All human tissues were obtained with informed consent. We utilized RNA-seq data publicly available from the Gene Expression Omnibus (GEO) database (accession number: GSE198618). Samples used for RT-qPCR and western blot validation were obtained by the Pulmonary Center of Excellence biorepository at UTHealth Houston following the institutional review board (IRB) approval (HSC-MS-08-0354 and HSC-MS-15-1049). All animal protocols were approved by the Institutional Animal Care and Use Committee (IACUC) according to the guidelines established by the international Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC, “Guide for the Care and Use of Laboratory Animals,” National Academies Press 2011). The rodents were euthanized in a carbon dioxide chamber followed by cervical dislocation and bilateral thoracotomy in accordance with protocols established by the American Veterinary Medical Association (AVMA POE 2020 Guidelines).

Gene expression data used in this study was obtained from the Gene Expression Omnibus (GEO) database (accession number: GSE198618). The dataset includes RNA-sequencing data from RV tissues of human subjects clinically categorized into three groups: compensated RV function (n = 11), decompensated RV function (n = 7), and donor controls (n = 14). A compensated RV is defined by a state in which the RV maintains adequate cardiac output despite increased pressure or volume load, through adaptive mechanisms such as hypertrophy, enhanced contractility, or augmented preload, without overt signs of systemic congestion or low perfusion. A decompensated RV is defined by a condition characterized by the inability of the RV to sustain effective forward flow, resulting in elevated right-sided filling pressures, systemic venous congestion, and impaired end-organ perfusion [45].

The dataset is publicly available through the GEO database at https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE198618.

Total RNAs from human and rat heart tissue were extracted using Trizol (Invitrogen), and1 µg of RNA was harvested. This RNA was subsequently reverse transcribed with a partial p7 adaptor (Illumina_4N_21T: GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTNNNNTTT TTT​TTT​TTT​TTT​TTT​TTT) and dNTPs with the additional of spiked-in azido-nucleotides (AzVTPs) at a 5:1 ratio. Subsequently, we utilized copper(I)-catalyzed azide-alkyne cycloaddition (CuAAC) to click the P5 adaptor (5′Hexynyl-NNNNAGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGTAGAT CTC​GGT​GGT​CGC​CGT​ATC​ATT, IDT) to the 5′ end of the cDNA. We next amplified the libraries with a universal primer (AAT​GAT​ACG​GCG​ACC​ACC​GAG) and 3′indexing primer (i.e., Index 1: CAA​GCA​GAA​GAC​GGC​ATA​CGA​GAT​CGT​GAT​GTG​ACT​GGA​GTT​CAG​A CGTGT). These libraries were then purified on a 2% agarose gel at 200–300 bps molecular sizes. Purified libraries were pooled and sequenced by Illumina NextSeq 500/550 High Output Kit v2.5 (150 cycles) (Cat# 20024907).

Differential expression analyses were performed with DESeq2 (v1.36.0), as described in [46], a variance-analysis package developed to infer statistically significant differences in RNA-seq data using a Negative Binomial GLM. Genes are called differentially expressed (DE) when having a Fold-Change (FC) >0.263 or <−0.263, and FDR <0.05 (using Benjamini-Hochberg method to control false discovery rate). For the differential expression analysis of the human data, raw count matrices were directly downloaded from the GEO database (GSE198618). APA dynamics from bull RNA-Seq data were analyzed using PolyAMiner-Bulk [47]. Data was processed with the following parameters: -a 0.65 -pa_p 1 -pa_a 5 -pa_m 5 -outPrefix 3UTROnly -expNovel 1 -s 2 - ignore UTR5, CDS, Intron, UN -apriori_annotations -modelOrganism human. PAC-Seq data were analyzed using PolyA-miner [48]. Read counts for each Cleavage and (PAS) were calculated as the total number of reads mapped to the site. Total gene counts were then determined as the sum of all C/PAS counts across the individual features within the gene. PolyAMiner-Bulk is adept at inferring changes in alternative polyadenylation (APA) from bulk RNA-Seq datasets. It enhances the detection process by integrating both established reference polyadenylation sites and newly detected C/PASs. It further refines the candidate C/PASs by applying a deep learning model called C/PAS BERT. PolyA-miner uses vector projections and beta-binomial test to analyze APA patterns within genes. To perform the Gene Ontology (GO) enrichment analysis, we used the gprofiler2 R package to query a list of genes against the Biological Process (GO:BP) database. Results were filtered by an adjusted p-value cutoff (≤0.05) and visualized.

MCT model: For the animal studies, male and female Sprague Dawley rats (Rattus norvegicus) at age 8–12 weeks (Charles River Laboratories) were subjected to the treatment protocol as previously described [49]. In brief, for the MCT model, a single subcutaneous injection of MCT (60 mg per kg body weight; n = 4) was applied (control rats received saline; n = 4), and the RV function was monitored. The whole experiment procedure was 6 weeks for the MCT rat model. For the PAB model, following anesthesia, the PA was separated from the aorta and left atrium, was tied against a 19-gauge needle and then released quickly, as previously described [49]. PAB-operated rats (n = 4) and sham-operated rats (n = 4) without tying the pulmonary trunk were included for the study. Rats were studied between weeks 3 and 8 following PAB operation. Finally, for the hypoxia/Sugen model, rats received a single IP injection of Sugen5416 (20 mg/kg) followed by exposure to 10% oxygen for 3 weeks and returned to normoxia for an additional 2 weeks. Rats housed in normoxia conditions throughout (n = 5) and subjected to SUHx (n = 4) were assessed for PH. For right heart catheterization, rats were anesthetized with continuous inhalation of 1.5%–2.0% isoflurane under a nosecone.

Flash frozen RVs were pulverized with a mortar and pestle. RNA was extracted with Triazole and chloroform. RNA was obtained by addition of Cotrimoxazole lysis reagent (QIAzole, QIAGEN; Hilden, Germany) followed by extraction with chloroform added in a 1:5 ratio. Following precipitation with 100% ethanol, samples were passed through an RNA-binding spin column (miRNeasy QIAGEN). The column was washed and mRNA eluted with water. Total RNA was quantified by A260/280 ratio via NanoDrop. cDNA synthesis was performed by reverse transcription reaction (iScript, BioRad) using 1000 ng of template. Fold change against the housekeeping gene 18S was determined via the double delta C_t method. Fluorometric amplification signal was monitored in real time with an ABI 7500 Real-time PCR system (Applied Biosystems; Foster City, CA).

Pulverized heart tissue was dissolved in RIPA buffer (Thermo Scientific, Rockford, IL) containing 1 mM protease and phosphatase inhibitor (Sigma-Aldrich). 30 mcg of protein was mixed with 6X SDS-Sample buffer (Boston BioProducts) and loaded into 4%-12% Mini-Protean TGX gels (Bio-Rad, Hercules, CA). These gels were then transferred to PVDF membranes (GE Healthcare, Piscataway, NJ), which were subsequently blocked with 5% milk in TBS-T(Sigma-Aldrich) for 1 h and incubated with the primary antibody overnight at 4 °C. Secondary antibodies were incubated for 1 h at RT and the blots were visualized with Clarity ECL (Bio-Rad). Densitometries were performed with the Bio-Rad Image Lab software.

We detected a significant change in the 3’UTR landscape and APA genes (DAGs) in decompensated RVs compared to control RVs (Figure 1A) that was more pronounced in comparisons between decompensated RVs compared to compensated RVs (Figure 1B), reflecting minimal changes in gene regulation between compensated RVs and controls. Only 18 significantly differentially regulated genes were observed in compensated RVs (Figure 1C). 3’UTR landscape of decompensated RVs reflected 3’UTR elongation as the predominant shift in APA, with 944 transcripts with longer 3’UTRs and 116 mRNAs with shorter 3’UTRs relative to control RVs (Figure 1A). Comparisons between compensated vs. decompensated RVs yielded 816 significant transcripts with longer 3’UTRs and 92 transcripts with shorter 3’UTRs compared to compensated RVs (Figure 1B). Comparisons between controls vs. decompensated and compensated vs. decompensated revealed 65 common genes that were shortened (Figure 1D) and 704 elongated transcripts (Figure 1E). Thus, global 3'UTR lengthening appeared specific to RV decompensation and may be a distinctive marker for progressed RV disease. Previous studies have identified, 3’UTR shortening and concomitant reduced expression of CPSF5 [36, 39, 50] as a pathological mechanism. In line with this, we aimed to assess expression levels of CPSF5 in decompensated human RVs compared to compensated RVs. Herein we report increased CPSF5 levels, consistent with 3’UTR elongation (Figures 2A,B). This is significant since 3’UTR shortening is often associated with loss of CPSF5(39) and an increase of this RNA binding protein may imply 3’UTR elongation. A complete list of APA analyses is provided in Supplementary Table S1.

Enrichment analyses comparisons between control RV Vs. decompensated RV and compensated RV vs. decompensated RV revealed enrichment of several key processes, including the Longevity Regulating pathway, AMP-activated protein kinase (AMPK), p53, insulin, Forkhead box O (FoxO) signaling pathways and lysine degradation (Supplementary Figures S1A, B). Pathways that were unique to control RV vs. decompensated RV included cellular senescence, human papillomavirus infection, phosphatidylinositol 3-kinase (PI3K-Akt) and pathways in cancer (Supplementary Figure S1A). Insulin resistance, HIF-1A, adiponectin signaling ad Mannose Type O-glycan synthesis were unique to compensated RV vs. decompensated RV comparisons (Supplementary Figure S1B).

The length of the 3’UTR significantly impacts mRNA fate, thereby influencing its stability [51]. To dissect this further, we assessed the differential expression of these genes to identify differentially expressed genes (DEG) in healthy vs. diseased RVs. Decompensated RVs displayed 8,631 DEGs compared to control RVs, of which 4,320 were upregulated and 3,412 were downregulated (Figure 3A). Similarly, decompensated RVs exhibited 5,263 DEGs relative to compensated RVs, with 2,648 upregulated and 2,270 downregulated (Figure 3B). There were also a notable number of DEGs in compensated RVs compared to control RVs, though far fewer than in decompensated RVs, with 408 genes upregulated and 291 genes downregulated (Figure 3C). We also analyzed the biological pathways associated with the DEGs to gain deeper insight into the mechanisms driving decompensation and RV failure. In the control vs. decompensated RV comparison, the upregulated DEGs were enriched for pathways related to stress-responsive signaling, growth and developmental processes, vascular remodeling, cytoskeletal organization, and inflammatory responses, consistent with an active but maladaptive attempt to preserve cardiac output through hypertrophic and remodeling mechanisms (Supplementary Figure S2A). In contrast, the downregulated DEGs were predominantly associated with mitochondrial and metabolic energy pathways, including fatty acid oxidation, oxidative metabolism, protein turnover, autophagy, and intracellular transport, indicating impaired energetic capacity and disrupted cellular homeostasis in the decompensated RV (Supplementary Figure S2B). In the compensated vs. decompensated RV comparison, the upregulated DEGs were primarily enriched for adaptive signaling pathways, developmental and morphogenetic programs, cell migration, angiogenesis, and vasculature development, reflecting the relative preservation of compensatory remodeling responses in compensated RVs (Supplementary Figure S2C). Conversely, the downregulated DEGs were strongly associated with core metabolic and bioenergetic processes, as well as cardiac contractile, electrophysiological, and conduction pathways, suggesting progressive loss of cardiomyocyte functional integrity and energetic efficiency during RV decompensation (Supplementary Figure S2D). A complete list of enrichment analyses is provided in Supplementary Table S2. Together, these findings indicate that RV decompensation is characterized by activation of stress and remodeling pathways alongside a broad suppression of metabolic, mitochondrial, and cardiomyocyte functional programs, distinguishing maladaptive failure from compensated remodeling.

To further investigate the mechanisms underlying RV failure and establish a solid platform for developing and testing therapeutic interventions, we utilized three animal models of PH-associated RV failure: the MCT, the PAB and the SuHx models of PH. The MCT model involves administering monocrotaline, a toxic pyrrolizidine alkaloid, which induces pulmonary vascular injury and subsequent PH (Figure 4A). This model is popular due to its simplicity and ability to cause severe PH and RV failure [52]. DAG analysis revealed 101 elongated transcripts and 257 transcripts with shorter 3’UTRs (Figure 4B). The PAB model surgically constricts the pulmonary artery, increasing RV afterload without affecting the pulmonary vasculature, enabling researchers to isolate the effects of increased afterload on the RV. It also allows for variable levels of constriction or loosening of the PA, providing insight into afterload effects on RV remodeling (Figure 4C) [52]. Using this model, we reveal 74 transcripts with elongated 3’UTRs and 162 transcripts with shorter 3’UTRs (Figure 4D). The SuHx model combines a vascular endothelial growth factor receptor (VEGFR) inhibitor, Sugen 5416, with chronic hypoxia, closely mimicking the progressive and severe nature of human PAH, resulting in significant pulmonary vascular remodeling and severe RV failure (Figure 4E) [52]. In this model, we identified 76 transcripts with longer 3’UTRs and 234 transcripts with shorter 3’UTRs (Figure 4F). A complete list of APA analyses is provided in Supplementary Table S3.

Subsequent pathway analysis revealed parallels with human decompensated RVs. DAGs in both cases were significantly associated with pathways, including the longevity regulating pathway, Mitogen-activated protein kinases (MAPK) signaling, and adrenergic signaling in cardiomyocytes and ECM organization (Supplementary Figure S3). However, some key differences revealed how, in the MCT model, DAGs were implicated in pathways such as ubiquitin-mediated proteolysis and cardiomyopathy (Supplementary Figure S3A). In the PAB model, DAGs mapped to pathways such as neurotrophin signaling and ErbB (receptor tyrosine kinase) signaling (Supplementary Figure S3B). As seen in the MCT model, there were notable overlaps with human RV failure, including involvement in insulin signaling and MAPK signaling pathways. In the SUHx model, with significant DAGs enriched in pathways such as protein processing in the endoplasmic reticulum and the Citrate cycle (TCA cycle) (Supplementary Figure S3C). In line with aspects of human RV failure, these DAGs also influenced pathways like insulin signaling and cardiomyopathy. Interestingly, while similar pathways were altered, the general trend of altered transcripts was flipped, where the human disease had more transcripts with lengthened 3’UTRs. Yet, the rat models demonstrated more transcripts with shortened 3’UTRs.

Significant differential gene expression was observed across the three rat models of PH-associated RV failure. In the MCT model, 1019 DEGs were identified, with 551 genes upregulated and 468 downregulated (Figure 5A). The PAB model displayed 482 DEGs, consisting of 276 upregulated and 206 downregulated genes (Figure 5B). The SUHx model exhibited the highest number of DEGs, with 3570 total DEGs, including 1817 upregulated and 1753 downregulated genes (Figure 5C).

Pathway enrichment analysis of the DEGs revealed significant similarities between the three rat models and human RV failure. In all models, upregulated genes were consistently associated with pathways involved in Extracellular Matrix (ECM) Organization and inflammation, reflecting the pathological remodeling processes observed in human disease (Supplementary Figures S4–S6). These changes highlight the shared mechanisms of structural remodeling and immune response activation in both experimental and clinical RV failure. Conversely, downregulated genes were predominantly linked to pathways involving metabolism and mitochondrial respiration, suggesting a disruption in energy production and metabolic homeostasis, which are critical for maintaining cardiac function under stress (Supplementary Figures S4–S6).

These findings emphasize the utility of the three rat models in recapitulating key molecular aspects of human RV failure. Additionally, the convergence of molecular pathways across both species provides critical insights into the pathological mechanisms of RV failure, supporting the use of these models for future therapeutic exploration.

APA alters the length of mRNA, influencing their stability, localization, and translation, affecting gene expression profiles during RV decompensation. Comparative analysis of rat models of RV failure (Figure 6A) and human decompensated RVs revealed common APA changes (Figure 6B), pointing to conserved mechanisms underlying disease progression. Notably, we identified 15 genes with APA dysregulation common to both rat models and human RV decompensation (Table 1). The 15 genes—AGPAT3, ALDH5A1, ANGEL2, AZIN1, DLAT, DNAJA2, EPB41L1, FDFT1, MAPK9, MXRA7, NAMPT, SERBP1, SOD2, SPRYD7, and STXBP5—play crucial roles in regulating metabolism and stress responses in RV failure (Figure 6C). These genes are involved in lipid and energy metabolism, protein folding, and antioxidant defense and could collectively lead to exacerbating disease progression. For example, AGPAT3 and FDFT1 affect lipid profiles and membrane integrity [53, 54], DLAT and NAMPT are key players in energy production [55, 56], DNAJA2 and SOD2 provide protective mechanisms against oxidative stress [57, 58], and MAPK9 mediates inflammatory and hypertrophic responses [59]. Dysregulation of these genes by APA can impair the RV’s ability to adapt to stress, leading to metabolic disturbances and promoting dysfunction. Therefore, these insights not only enhance our understanding of the disease but also present promising molecular targets for therapeutic intervention while also offering new opportunities to modulate disease progression at the post-transcriptional level.

Our data demonstrates significant changes in the 3’UTR landscape in RV dysfunction in both patients and in experimental models of disease. Given that alternative polyadenylation is orchestrated by the cleavage and polyadenylation (CPA) machinery, we examined whether select components of this complex are implicated in the observed APA dysregulation. Previous studies have demonstrated the role of CPA complex proteins in regulating APA. For example, knockdown of NUDT21 leads to global 3' UTR shortening and promotes disease-associated gene expression changes [32, 39]. Consistent with this framework, we observed altered abundance of the CPA component CPSF5 at the protein level in RV dysfunction (Figure 2), providing orthogonal evidence that dysregulation of the polyadenylation machinery accompanies APA remodeling in disease. Together, these findings support the notion that APA changes reflect underlying perturbations in CPA-mediated RNA processing and highlight APA as a reliable indicator of RV dysfunction.

Given the widespread APA dysregulation observed in RV decompensation, we investigated if the core cleavage and polyadenylation (CPA) machinery presented with altered gene expression. Supporting the dysregulated APA, we found that several components of the CPA complex are differentially expressed in decompensated RVs (Figures 7B,C). Eight CPA complex components are upregulated and five are downregulated when comparing decompensated RVs with control RVs (Figure 7C). Among the most prominent changes, PABPC1, SYMPK, PABPN1, PABPC4, and CPSF1 are upregulated in decompensated RVs, whereas PPP1CB, CSTF1, CPSF2, CPSF6 and CSTF2T are reduced. By comparison, analysis of compensated vs. decompensated RVs reveals 5 upregulated and 3 downregulated CPA components (Figure 7B). The upregulated genes include PABPC1, PABPC4, PABPN1, SYMPK, and FIP1L1, whereas PPP1CB, CSTF1, and CPSF2 are downregulated. In line with the minimal dysregulation of APA observed in compensated RVs, we detected no significant changes to the CPA complex in control vs. compensated groups (Figure 7A). Further analysis of DEGs of the CPA complex components shows that CPSF6, and CSTF2T are significantly downregulated in the decompensated RVs. At the same time, CPSF1, RBBP6 and PCF11 are significantly upregulated in the decompensated RV, but not in compensated or control RV. Our data shows that APA serves as a reliable indicator for understanding RV dysfunction.

Next, we validated gene and protein levels of differential APA targets identified in by qPCR and Western blot in both human and rat RVs. We utilized control RV (n = 11), compensated RV (n = 11) and decompensated RV (n = 12) tissues from the Biorepository of Centre de recherche de l'Institut universitaire de cardiologie et de pneumologie de Québec; these represent distinct samples from the RVs used for bioinformatic analyses. Herein, we identified reduced expression of AZIN1, FDFT1, SOD2, DLAT, SPRYD7, and EPB41L1 in decompensated RV compared to control or compensated RV groups (Figures 8A–F). STXBP5 expression was also reduced in decompensated TV compared to compensated RV (Figure 8G). However, NEDD9 expression levels were elevated in decompensated RVs compared to both control and decompensated RVs (Figure 8H). From the 15 conserved APA targets, we prioritized FDFT1 and SOD2 for protein validation because they map to two core RV decompensation hallmarks: lipid/sterol metabolism and mitochondrial oxidative stress, while also being highly plausible points of APA-driven post-transcriptional control. Herein, we report increased expression of these mediators in decompensated RVs compared to control or compensated RVs (Figure 9A,B). We also determined expression of these targets in our rat models of RV dysfunction (Figure 10). Herein, Azin1 was upregulated only in compensated RVs in MCT model with no changes in either the PAB or SuHx models (Figures 10A–C). Fdft1 and Sod2 expression were both reduced in decompensated RVs in the MCT and in the SUHX models, with no changes seen in the PAB model (Figures 10D–I). Dlat expression levels were only reduced in the SUHx-treated mice with no changes in either MCT or PAB models (Figures 10J–L). Stxb5 expression levels were unaltered in the MCT model, elevated in the decompensated RV from the PAB model, and reduced in SUHx-treated rats (Figures 10M–O). No changes in Nedd9 expression were reported in any experimental model (Figures 10P–R). Taken together, these results demonstrate differences in gene expression between experimental models of RV dysfunction.