Myocardial recovery with mechanical circulatory support is linked to alternative splicing and subcellular localization of CAMK2D

Abstract

Background:

Cardiac reverse remodeling occurs in a small subset of heart failure patients treated with guideline-directed therapies. This phenomenon, which is defined by reduced ventricular dilatation and improved systolic function, is most common in patients receiving left ventricular assist device (LVAD) therapy. Identifying therapeutic targets for initiating reverse remodeling is an area of great clinical interest, as these patients experience improved outcomes and quality of life. Targets may be discovered among the unique molecular changes associated with LVAD-induced partial myocardial functional recovery; however, the mechanisms underlying this favorable response are incompletely understood.

Methods:

To identify molecular signatures of recovery, we studied paired pre- and post-LVAD myocardial samples from patients with heart failure who received LVAD as a bridge-to-transplant (10 responders, 9 non-responders) and non-failing controls. We performed bulk RNA-sequencing, tandem-mass-tag (TMT) quantitative proteomics, and TMT quantitative phospho-proteomics with follow-up mechanistic and functional investigations in primary rat cardiomyocytes and human engineered heart tissues (EHTs).

Results:

Alternative RNA splicing was the leading pathway associated with a favorable response to LVAD. Responders had increased RNA splicing factor expression and unique gene splice variant expression compared with non-responders. Alternative splicing of CAMK2D was a particularly strong predictor of recovery where increased exon 14 inclusion, which encodes the nuclear splice variant (CAMK2D-B), inversely correlated with functional recovery. Notably, non-responders also displayed hyperphosphorylation near the nuclear localization signal (NLS) in CAMK2D-B. Investigations in primary cardiomyocytes and subcellular organelle fractions from the human hearts revealed that NLS phosphorylation prevented adrenergic stress-dependent nuclear targeting of CAMK2D-B. Expression of a cytoplasm-restricted CAMK2D-B uniquely remodeled the phospho-proteome of primary rat cardiomyocytes compared with a nuclear-competent version and blunted calcium transients in EHTs.

Conclusions:

This is the first study to integrate transcriptome, alternative transcriptome, proteome, and phospho-proteome analyses of heart samples from LVAD-supported patients to investigate myocardial recovery. We identified that increased expression and phosphorylation of the nuclear CAMK2D splice variant predicted poor outcomes. Further, this phosphorylation restricted CAMK2D-B to the cytosol, leading to impaired cardiomyocyte calcium handling. These findings suggest LVAD non-responder patients may benefit from therapies that modulate subcellular localization of CAMK2D or inhibit its activity.

INTRODUCTION

Heart failure is one of the leading causes of morbidity and mortality in the developed world¹; however, many current therapies treat heart failure symptoms and do not target the underlying molecular causes of the disease^2–4. Thus, while patients with chronic heart failure are living longer than ever before⁵, their quality of life remains poor and the burden of heart failure on the healthcare system is increasing^1,6,7. The only current ‘cure’ for heart failure is cardiac transplant, access to which is limited by a shortage of healthy donor hearts⁸. However, substantial clinical evidence suggests that heart failure may be at least partially reversible, as a subset of patients treated with guideline-directed therapies experience modest recovery from heart failure, which is characterized by reduced left ventricular (LV) dilatation and improved LV systolic function (i.e., reverse remodeling)⁹. Reverse remodeling is associated with dramatic improvements in patient survival and quality of life^9–11. However, despite considerable interest from both clinical and basic science perspectives in identifying therapeutic targets for reverse remodeling and recovery, the molecular mechanisms that regulate this process remain poorly understood.

Left ventricular assist device (LVAD) therapy, which mechanically unloads the LV and assumes control of cardiac circulatory function¹², is the treatment that is most often associated with reverse remodeling and partial functional recovery^9,13–15. LVAD therapy also necessitates the removal of some LV myocardium during device placement, thus generating a pre-treatment sample for future paired comparisons. A recent investigation into the molecular biology of recovery by bulk RNA-sequencing found few pre-LVAD gene expression differences between patients who went on to recover and those who did not¹⁴, indicating a striking transcriptional similarity between response sub-groups before treatment. This finding suggested that molecular features of a favorable response might only manifest after treatment (i.e., they are induced by LVAD). In support of this, single-nucleus RNA-sequencing of pre- and post-LVAD myocardium from responders and non-responders identified cell type-specific recovery signatures post-LVAD, where decreased expression of pro-inflammatory genes in cardiac macrophages and fibroblasts were the most discriminating features¹⁵. However, as is true with inflammation and heart failure¹⁶, whether changes in pro-inflammatory gene expression are a cause or consequence of the partial functional recovery induced by LVAD is not clear. Further, this earlier study found that, despite improved gross cardiac contractility denoted by increased systolic function in responders post-LVAD, cardiomyocytes did not revert to a ‘healthy’ transcriptional state in recovery¹⁵. These findings support the conclusion that gene expression changes alone cannot likely explain the different patient outcomes with LVAD. Further, they suggest that other molecular factors contribute to improved outcomes in the subset of advanced heart failure patients who experience reverse remodeling and partial functional recovery.

METHODS

Complete experimental methods are included in the associated Supplementary Material.

Ethical Oversight

Paired pre-LVAD cores and post-LVAD explanted hearts were obtained with informed consent at the University of Colorado Anschutz Medical Center. Animal experiments were approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Colorado Boulder.

Clinical Definition of Responders and Non-responders

Among 41 available paired pre- and post-LVAD patient samples in the University of Colorado Anschutz Medical Center biobank at the time of the study (Data S1), responder (R) classification (n = 10) was assigned if both of the following criteria were met: 1. The LV ejection fraction (LVEF) increased by an absolute value of >15 (e.g., from 20% to 35%), and 2. The LV end diastolic dimension (LVEDD) decreased by >15% following mechanical circulatory support. Non-responder (NR) classification (n = 9) was assigned if both of the following criteria were met: 1. LV ejection fraction increased by an absolute value <5 or decreased, and 2. LV end diastolic dimension decreased by <15% or increased following mechanical circulatory support. These definitions were specified prior to downstream analyses. Patients whose responses did not meet the criteria for either group (n = 22) were defined as intermediate responders and were not included in the study. The echocardiographic measures of LV structure and function were obtained as part of standard clinical care and reflect the clinical condition of the patient at the time of the study. The pre-LVAD echocardiograms were obtained at the closest time frame prior to LVAD implantation. The post-LVAD echocardiograms were obtained at the time closest to LVAD explant (i.e., at cardiac transplant). The post-LVAD echocardiograms for all patients in the study were obtained with the LVAD settings clinically required for the patient and reflect an unloaded condition.

Statistical Analysis and Data Presentation

Sample size calculations were not performed for deciding on the analysis of patient groups as this was limited to the number of samples available. All available samples that met the pre-defined criteria for responders and non-responders were included in the study. Sample sizes for NRVM and EHT experiments were calculated based on previous experience using these models. NRVMs from at least three independent biological replicates were used in these studies. Groups were blinded to the experimenter for the proteomics, phospho-proteomics, and RNA-seq sample preparation and data collection. Groups were otherwise not blinded for the remainder of the experiments. Comparisons of two groups were performed by paired or unpaired two-tailed t-test, as described in the figure legends. Paired analyses were performed for pre- versus post-LVAD comparisons and unpaired analyses for pre- versus pre-LVAD and post- versus post-LVAD comparisons pertaining to the responder and non-responder groups. Comparisons of more than two groups were performed by one- or two-way ANOVA, as described in the figure legends. Comparisons of categorical data between responders and non-responders pre- and post-LVAD were performed using the Chi-squared test. All data analyses and graphical representations were performed using GraphPad Prism version 10. The data throughout the manuscript are presented as the mean ± standard error, mean ± SD, or median and range as detailed in the figure legends and table footnotes. For the RNA-seq and proteomics data and the corresponding gene ontology enrichment analyses, an adjusted p-value of < 0.05 was considered statistically significant. For all other data analyses, a p-value of < 0.05 was considered statistically significant.

Data and Materials Availability

The raw RNA sequencing data were deposited at NCBI (BioProject ID: PRJNA1312133). The raw proteomics and phospho-proteomics data were deposited in the Proteome Xchange Consortium via the PRIDE¹⁷ partner repository (Dataset IDs: PXD068395, PXD068397, PXD068672). All other data are available in the main text or associated supplementary files. Requests for materials associated with this study should be directed to the corresponding author.

RESULTS

Ventricular transcriptome remodeling with mechanical circulatory support

Among 41 available paired pre- and post-LVAD patient samples (Data S1), we identified 10 responders and 9 non-responders based on LV structural and functional outcomes obtained by echocardiography (see methods) (Fig. 1A–B). It is important to note that the LV structural and functional improvements exhibited by the responder group, while significant, represent partial myocardial recovery that was insufficient to obviate the need for cardiac transplant. All patients had non-ischemic dilated cardiomyopathy and were classified as having New York Heart Association (NYHA) class IV (end stage) heart failure at the time of LVAD placement. Patients in both outcome groups displayed similar age ranges, medication histories, and clinical characteristics before starting therapy (Table 1, Table 2), suggesting that molecular differences likely contributed to their bifurcated functional responses to LVAD therapy. To identify potential transcriptional differences, we performed bulk RNA sequencing on heart failure (pre-LVAD), post-LVAD, and non-failing control samples. We found that mechanical circulatory support partially reversed heart failure-associated gene expression changes to non-failing control levels (Fig. S1), which included reduced pro-inflammatory gene expression (Fig. S1I). However, few genes were significantly differentially expressed between responders and non-responders pre-LVAD, in agreement with previous work¹⁴, or post-LVAD (Fig. S2). Although total gene expression levels were not different between groups, we reasoned that a paired analysis of gene expression changes from pre- to post-LVAD might reveal transcriptional signatures of recovery. We therefore analyzed the RNA-seq data for LVAD treatment effects between responders and non-responders. Gene set enrichment analysis revealed that increased RNA splicing factor gene expression was positively associated with recovery (Fig. 1C), while genes involved in the immune response were negatively associated (Fig. 1D).

Fig. 1. Partial myocardial recovery with mechanical circulatory support is linked to RNA alternative splicing.

A-B. Left ventricular ejection fraction (LVEF) (A) and LV end diastolic dimension (LVEDD) (B) in pre- and post-LVAD in responders (R) and non-responders (NR); n = 10 R, 9 NR; paired two-tailed t-test. C-D. GSEA gene ontology (GO) Biological Process (BP) analysis of bulk RNA-seq data for genes with expression changes pre- to post-LVAD that were positively (C) or negatively (D) associated with a favorable response to LVAD. E. Volcano plot displaying quantitative proteomics results from post-LVAD responder (R) and non-responder (NR) myocardium; n = 9/group. F. Gene ontology (GO) Biological Process pathway enrichment analysis of the significantly upregulated proteins in Responders. G. Heat map displaying the significantly differentially expressed RNA splicing proteins between post-LVAD R and NR. H. Venn diagram displaying the number of local splicing variations identified between HF (n = 19) and control (n = 5) by RNA-seq and the differential alternative splicing responses post-LVAD in R (n = 10) and NR (n = 9) patients. I. TTN exon 242 alternative splicing map and quantification between groups; displayed as median and range. J. CAMK2D exon 16 and 14 alternative splicing maps and quantification between groups; PSI = proportion spliced in; displayed as median and range; n = 5 control, 9 HF-NR and LVAD-NR, 10 HF-R and LVAD-R; *p < 0.05, **p < 0.01, ***p < 0.001 by an independent two-sample Mann-Whitney U-test (Wilcoxon test in MAJIQ).

Table 1.

Patient clinical characteristics.

Characteristic	Non-Failing Donors	HF/LVAD Patients	p-value
Number of patients	6	19	n.a.
Age (years)	51.0 ± 16.9	48.1 ± 13.3	0.660
Sex (% male)	100	84.2	0.299
Race (% white)	83.3	68.4	0.478
LVAD duration (days)	-----	267 ± 253	n.a.
Clinical Data	HF (pre-LVAD)	Post-LVAD	p-value
LVEF (%)	14.1 ± 9.8	27.7 ± 15.6	0.003
LV Fractional Shortening (%)	7.2 ± 4.2	13.8 ± 8.3	0.003
LVESD (cm)	6.4 ± 0.9	5.0 ± 1.6	0.0001
LVEDD (cm)	6.9 ± 1.0	5.8 ± 1.5	0.0003
Serum BNP (pg/mL)*	1993 ± 1214	244 ± 196	0.0001
Medications
ACE inhibitor, n (%)	1 (5.3)	10 (52.6)	0.001
Aldosterone antagonist, n (%)	15 (78.9)	14 (73.7)	0.703
Anti-arrhythmic, n (%)	4 (21.1)	5 (26.3)	0.703
Beta blocker, n (%)	3 (15.8)	10 (52.6)	0.017
Digoxin, n (%)	9 (47.4)	5 (26.3)	0.179
Dobutamine, n (%)	11 (57.9)	1 (5.3)	0.001
Loop diuretic, n (%)	15 (78.9)	6 (31.6)	0.003

Data are presented as the mean ± SD.

All clinical data were analyzed by two-tailed paired t-test.

Categorical data (Sex, Race, Medications) were analyzed by Chi-squared test.

^*Values not available for 4 pre-LVAD and 4 post-LVAD patients

Table 2.

Clinical characteristics of responders and non-responders pre- and post-LVAD.

Characteristic	Responders	Non-responders	p-value
Number of patients	10	9	n.a.
Age (years)	46.3 ± 15.4	50.0 ± 10.9	0.559
Sex (% male)	80.0	88.9	0.596
Race (% white)	90.0	44.4	0.033
Ethnicity (% non-Hispanic)	80.0	100	0.156
LVAD duration (days)	229 ± 268	308 ± 245	0.517
	HF (pre-LVAD)	HF (pre-LVAD)	p-value
LVEF (%)	10.4 ± 8.6	18.1 ± 9.9	0.086
LVFS (%)	6.1 ± 3.0	8.3 ± 5.2	0.272
LVESD (cm)	6.2 ± 0.8	6.6 ± 1.0	0.360
LVEDD (cm)	6.6 ± 0.8	7.1 ± 1.1	0.206
LV Systolic Volume (mL)	176 ± 53	227 ± 121	0.247
LV Diastolic Volume (mL)#	195 ± 53	273 ± 131	0.105
Serum BNP (pg/mL)*	1645 ± 1002	2517 ± 1404	0.182
Medications
ACE inhibitor, n (%)	1 (10.0)	0 (0.0)	0.330
Aldosterone antagonist, n (%)	7 (70.0)	8 (88.9)	0.313
Anti-arrhythmic, n (%)	4 (40.0)	0 (0.0)	0.033
Beta blocker, n (%)	3 (30.0)	0 (0.0)	0.073
Digoxin, n (%)	5 (50.0)	4 (44.4)	0.809
Dobutamine, n (%)	5 (50.0)	6 (66.7)	0.463
Loop diuretic, n (%)	7 (70.0)	8 (88.9)	0.313
	Post-LVAD	Post-LVAD	p-value
LVEF (%)	37.0 ± 14.4	17.4 ± 9.4	0.0029
LVFS (%)	17.2 ± 10.1	10.0 ± 3.1	0.0547
LVESD (cm)	3.9 ± 1.0	6.3 ± 1.0	3.21E-5
LVEDD (cm)	4.7 ± 0.8	7.0 ± 1.0	2.66E-5
LV Systolic Volume (mL)#	50 ± 24	168 ± 81	0.0007
LV Diastolic Volume (mL)#	77 ± 35	202 ± 93	0.0017
Serum BNP (pg/mL)*	164 ± 82	335 ± 252	0.0924
Medications
ACE inhibitor, n (%)	7 (70.0)	3 (33.3)	0.110
Aldosterone antagonist, n (%)	8 (80.0)	6 (66.7)	0.510
Anti-arrhythmic, n (%)	2 (20.0)	3 (33.3)	0.510
Beta blocker, n (%)	6 (60.0)	4 (44.4)	0.498
Digoxin, n (%)	3 (30.0)	2 (22.2)	0.599
Dobutamine, n (%)	1 (10.0)	0 (0.0)	0.305
Loop diuretic, n (%)	2 (20.0)	4 (44.4)	0.252

Data are presented as the mean ± SD.

Categorical data were analyzed by Chi-squared test. All other data were analyzed by two-tailed unpaired t-test.

^#Values not available for 1 pre-LVAD and 1 post-LVAD patient

^*Values not available for 4 pre-LVAD and 4 post-LVAD patients

Partial functional recovery with LVAD is associated with unique RNA alternative splicing changes

The generation of multiple transcripts from a single gene by alternative splicing increases proteome diversity and thus serves as a mechanism to modulate cellular function¹⁸. Over 90% of the ~20,000 protein-coding genes in the human genome undergo alternative splicing, which increases potential proteome complexity by an order of magnitude¹⁹. To examine whether the transcriptional differences in RNA splicing factor expression translated to meaningful changes at the protein level, we performed tandem-mass-tag (TMT) quantitative proteomics on post-LVAD responder and non-responder samples. This approach identified 154 differentially expressed proteins between the groups, including increased abundance of RNA splicing factors in responders (Fig. 1E–G, Fig. S3–4). Given this support for the role of alternative splicing in heart failure development and recovery, we next mined the RNA-seq dataset to detect and quantify local splicing variations (LSVs)²⁰. We identified 1,235 LSVs that were significantly impacted in heart failure versus control samples and 411 that changed with mechanical unloading (Fig. S5–7). Among the heart failure-associated splicing changes, 34% of altered cassette exon events (i.e., exon-skipping) were predicted to affect protein domains involved in cellular signal transduction (Fig. S8–9, Table S1), suggesting that alternative splicing contributes to cell functional changes in heart failure. To determine if unique alternative splicing changes were induced in LVAD responders, we examined LSVs in the responder and non-responder groups compared to heart failure. This analysis revealed that responders and non-responders had distinct alternative splicing responses to LVAD, with 351 and 236 LSVs unique to each group, respectively (Fig. 1H).

Among 73 heart failure-associated LSVs that were modified in responders, one notable example was exon 242 inclusion in TTN (Fig. 1I). TTN encodes the giant protein titin, which regulates myofibril passive tension²¹. Exon 242 encodes an 89 amino acid immunoglobulin domain that mediates protein-protein interactions (Fig. S9) in the I-band region of titin, which is the primary determinant of titin’s elastic properties²². Thus, alternative splicing in this region is expected to impact titin-based myofibril passive tension. Our follow-up analysis of this event using targeted assays revealed that, while exon 242 inclusion increased across all patients after LVAD compared with heart failure, variability between responders and non-responders contributed to a modest, but non-significant change between these groups as measured by exon-specific qPCR (Fig. S10).

Partial myocardial functional recovery with LVAD coincides with decreased CAMK2D exon 14 inclusion

Among the most significant recovery-associated splicing changes identified in the RNA sequencing dataset were increased exon 16 and decreased exon 14 inclusion in CAMK2D (Fig. 1J), which encodes Ca²⁺/calmodulin-dependent protein kinase IIδ. CAMK2D is a serine/threonine kinase that regulates Ca²⁺-dependent contractility and pro-inflammatory signaling in cardiomyocytes and its chronic hyperactivation is implicated in multiple forms of heart disease^23–25. Alternative splicing of CAMK2D exons 14-16 generates four cardiac splice variants (Fig. 2A), which differ in their subcellular localization^26,27. Exon-specific qPCR analysis corroborated the RNA-seq finding that the B splice variant, which includes a nuclear localization signal (NLS) encoded by exon 14, increased in heart failure (Fig. S11). Notably, exon 14 inclusion was increased in non-responder patients both pre- and post-LVAD (Fig. 2B) and displayed a strong inverse correlation with LV functional and structural outcomes (Fig. 2C–D). Targeted RT-PCR and qPCR analyses further validated that there was a shift in the dominant CAMK2D splice variant from 9 (exons 13-16-17) in non-failing controls to B (exons 13-14-17) in heart failure, which was reversed only in patients who exhibited partial functional recovery after receiving LVAD support (Fig. 2E–G).

Fig. 2. Increased CAMK2D exon 14 inclusion in heart failure is reversed only in patients who experience functional recovery with LVAD.

A. Graphical representation of cardiac CAMK2D splice variants and their respective variable region (exons 13-17) alternative splicing; NLS = nuclear localization signal. B. CAMK2D exon 14 proportion spliced in (PSI) in pre- and post-LVAD responders and non-responders from the RNA-seq data; data analysis by paired t-test within groups (pre- vs. post-LVAD) and unpaired t-test between groups (NR vs. R). C-D. Linear regression analysis of exon 14 PSI post-LVAD vs. absolute change in LVEF (C) and LVEDD (D) after LVAD therapy; confidence interval = 95%. E. Representative RT-PCR acrylamide gel for CAMK2D cardiac splice variants and PGK1 housekeeping gene in non-failing controls, HF, and post-LVAD R and NR; BP = base-pairs. F. Proportion CAMK2D splice variant expression from the RT-PCR; displayed as mean ± SEM; n = 6 Control, 10 HF and LVAD R, 9 HF and LVAD NR. G. Quantitative RT-PCR analysis of CAMK2D-B in Control, HF (pre-LVAD), and post-LVAD normalized to the housekeeping gene PGK1 and plotted as a fold-change vs. non-failing control expression; displayed as median and range; n = 10 R, 9 NR; one-way ANOVA with Tukey’s post-hoc test.

CAMK2D-B NLS phosphorylation correlates with LVAD-induced partial functional recovery

To identify potential additional molecular factors uniquely regulated in recovery, we performed TMT quantitative phospho-proteomics on the post-LVAD responder and non-responder samples. Remarkably, the most significantly differentially phosphorylated sites identified were three serine residues in CAMK2D (S332, S333, and S334), which increased in non-responders (Fig. 3A–B). These residues are located immediately downstream of the NLS encoded within exon 14 (Fig. 3A) and their phosphorylation was previously identified as a mechanism for regulating nuclear localization of CAMK2D^28–30. Western blot analyses did not identify differences in CAMK2D regulatory domain T287 autophosphorylation between groups (Fig. 3C, Fig. S12), suggesting similar kinase activity between responders and non-responders³¹. However, in agreement with the phospho-proteomics findings, we found S332 phosphorylation increased in heart failure and then was fully restored to non-failing control levels in patients who exhibited functional recovery (Fig. 3C, 3D). In non-responders, pre-LVAD p-S332 levels were also ~2-fold higher than in responders and remained elevated following mechanical unloading (Fig. 3C, 3D). Pre- and post-LVAD S332 phosphorylation displayed significant inverse correlations with recovery of LV function (Fig. 3E–F), supporting the potential utility of this event as a biomarker to predict which patients are poised for partial recovery from heart failure with LVAD.

Fig. 3. CAMK2D-B S332-S334 phosphorylation predicts poor functional recovery with LVAD therapy.

A. Volcano plot displaying phospho-proteomics results from post-LVAD responder (R) and non-responder (NR) myocardium; n = 9/group. B. CAMK2D-B S332, S333, and S334 phosphorylation between R and NR post-LVAD; two-tailed t-test; *p < 0.05, **p < 0.01, ****p < 0.0001. C. Representative western blots for total CAMK2D, p-T287, and p-S332 in non-failing control, pre-LVAD (HF) R and NR, and post-LVAD R and NR. D. p-S332 CAMK2D normalized to total protein; n = 6 Control, 10 HF and LVAD R, 9 HF and LVAD NR; one-way ANOVA with Tukey’s post-hoc test. E-F. Linear regression analysis of post-LVAD (E) and pre-LVAD (F) S332 phosphorylation vs. absolute change in LVEF on LVAD therapy; confidence interval = 95%. G. Representative images for NRVMs 24-hours post-transduction with GFP-CAMK2D-B adenoviruses ± PE (20 μM); B_SSS = wildtype, B_AAA = Serine 332-334 mutated to Alanine (Phospho-null), B_DDD = Serine 332-334 mutated to Aspartic Acid (Phospho-mimetic). H. Nuclear GFP normalized to whole-cell GFP; Vehicle: n = 75 B_SSS, 82 B_AAA, and 72 B_DDD. PE: n = 57 B_SSS + PE, 72 B_AAA, 91 B_DDD from three biological replicates; displayed as median and range; two-way ANOVA with Tukey’s post-hoc test. I. Representative western blots for CAMK2D in subcellular fractions from post-LVAD responders and non-responders. J-M. CAMK2D expression in the cytosol (J), membrane (K), nucleus (L), and insoluble (M) fractions normalized to corresponding loading controls; ATP1A2 = Na⁺/K⁺ ATPase subunit α-2; NPM1 = nucleophosmin; two-tailed unpaired t-test. For all except I, the data are presented as the mean ± SEM.

Regulation of CAMK2D-B subcellular localization

Since phosphorylation at S332-S334 had previously been shown to prevent nuclear translocation of CAMK2D-B in other cell types^28,29, we next sought to determine the role of phosphorylation at these sites in cardiomyocytes. We generated adenovirus vectors expressing green fluorescent protein (GFP)-tagged wildtype (B_SSS), phospho-null (serine to alanine mutations at 332-334: B_AAA), or phospho-mimetic (serine to aspartate mutations at 332-334: B_DDD) CAMK2D-B and transduced neonatal rat ventricular myocytes (NRVMs). As expected, B_DDD displayed cytoplasmic localization (Fig. 3G–H). However, unlike in non-cardiomyocytes where phospho-null mutations alone led to a complete translocation of CAMK2D-B to the nucleus²⁸, the B_AAA construct displayed only modest nuclear localization in NRVMs (Fig. 3G–H). Notably, when NRVMs were treated with the adrenergic agonist phenylephrine (PE), nuclear localization of B_AAA was potently induced, while B_DDD remained cytoplasmic (Fig. 3G–H). Nuclear translocation of B_AAA was also induced with endothelin-1, but not caffeine or insulin-like growth factor-1 (Fig. S13). To determine whether subcellular localization of CAMK2D was also altered in the hearts of LVAD non-responder patients, we performed subcellular fractionation on post-LVAD samples to isolate the cytosol, membrane, nucleus (soluble/nucleoplasm), and insoluble (myofilament and chromatin-associated proteins) fractions (Fig. S14). Western blot for CAMK2D within these fractions revealed that non-responder patients had modestly increased cytosolic CAMK2D, similar membrane-associated CAMK2D, and reduced CAMK2D abundance in the nuclear and insoluble fractions compared to responders (Fig. 3I–M). These findings suggest that the increased CAMK2D-B in non-responders failed to translocate to the nucleus due to the NLS phosphorylation, which led to increased total cytosolic CAMK2D. Analyses of subcellular fractions from the NRVMs transduced with B_AAA and treated with PE further corroborated the finding that NLS de-phosphorylated CAMK2D-B translocated to the nucleus with adrenergic agonism (Fig. 4A–B).

Fig. 4. Phosphorylation at S332-S334 prevents autoactivation-dependent CAMK2D-B nuclear translocation.

A. Representative western blot for GFP-B_AAA in the nuclear/insoluble and cytosolic fractions of NRVMs treated with vehicle or PE. B. Nuclear/insoluble GFP-B_AAA normalized to Histone H3; n = 4/group; two-tailed t-test. C. Representative western blot for T287 phosphorylated GFP-CAMK2D-B in adenovirus-transduced NRVMs ± PE. D. p-T287 GFP-CAMK2D-B normalized to GAPDH; n = 6/group; two-way ANOVA with Tukey’s post-hoc test. E. Representative images for NRVMs 24-hours post-transduction with B_AAA or B_AAA-T287A/D mutants. F. Nuclear GFP normalized to whole-cell GFP; Vehicle: n = 16 B_AAA, 82 T287A, and 68 T287D. PE: n = 66 B_AAA, 91 T287A, and 117 T287D from three biological replicates; two-way ANOVA with Tukey’s post-hoc test. G. Representative images for NRVMs 24-hours post-transduction with B_DDD or B_DDD-T287A/D mutants. H. Nuclear GFP normalized to whole-cell GFP; Vehicle: n = 52 B_DDD, 40 T287A, and 82 T287D. PE: n = 47 B_DDD, 67 T287A, and 83 T287D from three biological replicates. Data in B and D are presented as the mean ± SEM; data in F and H are plotted as the median and range. Scale bars for microscopy images = 15 μm.

Given the above finding that adrenergic and endothelin receptor agonism induced nuclear translocation of the B_AAA construct, we hypothesized that both NLS dephosphorylation and CAMK2D constitutive activity were required for nuclear translocation in cardiomyocytes, as activation of these receptors increases cytosolic Ca²⁺, leading to CAMK2D activation by autophosphorylation at T287^32–34. We measured CAMK2D T287 phosphorylation of the GFP-tagged CAMK2D constructs by western blot and found it expectedly increased with PE treatment (Fig. 4C–D). Next, to test the hypothesis that both T287 autophosphorylation and S332-334 dephosphorylation were required for nuclear translocation, we generated adenovirus expression vectors from the B_AAA and B_DDD constructs where phospho-null (T287A) or phospho-mimetic (T287D) mutations were introduced in the regulatory domain. As hypothesized, B_AAA-T287D displayed near-complete nuclear localization that was PE-independent (Fig. 4E–F). Meanwhile, B_DDD localized to the cytoplasm and was unaffected by T287A/D modifications or PE treatment (Fig. 4G–H), indicating that NLS de-phosphorylation is a pre-requisite for autoactivation-dependent nuclear translocation of CAMK2D-B. Interestingly, the B_AAA-T287A construct did exhibit a small, but significant, increase in nuclear localization with PE treatment, although the effect was decreased compared to B_AAA alone (Fig. 4E–F). The reason for this is unknown, but we hypothesize that exposure of the CAMK2D-B NLS sequence may be dependent on structural changes associated with the binding of calmodulin, which should still be able to bind the T287A mutant, albeit with much lesser affinity than in the T287 phosphorylated or phospho-mimetic (T287D) states³⁵.

A previous study of CAMK2D-B NLS regulation in a non-cardiomyocyte cell line identified that CAMKI and CAMKIV could phosphorylate the NLS regulatory serine residues²⁸. To determine if these kinases might be implicated in the dysregulation of CAMK2D-B localization in non-responder patients, we first examined their tissue-specific RNA expression across hundreds human samples in the GTEx dataset³⁶. While CAMKI and CAMKIV are highly expressed in the brain, where CAMK2D plays important roles in memory consolidation and synaptic plasticity³⁷, we found that they were expressed lowly (CAMKI) or not at all (CAMKIV) in the heart (Fig. S15A–C). To examine whether CAMKI might still regulate CAMK2D-B despite its low expression, we treated NRVMs with a CAMKI inhibitor during viral transduction with the GFP-B_SSS construct and examined localization via microscopy. CAMKI inhibition had no effect on nuclear GFP signal, with or without PE treatment, suggesting that this kinase does not regulate the CAMK2D-B NLS in cardiomyocytes (Fig. S15D). Previous studies have identified that protein phosphatase 1 and calcineurin can each de-phosphorylate the S332 site of CAMK2D-B^38,39 and both are highly expressed in the heart (Fig. S16), suggesting they may have overlapping functions with respect to CAMK2D-B NLS regulation. Identification of the upstream kinase(s) in cardiomyocytes is an important area for future investigation.

Cellular signaling and functional consequences of cytoplasmic CAMK2D-B

Due to the variability associated with human samples and the stringent statistical cutoffs employed, our phospho-proteomics analysis identified few significant hits, thus limiting insight into potential downstream cellular signaling consequences of increased cytoplasmic CAMK2D-B in heart failure. However, we reasoned that the phosphorylation events mediated (directly or indirectly) by cytoplasmic CAMK2D-B would correlate with p-S332-S334 levels. We therefore performed linear regression analysis for phospho-peptide intensities across the entire phospho-proteomics dataset versus p-S332-S334 CAMK2D-B, which identified several previously established CAMK2D substrates among the strongest correlated events (Table S2). Pathway over-enrichment analysis of phospho-sites with an r² > 0.50 versus p-S332-S334 CAMK2D-B identified Rho GTPase signaling, cardiac conduction, and regulation of cardiac hypertrophy as the top affected pathways (Fig. S17).

To further explore the cellular signaling changes regulated by cytoplasmic CAMK2D-B using a more systematic approach, we performed TMT quantitative phospho-proteomics on NRVMs transduced with B_AAA or B_DDD and treated with PE or vehicle. This approach identified the baseline phospho-proteome effects of nuclear-competent and cytoplasm-restricted CAMK2D-B, as well as activation-dependent effects (Fig. 5A–E). At baseline, phosphorylation events that increased in NRVMs transduced with B_AAA were enriched with cell surface and cytoskeletal proteins, while the downregulated phosphorylation events included nuclear proteins involved in alternative splicing (Fig. S18). With PE treatment, B_AAA led to increased phosphorylation of nuclear-localized RNA processing factors (Fig. 5D, Fig. S19). PE-dependent phosphorylation events that increased in B_DDD-transduced cells were enriched with proteins involved in Rho GTPase regulation (Fig. 5E), matching the findings in the human phospho-proteome (Fig. S17). Additionally, when we mined the human and NRVM phospho-proteomics datasets for shared features, we identified multiple individual peptide examples with similar phosphorylation differences between non-responders/B_DDD and responders/B_AAA, including proteins involved plasma membrane calcium transport (ATP2B1), plasma membrane ion channel anchoring (ANK3), and protein homeostasis (RPL19 and USP47) (Fig. S20). These findings further support the conclusion that increased cytoplasm-restricted CAMK2D-B remodeled the phospho-proteome of non-responders.

Fig. 5. Cytoplasm-restricted CAMK2D-B remodels the phospho-proteome and blunts cardiomyocyte calcium transients.

A. Heat map depicting the differentially expressed phospho-peptides in NRVMs transduced with empty vector (E), B_AAA, or B_DDD ± PE; p-value < 0.01, log₂ fold-change < −0.3, > 0.3. B. Volcano plot displaying differential phospho-peptide expression in NRVMs transduced with B_AAA or B_DDD. C. Volcano plot displaying differential phospho-peptide expression in NRVMs transduced with B_AAA or B_DDD and treated with PE. D-E. Reactome Pathway over-enrichment for the differentially expressed phospho-peptides in B_AAA + PE (D) versus B_DDD + PE (E). F. Representative images of EHTs 24 hours post-transduction with Empty vector, GFP-B_AAA, or GFP-B_DDD adenoviruses. G. Representative Ca²⁺ transient traces for EHTs loaded with an intracellular calcium indicator (Cal 520, AM) and paced at 1 Hz and treated with vehicle or a combination of the α and β adrenergic agonists PE (50 μM) and isoproterenol (ISO, 2 μM). Traces are plotted as % change in Cal 520, AM fluorescence (ΔF) over time. H. Normalized EHT mean calcium transient amplitude (ΔF/F at time 0, F0); displayed as median and range; Vehicle: n = 41 Empty vector, 43 B_AAA, 41 B_DDD EHTs from three biological replicates. ISO/PE: n = 14 EHTs/group from two biological replicates; two-way ANOVA with Tukey’s post-hoc test.

To test for potential functional consequences of increased cytoplasmic CAMK2D-B in human cardiomyocytes, we employed a human induced pluripotent stem cell (iPSC)-derived cardiomyocyte engineered heart tissue (EHT) model⁴⁰. EHTs were transduced with Empty vector control, B_AAA, or B_DDD adenoviral vectors (Fig. 5F) and then Ca²⁺ transients were measured in the absence and presence of adrenergic stimulation. EHTs transduced with B_DDD had reduced Ca²⁺ transient amplitude compared to B_AAA at baseline (Fig. 5G–H). As expected, adrenergic agonism caused a significant increase in Ca²⁺ transient amplitude in EHTs transduced with the Empty vector control (Fig. 5G–H). While the Ca²⁺ transient amplitude in B_AAA EHTs did not increase to the same extent as with Empty vector, the response to adrenergic agonism was significantly improved compared with B_DDD EHTs, which were completely unresponsive to the stimulus (Fig. 5G–H). Other parameters were also significantly impacted by B_DDD, including the area under the curve of the Ca²⁺ transient, contraction speed, and relaxation speed (Fig. S21). Given the previously identified positive relationship between Ca²⁺ transient amplitude and contraction force at physiological beat-rate frequencies in other human iPSC-CM EHT models⁴¹, these data support the conclusion that increased cytoplasmic CAMK2D-B impairs cardiomyocyte contractility through dysregulation of Ca²⁺ transient kinetics.

DISCUSSION

Partial restoration of cardiac function in heart failure, while rare and unpredictable, does occur in a small subset of patients treated with current therapies⁹. Given this evident reversibility of what was long thought to be a static disease state, there is considerable interest in identifying new therapeutic targets that promote reverse remodeling and functional recovery^9,42–45. However, a lack of human heart tissue studies paired with molecular mechanistic investigations has resulted in the underlying biology that mediates reverse remodeling and recovery remaining poorly understood. Previous investigations using multi-omics analyses to identify molecular features of partial myocardial recovery with LVAD identified modest differences in cardiac cell type proportions and gene expression between responders and non-responders^14,15. Our RNA-seq analysis of pre- and post-LVAD samples further corroborates the findings of these previous studies indicating that the cardiac transcriptome is remarkably similar between these response sub-groups¹⁴, although modest differential expression of inflammatory genes was observed, in line with previous findings¹⁵. However, a deeper analysis of protein and splice variant expression identified dramatic changes in RNA alternative splicing in heart failure and unique alternative splice variant expression changes in patients who experienced partial functional recovery with LVAD. Among these, alternative splicing and phosphorylation of CAMK2D were strong predictors of LV structural and functional outcomes with LVAD therapy. Notably, this was true whether we used our criteria for reverse remodeling/recovery or the criteria established previously by other groups who implanted LVADs at earlier stages of systolic heart failure (Fig. S22)¹⁰. Further, our mechanistic investigations using primary rodent cardiomyocytes and human EHTs identified that increased cytoplasm-restricted CAMK2D-B expression in cardiomyocytes, mimicking the non-responder state, dramatically remodeled the phospho-proteome and blunted cardiomyocyte Ca²⁺ transients. The reduced Ca²⁺ transients are consistent with previous observations with hyperactivation of cytoplasmic CAMK2D, which resulted from increased SR Ca²⁺ leak⁴⁶.

Two mouse studies from separate groups published back-to-back in 2009 confirmed the significance of CAMK2D activation for heart failure progression that had been implied from findings in the human heart failure myocardium^47,48, as CAMK2D knockout mice were found to be protected from heart failure with chronic pressure overload^49,50. Given this link between CAMK2D and heart disease progression²⁶, the development of therapeutic strategies that inhibit CAMK2D has long been a goal of heart disease research²⁴. However, despite the clear role of CAMK2D in driving cardiac pathology, small molecule inhibitors have yet to emerge in clinical practice²⁴. Some reasons for this may include off-target inhibition of this ubiquitously expressed kinase in other tissues, inhibition of beneficial cardiac CAMK2 functions, and failure to discriminate between different CAMK2D splice variants or CAMK2 gene isoforms²⁴. Appropriate patient stratification is likely another factor influencing the varied clinical outcomes with CAMK2 inhibitors and our findings suggest that LVAD non-responders are one sub-population that might be particularly well-suited for a CAMK2 inhibitor. Thus, this study may help to inform future clinical trial design to select appropriate heart failure patient sub-populations for CAMK2 inhibitors.

Nevertheless, alternative strategies to regulate CAMK2 may prove more effective than inhibitors due to their above-described drawbacks, although development of such therapies is admittedly time-consuming and high-risk. For example, recent evidence from pre-clinical studies identified that inhibiting CAMK2D constitutive activation by adenine base editing of key residues in the regulatory domain (M281/M282, T287) could successfully prevent heart failure development after cardiac injury in mice^25,51,52. While these findings are exciting, hesitancy around the use of genome editing technology and challenges with vector delivery to the human heart make rapid clinical implementation unlikely, thus warranting the development of additional therapeutic modalities. Our findings indicate that the subcellular localization of CAMK2D – specifically mis-localization of the nuclear splice variant to the cytoplasm – plays an important role in the cardiac pathology mediated by this kinase that has previously been largely overlooked. Therefore, modulating CAMK2D subcellular localization may represent a new therapeutic strategy for advanced heart failure, particularly in patients with non-ischemic dilated cardiomyopathy receiving mechanical circulatory support. The specificity of such an approach would also mitigate the off-target inhibition concerns posed by current therapeutic strategies, as the B splice variant of CAMK2D is almost exclusively expressed in cardiomyocytes⁵³. However, further molecular insights into the upstream regulation of the NLS regulatory serine residues will be crucial before such a therapy can realistically be pursued.

Alternative splicing can generate at least 11 different splice variants from the CAMK2D gene²⁶, three of which are highly expressed in cardiomyocytes: B, C, and 9^26,54. In the human heart, the B and 9 splice variants comprise ~90% of total CAMK2D⁵⁴. The C and 9 splice variants localize to the cytoplasm and regulate the activity of proteins involved in Ca²⁺-handling and inflammation^26,55. Overexpression of the C and 9 splice variants in the mouse heart induces rapid heart failure onset^56–58. In contrast, multiple studies have shown that the B splice variant, which is the only variant that includes the NLS encoded by exon 14, has cardioprotective properties. These include prevention of cardiomyocyte apoptosis, inhibition of pro-inflammatory signaling, and induction of mitochondrial Ca²⁺ uptake in conditions of Ca²⁺ overload^59–62. Overexpression of CAMK2D-B does induce hypertrophic cardiac remodeling in mice, but disease onset occurs much later than with overexpression of the cytoplasmic splice variants^53,55. However, considering our findings, it should be noted that simply overexpressing CAMK2D-B does not ensure nuclear translocation in cardiomyocytes. Increased CAMK2D-B expression in human heart failure, identified herein and in one previous study⁴⁷, suggests a compensatory shift in alternative splicing to a protective splice variant. However, our findings show that hyperphosphorylation at the NLS in LVAD therapy non-responsive patients prevented CAMK2D-B from reaching the nucleus, resulting in increased cytoplasmic CAMK2D. Meanwhile, in heart failure patients who experienced partial functional recovery with LVAD support, exon 14 inclusion decreased to healthy control levels. We speculate that this is because these patients were no longer in the dysfunctional state that triggered this attempted cardioprotective post-transcriptional response. Our findings suggest that nuclear targeting of CAMK2D-B represents a molecular strategy to sequester the kinase away from cytoplasmic targets associated with cardiac pathology and, moreover, that restricting this splice variant to the cytoplasm may prevent functional recovery from advanced heart failure in patients receiving mechanical circulatory support.

Clinical Perspective.

What Is New?

• Left ventricular structural and functional non-response to LVAD therapy is linked to increased expression and S332/S333/S334 phosphorylation of the nuclear (B) CAMK2D splice variant.

• Phosphorylation near the CAMK2D-B nuclear localization signal (NLS) prevents adrenergic stress-dependent nuclear translocation, restricting localization to the cytoplasm.

• Increased cytoplasm-restricted CAMK2D-B remodels the phospho-proteome and impairs cardiomyocyte calcium handling.

What Are the Clinical Implications?

• The identification of CAMK2D alternative splicing and phosphorylation as biomarkers to potentially predict which patients are poised for a favorable response to LVAD therapy represents a major step forward in our understanding of the molecular biology of cardiac recovery.

• These findings form the basis for future investigations into heart failure therapeutics that target CAMK2D subcellular localization and position LVAD non-responders as a potentially well-suited patient sub-population for previously developed CAMK2 inhibitors.

Non-standard Abbreviations and Acronyms

BP

biological process

CAMK2D

calcium/calmodulin-dependent protein kinase 2 delta

EHT

engineered heart tissue

GFP

green fluorescent protein

GO

gene ontology

GSEA

gene set enrichment analysis

iPSC-CM

induced pluripotent stem cell-derived cardiomyocyte

LSV

local spicing variation

LVAD

left ventricular assist device

LVEDD

left ventricular end diastolic diameter

LVEF

left ventricular ejection fraction

NLS

nuclear localization signal

NR

non-responder

NRVM

neonatal rat ventricular myocyte

NYHA

New York Heart Association

R

responder

TMT

tandem mass tag