Mechanisms of RBM20 Cardiomyopathy: Insights from Model Systems

Abstract

RNA binding motif 20 (RBM20) is a vertebrate- and muscle-specific RNA-binding protein that belongs to the serine-arginine-rich (SR) family of splicing factors. The RBM20 gene was first identified as a dilated cardiomyopathy (DCM)-linked gene over a decade ago. Early studies in Rbm20 knockout (KO) rodents implicated disrupted splicing of RBM20 target genes as a causative mechanism. Clinical studies show that pathogenic variants in RBM20 are linked to aggressive DCM with early onset heart failure and high mortality. Subsequent studies employing pathogenic variant knock-in (KI) animal models revealed that variants in a specific portion of the RS domain in RBM20 not only disrupt splicing, but also hinder nucleocytoplasmic transport and lead to the formation of RBM20 biomolecular condensates in the sarcoplasm. Conversely, mice harboring a disease-associated variant in the RNA recognition motif (RRM) do not show evidence of adverse remodeling or exhibit sudden death despite disrupted splicing of RBM20 target genes. Thus, whether disrupted splicing, biomolecular condensates, or both contribute to DCM is in debate. Beyond this, additional questions remain, such as whether there is sexual dimorphism in the presentation of RBM20 cardiomyopathy? What are the clinical features of RBM20 cardiomyopathy and why do some individuals develop more severe disease than others? In this review, we summarize the reported observations and discuss potential mechanisms of RBM20 cardiomyopathy derived from studies employing in vivo animal models and in vitro human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs). Potential therapeutic strategies to treat RBM20 cardiomyopathy are also discussed.

Introduction

Dilated cardiomyopathy (DCM) is a heart muscle disease characterized by left or biventricular dilation and impaired systolic function that cannot be explained by hypertension, valvular, congenital, or ischemic heart disease.¹ The prevalence of DCM may be as high as 1 in ~250 individuals.² DCM is a disease with diverse etiologies, including genetic variants, infection, inflammation, autoimmune diseases, and exposure to certain toxins.³ Genetic forms of DCM account for approximately 40% of reported cases.² Yet, unlike hypertrophic cardiomyopathy, which is caused almost exclusively by pathogenic variants in a small number of sarcomeric genes,⁴ the genetics of DCM are vastly more complex. To date, variants in over 50 genes encoding proteins belonging to diverse cellular structures (e.g., the sarcomere, nuclear envelope, cytoskeleton, sarcolemma, and intracellular junctions) have been implicated in the development of this disease. For lists of DCM-associated genes, the reader is referred to several excellent reviews that have been published on this topic.^2,5–8 Among DCM-associated genes, RBM20, which encodes RNA binding motif protein-20 (RBM20), is relatively unique. It is presently the only RNA-binding protein for which variants identified in human patients have been causally linked to disease.^9–18

RBM20 was first identified in two cases of aggressive familial DCM, now often referred to as RBM20 cardiomyopathy.¹⁹ At the time, the function of RBM20 was unknown, but structural similarities between RBM20 and proteins belonging to the serine-arginine-rich (SR) family of splicing factors led to the proposal that the protein was involved in pre-mRNA splicing.¹⁹ Soon after this discovery, we identified a naturally occurring deletion of Rbm20 as the cause of DCM in a rat strain deficient in Ttn splicing.²⁰ This rat strain was the first Rbm20 knockout (KO) model, confirmed that the protein regulates alternative splicing, and aided the identification of a set of 31 RBM20-regulated genes.²⁰ These genes included Ttn, which encodes the giant sarcomeric protein titin, as well as several other genes important for Ca²⁺-handling, such as Ryr2, Camk2d, and Cacna1c.^20,21 Thus, our findings provided a putative mechanistic basis for RBM20 cardiomyopathy. This disease was presumed to arise because of 1) reduced ventricular wall tension resulting from changes in titin splicing and 2) impaired contractility secondary to altered splicing of Ca²⁺-handling genes. This mechanism received further support with the demonstration that Rbm20 KO mice phenocopy KO rats with DCM and pro-arrhythmic changes in Ca²⁺-handling.²² Yet, it was apparent that something was missing. For one, the early onset and fast progression to heart failure observed in human patients with pathogenic variants in RBM20^19,23 was not recapitulated in Rbm20 KO rodents.^20,22 Moreover, when we generated mice lacking the RNA-binding domain in RBM20 (Rbm20^ΔRRM), which rendered the protein splicing-deficient, we made the startling discovery that these animals did not exhibit adverse remodeling despite impaired systolic function.²⁴ Consequently, it seemed disrupted splicing of RBM20 target genes may not be sufficient to cause RBM20 cardiomyopathy in rodents.

A contributing disease mechanism was hinted at when it was demonstrated that variants in a stretch of five amino acids within the arginine-serine-rich (RS) domain promoted redistribution of the protein from the nucleus to the cytoplasm in transfected cells.²⁵ This phenomenon was later confirmed in the myocardium of patients with RBM20 cardiomyopathy and Rbm20 pathogenic variant knock-in (KI) animal models with the striking finding that accumulation of RBM20 in the sarcoplasm promoted liquid-liquid phase separation of the protein in biomolecular condensates.^11,12,26 In the years since this discovery, these findings have been confirmed by our lab and others in additional rodent models and human induced pluripotent stem cell (hiPSC)-derived cardiomyocytes (CMs).^{9,10,13–18,27,28} The finding that RS domain pathogenic variant KI animals recapitulate aspects of RBM20 cardiomyopathy not observed in KO rodents was a significant breakthrough in the field as it not only unambiguously confirmed that these variants are causative in aggressive DCM, but also implicated sarcoplasmic RBM20 condensates in the development of this devastating disease. Nevertheless, many questions remain to be answered. In this review, we will highlight important questions that have been answered, as well as those still remaining, in light of evidence provided by recently developed RBM20 model systems.

RBM20 FORM AND FUNCTION

The RBM20 gene is located on the long arm of chromosome 10 in humans. It is comprised of 14 exons and encodes a protein 1227 amino acids in length. RBM20 belongs to the SR protein family of splicing factors.²⁹ Like other SR family proteins, RBM20 contains a C-terminal domain rich in arginine and serine dipeptides, termed the RS domain (encoded by exon 9) and an N-terminal RNA-binding domain of the RNA recognition motif (RRM) type (encoded by portions of exons 6 and 7) (Figure 1).

Figure 1.

Domain structure and pathogenic sequence variants in human RBM20. A. Schematic showing structure of human RBM20 with the amino acid positions of exon boundaries and domains/amino acid-rich regions indicated. B. Sequence alignment of the RS domain with pathogenic variants reported in the literature listed. *signifies residue that is not conserved throughout vertebrate evolution. The positions of the NLS, identified in our recent work,³⁰ and the DCM-associated variant hotspot in exon 9 (c.1881–1920) are indicated. C. Sequence alignment showing portion of the glutamate-rich region in RBM20 with pathogenic variants reported in the literature listed. The DCM-associated variant hotspot in exon 11 (c.2721–2760) is shown. Residues conserved throughout vertebrate evolution are highlighted in red. Residues highlighted in blue and green denote conserved positive and negative charge, respectively. For panels B and C, sequences for species at different points throughout vertebrate evolution from zebra fish (D. rerio) to humans (H. sapiens) were downloaded from the NCBI database. Sequence alignments were constructed with the aid of Multalin.³¹ References reporting pathological variants listed in panels B and C are provided in Table 1.

RRMs represent the most abundant RNA-binding domain in higher vertebrates and are often found in multiple copies in RRM-containing proteins.³² In particular, SR proteins contain at least one RRM, although several contain more than one.²⁹ Similar to the prototypical SR protein, SRSF2, RBM20 contains a single RRM (Figure 1). Evidence from prior in vitro studies suggested that the RRM is not essential for RBM20 splicing activity.²⁵ However, in vivo study with an in-frame targeted deletion of exons 6 and 7 (Rbm20^ΔRRM) revealed characteristic loss-of-function changes in the splicing of major RBM20 target genes such as Ttn and Camk2d.²⁴ Interestingly, the splicing of many other presumed RBM20 target genes was unaffected in the hearts of Rbm20^ΔRRM mice.²⁴ Thus, it appears that splicing of certain RBM20 target genes, such as Ttn and Camk2d, requires the RRM, at least in vivo, while the splicing of other RBM20 target genes may not. An unexpected finding of our work was that splicing changes in Rbm20^ΔRRM mice were insufficient to promote the development of a DCM-like phenotype similar to that in Rbm20-deficient mice and rats.^20,22 Whether this is due to differences in the magnitude of splicing alterations and/or lack of minor splicing changes in other RBM20 target genes beyond Ttn, Camk2d, and Ldb3/Cypher in Rbm20^ΔRRM mice is unclear at the present time.

The RS domain in SR proteins mediates protein-protein interactions, such as those between different RS domain-containing proteins and between SR proteins and components of the general splicing machinery.^33–35 There is also evidence that RS domains interact with RNA and that this interaction is required for pre-spliceosome assembly.³⁶ Furthermore, protein-protein interactions between the RS domain and the nuclear import receptor transportin 3 (also called transportin-SR) are important for SR protein nuclear localization.^37–39 Filippello et al. were the first to confirm the importance of the RS domain in RBM20 nuclear targeting in vitro, and further suggested that the entire region of the protein spanning from the RRM to the RS domain is important for this function.⁴⁰ Recently, we generated mice expressing RBM20 with an in-frame deletion of the RS domain (Rbm20^ΔRS) and found that the protein was mis-localized to the sarcoplasm in the hearts of these mice,³⁰ verifying the nuclear targeting function of the RS domain in vivo. Nevertheless, our analysis of RBM20 localization in the hearts of Rbm20^ΔRRM mice failed to show evidence of RBM20 mis-localization.³⁰ Whether this was due to retention of critical signals located between the RRM and RS domain is unclear. Subsequent in vitro analyses employing sequence deletion and DCM-associated variant constructs corroborated the nuclear targeting function of the RS domain, and identified the critical nuclear localization signal (NLS) (Figure 1).³⁰ Taken together, these findings confirm that the RS domain plays an essential role in nuclear targeting through a NLS located within this domain.

Beyond these two hallmark domains, RBM20 contains two U1-type zinc finger (ZnF) domains encoded by exons 3–4 and exon 13, respectively (Figure 1). In vitro studies have assessed the function of the ZnF domains in RBM20 splicing regulation, however, the results have been inconsistent. For example, results from in vitro splicing assays employing a series of RBM20 sequence deletion constructs and a TTN²⁴¹⁻³ expression construct indicated that, in addition to the RRM and RS domain, ZnF2 is important for splicing repression.⁴¹ Conversely, Murayama et al. failed to identify a role for either ZnF domain in RBM20 splicing regulation using RBM20 constructs lacking these domains and a Ttn reporter minigene consisting of exons 50, 51, 218, and 219 and the intervening introns located between exons 50–51 and 218–219.²⁵ Regardless, in light of findings related to the function of the RRM, in vivo experiments will likely be necessary to elucidate the function of the ZnF domains in RBM20.

In addition to these domains, RBM20 also contains regions rich in proline and leucine in exons 1 and 2, respectively, as well as a glutamate-rich region that spans from the end of exon 9 into exon 11 (Figure 1). To date, no function has been ascribed to these regions, although evidence suggests that the glutamate-rich region may be important for RBM20 protein stability.⁴² Specifically, analysis of tissue from a DCM patient heterozygous for the pathogenic E913K variant, which is located within the glutamate-rich region, uncovered significant downregulation of RBM20 at the protein level.⁴² Analysis of RBM20 mRNA levels revealed that they were not impacted by this variant.⁴² Therefore, at the present time it is thought that the glutamate-rich region is important for the stability of RBM20 and that variants in this region destabilize the protein, leading to its degradation.

In the healthy adult heart, RBM20 is localized to two speckles in the nucleus that we previously showed are active sites of transcription for the Ttn gene.⁴³ These speckles do not overlap with other nuclear bodies, such as paraspeckles, Cajal bodies, or PML bodies.⁴³ More recently, it was shown that RBM20 speckles represent cardiac-specific trans-interacting chromatin domains where certain chromosomes containing RBM20 target genes (e.g., titin, etc.) are brought into close proximity.⁴⁴ Localization to the nucleus is essential for RBM20 splicing function.^20,25 Our lab was the first to demonstrate that RBM20 regulates Ttn alternative splicing by acting as a splicing repressor.⁴³ Genome-wide analysis of RBM20-regulated exons by Maatz et al. revealed that four times more exons were repressed than activated.²¹ Taken together these results indicate that RBM20 functions primarily as a splicing repressor in the heart, although this function is gene-dependent. Studies have established that RBM20-mediated exon repression occurs through binding of the protein to intron sequences downstream of exons subject to alternative splicing, although the precise mechanism has not been determined.⁴¹

Beyond splicing, RBM20, along with the Ttn pre-mRNA, has been shown to increase the interchromosomal association of multiple RBM20-regulated genetic loci, forming splicing factories.⁴⁴ Additionally, recent studies uncovered marked changes in alternative polyadenylation in RBM20 KO and pathogenic variant KI hiPSC-CMs,¹⁸ suggesting that RBM20 may also play a role in regulating this process. These results suggest that RBM20 has functions beyond splicing in the nucleus and additional studies to not only confirm these functions, but also establish the mechanistic bases, are needed.

RBM20 CARDIOMYOPATHY: COMPLEX AND VARIABLE CLINICAL COURSE WITH POSSIBLE SEXUAL DIMORPHISM

The first cases of RBM20-associated DCM were reported in 2009 in two families with an aggressive autosomal dominant DCM.¹⁹ Both families had a strong history of sudden death with multiple members receiving an implantable cardiac defibrillator (ICD), as well as many members developing progressive heart failure requiring heart transplantation.¹⁹ Genetic linkage analysis with DNA sequencing revealed that both of these families carried pathogenic variants in the RBM20 gene.¹⁹ Screening of additional patients identified a further three pathogenic variants in six other cases of familial DCM.¹⁹ This report established RBM20 as a DCM gene and described several of characteristic features of RBM20 cardiomyopathy that make it a particularly malignant disease, including high penetrance and increased risk for heart failure and/or sudden death.

Additional insight from genotype-phenotype association analysis revealed that patients with pathogenic variants in RBM20 require heart transplantation at a markedly younger age than those with pathogenic variants in other DCM genes.²³ Analysis of a registry of patients with pathogenic variants in RBM20 uncovered increased risk for composite arrhythmias, including atrial fibrillation, non-sustained ventricular tachycardia, ICD discharge, and sudden cardiac arrest in carriers of pathogenic variants.⁴⁵ Atrial fibrillation was also found to be more common in individuals with pathogenic variants in RBM20 in the overall Genetic Risk Assessment of Defibrillator Events (GRADE) study cohort, as well as specifically in subjects with RBM20-associated DCM within this cohort.⁴⁶ The risk associated with pathogenic variants in RBM20 is highlighted by the recently updated ESC guidelines for the management of patients with ventricular arrhythmias and prevention of sudden cardiac death which now recommends considering ICD implantation in DCM patients with left ventricular ejection fraction <50% and greater than 2 risk factors (syncope, late gadolinium enhancement on cardiovascular magnetic resonance imaging, inducible sustained monomorphic ventricular tachycardia upon programmed electrical stimulation, and pathogenic variants in certain genes including RBM20).⁴⁷

An interesting question is whether there is sexual dimorphism in the presentation of RBM20 cardiomyopathy. A study of 111 patients with DCM living in Denmark suggested that the presentation of RBM20 cardiomyopathy is sexually dimorphic with males carrying pathogenic variants developing more severe disease than female carriers.⁴⁸ However, the findings of a recently published article from Lennermann et al. suggest that the issue of sexual dimorphism requires further investigation.¹³ Specifically, comparison of male and female RBM20 KO mice failed to identify sex-specific differences in baseline cardiac function, susceptibility to lethal arrhythmia, or response to acute catecholamine-induced cardiac stress.¹³ Intriguingly, while differences in organ-level function were not detected, the authors did uncover sex-specific differences at the molecular level.¹³ These differences included greater dysregulation of transcript levels and splicing, as well as increased phosphorylation of sarcomeric and cytoskeletal proteins, in male versus female Rbm20 KO mice.¹³ Moreover, the authors did not detect differences in baseline cardiac function in male versus female R636Q (analogous to R634Q in humans) KI mice; or disease severity or outcome in a cohort of patients carrying pathogenic variants in RBM20.¹³ Based on these findings the authors conclude that sex-specific differences in disease severity may not be a common feature of RBM20 cardiomyopathy.¹³ Although this study does provide some counter evidence for sex-specific differences in the presentation of RBM20 cardiomyopathy, it is worth noting that findings in rodents do not necessarily translate to humans, due to species differences. Moreover, the author’s analyses focus predominantly on sex differences in Rbm20 KO mice.¹³ At this point it is well-established that Rbm20 KO mice do not fully recapitulate the phenotype of humans and animals harboring pathogenic variants in RBM20 and, thus, whether the presentation and severity of RBM20 cardiomyopathy are sexually dimorphic, or if this was a specific feature of the Denmark cohort, remains an open question.

In recent years, a number of RBM20 variants have been reported in individuals with apparent non-classical disease presentations, including left ventricular noncompaction cardiomyopathy (LVNC),^49–52 arrhythmogenic right ventricular cardiomyopathy/dysplasia,⁵³ hypertrophic cardiomyopathy (HCM),^54,55 and in a patient with ventricular arrhythmias and a structurally normal heart.⁵⁶ The American Heart Association classified LVNC as a distinct primary genetic cardiomyopathy in 2006.⁵⁷ Nevertheless, there remains debate whether LVNC is simply an alternative manifestation of other cardiomyopathies, such DCM or HCM. Features such as right ventricular involvement or hypertrophy could occur normally during the progression of RBM20-associated DCM or in a subset of individuals. For example, the majority of juvenile Rbm20 R636S KI (Rbm20^R636S) pigs showed evidence of four chamber dilation and severe left ventricular hypertrophy with left ventricular wall thicknesses averaging around twice that of WT hearts.¹² Conversely, a subset of severely affected homozygous Rbm20^R636S pigs showed chamber dilation with dramatic thinning of the myocardial walls.¹² We have also observed this phenomenon sporadically in NLS variant KI and RS domain deletion mice (unpublished observation), although the cause is unknown. Differences in disease severity/presentation have even been reported in patients from the same family carrying the same pathogenic variant.⁵⁸ Thus, it seems likely that we have yet to fully appreciate the true clinical complexity of this disease. Additional studies with rigorous attention to clinical presentation and progression will be necessary to better establish the clinical features of RBM20 cardiomyopathy and determine why some individuals develop more severe disease than others.

VARIANTS IN THE RBM20 NLS ARE CAUSATIVE IN AGGRESSIVE DCM

The original report by Brauch et al. identified six pathogenic variants in RBM20 all located within the RS domain,¹⁹ in what, based on our recent work,³⁰ we now know to be the core NLS. Since this initial study, many additional RBM20 variants have been reported both in and outside the RS domain in association with disease,^{13,20,26,46,48,49,52,53,55,56,59–88} and the number continues to increase. Although variants have now been identified throughout the RBM20 protein, the windows c.1881–1920 (encoding the NLS in exon 9) and c.2721–2760 (encoding a portion of the glutamate-rich region in exon 11) represent the two currently accepted hotspots for DCM-associated variants (Figure 1B & 1C).⁴⁵ Indeed, most RBM20 variants confirmed or strongly suspected of being pathogenic are localized within these windows (Table 1). The exceptions are those in the latter portion of exon 9 (e.g., R716Q) (Table 1). Variants in this part of exon 9 were found to be common in the general population and, thus, this region of the protein is not considered a pathogenic window.⁴⁵ Regardless, the identification of variants in this region in familial DCM suggests that certain variants may still be pathogenic despite greater variant tolerance in this region.

Table 1.

Strongly suspected/confirmed pathogenic RBM20 variants reported in the literature, as well as variant previously suspected to be pathogenic (I536T) that was recently found not to be so.

Domain	Nucleotide change	Amino acid change	Exon	Reported phenotypes	Evolutionary conserved residue	Familial	References	Model(s) available
RRM	c.1607T>C	p.I536T	6	SD without CM	Yes	Yes	76	Mouse89
RS	c.1898C>T	p.P633L	9	DCM	No	Yes	13,15,72,78	Mouse27; hiPSC-CM15,27
RS	c.1901G>T	p.R634L	9	DCM/LVNC	Yes	Yes	13,52	No
RS	c.1901G>A	p.R634Q	9	DCM	Yes	Yes	19,48,59,73,78	Mouse13,14,27; hiPSC-CM14,15,27,90
RS	c.1900C>T	p.R634W	9	DCM	Yes	Yes	13,25,55,58,59,68,69,74	hiPSC-CM91
RS	c.1903T>G	p.S635A	9	DCM	Yes	Unknown	20	Mouse10,11; hiPSC-CM16
RS	c.1904C>G	p.S635C	9	DCM	Yes	Yes	67	No
RS	c.1906C>T	P.R636C	9	DCM	Yes	Yes	59,74,77,78,92	No
RS	c.1907G>A	p.R636H	9	DCM/LVNC/HCM	Yes	Yes	19,48,50,54,59,60,62,74,75,93–95	No
RS	c.1906C>A	p.R636S	9	DCM	Yes	Yes	19,48	Pig12; hiPSC-CM14,17,18
RS	c.1909A>G	p.S637G	9	DCM/LVNC	Yes	Yes	19,50,96	Mouse9
RS	c.1912C>G	p.P638A	9	HCM	Yes	Yes	49	No
RS	c.1913C>T	p.P638L	9	DCM/LVNC	Yes	Yes	13,19,26,46,48,55,60,67,78	No
RS	c.1912C>T	p.P638S	9	DCM	Yes	Unknown	81	No
	c.2017C>T	p.R673W	9	DCM	No	Yes	60,62	No
	c.2062C>T	p.R688X	9	DCM	---	Yes	63,66,75,84	No
	c.2147G>A	p.R716Q	9	DCM	No	Yes	55,59,60,68,69,84	No
	c.2282G>A	p.R761Q	9	DCM	Yes*	Yes	53	No
E-rich	c.2714T>A	p.M905K	11	DCM	Yes	Yes	80	No
E-rich	c.2723T>C	p.L908P	11	DCM	Yes	Yes	74	No
E-rich	c.2737G>A	p.E913K	11	DCM	Yes	Yes	42,48,60,74	No
E-rich	c.2741T>C	p.V914A	11	DCM	Yes	Yes	26	No

^*positively charged residue is conserved at the position. Familial refers to identification in association with familial DCM. RRM, RNA recognition motif; RS, arginine-serine-rich; E-rich, glutamate-rich; DCM, dilated cardiomyopathy; LVNC, left ventricular non-compaction cardiomyopathy; HCM, hypertrophic cardiomyopathy; SD, sudden death; CM, cardiomyopathy; hiPSC-CM, human induced pluripotent stem cell-derived cardiomyocyte.

Recent studies from our lab and others have confirmed that NLS-disrupting variants in exon 9 are causative in RBM20 cardiomyopathy using in vivo and in vitro model systems (Table 1).^{9,10,12–18,25,27} For example, our lab generated mice harboring the S639G variant (analogous to S637G in humans) and showed that these mice develop a severe DCM with early onset heart failure that phenocopies disease in human patients with the same variant.⁹ We reported similar findings for mice carrying the S637A variant (analogous to S635A in humans).¹⁰ Analysis of other variants in the NLS, including P633L, R634Q, and R636S (number based on the human sequence), in animal models and hiPSC-CMs,^{12–18,25,27} have provided further support for the pathogenesis of variants in this portion of the protein.

The c.1601–1640 window in exon 7, which encodes a portion of the RRM (Figure 1), was the third most confidently ranked pathogenic window in RBM20.⁴⁵ Several lines of evidence now indicate that variants in this domain may not be sufficient to cause disease. Yamamoto et al. recently generated mice harboring the I538T (analogous to I536T in humans) variant (Rbm20^I538T) in the RRM.⁸⁹ This variant was previously identified in an apparent familial case of sudden cardiac death,⁷⁶ however, Rbm20^I538T mice did not develop cardiomyopathy or exhibit sudden death until late in life.⁸⁹ We have also previously demonstrated that Rbm20^ΔRRM mice display splicing defects in major RBM20 target genes, but do not show evidence of pathological remodeling.²⁴ Moreover, systolic dysfunction was only detected in homozygous Rbm20^ΔRRM mice,²⁴ whereas the majority of patients with DCM-associated variants in RBM20 carry the variants in a heterozygous state. These findings suggest that loss-of-function variants in the RNA-binding domain in RBM20 alone are not sufficient to cause cardiomyopathy, at least in mice. Whether this is also the case in humans remains an open question.

Importantly, even though there is a lack of strong evidence (such as co-segregation in familial DCM) to support a role for variants outside the exon 9 and 11 hotspots in the pathogenesis of RBM20-associated DCM, this is not to say that these variants are definitively non-pathogenic. Instead, additional studies and animal models will be necessary to establish the pathogenicity and disease mechanism(s) associated with these variants.

PREVALENCE OF RBM20 CARDIOMYOPATHY

Pathogenic variants in RBM20 have been implicated in between 2–6% of DCM cases based on population-based data published in earlier reports.^{19,46,59,61,96} Here, we sought to provide an updated estimate of the prevalence of RBM20 cardiomyopathy. An important caveat to consider when attempting to determine the prevalence of this disease is that many of the identified variants are unlikely to be pathogenic. There are currently over 900 variants in RBM20 reported in the ClinVar database and the pathogenicity of most of these variants has not been established. Thus, we took advantage of the fact that pathogenic variants in RBM20 are often reported in association with familial DCM due to autosomal dominant inheritance and high penetrance.^{19,49,52,59,60,63,67,74,92,94–98} While not without drawbacks, assessing the prevalence of RBM20 cardiomyopathy based on the frequency of RBM20 variants in familial DCM cases has the advantage that such variants have an increased likelihood of being pathogenic, particularly in cases where the variant co-segregates with disease. Additionally, we focused on the prevalence of those variants either confirmed or highly suspected of being pathogenic (i.e., those variants listed in Table 1). Fifteen studies from the literature including 607 cases of verified/suspected familial DCM/LVNC screened for variants in RBM20 were considered. Based on the data provided in these studies an RBM20 variant frequency of 4.2% (95% CI 1.8–7.2%) was calculated (Figure 2A). Among these studies, RBM20 variant co-segregation with disease could only be verified in five of the thirteen studies reporting variants (Figure 2B), so it is possible that his calculation overestimates the true frequency. A better understanding of which RBM20 variants are pathogenic will enable more reliable estimation of the true prevalence of RBM20 cardiomyopathy.

Figure 2.

Forest plot of RBM20 variant frequency in familial DCM/LVNC cases. A. An estimated 4.2% of familial DCM/LVNC cases are associated with RBM20 variants. “Events” refers to the number of individuals in each cohort with pathogenic variants in RBM20 (only variants listed in Table 1 were considered pathological variants) and the “total” refers to the total cohort size (i.e., individuals with DCM/LVNC screened in the study). The red boxes denote a point estimate of the study result (i.e., the proportion of individuals in the given cohort carrying pathogenic variants in RBM20) and the lines represent the 95% confidence intervals of the study result, with each end of the line denoting the boundaries of the confidence interval. CI, confidence interval. B. Additional study details including reported variants that were considered in this analysis and whether the variants co-segregated with disease in the affected families.

LOSS OF RBM20 SPLICING ACTIVITY GIVES RISE TO MILD DCM IN RODENTS

Although much work has focused on the regulation of Ttn splicing by RBM20, many additional target genes have been reported by our lab and others.^18,20,21 It is accepted that the DCM-like phenotype in Rbm20 KO rodents results from changes in the splicing of these genes (Figure 3).^20,22 Yet, the cardiomyopathy phenotype in these animals is relatively mild compared to that in humans and animals harboring pathogenic variants in RBM20 (Figure 3). For example, even homozygous Rbm20 KO rats do not exhibit signs of systolic dysfunction and approximately 83% of rats live over 18 months without developing heart failure.²⁰ This stands in stark contrast to the high penetrance and early onset heart failure observed in patients with NLS variants in RBM20.^19,23 In addition, RBM20 splicing deficient rodents, such as Rbm20^ΔRRM and Rbm20^I538T mice, do not show evidence of pathological remodeling despite aberrant splicing of major RBM20 target genes, such as Ttn and Camk2d.^24,89 Thus, the rodent data strongly suggests that loss of RBM20 splicing activity causes only mild DCM and cannot explain the severe disease that develops in variant KI animals. However, it is unclear at present whether this is also the case in humans.

Figure 3.

Graphical summary of current RBM20 models, contributing mechanism(s), and phenotypes. Rbm20 KO rodents develop a mild DCM caused by loss of nuclear RBM20 and mis-splicing of RBM20 target genes, such as Ttn and Ca²⁺-handling genes (e.g. Camk2d, Ryr2, etc.). In contrast to RBM20 KO animals, NLS variant KI and RS domain deletion animals develop a more severe DCM phenotype mimicking that in human patients. In addition to RBM20 target gene mis-splicing, these animals also exhibit abnormal accumulation of RBM20 in the sarcoplasm in RBM20 biomolecular condensates (shown as blue circles in the cell diagram). Surprisingly, despite mis-splicing of major RBM20 target genes (i.e., Ttn, Camk2d, Ldb3), RRM deletion and variant KI models show only systolic dysfunction without evidence of adverse remodeling. Notably, there is no evidence for RBM20 mis-localization in the hearts of these animals. NLS variant KI hiPSC-CMs show evidence of a DCM-like phenotype with impaired contractility and Ca²⁺-handling abnormalities along with sarcoplasmic RBM20 condensates and mis-splicing consistent with that in NLS variant KI animals.

TTN Splicing

Structurally, titin spans half of the sarcomere and connects the Z-disk to the M-line, making it an important structural component of the sarcomere. Titin is also a major determinant of myocardial stiffness due to its elastic properties.⁹⁹ The TTN gene is comprised of 364 exons and is located on the long arm of chromosome 2 in humans.¹⁰⁰ Titin is typically subdivided into four regions: the Z-disk, I-band, A-band, and M-band (Figure 4). The I-band, which contains the elastic region of the protein, can be further broken down into six domains: the proximal immunoglobulin (Ig) repeat domain (exons 29–47), N2B element (exon 49), middle Ig repeat domain (exons 51–101), N2A element (exons 102–108), proline-glutamate-valine-lysine (PEVK) domain (exons 109–224), and distal Ig repeat domain (exons 225–251) (Figure 4).¹⁰¹ As sarcomeres are stretched, the tandem Ig segments extend, followed by extension of the PEVK and N2B-unique segments.^102–104 The extension of these segments, underlies the increase in passive tension with sarcomere stretch.

Figure 4.

RBM20-dependent regulation of TTN alternative splicing. The TTN gene consists of 363 exons. Alternative splicing occurring in exons encoding the Z-disk (1–28), A-band (252–357), and M-band (358–363) does not significantly alter the size of the titin protein. Exons encoding the I-band region, which consists of the proximal Ig, N2B, middle Ig, N2A, PEVK, and distal Ig segments, are subject to significant alternative splicing under the control of RBM20 that alters titin size to such an extent that it is detectable by gel electrophoresis. At normal levels of RBM20, the N2B pathway is favored such that the ratio of N2B-to-N2BA isoforms in the adult heart is approximately 3-to-7. When RBM20 levels are reduced, N2BA splicing pathways are increased thereby decreasing the ratio of N2B-to-N2BA isoforms in the adult heart. When RBM20 is completely absent as in Rbm20 KO rodents, exon skipping in the middle Ig and PEVK regions does not occur leading to inclusion of these exons and expression of a giant titin isoform (N2BA-G). Adapted from Guo et al.¹⁰⁵ Arrows indicate exons spliced together and solid line connections denote consecutive exons. Arrows with solid and dashed lines indicate validated splicing patterns and putative splicing pathways, respectively.

Exons encoding the middle Ig repeat domain, N2A-unique element, and majority of the PEVK domain are known to undergo extensive alternative splicing.^100,105 In particular, extensive exon skipping occurs for a portion of the exons encoding the middle Ig repeat domain (exons 50–96) and PEVK segment (exons 115–225) in the healthy adult heart.⁴³ In this way, exclusion of exons in these regions decreases the number of extensible domains in titin, thereby reducing the ability of this molecule to buffer tension in response to stretch. Soon after our discovery that titin alternative splicing is regulated by RBM20, we demonstrated that RBM20 directly controls these exon skipping events.⁴³ In the presence of RBM20, skipping of exons between 50 and 219 is increased, favoring production of the shorter N2B isoform of titin that lacks the N2A element, as well as the majority of the middle Ig and PEVK domains (Figure 4). Conversely, when RBM20 levels are reduced, inclusion of these exons is favored and longer N2BA isoforms, which include the N2A element and a greater number of exons encoding the middle Ig repeat and PEVK domains, are expressed (Figure 4).

Given that pathogenic variants in TTN are a well-established cause of DCM,¹⁰⁶ and that RBM20 modulates titin isoform expression,²⁰ it is not surprising that much of the early work on RBM20 cardiomyopathy focused on altered titin isoform expression as a major mechanism of disease. At the molecular level, RBM20 splicing deficiency produced by genetic ablation or loss of nuclear localization induces a characteristic shift in titin isoform expression from the dominant N2B isoform to the longer and more compliant N2BA isoforms.^{9–12,20,22,89} This shift decreases the passive tension of the sarcomere and can, via the effect of passive tension on calcium sensitivity,^24,107 explain impaired systolic function in Rbm20 KO rodents.¹⁰⁸

Ca²⁺-handling Gene Splicing

In addition to titin, RBM20 animal models also exhibit disrupted splicing of Ca²⁺-handling genes.^{9–12,20,22,30,89,109} One such change is exclusion of exon 14 and inclusion of exons 15 and 16 in Camk2d (Figure 5). At the protein level, these changes results in a near complete switch from the CaMKII-δB isoform to the CaMKII-δA and CaMKII-δ9 isoforms.²⁰ Unlike CaMKII-δB, the latter isoforms lack the nuclear localization signal encoded by exon 14 and, thus, are concentrated at the intercalated discs and T-tubules where the Ca²⁺ channels are located rather than in the nucleus.¹¹⁰ Beyond changes in splicing, studies have also shown that the expression of CAMKII is increased at the intercalated discs in Rbm20 KO cardiomyocytes.²² Given that CAMKII can augment the L-type Ca²⁺ current through Ca²⁺-dependent facilitation,¹¹¹ this change has been proposed as the reason for Ca²⁺ overload and increased arrhythmia risk in Rbm20 KO animals.²² Beyond changes in Ca²⁺-handling associated with altered Camk2d splicing, evidence in the literature suggests that increased expression of CaMKII-δ9 in the heart leads to cardiomyopathy and heart failure through degradation of ubiquitin-conjugating enzyme E2T (UBE2T), accumulation of DNA damage, and genome instability.¹¹² Taken together, these findings suggest that splicing changes in Camk2d are likely involved to some extent in cardiomyopathy and/or sudden death in individuals with pathogenic variants in RBM20. Yet, it must be mentioned that we detected similar changes in Camk2d splicing in the hearts of Rbm20^ΔRRM (Table 2), which exhibit impaired systolic function but do not show evidence of Ca²⁺-mishandling and do not develop cardiomyopathy or exhibit sudden death.²⁴

Figure 5.

RBM20-dependent regulation of CAMK2D alternative splicing. The CAMK2D gene consists of 22 exons. Exons encoding the kinase domain (1–10), regulatory segment (11–12), and hub domain (20–22) are constitutively expressed. Three exons encoding the variable linker (13–19) are subject to alternative splicing (exons 14–16) under the control of RBM20. At normal levels of RBM20, inclusion of exon 14 is favored resulting in CaMKIIδ_B being the dominant isoform expressed in the heart with less of the CaMKIIδ_A, CaMKIIδ_C, and CaMKIIδ₉ isoforms. Conversely, when RBM20 levels are decreased or the protein is absent, exclusion of exon 14 is favored resulting in a shift to CaMKIIδ_A, which contains the NLS, as the dominant isoform expressed in the heart. Expression of the CaMKIIδ₉ isoform is also increased in the hearts of Rbm20 KO animals with less of the CaMKIIδ_B and CaMKIIδ_C isoforms.

Table 2.

List of RBM20 target genes modified from Rexiati et al.¹¹³ with encoded protein and detection in different model systems through RNA-seq analysis indicated.

Gene	Protein	Reference(s)	Splicing changes detected in individual models
			KO20	ΔRRM24	ΔRS30	S639G9
APTX	Aprataxin	20	Yes
CACNA1C	Calcium voltage-gated channel subunit alpha 1C	20	Yes		Yes
CAMK2D	Calcium/calmodulin dependent protein kinase II delta	20,21	Yes	Yes	Yes	Yes
CAMK2G	Calcium/calmodulin dependent protein kinase II gamma	20	Yes
DAB1	DAB adaptor protein 1	20	Yes
DNM3	Dynamin 3	20	Yes
DST	Dystonin	21	Yes
DTNA	Dystrobrevin alpha	20	Yes			Yes
ENAH	ENAH actin regulator	21	Yes			Yes
FHOD3	Formin homology 2 domain containing 3	20	Yes
FNBP1	Formin Binding Protein 1	20	Yes
GIT2	GIT ArfGAP 2	20	Yes
IMMT	Inner membrane mitochondrial protein	21	Yes		Yes
KALRN	Kalirin RhoGEF kinase	20	Yes
KCNIP2	Potassium voltage-gated channel interacting protein 2	20	Yes		Yes
LDB3	LIM domain binding 3	20,21	Yes	Yes	Yes
LMO7	LIM domain 7	21	Yes
LRRFIP1	LRR binding FLII interacting protein 1	21	Yes		Yes
MECP2	Methyl-CpG binding protein 2	20	Yes
MLIP	Muscular LMNA interacting protein	21	Yes
MTMR1	Myotubularin related protein 1	20	Yes
MYH7	Myosin heavy chain 7	21	Yes		Yes
MYOM1	Myomesin 1	21	Yes
NEXN	Nexilin F-actin binding protein	21	Yes
NFIA	Nuclear factor I A	20	Yes
NPRL3	NPR3 like, GATOR1 complex subunit	20	Yes
NTRK3	Neurotrophic receptor tyrosine kinase 3	20	Yes
OBSCN	Obscurin	21	Yes
PDLIM3	PDZ and LIM domain 3	21	Yes
PDLIM5	PDZ and LIM domain 5	20	Yes		Yes	Yes
PLEKHA5	Pleckstrin homology domain containing A5	20	Yes
RALGPS1	Ral GEF with PH domain and SH3 binding motif 1	20	Yes
RTN4	Reticulon 4	21	Yes
RYR2	Ryanodine receptor 2	21	Yes		Yes	Yes
SEMA6D	Semaphorin 6D	20	Yes
SH3KBP1	SH3 domain containing kinase binding protein 1	20	Yes
SLC38A10	Solute carrier family 38 member 10	20	Yes
SORBS1	Sorbin and SH3 domain containing 1	20,21	Yes
SPEN	Spen family transcriptional repressor	20	Yes
TNNT2	Troponin T2, cardiac type	21	Yes		Yes	Yes
TPM1	Tropomyosin 1	20	Yes
TRDN	Triadin	20,21	Yes		Yes
TTN	Titin	20,21	Yes	Yes	Yes	Yes
UBE2F	Ubiquitin conjugating enzyme E2 F	20	Yes
ZNF451	Zinc finger protein 451	20	Yes

KO, knockout; ΔRRM, RNA recognition motif deletion; ΔRS, arginine-serine-rich domain deletion.

Another splicing change that was observed in Rbm20 KO rodents and several pathogenic variant KI models,^9–11,20,89 is inclusion of a 24-bp exon in Ryr2 that targets the protein to the nucleus instead of the SR.¹¹⁴ Decreasing the proportion of RyR2 targeted to the SR has the potential to interfere with cellular Ca²⁺ homeostasis. In line with this, we found that treatment of isolated cardiomyocytes from Rbm20 KO rodents with the RYR2 stabilizer S107 is sufficient to improve contractile function or Ca²⁺-handling in cardiomyocytes isolated from these animals.¹¹⁵ Prior studies have shown that S107 inhibits SR Ca²⁺ leak by inhibiting dissociation of Calstabin 2 (also known as FKBP12.6) from RyR2.¹¹⁶ Unfortunately, little is known about the RyR2 complex in the nucleus and, thus, it is unclear whether improved Ca²⁺-handling upon S107 treatment in isolated Rbm20 KO cardiomyocytes is due to action on the remaining channels in the SR or those in the nuclear membrane. It is also important to note that administration of S107 did not rescue cardiac function in 12-month-old Rbm20 KO rats, which we speculate is the result of multiple pathological mechanisms contributing to disease in these animals. Consequently, it seems that splicing changes in Ryr2 may account for only a portion of the Ca²⁺-handling defects in Rbm20 KO and pathogenic variant KI models.

Other Ca²⁺-handling genes exhibiting altered splicing in certain RBM20 models including Cacna1c, Kcnip2, and Trdn.^20,21 Specifically, Rbm20 KO in rodents results in inclusion of exon 9* in Cacna1c, which leads to hyperpolarization of L-type Ca²⁺ channels,¹¹⁷ and may contribute to the detected increase in L-type Ca²⁺ current in Rbm20 KO mouse cardiomyocytes. Indeed, studies have shown that treatment with the Ca²⁺ channel blocker verapamil is sufficient to improve Ca²⁺-handling in cardiomyocytes isolated from Rbm20 KO mice.²²

Intriguingly, changes in the splicing of Ryr2 and Cacna1c are absent in Rbm20^ΔRRM mice (Table 2).²⁴ The absence of splice changes in Ryr2 and Cacna1c in Rbm20^ΔRRM mice is notable as no differences in Ca²⁺-handling were detected in isolated cardiomyocytes from these mice despite characteristic changes in the splicing of Camk2d.²⁴ Based on this evidence two conclusions can be reached. The first is that aberrant Camk2d splicing alone is unlikely to explain severe disease and high mortality in patients and animals with NLS variants in RBM20. The second is that, even in Rbm20 KO animals, changes in Camk2d splicing and isoform expression are unlikely to be the sole cause of Ca²⁺-mishandling with changes in other genes, such as Ryr2 and Cacna1c, likely being important for this phenotype.

Given that changes in the splicing of Ca²⁺-handling genes are an accepted cause of sudden death in Rbm20 KO rodents, it seems likely that these splicing changes also contribute to premature mortality in NLS variant KI animals. However, since mortality is relatively low in Rbm20 KO rodents, these changes would only be expected to account for a small proportion of the premature mortality cases in RBM20 NLS variant KI animals.

Other Target Genes

Aside from Ttn, Camk2d, Ryr2, and Cacna1c, 41 additional genes are considered RBM20 target genes based on altered splicing in Rbm20-deficient animal models or identification in cross-linking and immunoprecipitation (CLIP) sequencing experiments (Table 2).^20,21 In general, splicing changes in genes besides Ttn and Camk2d are relatively minor by comparison, and the functional significance of many of these changes remains to be investigated. Moreover, although splicing changes in Ttn and Camk2d have been universally detected across RBM20 models, this is not the case for many other RBM20 target genes (Table 2). The reason for inconsistencies in splicing changes in Rbm20 KO and NLS variant KI animals was previously unknown but emerging evidence may provide some insight. In an elegant study conducted by Fenix et al., the authors employed enhanced CLIP sequencing to probe splicing changes in RBM20 KO and NLS variant KI versus WT hiPSC-CMs.¹⁸ Surprisingly, they not only identified novel RBM20 target genes, as well as confirmed previous targets, but also uncovered splicing changes that were unique in R636S KI versus KO hiPSC-CMs.¹⁸ This finding is important as it challenges the traditional view that NLS variants in RBM20 phenocopy RBM20 genetic ablation with respect to splicing. Undoubtedly, the use of new techniques such as single-molecule, full-length transcript isoform sequencing, which enables the detection and quantification of transcript isoforms on a genome-wide scale without prior annotation, will aid our understanding of how aberrant exon splicing induced by genetic ablation or pathogenic variants in RBM20 affects protein isoform expression.⁹⁰ Additional studies focused on not only defining the mechanistic basis for splicing differences in RBM20 KO and NLS KI animals, but also assessing the contribution of these differences to the RBM20 cardiomyopathy phenotype, are urgently needed.

RBM20 MIS-LOCALIZATION PRECIPITATES AGGRESSIVE DCM

In 2020, the RBM20 field took a giant leap forward. With the establishment of the first Rbm20 variant KI animal models it was unambiguously demonstrated that NLS variants in RBM20 are causative in severe DCM (Figure 3).^11,12 The discovery that these variants promote nuclear exclusion and sarcoplasmic accumulation in biomolecular condensates not only confirmed prior in vitro data,²⁵ but also provided a novel putative disease paradigm in RBM20-associated DCM based on the formation of pathological biomolecular condensates (Figure 3). The formation of sarcoplasmic RBM20 condensates secondary to NLS variants has now been validated in human patient tissue,²⁶ as well as a growing list of additional models including gene-edited animals and hiPSC-CMs;^{9,10,13–18,27,28} yet many questions remain to be answered chief among which is to what extent these condensates contribute to the pathogenesis of RBM20 cardiomyopathy.

NLS, But Not RRM, Variant KI Animals Phenocopy RBM20 Cardiomyopathy

To date, the majority of RBM20 variants that have been engineered in laboratory animals have been in the NLS (Table 1). In total, five different NLS variants have been knocked into animals, including mice and pigs (Figure 3). Our lab generated S637A and S639G (analogous to the S635A and S637G variants in humans, respectively) KI mice (Rbm20^S637A and Rbm20^S639G).^9,10 Rbm20^S637A mice were also generated previously by the Kuroyanagi group.¹¹ Rbm20^R636S pigs were developed by the Schneider group.¹² Three groups have independently generated Rbm20^R636Q mice (analogous to R634Q in humans).^13,14,27 More recently, Rbm20^P635L mice (analogous to P635L in humans) have been generated.^27,28

In our own studies, analysis of Rbm20^S637A mice revealed marked systolic dysfunction and dilation of the left ventricle in both heterozygous and homozygous variant carriers by 8-weeks-of-age, consistent with the development of a DCM-like phenotype in these mice.¹⁰ Homozygous Rbm20^S637A mice also exhibited premature mortality with approximately 34% of mice dying within 100 days of birth, which mimics the early onset and high mortality associated with pathogenic variants in RBM20 in humans. Our analysis of Rbm20^S639G mice yielded similar findings with the development of severe DCM in homozygous, as well as heterozygous variant carriers.⁹ Surprisingly, survival analysis indicated that Rbm20^S639G mice develop even more severe disease than that in Rbm20^S637A mice as evidenced by an increase in the percentage of mice that die within the first 100 days after birth (48% in Rbm20^S639G versus 34% in Rbm20^S637A mice). This finding is intriguing and suggests that subtle differences exist in the pathogenicity of different NLS variants although additional studies will be necessary to validate this finding and elucidate the molecular mechanisms.

Similar to our Rbm20^S637A mice, the homozygous and heterozygous Rbm20^S637A mice generated by the Kuroyanagi group also develop severe cardiac dysfunction consistent with DCM.¹¹ Analysis of Rbm20^S637A mice at a slightly later timepoint than in our analysis (12–16-weeks-of-age in their study versus 8-weeks-of-age in ours) uncovered greater cardiac dysfunction and remodeling in homozygous versus heterozygous variant carriers.¹¹ Additionally, ECG analysis revealed increased incidence of atrial fibrillation in Rbm20^S637A mice,¹¹ which is common in patients with RBM20 cardiomyopathy.^45,46 Importantly, the authors also compared their Rbm20^S637A mice with Rbm20 KO mice that were generated through targeted deletion of a 26-nucleotide stretch encoding a portion of exon 9.¹¹ This comparison confirmed the development of a relatively mild DCM phenotype in homozygous Rbm20 KO mice and more severe disease in NLS variant KI animals.¹¹

Lennermann et al. generated Rbm20^R636Q mice.¹³ Although comprehensive characterization of these mice was not performed in their study, the authors show data indicating that both homozygous and heterozygous KI mice develop cardiac dysfunction by 14–16-weeks-of-age similar to other NLS variant KI rodent models.^9–11,14 Nishiyama et al. independently generated Rbm20^R636Q mice and showed that these mice, similar to other NLS variant KI models, develop a DCM-like phenotype with impaired systolic function and left ventricular dilation that is more severe in homozygous versus heterozygous animals.¹⁴ Evidence of atrial dilation was apparent, particularly in homozygous Rbm20^R636Q animals, which we also reported previously in Rbm20^S637A and Rbm20^S639G mice.^9,10 Recently, Rbm20^R636Q mice, as well as Rbm20^P635L mice, were also generated and characterized by the Steinmetz group with similar findings of a DCM-like phenotype with cardiac dysfunction and premature mortality.²⁷ Interestingly, comparison of heterozygous Rbm20^P635L and Rbm20^R636Q mice revealed greater cardiac dysfunction in the latter, which correlated with RBM20 nuclear exclusion and sarcoplasmic granule formation.²⁷ This is noteworthy as it may explain our previous observation that Rbm20^S639G mice develop more severe disease than Rbm20^S637A mice. Specifically, it may be that the S639G variant disrupts RBM20 nuclear localization to a greater extent than the S637A variant and, thus, leads to more severe disease because of increased accumulation of RBM20 in sarcoplasmic granules.

In a monumental effort, Rbm20^R636S pigs were generated by the Schneider lab.¹² These pigs were one of the first variant KI animal models and are the only RBM20 large animal model established to date.¹² Analysis of cardiac function in homozygous pigs confirmed impaired systolic function and chamber dilation in these animals.¹² Moreover, homozygous Rbm20^R636S pigs demonstrated high early mortality with approximately 70% of animals dying within 50 days of birth.¹²

Recently, Rbm20^I538T mice were generated.⁸⁹ Similar to RRM deletion and NLS variant KI mice, these mice exhibit altered splicing of RBM20 target genes, including Ttn, Ldb3/Cypher, Camk2d, and Ryr2.⁸⁹ However, in contrast to NLS variant KI models, Rbm20^I538T mice did not show evidence of systolic dysfunction and did not develop cardiomyopathy.⁸⁹ Moreover, even though this variant was identified in a proband that died in his early 20s from fatal arrhythmia,⁷⁶ neither homo- nor heterozygous Rbm20^I538T mice exhibited sudden death until late in life.⁸⁹ Thus, it can be concluded that this variant does not cause disease despite similar changes in the splicing of major RBM20 target genes to those in Rbm20-deficient and NLS variant KI animals. Further study of other DCM-associated variants in the RRM, such as the V535I variant,⁵⁹ will be necessary to definitively rule out pathogenicity.

KI hiPSC-CMs Corroborate Pathogenicity Of RBM20 NLS Variants In Humans And Provide A Platform For Therapeutic Testing

In addition to RBM20 animal models, RBM20 KO and variant KI hiPSC-CMs have also been generated (Figure 3).^{14–18,27,90,91,118} To date, a total of five different NLS variant hiPSC lines have been derived from patient cells or produced via CRISPR-Cas9 genome editing of commercially available hiPSCs derived from healthy individuals. The first RBM20 hiPSC line was generated by the Nelson lab in 2016 from a patient heterozygous for the R636S variant.^17,118 RBM20^R636S hiPSCs have also been generated through CRISPR-Cas9 genome editing in healthy control iPSCs by two different labs.^14,18 Rebs et al. derived hiPSCs from a DCM patient heterozygous for the R634W variant.⁹¹ RBM20^R634Q hiPSC-CMs were established by two different labs using CRISPR-Cas9 genome editing.^14,15 In addition to generating one of the aforementioned R634Q hiPSC lines, Briganti et al. also generated RBM20^P633L hiPSCs using a similar strategy.¹⁵ Additionally, hiPSCs were derived from a patient heterozygous for the S635A variant.¹⁶

Fenix et al. directly compared RBM20^R636S hiPSC-CMs to a previously generated RBM20 KO hiPSC-CMs.¹⁸ Their comparison revealed that twitch force in three-dimensional engineered heart tissue generated from KO and RBM20^R636S hiPSC-CMs was significantly decreased in NLS variant KI, but not KO, hiPSC-CMs in comparison to WT hiPSC-CMs.¹⁸ Although it is widely acknowledged that the DCM phenotype resulting from NLS variants is more severe than that associated with genetic ablation, this study, along with the study by Ihara et al.,¹¹ is one of the few studies to provide evidence for this discrepancy through direct comparison of the associated disease phenotypes. Moreover, further analysis of RNA binding and processing revealed unique alterations not only in splicing, but also circular RNA production and alternative polyadenylation, in RBM20^R636S versus KO hiPSC-CMs.¹⁸ The consequences of these unique alterations are unclear but may play a part in explaining phenotypic differences between pathogenic variants in RBM20 and KO, and warrant further investigation.

hiPSC-CMs are a valuable model for therapeutic development and testing with direct translational relevance to human disease.¹¹⁹ Several studies have assessed different treatment methods using RBM20 NLS variant KI hiPSC-CMs.^14,118 Wyles et al. showed that treatment with either the β-adrenergic receptor antagonist carvedilol or the Ca²⁺ channel blocker verapamil reduced sarcomere disarray upon β-adrenergic challenge with norepinephrine.¹¹⁸ In addition, the authors found that treatment either reduced (carvedilol) or eliminated (verapamil) arrhythmic beating in RBM20^R636S hiPSC-CMs.¹¹⁸ These results indicate that Ca²⁺ channels may be a therapeutic target in RBM20 cardiomyopathy.

Briganti et al. used CRISPR-Cas9 genome editing to generate hiPSCs with the P633L, R634Q, and S635 frame shift variants in RBM20.¹⁵ The authors demonstrated improved RBM20 target gene splicing, Ca²⁺ upstroke time, and contractility in cardiomyocytes derived from all three RBM20 variant hiPSCs following treatment with trans-retinoic acid.¹⁵ Surprisingly, partial rescue of function appeared to result from increased expression of the RBM20 protein with the pathogenic variants in treated hiPSC-CMs.¹⁵ The reason for rescue of splicing and contractile function upon upregulation of RBM20 with these variants is not immediately forthcoming as this would be expected to increase RBM20 granule burden. Using super-resolution microscopy, Fenix et al. showed that NLS variants do not completely abolish RBM20 nuclear localization.¹⁸ Moreover, it was demonstrated that overexpression of TNPO3, the recently identified RBM20 nuclear import receptor, in hiPSC-CMs and mice with the P633L variant (P635L in mice) improves nuclear localization of the mutant protein and rescues the splicing of RBM20 target transcripts.²⁸ Thus, improved function following RBM20 upregulation may be due to partial rescue of splicing defects by increasing the small pool of mutant RBM20 in the nucleus.

Recent studies from two labs have demonstrated the clinical promise of CRISPR-based editing strategies to correct pathogenic RBM20 variants in vivo.^14,27 Nishiyama et al. utilized adenine base editing (ABE) and prime editing CRISPR-Cas9 approaches to correct NLS variants in RBM20^R634Q and RBM20^R636S hiPSC-CMs, respectively.¹⁴ Using ABE to correct the R634Q variant, the authors achieved 92% efficiency of A-to-G editing, which rescued alternative splicing of RBM20 target genes, restored RBM20 nuclear localization, and eliminated RBM20 granule formation.¹⁴ The authors also developed a second prime editing strategy to correct the RBM20^R636S variant in hiPSCs.¹⁴ This strategy achieved 40% efficiency in A-to-C editing and was sufficient to correct RBM20 nuclear localization in gene-edited cells.¹⁴ Importantly, the authors validated their correction strategy in Rbm20^R636Q mice, which showed improved cardiac function and restored transcript expression following AAV9-mediated delivery of ABE components.¹⁴ Similarly, Grosch et al. optimized base editors and gRNAs to correct pathogenic variants in RBM20^P633L and RBM20^R634Q hiPSC-CMs and then moved on to correct the P635L variant in KI mice.²⁷ Importantly, the authors employed a synthetic variant of AAV9, called AAVMYO, which has improved targeting to cardiomyocytes for their in vivo correction experiments.²⁷ Using this delivery approach, the authors achieved restoration of RBM20 nuclear foci in approximately 75% of cardiomyocytes, rescue of RBM20 target transcript splicing, and cardiac function approaching that in WT mice by 12 weeks following editing all with no apparent off-target editing.²⁷ Collectively, these studies establish CRISPR-Cas9-mediated correction of pathogenic NLS variants in RBM20 as a promising therapeutic strategy to treat RBM20 cardiomyopathy.

Notably, a second paper published by the Steinmetz group not only identified the RBM20 nuclear import receptor (TNPO3) but also showed that overexpression of this protein was sufficient to at least partially rescue nuclear localization of the P633L variant in hiPSC-CMs and mice (P635L in mice).²⁸ The observed restoration of variant RBM20 nuclear localization was associated with improved splicing of RBM20 target genes both in vitro and in vivo.²⁸ This finding is important as it suggests that the variant protein functions similar to WT, at least with respect to splicing, and thus strategies aimed at restoring nuclear localization of variant RBM20 may also be viable for the treatment of RBM20 cardiomyopathy. Nevertheless, additional studies, in particular those assessing not only splicing but also cardiac function as an endpoint, will be necessary to confirm that such strategies are worth pursuing alongside CRISPR-based therapies.

RS Domain Deletion Recapitulates Aspects Of The NLS Variant Phenotype In Mice

Our lab sought to clarify the role of the RS domain in RBM20-associated DCM. Towards this end, we recently generated Rbm20^ΔRS mice via in-frame deletion of a 34 amino acid sequence in exon 9 that encompasses the RS domain sequence, using CRISPR-Cas9 genome editing.³⁰ Similar to NLS variant KI mice, Rbm20^ΔRS mice develop DCM with increased left ventricular dimensions, systolic dysfunction, and changes in gene expression consistent with the development of cardiac stress and failure (Figure 3).³⁰ Importantly, although RBM20 target gene splicing was disrupted in the hearts of Rbm20^ΔRS mice, analysis of RBM20 localization revealed that RBM20 was mis-localized in Rbm20^ΔRS, but not Rbm20^ΔRRM, mouse hearts.³⁰

Although prior evidence from in vitro studies indicated that sequences spanning from the RRM to the RS domain are required for full nuclear localization of RBM20,⁴⁰ this finding suggested to us that the core NLS in RBM20 must be located within the RS domain. In vitro experiments employing sequence deletion plasmids and immunocytochemical staining revealed that the first nine amino acids in the RS domain constitute the core NLS in RBM20.³⁰ This finding is supported by the fact that all of the validated DCM-causing variants in the RS domain are located in this segment (Table 1).^9–14 Further analysis of plasmids containing DCM-associated variants in other domains/segments of RBM20, as well as in the non-NLS RS domain sequence, further confirmed that only variants in the NLS facilitate re-localization and granule formation.³⁰ Collectively, our findings identify the critical NLS in RBM20 and demonstrate that variants in this NLS cause DCM through loss of RS domain-mediated nuclear localization.

Our findings also provide insight into the domains of RBM20 responsible for sarcoplasmic granule formation. Prior studies mapped the sequence required for granule assemble to residues 130–658 in human RBM20,¹² which includes ZnF1, the RRM, and majority of the RS domain (Figure 1). The fact that perinuclear RBM20 condensates, similar to those reported in the hearts of animals with NLS variants in RBM20,^9–12,14 were detected in the hearts of Rbm20^ΔRS mice, indicates that the RS domain of RBM20 is not the primary catalyst for granule formation.³⁰ Instead, based on these findings, it seems that ZnF1, the RRM, or both are likely involved in this function. Involvement of the RNA-binding domain (RRM), in particular, seems likely as studies have shown that RNA plays an integral role in the formation of biomolecular condensates.¹²⁰

While the RS domain is not required for the formation of sarcoplasmic RBM20 condensates, this domain may still play an important role in determining disease severity by influencing the composition of these condensates. Indeed, the RS domain in SR proteins is known to mediate a host of protein-protein interactions.³³ Our current hypothesis is that residual RS domain function in the context of NLS variants enables the recruitment of a specific complement of proteins and/or RNAs to sarcoplasmic RBM20 condensates, leading to a more severe DCM phenotype. In support of this hypothesis, premature mortality was delayed in Rbm20^ΔRS mice compared to that in mice harboring pathogenic NLS variants even though sarcoplasmic granules were present.³⁰ Regardless, additional studies will be necessary to determine the specific sequences/domains in RBM20, as well as the role of the RS domain, in the formation and pathogenicity of sarcoplasmic RBM20 condensates.

Summary

The last three years have seen the establishment of the first RBM20 variant KI animal models.^9–14,27,89 In particular, NLS variant KI animal models have unequivocally linked these variants to aggressive DCM and, thus, proved their pathogenicity (Figure 3). These animal models, as well as hiPSC-CM models, offer an unprecedented opportunity to not only identify disease mechanisms but also test new therapeutic strategies with translational relevance to human patients. On the other hand, recently described RRM variant KI mice do not develop a cardiac phenotype—a finding that, in combination with data from RRM deletion mice, suggests that disrupted splicing of RBM20 target transcripts alone does not cause overt disease, at least in rodents. Investigation of additional disease-associated variants in the RRM will be necessary to confirm these findings. Moreover, whether this is also the case in humans is uncertain. Finally, disease-associated variant KI models in other regions of the RBM20 protein (e.g., variants encoded by the latter portion of exon 9 and those in the glutamate-rich region) are urgently needed to confirm the pathogenicity of these variants and elucidate the mechanism(s) of disease.

On the topic of translational potential, the majority of studies describing NLS variant KI animal models thus far have focused on homozygous variant carriers due to the fast progression of disease in these animals.^9–14 Yet, it must be kept in mind that the majority of human patients carry RBM20 variants in a heterozygous state. It has been shown that in animals and hiPSC-CMs heterozygous for NLS variants in RBM20, the protein is distributed in both the nucleus and cytoplasm.^{9,10,12,14,18} Consequently, in addition to reduced RBM20 levels in the nucleus, RBM20 condensates will also be present in the sarcoplasm. This mix of nuclear and sarcoplasmic RBM20 is expected to have unique consequences with regards to the phenotype. Indeed, emerging evidence supports dosage-dependent effects in the splicing of a subset of RBM20 target genes in hiPSC-CMs homo- and heterozygous for NLS variants in RBM20.¹⁸ Thus, longer-term studies using heterozygous models should provide the greatest disease relevance and translational potential.

Study of NLS variant KI models have brought to light a novel putative mechanism of pathogenicity in RBM20 cardiomyopathy; sarcoplasmic RBM20 condensates. The fact that NLS variant KI animals appear to phenocopy patients with RBM20 cardiomyopathy while RRM deletion or variant mice do not, suggests that RBM20 mis-localization is the driving pathological force in RBM20-associated DCM. Based on rodent data, it seems clear that the mis-splicing of RBM20 target genes alone is insufficient to cause severe disease. Instead, contributions from sarcoplasmic granules or splicing alterations unique in NLS variant KI models may be very important and warrant further investigation as do variant-associated alterations in emerging non-splicing functions of RBM20, such as circular RNA production and alternative polyadenylation. A better understanding of whether sarcoplasmic granules and/or the disruption of non-splicing functions of RBM20 also contribute to human disease will be imperative.

Sarcoplasmic RBM20 biomolecular condensates have gained attention as a novel putative pathogenic mechanism in RBM20 cardiomyopathy. Evidence supporting the pathogenicity of RBM20 condensates is that firstly, differences in the phenotypes of Rbm20 KO animals and patients/animals with RBM20 cardiomyopathy are indicative of a gain-of-function mechanism in the latter. The second is that RNA-binding protein mis-localization and cytoplasmic biomolecular condensate formation has been implicated in the development of other diseases. Specifically, cytoplasmic biomolecular condensates formed secondary to variants in the RNA-binding proteins TDP-43 and FUS are hallmark pathogenic features of the neurodegenerative diseases amyotrophic lateral sclerosis and frontotemporal dementia.¹²¹

Even though sarcoplasmic RBM20 condensates are suspected of being pathogenic, it bears mentioning that some data do not support this. Specifically, pathogenic variants in the exon 11 hotspot (Figure 1C) cause RBM20 cardiomyopathy even though they are not expected to affect RBM20 nucleocytoplasmic trafficking. Intriguingly, limited clinical description of individuals with pathogenic variants in exon 11 suggest similar disease presentation to NLS variant carriers with early age at diagnosis, heart failure, and arrhythmia.^26,42,80 Such a finding is unexpected if RBM20 condensates are the primary pathogenic mechanism in RBM20 cardiomyopathy. A possible explanation is that sarcoplasmic condensates are pathogenic and that subtle differences exist in the phenotypes of individuals with NLS and non-NLS variants that remain to be appreciated due to the paucity of clinical data and/or the confounding heterogeneity in disease presentation that exists (even among individual with the same genetic variant). Regardless, additional models to not only validate the pathogenicity of RBM20 condensates, but also confirm the pathogenicity of non-NLS variants and determine the associated disease mechanisms, are urgently needed. Sarcoplasmic RBM20 biomolecular condensates are set to shift the paradigm of pathobiology in RBM20 cardiomyopathy caused by NLS variants; however, to what extent these membrane-less organelles contribute to this disease only time will tell.

Nonstandard Abbreviations and Acronyms:

DCM

dilated cardiomyopathy

RBM20

RNA binding motif protein-20

SR

serine-arginine-rich

KO

knockout

RS

arginine-serine-rich

KI

knock-in

hiPSC

human induced pluripotent stem cell

CMs

cardiomyocytes

RRM

RNA recognition motif

NLS

nuclear localization signal

ZnF

zinc finger

ICD

implantable cardiac defibrillator

LVNC

left ventricular noncompaction cardiomyopathy

HCM

hypertrophic cardiomyopathy

PEVK

proline-glutamate-valine-lysine

UBE2T

ubiquitin-conjugating enzyme E2T

CLIP

cross-linking and immunoprecipitation

ABE

adenine base editing