Development of calcific aortic valve stenosis and phenotypic characterization studies
Initially, we attempted to conduct the Drolet et al.12 experiment in New Zealand rabbits but failed to produce echocardiographically confirmed aortic valve stenosis, thus replicating the study of Hekimian et al.14. We then demonstrated that when water was used as a carrier, vitamin D2 was not increased in the animal serum (Supplemental Information, Tables S1, S2). We redesigned the experiment using oil as a carrier for vitamin D2, investigated the optimal protocol in a preliminary dose-escalation experiment, and validated vitamin D2 concentrations in serum and aortic valve tissue with LC/MS-MS (Table S3). Aortic valve stenosis was reached at 7 weeks using 3.500 International Units (I.U.s) of ergocalciferol/kg/day plus a diet supplemented with 1% w/w cholesterol (Table S4). After approximately 8 weeks of intervention, the animals died with symptoms of pulmonary edema.
We then designed a longitudinal experiment in which a stage of 2 weeks was arbitrarily chosen as an early CAVD phase and a stage of 4 weeks was chosen as an intermediate phase. The novel protocol induced significant increases in serum vitamin D metabolites at 2 and 7 weeks. In the aortic valve, only 25-OH-vitamin D2 was significantly increased (Table S3, Fig. 1a, b). Moreover, the diet resulted in very high levels of cholesterol, mild renal impairment, and mild hypercalcemia at 4 weeks (Fig. 1, Table S5). Evidence indicated a gradual weight loss, mild hypertension, and the development of heart failure: liver congestion with increased liver enzymes, progressive hyponatremia, and increased troponin-T (Table S5,S6).
The intervention reduced the aortic valve area (AVA) by 30% at 4 weeks and 7 weeks, increased the transaortic mean gradient by 19.5% at 4 weeks and further increased it by 51% at 7 weeks (Fig. 1d,e, Table S4). Calcification was significantly increased in the valve stenosis group, as quantified by ex vivo [18F]-NaF microPET/CT and confirmed by γ-counting, where a 5.62-fold increase in the SUVmax was observed (Fig. 1k, Table S7). Histology revealed calcification of the aortic valve at its tip (Fig. 1r) and its cusp.
FT-IR spectroscopy indicated chemical and structural changes between the control and calcified cusps at 7 weeks. In particular, the presence of apatitic structures in the 7-week aortic valve, along with changes in protein structure conformation, lipids, nucleic acids (DNA, RNA), and carbohydrates, was detected (Fig. 1y, Table S8)15. Of note, a low degree of crystallinity in apatites is typically associated with water molecules, resulting in pronounced absorption within the 3000–3700 cm− 1 range16.
The Raman spectra of calcified valve areas at 7 weeks are characterized by new vibrations at frequencies of 426, 606, 958 and 1085 cm− 1, which match the four internal modes of A- and B-type carbonated apatites17 even though only a weak peak at 1085 cm− 1 is observed (Fig. 1z, Figure S1). The bands at 743 and 1037 cm− 1 match those of calcium pyrophosphate (PPi). Spectral features suggestive of an increase in lipid/cholesterols content are present which can be considered a condition of disease progression: the band at 700 cm− 1, the slightly leftward position of the band at 1445 cm− 1, and the increased intensity of the band at 2855 cm− 1 9. Another important aspect is the emergence of the hydroxyproline band at 877 cm− 1 18 in the 7 weeks samples. The Raman spectra of the valve tissues at 2 weeks of CAVD’s progression (Fig. 1, green spectrum) are quite similar to the control spectra, with only slight variations of the collagen bands at 817, 920 cm− 1 and a weak signal at 1066 cm− 1, along with changes in the frequency range of magnesium pyrophosphate vibrations19, around the 1060 cm− 1.
Scanning electron microscopy (SEM) revealed dispersed dense structures with irregular shapes located within the collagen fibers (Fig. 1A, B). Elemental compositional analysis indicated that they primarily consisted of calcium and phosphate, resembling the chemical composition of hydroxyapatite, along with trace amounts of sodium (Na), magnesium (Mg), silicon (Si) and sulfur (S) (Fig. 1AB). Similarly, bright-field transmission electron microscopy (TEM) imaging, elemental analysis, and diffraction patterns revealed spherical nanometer particles carrying amorphous calcium phosphate within the valves of the stenosis group; moreover, the ordinal collagen fibrils packing was replaced by a randomly disorganized and fragmented collagen matrix (Fig. 1AC, AD).
We then assessed known basic triggers of osteoblastic transformation4. Serum phosphate (Pi) and the PPi were increased in the experimental animals with valve stenosis. In comparison, PPi levels were reduced in the valve stenosis group tissue supernatants (Fig. 1m-o). The tissue levels of IL1-b were increased by 6-fold in the valve stenosis group (Fig. 1l). Active enzymatic-mediated mineralization was demonstrated as total endogenous phosphatase activity localized in the valve according to the BCIP/NBT assay but not in the control valves (Fig. 1s). Similarly, the tissue levels of RANKL, a critical regulator of osteogenesis, were significantly increased by 33% at 2 weeks and returned to baseline at 7 weeks (Fig. 1p). The key osteogenic transcription factor RUNX2, tissue non-specific alkaline phosphatase TNAP (the main known phosphate-liberating enzyme), and osteopontin (a major regulatory protein of calcification), were also validated in late-stage lesions (Fig. 1q, t-x, Figure S4, S5). Compared to controls, the experimental aortic valves were strongly positively stained for wheat germ agglutinin (WGA), indicating increased cellular infiltration/proliferation and/or increased N-acetyl-glycosaminoglycan content (Fig. 1t). To demonstrate the translational potential of the model, three patients with CAVS were imaged with [18F]-NaF PET/CT, and the standard uptake values were calculated: the mean NaF SUVmax was 2.56, the mean peak SUVmax was 1.40, and the mean target to background (TBR) was 2.38 (Fig. 1j), equivalent to established studies6.
Investigation of molecular players during experimental aortic valve stenosis progression and validation studies
To identify the molecular players at the early stage, we comprehensively analyzed the tissue transcriptome, proteome, and metabolome at 7 weeks, and additionally the transcriptome at 2 and 4 weeks. We examined the differential expression of mRNAs and microRNAs at each stage of intervention. Analysis of the differential expression at 7 versus 2 weeks revealed that the top 5 pathways were involved in oxidative phosphorylation, diabetic cardiomyopathy, thermogenesis, prion disease, and nonalcoholic fatty liver disease (Fig. 3a, Figure S6). The earliest deregulated pathway at 2 weeks was only the peroxisome pathway, while at 4 weeks, the top 5 pathways were involved in tuberculosis, Staphylococcus aureus infection, osteoclast differentiation, leishmaniasis, and the phagosome (Figure S7). The significantly deregulated transcripts exhibited coexpression patterns. In the longitudinal analysis, 20 gene clusters emerged (Fig. 3). Among the largest groups, cluster 6 (32 pathways) demonstrated coexpression of ALPL, IBSP, ANXA8, MMP14, MMP17, TIMP1, and POU2F2; immune markers (AIF1, C1RL, CCL19, CD22, CD74, CXCR1, IL10RB, IL21R, Ly86, MCP1, MCP2, RLA-DMA, RLA-DR-A, TLR1, TNFAIP6, and TNFRSF1B); ECM genes (TNC, TNN, VCAM1); and kinases (such as PRKCB and TYROBP); cluster 5 (31 pathways): coexpression of APOBR, C1R, C3AR1, C5AR1, and CLIC2; immune markers (CXCL16, FCER1G, FCGR1A, IFNGR1, IL4R, IL6R, IL7R, and IL10RA); DNA damage-related RAD51, key kinases (JAK3, MAPK4K1, PIK3R5, PIK3R6, and PTPRC); and ECM genes (CORO1A and GALNT6).
At the microRNA level, 10 differentially deregulated miRNAs, namely, rabbit miRNAs 21-5p, 204-5p, 146a-5p, 143-3p, 16a-5p, 2387-3p, 499-5p, 133a-3p, 23b-3p, and miR-125b-5p, were identified (Table S9). Among the top three miRNAs, miR-21-5p and miR-146a-5p were longitudinally increased, while miR-204-5p was decreased in the experimental animals (Fig. 3e). We investigated the top three significant microRNAs in human valves of patients with severe CAVS against control valves obtained from patients undergoing heart transplantation due to advanced heart failure (Fig. 3d). MiR-21-5p was upregulated by 4.45-fold in CAVS tissue (Mann‒Whitney p < 0.001). MiR-204-5p was downregulated by 2.66-fold in CAVS tissue (Mann‒Whitney p = 0.001). MiR-146a-5p was not validated in human valves since its expression was too low in CAVS patients and controls (Mann‒Whitney test p = 0.706) (n = 17 patients, 10 controls).
The longitudinal expression of the top three miRs, as expected, had the opposite pattern in comparison to their predicted targets (Fig. 3e). The hsa-miR-21-5p had a predicted interaction with 346 transcripts with a microT score > 0.50 (considering the significant conservation of miR sequences and their targets between rabbits and humans20). These transcripts are related to the polycomb repressive complex, parathyroid hormone, focal adhesion, neurotrophin, and actin regulation pathways. Similarly, hsa-miR-204-5p is predicted to interact with 79 transcripts related to hematopoietic cells, chemokine signaling, cytokine–cytokine signaling, primary immunodeficiency, Fc gamma receptor, JAK–STAT, and viral protein–receptor interaction pathways. hsa-miR-146a-5p is predicted to interact with 485 transcripts related to the polycomb repressive complex, parathyroid hormone, focal adhesion, neurotrophin, actin regulation, depression, longevity, and ECM-receptor interaction pathways.
We then compared the transcriptomic profile of the lesion to a selected set of genes implicated in several underlying mechanisms in the human disease (Fig. 4, Table S10): inflammatory10,21,22, DNA damage-response23,24, hypoxia25, mechanotransduction3, EndMT26, the fibrotic and calcified parts of diseased valves10,27, and the relevant VIC markers28,29. The inflammatory genes IL10RA, CXCR1, CXCR2, C5AR1, and most prominently CXCL14 were significantly overexpressed at 4 weeks, 7 weeks, or longitudinally. Further, examining on the whole the inflammatory pathways activation (Supplemental Figure S6), they were overall increased in the intermediate stage, and relatively reduced thereafter, towards the valve stenosis stage where calcification is increased.
Among the hypoxia-responsive genes, PERM1 and HIGD1B were upregulated > 2-fold. The DNA damage response gene Rad51 was significantly upregulated early at 4 weeks, and to a lesser extent, the PARPBP gene. The MAPK4 gene is the principal mechanotransduction gene whose expression is longitudinally downregulated.
Among EndMT-related genes, ACTA2 expression was significantly suppressed longitudinally, CDH2 expression was increased longitudinally, and MMP13 expression was increased > 5-fold longitudinally. Comparison to the human fibrotic layer signatures replicated the significant overexpression of the PRG4 and SERPINA3 genes. Comparison to the calcified tissue replicated the significant induction of the CORO1A, HCLS1, LCP1, CD74, and CD79A genes at 7 weeks and longitudinally. The expression of DES was significantly induced longitudinally, and SPP1 was overexpressed at 7 weeks.
On the other hand, the ELN1 gene was significantly upregulated early at 2 weeks and significantly downregulated thereafter. BGLAP was suppressed, while other known VIC markers, such as ALPL, ENG, ITGB1, MEOX1, NT5E and VIM, demonstrated changes of lower magnitude. Consistent with phosphate-induced calcification, the transcript levels of Pit-2 (POU2F2), a phosphate cotransporter implicated in vascular calcification30, were increased in the lesion. Of the implicated ectonucleotidases in calcification31, only the transcript of NT5E changed significantly (Table S10).
Next, we focused on the earliest deregulated gene transcripts in a time-course analysis using the ImpulseDE2 model. In total, 626 genes were significantly deregulated at 2 weeks, 543 at 4 weeks, and only 31 emerged at 7 weeks (Supplementary file). The analysis corroborated
the finding that the majority of the inflammatory genes were altered after 4 weeks. Instead, the early stage is dominated by a myofibroblastic and osteogenic fingerprint (including TGF-β, NOTCH2, BMP/smad, and Wnt signaling) in interplay with genes belonging to ECM, mechanotransduction, peroxisome, autophagy, DNA damage response, and phospholipid signaling, before inflammation occurs. Among the genes related to early inflammation, being very scarce, were TLR6, C3d receptor, and immunoglobulin J chain. Perlecan, fibulin, sclerostin, collagen isoforms, laminins, osteoglycin, MFAP, and aggrecan were among the native ECM transcripts demonstrating early changes. Moreover, LOX and LOXL4 lysyl oxidases were early implicated, which mediate hydroxyproline formation and the stabilization of collagen and elastin fibers through cross-linking, increasing matrix stiffness3. Among the mechanotransduction signaling-associated genes were PKD1, MAPK4, DDR, ILK, and TEAD3. Peroxisome-associated genes were ALDH18A1, DNM1L, EPHX2. Autophagy and phagosome-associated genes were USP20, SYT11, FBXL2, WDFY3, DEPP1, PLEKHG5, RAB7A, RAB20, RAB39A, TICAM2, GABARAP, and ATG4A. DNA damage-response-associated genes were Rad51, PARP3, GADD45A, ABL1, EGR1, RPA3, PHF1, FOXN3, DAZAP2, CCN4, and DDIT4. Phospholipid signaling-associated genes were ATP10A, CDS2, EFR3A, INPP4A, AXL, INPP5F, ITPR1, IRAG1, LPCAT3, PLCB4, NYAP1, and PIP5K1A. Additionally, amyloid beta precursor transcripts were also activated early (APBB1, APPL2, APLP2), along with extracellular adipokines C1QTNF2, C1QTNF7.
Next, we investigated the aortic valve tissue proteome using untargeted proteomic analysis at 7 weeks, which revealed 95 differentially expressed proteins in the cases versus controls at a 0.05 level of significance (n = 5 experimental, 6 controls, Table S11). Among them, the intervention induced a set of proteins uniquely detected in the experiments: APOB, APOD, B2M, YBX1, HMGCR, C8B, ATP6V1D, H13_RABIT, SORL1, SPARC, C8G, CRYBA2, KV05_RABIT, JCHAIN, T5, SLC9A3, CYP2J1, and CYP2A10. Other proteins with previously identified roles in the disease10 were significantly increased in the lesion: CLU, VTN, FN1, AHSG, HRG, C3, LGALS3, COL12A1, FLNB, MPG, ANXA8, MMP2, TIMP, CD63, F10, SERPIND1, CD14, TGFBI, PON1, FTL, and FTH1. The model is characterized by a baseline smooth muscle cell phenotype, with a significant shift from MYH11 to MYH4 in the experimental group. Protein-derived pathway enrichment revealed that the complement and coagulation cascade was the most significantly deregulated pathway (Fig. 5a). The top 5 deregulated pathways included salivary excretion, arrhythmogenic right ventricular cardiomyopathy, circadian entrainment, and oxytocin signaling (Fig. 5). Tracking the respective mRNA‒protein differential expression overlap across stages revealed that the earliest changes were in CD14, TIMP1, RAD51, and LGALS3 at 4 weeks (Fig. 5b).
Along with CLU, ANXA8, FTH1, FTL, JCHAIN, APOE, APOD, CST3, CYP2J1, and AHSG were significantly upregulated at 7 versus 2 weeks, while PFKP, SLMAP, TES, ACTA2, BCHE, MYH11, OGN, PRKG1, SOD3, MYLK, GOT2, DAG1, ANXA11, PYGM, CAPZA2 and GYG1 were downregulated (Fig. 5a).
Since the action of miRs impacts protein expression, we investigated the predicted interactions of the proteins with the differentially expressed microRNAs. Among the upregulated proteins, ETF1, SSR1, and FN1 were the most highly predicted targets, while among the downregulated were NOS1, MYLK, RYR2, ADCY5, and CACNA2D1. Of note, 20 out of 32 proteins whose mRNAs were not significantly deregulated longitudinally were predicted to be targets of the identified microRNAs (Fig. 5c, d).
Inspired by the results showing increased α-helix formation around the calcified ECM and the emerging amyloid-related pathways (Fig. 3) and early genes, we used the AMYPred-FRL tool32 to calculate the probability of amyloid formation by the overexpressed proteins. Immunoglobulins, filamin-B, beta-crystallin A2, apolipoproteins, ferritin heavy chain, and beta-2-microglobulin had a > 0.9 probability for amyloid formation (Fig. 5e, Supplementary File).
We then investigated selected key proteins for validation in the aortic valve (Fig. 6a-c). Firstly, SLMAP was expressed in the control aortic valves, only in their cusp, and in the experimental animals throughout the valve including the tip, but was absent from the myocardium. Thus, the basic smooth muscle phenotype of the model was further induced by the intervention. Among VIC markers, desmin and vimentin were validated in the valve cusp, while desmin only in the tip of experimental animals at 7 and 2 weeks.
Populations of desmin + and vimentin + cells were located at the base of the valve cusp, with a group of vimentin+/desmin- cells infiltrating the fibrosa (Fig. 6d-h). The deposition of complement C3 and immunoglobulin G in the valve cusp and its tip was validated at 2 and 7 weeks of intervention. Moreover, CD8 + and TGF-β cells were present in the valve at 7 weeks (Fig. 6i-q).
Next, we investigated the relative changes in the metabolome content of the valve using targeted and untargeted metabolomic analysis, which revealed 61 metabolites that were significantly differentially expressed among the groups, with an area under the curve (AUC) > 0.8 (Table S12). Among the top metabolites, hexanoylglycine, 7,8-dihydroneopterin, neopterin, and L-pyroglutamic acid are related to reactive oxygen species (ROS) production, the respiratory chain and mitochondrial function33,34.
Hexanoylglycine and neopterin increased, while 7,8-dihydroneopterin decreased at 7 weeks. 7,8-Dihydroneopterin is a potent ROS scavenger produced by IFN-γ-stimulated macrophages and is transformed by superoxide and hypochlorite to neopterin34, which in turn may increase inducible NOS levels. Reduced pyroglutamic acid is also linked to ROS, since it is derived from oxidized glutathione. In this context, we evaluated the relative expression of ROS-related genes35,36 (Fig. 7a). Indeed, NDUFA4L2 and inducible NOS2 were significantly overexpressed, while other transcripts demonstrated less potent changes.
Gamma-glutamyl L-putrescine and gamma-glutamyl L-ornithine indicate deregulated glutamylation coupled to the deregulation of putrescine/ornithine metabolism. The dipeptide putrescine is increased in myocardial and cancer cells in response to ROS-driven mitochondrial DNA action through the CGAS-NFK-β-ODC1 axis37,38.
The CGAS transcript demonstrated a > 2-fold change longitudinally, while AZIN2, an inhibitor of the antizyme inhibitor (which counteracts ODC1), was significantly increased. Furthermore, the expression of the known platelet-derived transglutaminase39,40 EPB42, was significantly increased at the mRNA level (Fig. 7b).
The tissue level of 5-disphosphomevalonic acid, an organic pyrophosphate, decreased significantly at 7 weeks, while the levels of mevalonate kinase (MVK) and phosphomevalonate kinase (PMVK), key upstream enzymes of its metabolism41, were not significantly altered at the mRNA level, as was the case for downstream mevalonate pyrophosphate decarboxylase (MVD). This suggests that 5-disphosphomevalonic acid could be an unrecognized organic PPi species participating in calcification pathophysiology (Fig. 7c).
Lysophosphatidylethanolamine [LysoPE(0:0/18:0)], phosphatidylinositol [PI(18:0/16:0)], and phosphatidylcholine [PC(22:5/22:6)] are phospholipid metabolites. LysoPE decreased at 7 weeks, consistent with a decrease in PC (22:5/22:6), since oxidized PC
provides LysoPE via the action of the PLA2G7 enzyme42 and phospholipids are prone to degradation due to direct ROS-mediated oxidation43. On the other hand, the PI (18:0/16:0) significantly increased. The relative expression of the related genes demonstrated a significant longitudinal upregulation of PIK3R5 and PIK3CB and a less significant downregulation of CDS2, the rate-limiting enzyme of PI44, suggesting the action of excessive PI on a negative feedback loop (Fig. 7d). 3-cis-Hydroxy-b,e-Caroten-3'-one was significantly reduced in the experimental valves (Table S12).
Figure 8. (a) Pathway enrichment analysis based on the differentially abundant metabolites, except for phospholipids. Enrichment was performed using the STICTH database with the human genome as a reference. (b) Lipidomic pathway enrichment analysis based on the differential phospholipids using the LIPEA database, with the human genome as a reference.
The top 5 metabolite-derived pathways included cancer, viral infection, Wnt, pyrimidine metabolism, and TNF (Fig. 8a). The separately analyzed lipid species correlated with glycerophospholipid and inositol phosphate metabolism and signaling pathways, autophagy, tuberculosis, and ferroptosis. Additionally, these pathways and their respective genes in majority were predicted targets of the differentially expressed miRs (Fig. 7).
By combining the data of the three omics analyses, we found a set of 62 genes that were commonly deregulated at 7 versus 2 weeks in at least 2 of them and, similarly, a set of 61 cellular pathways (Fig. 9). Thus, these genes and their related pathways are expected to have a broader impact on the development of aortic valve osteogenic metaplasia. By using only the early deregulated genes obtained from the time-course analysis, a set of 17 early common genes emerged. These included key genes of AXIN2, a regulator of the Wnt pathway, as well as FOS and JUNB, components of the AP-1 transcription factor.