis from peripheral blood samples to confirm the clinical relevance of variant 2129del4 in Ind1. Both methods reached the same outcome, reliably classifying Ind1 in the ARID1B-CSS group. Two additional alterations, the herein tested frameshift deletion 2188ter and the previously investigated variant C1945R(Aref-Eshghi et al. 2018a), also showed BAFopathy methylation profiles, whereas two others (E2011V, D1727N(Aref-Eshghi et al. 2018a)) did not. In all cases, our experimental findings confirmed the results of RNA-Seq and DNA-methylation episignatures, both of which have rapidly found their way into research and diagnostic contexts(Stenton and Prokisch 2020; Sadikovic et al. 2021). Taken together, both approaches can reliably be used for the classification of non-truncating ARID1B variants.
ARID1A/B subunits interface with SMARCA4 via their conserved EHD2 domain. SMARCA4 missense variants that impair the interaction with the EHD2 domain of the ARID1B paralog, ARID1A, have previously been linked to reduced BAF complex function(Mashtalir et al. 2020). Based on these data, compromised interaction with SMARCA4 was presumed initially as the mechanism of pathogenicity for EHD2 non-truncating variants in ARID1B. However, decreased interaction was observed only for the frameshift variant 2188ter, which deletes 61 amino acids from the EHD2 domain (roughly 10%), thus having the largest impact on the overall structure. These results indicate that loss of interaction with SMARCA4 is not the main cause of pathogenicity for EHD2 variants. This conclusion is in accordance with a systematic mutational screen, finding only few EHD2 variants with the potential to impact binding to SMARCA4(Mermet-Meillon et al. 2024).
Instead, we observed the formation of aggresomes in the majority of EHD2 and both ARID variants. The only alterations that did not exhibit aggregation were the two variants which did not show a BAFopathy methylation profile (D1727N, E2011V). Aggresomes, also known as microtubule-dependent cytoplasmic inclusion bodies, are pericentriolar structures owing their extreme stability to the surrounding vimentin cage. They arise when proteasome capacity is exceeded by an overload of misfolded, mostly poly-ubiquitinated proteins, subsequently leading to the accumulation of peripheral small protein aggregates proximal to the MTOC(Johnston et al. 1998; Ajmal 2023). Nuclear aggregates like the ones observed in ARID domain variants have also been associated with an abnormal protein conformation(Ajmal 2023). By measuring protein levels in the cell through a stability sensor assay, Mermet-Meillon and colleagues described a negative effect of EHD2 missense variants on protein stability. Nevertheless, the applied methodology could not differentiate between protein misfolding, destabilisation or mislocalisation(Mermet-Meillon et al. 2024). Our findings further elucidate the pathomechanism of the EHD2 non-truncating variants by revealing a loss-of-protein function due to misfolding and aggregation. Amino acid substitutions in the ARID domain exhibited the same effect. The formation of aggregation is most likely attributed to the exposure of amylogenic protein stretches of the EHD2 and ARID domain due to the nearby alterations(Teng and Eisenberg 2009). Interestingly, the amylogenic segments (Fig. 5 and Table S2) are part of the central helical structures within the EHD2 domain, which were previously reported to be particularly sensitive to pathogenic variants(Mermet-Meillon et al. 2024).
Furthermore, Mermet-Meillon and colleagues showed that some clinically relevant EHD2 variants from ClinVar caused decreased ARID1B protein levels(Mermet-Meillon et al. 2024). Their analysis included the missense variant I2031T, referred to as I2018T under their nomenclature, which was also examined in our study. The authors concluded that I2031T would lead to reduced protein levels, according to a FACS-based assay. On the contrary, quantitative western blot analysis in the present study showed that all aggregated variants, including I2031T, exhibit the same ARID1B protein levels as the wild type protein (Fig. S4). The difference between the two studies may likely be attributed to the different methodologies applied. Specifically, our study addressed protein levels of the entire wild type protein or I2031T variant, whereas the aforementioned study specifically addressed stability of the EHD2 domain fragment. Moreover, we demonstrated the formation of aggresomes, which are considered stable formations, possibly leading to the preservation of total protein levels.
The ARID1B-associated BAFopathy belongs to the mild CSS spectrum. Nevertheless, moderate and severe CSS cases have also been described(Hoyer et al. 2012; van der Sluijs et al. 2019; Vasko et al. 2021; Schmetz et al. 2024). Five of the herein described individuals with aggregating EHD2 variants and available clinical description as well as the carrier mother of the individual harboring the I2031T variant presented with mild DD/ID. No other serious malformations or congenital anomalies were noted(Mignot et al. 2016; Yan et al. 2019a; Miyamoto et al. 2021b). Four of them displayed corpus callosum agenesis (File S1 “clinical_table”, File S2 “clinical reports”). In the literature one additional de novo (likely) pathogenic EHD2 missense change (c.5855T > C p.Met1952Thr) in an individual with mild ID, epilepsy and marfanoid features has been reported(Chevarin et al. 2020). Unfortunately, ClinVar entries lack the corresponding clinical information. Although the initial clinical descriptions point to a mild CSS phenotype, a conclusive assessment concerning phenotypic severity requires a larger cohort of CSS cases with non-truncating EHD2 variants. For alterations in the ARID domain there is no clinical information reported to date.
So far, the pathogenicity of non-truncating ARID1B variants was controversial due to the lack of experimental data, as reflected in the respective ClinVar entries. Indeed, many of the listed EHD2 and ARID domain variants are interpreted as VUS, whereas some alterations outside of these domains have been classified as (likely) pathogenic without functional evidence. Our study together with the recently published findings of Mermet-Meillon and colleagues(Mermet-Meillon et al. 2024) experimentally confirms the pathogenicity of non-truncating or NMD-escaping EHD2 variants. To our knowledge, the effect of non-truncating ARID domain variants has not been previously investigated. Nevertheless, the functional studies presented here demonstrate robust pathological effects. We understand that missense variants are difficult to interpret and that not all EHD2 or ARID domain alterations will be causative. Structural predictions, like the AlphaMissense score, and methylation or transcriptome analysis constitute reliable supporting tools for further clarification, yet they are not always available in everyday practice. To enable an effective assessment in the diagnostic setting we propose that de novo variants (PS2_supporting) in the EHD2 or ARID domains (PM1_moderate), which are absent or infrequent (≤ 2) in population databases (PM2_supporting) and reach computational evidence thresholds (PP3_supporting or PP3 moderate) in individuals with CSS clinical features can be a priori interpreted as likely pathogenic.
In conclusion, we demonstrate the pathogenicity of yet unexplored non-truncating/NMD-escaping variants in the EHD2 domain of ARID1B. Additionally our study suggests the causality of variants located in the ARID functional domain of ARID1B. Experimental investigation reveals for the first time the underlying pathomechanism for both EHD2 and ARID domain variants, which involves protein accumulation in stable aggresomes or nuclear aggregates due to protein misfolding. Consequently, our findings set the starting point for (re-)evaluation of unclear non-truncating changes in the ARID1B gene.