Conserved transcriptional co-regulation of pyrophosphate homeostasis genes governs systemic mineralization factors in mice and humans

Abstract

Inorganic pyrophosphate (PPi) is a key inhibitor of ectopic calcification, yet transcriptional regulation of genes controlling its systemic production and degradation (ABCC6, ALPL, ANKH, and ENPP1) remains poorly understood. We hypothesized that PPi homeostasis is regulated by an evolutionarily conserved transcription factor (TF) network. Promoter motif analysis combined with ATAC-seq revealed conserved enrichment of TF binding sites, including FOXA1, HNF4A, and SREBF1, across mouse and human orthologues. Bulk and single-cell RNA-seq together with RT-qPCR analyses in wild-type and Abcc6^−/− mice showed hepatocytes as major cell type expressing the most relevant genes maintaining PPi homeostasis. Further inference analysis identifies a conserved transcriptional program that regulates systemic PPi balance across mice and humans. Functionally, mice showed an age-dependent inverse correlation between plasma PPi and serum alkaline phosphatase (AP) activity, strongest during early life. Abcc6^−/− mice displayed persistently reduced but gradually increasing PPi levels and altered Pi/PPi ratios during aging. In humans, plasma PPi correlated inversely with AP activity and positively with Pi in both controls and ABCC6-deficient pseudoxanthoma elasticum patients. Together, these findings support a conserved TF-associated regulatory program linking PPi homeostasis gene expression with circulating mineralization-related factors across physiological and pathological states.

HIGHLIGHTS

• •Integrated multiomics approaches suggest that murine and human PPi homeostasis is under conserved transcriptional control.
• •Alkaline phosphatase activity governs PPi balance in mice and humans.
• •Post-weaning kinetics in PPi and AP activity suggest a therapeutic window.
• •PPi increases with age in Abcc6^−/− mice, indicating compensation.

1. Introduction

Biomineralization is a tightly regulated physiological process essential for skeletal and dental development [1]. It involves a highly orchestrated mechanism by which calcium phosphate-containing minerals are deposited in specific regions of the extracellular matrix (ECM), forming a scaffold for crystallization. This high spatial precision is partly achieved through the localized enzymatic removal of ubiquitous mineralization inhibitors by their respective degrading enzymes—a concept referred to as the stenciling principle [2].

A prime example of this is tissue non-specific alkaline phosphatase (TNAP) encoded by the ALPL gene in human and Alpl (akp2) gene in mice, which plays a pivotal role by hydrolyzing extracellular pyrophosphate (PPi) [3], a potent inhibitor of biomineralization. PPi not only regulates skeletal mineralization but also prevents hydroxyapatite crystal growth in soft tissues by binding to calcification nidi [4,5]. Under pathological conditions, ectopic calcification, the abnormal deposition of calcium phosphate in soft tissues such as arteries and cartilage, can occur, mimicking physiological mineralization [6]. Notably, ectopic calcification remains a significant unmet clinical challenge: it is currently untreatable yet highly prevalent, especially with aging [7] and in common high-burden diseases such as cardiovascular and chronic kidney disease [8].

Circulatory PPi is generated from nucleotide triphosphates (NTPs), especially ATP. Most NTPs are released in a regulated manner dependent on the activity of the ATP-binding cassette transporter ABCC6 [9], expressed predominantly in the liver and kidney, and to a lesser extent by the broadly expressed progressive ankylosis protein (homologue), ANK(H) [10]. Extracellular ATP is then rapidly converted to PPi by ectonucleotide pyrophosphatase/phosphodiesterase 1 (ENPP1), the sole, ubiquitously expressed ectoenzyme capable of hydrolyzing NTPs to PPi and nucleotide monophosphates [11].

In its membrane-bound form, TNAP is attached via a GPI anchor, while PIPLC can release it into circulation [12]. Either way, the enzyme degrades PPi [13], promoting physiological or pathological mineralization. TNAP activity is a critical determinant of bone growth and remodeling, primarily through its local effects, as osteoblast-derived AP facilitates bone mineralization by removing the protective PPi layer from hydroxyapatite crystals, enabling further crystal growth and ECM mineralization [14]. In contrast, non-bone-resident and circulating TNAP, reflected in serum alkaline phosphatase (AP) level and activity [15], may significantly impact circulatory PPi levels, as reflected in the high plasma PPi levels observed in hypophosphatasia [16,17].

Thus, circulating levels of Pi, PPi, and the PPi degrading AP activity are fundamental determinants of both pathological and physiological mineralization [18], a process that is evolutionarily conserved in vertebrates.

While age is a major factor in the progression of vascular calcification, data on the kinetics of circulatory PPi levels in both humans and animal models are limited and inconsistent. Similarly, the relationship between AP activity and plasma PPi levels has not been systematically investigated. Mendelian disorders serve as valuable models for understanding the molecular mechanisms of ectopic mineralization. One such rare disease is pseudoxanthoma elasticum (PXE, OMIM #264800), an autosomal recessive multisystem disorder characterized by slowly progressive and prototypical dermatological, ocular, and vascular calcification [19]. PXE generally results from biallelic fully or incompletely penetrant pathogenic variants in ABCC6 [20,21], though variants in ENPP1 rarely also cause the disease [22,23]. Due to impaired ATP release from hepatocytes, PXE patients exhibit markedly reduced circulatory PPi levels [[24], [25], [26]].

Since genes involved in the same biological processes are often co-regulated at the transcriptional level via evolutionarily conserved mechanisms [27,28], we hypothesized the existence of a conserved transcriptional network that governs systemic PPi levels. To explore the physiological and pathological aspects of PPi homeostasis, we first investigated whether evolutionarily conserved transcription factors have predicted binding sites in the regulatory regions of the human ABCC6, ANKH, ENPP1, and TNAP genes and their murine orthologues. We then assessed gene expression, correlation, and kinetics, along with the abundance and activity of circulating products, in wild-type and Abcc6^−/− mice as well as in human control individuals and PXE patients.

2. Materials and Methods

2.1. Transcription factor binding site prediction

2.1.1. Motif analysis

Transcription factor (TF) motifs were identified using Hypergeometric Optimization of Motif EnRichment (HOMER) v4.11 [29] and GimmeMotifs v0.24.1 [30]. Analyses were based on mm10/GRCm38 (mouse) and hg38/GRCh38 (human) assemblies. Transcription start sites (TSS) for Abcc6, Alpl, Enpp1, and Ank (and human orthologs) were extracted from Ensembl v109, with promoter regions defined as −50 kb to +1 kb relative to the TSS. Motif scanning was performed in Python 3.10 using GimmeMotifs, with hits retained at p ≤ 1 × 10⁻⁴ (FPR 0.001, the default “stringent” setting, to keep the per-site false-positive rate below 0.1 %, and has been benchmarked to give very high concordance with experimentally validated TF binding events, even when scanning large (∼10 kb) genomic regions). HOMER motif scores were converted to one-sided p-values (scoreMotifs.pl pvalue) using the same threshold. Ambiguous motif families (e.g., RXR [A/B/G]) were collapsed to a shared HGNC symbol. Only motifs detected by both tools (∼99% concordance) were considered for downstream analysis. Results were handled in pandas v2.2. TFs were considered “common” if ≥ 1 hit appeared in all four genes per species; those shared across both species were labeled “8/8 shared.”

To validate findings, promoter motif scans were repeated using the JASPAR 2024 core vertebrate database [31]. For each gene, a −10 kb to +1 kb region (Ensembl v109 TSS) was extracted, and sequences were scanned with MEME Suite v5.5.3 (FIMO p ≤ 5 × 10⁻⁵; ≈ FIMO score >10 for 10-mers), using a first-order Markov background. Both DNA strands were searched; overlapping hits (same PWM, center ≥4 bp apart) were retained. PWM-to-gene mapping used the official JASPAR2024_vertebrates.motif2gene file. Paralogous motifs (e.g., RXRA vs RXRB) were merged if binding sites showed ≥80% identity (TomTom q < 0.05). For each TF, the best p-value per promoter and total significant hits were recorded. A TF was classified as “high confidence” if it met the p-value threshold (≤1 × 10⁻⁵) in ≥1 promoter and appeared in ≥3 of 4 promoters in at least one species.

2.1.2. Transcription factor binding site prediction using mouse and human ATAC-seq datasets

Publicly available ATAC-seq datasets were retrieved from mouse and human liver samples GSM8257061, GSM8257062, GSM8257063, and GSM5214557, respectively. Reads were mapped to the mouse mm10 and human hg38 reference genomes using Bowtie2. Accessible chromatin peaks were called from the aligned BAM files using MACS3 (callpeak). Peak outputs were converted and retained in BED format for downstream analyses. For the final gene-centered regulatory analysis, accessibility was evaluated within 50-kb windows around the target loci. In this step, peaks were determined using a 3 × background threshold and overlapping or adjacent qualifying intervals were merged into consolidated peak regions. The analysis focused on the: Alpl/ALPL, Abcc6/ABCC6, Enpp1/ENPP1, and Ank/ANKH genes. To account for alternative promoter usage, TSS coordinates were derived from Ensembl annotations using hg38 and mm10. Target-gene accessibility was quantified in 1-kb bins across the TSS-centered windows.

For motif analysis only the transcription factor set identified in the previous analysis was considered. Motif occurrence analyses were performed with HOMER on the accessible peak sets associated with the target genes. No significance threshold was applied for the individual TF binding sites. The final analysis, for each of the four target genes in mouse and human liver, specifically quantified the number of accessible peaks, their positions relative to annotated TSSs, and the occurrence of TF motifs within those accessible regions.

2.2. In silico gene expression correlation analysis

We collected bulk-RNA-seq liver samples from previously published datasets [[32], [33], [34], [35], [36], [37]]. We selected GSE211975, GSE130913, GSE143206, GSE124214 and GSE174535 datasets from GEO database. Each dataset was screened manually; any treatment, genetic modification or quality failure (adapter >2 %, mapping rate <60 %) led to exclusion of the sample. In total, 63 samples remained from 8-week-old males. During raw read processing we did QC and trimming using fastp 0.23.4 (automatic adapter removal, Q ≥ 20, reads ≥30 nt). We also did quality summary using multiqc 1.13; all retained runs showed mean Q ≥ 34 and adapter contamination <1 %. For quantification, we used Salmon v1.10.1. And GENCODE M25 (release 2021–03) transcriptome as index on mm10/GRCm38. p6, decoy-aware (salmon index --keepDuplicates, k-mer = 31). The gene-level summarization and normalization were executed in R 4.3.3. Filtering and grouping were performed with dplyr. For Spearman correlation and statistics, the following thresholds were used: |ρ| ≥ 0.6 and BH-adjusted p < 0.05.

2.3. Reagents

Unless otherwise indicated reagents were obtained from Merck/MilliporeSigma (Burlington, Massachusetts, USA).

2.4. Mouse experiments

2.4.1. Laboratory animals

Wild-type (C57BL/6 J) mice were derived from animals originally purchased from Charles River Laboratories. Abcc6^−/− mice, a generous gift from prof. Arthur Bergen, were initially generated on a 129/Ola background and subsequently backcrossed to the C57BL/6 J strain for more than 10 generations. [38]. Mice were housed in IVC with unrestricted access to standard mouse chow (VRF1, Special Diets Services) and water according to standard housing conditions (12:12-h light:dark cycle, room temperature of 22 ± 2 °C). Genotyping biopsies also suitable for identification of the individuals were collected between postnatal days 8–14. Genotyping was performed using KAPA HotStart Mouse Genotyping Kit (Merck Life Sciences) following the manufacturer's protocol with primers against the mouse phosphoglycerate kinase 1 promoter or Abcc6 sequence; m PGK1 reverse: ATGTGGAATGTGTGCGAGGCC; m Abcc6 forward: TGAATCTTTCTGGGGGCC AG, m Abcc6 reverse: GTACCCTGGAGCAATCCACT. Genotyping PCR reaction was performed in a T100 Thermal Cycler (Bio Rad), with the following protocol: initial denaturation: 95 °C for 3 min; 30 cycles of denaturation: 95 °C for 15 s, annealing: 63 °C for 10 s; extension: 72 °C for 10 s; final extension: 72 °C for 30 s; Amplicons were analyzed with electrophoresis in 2% agarose gels, wild type alleles provide 163 bp, knockout 264 bp amplicons. Body weight and length (from the tip of the nose to tail base) was determined after euthanized. Experiments were carried out in accordance with the protocols approved by the Food Chain Safety and Animal Health Directorate of the Government Office of Pest County, Hungary under permit number PE/EA/01175–6/2023. Similar number of male and female mice were used in the experiments.

2.4.2. Analysis of mRNA expression

Animals were anesthetized, and up to 1 ml of blood was drawn via cardiac puncture to prepare plasma and serum samples (see later). Immediately thereafter, the circulation was perfused with 10 ml PBS to remove residual blood. Kidney and liver were directly snap frozen in liquid N₂ and stored until processing. Femur and tibia were collected on a pre-chilled aluminum platform on dry ice, and all steps were performed rapidly to preserve RNA integrity. Soft tissues were removed, and bones were cleaned in cold PBS. Trabecular bone and bone marrow were removed, the latter by centrifugation at 12,000×g for 30 s at 4 °C. Cortical bone was then snap-frozen in liquid nitrogen and stored until further processing at −80 °C. Later frozen bones were ground under liquid nitrogen to a fine powder using pre-chilled mortars and pestles, finally pulverized bone was transferred into 500 μL TRIzol reagent. Kidney and liver samples were homogenized directly in 500 μL TRIzol reagent using a teflon–glass tissue homogenizer. Total RNA from liver, kidney, and bone was extracted using the Direct-zol RNA MiniPrep Plus kit (Zymo Research) according to the manufacturer's protocol. RNA concentration and purity were evaluated using a NanoDrop One spectrophotometer (Thermo Fisher Scientific), before assessing RNA integrity on 1.5% agarose gel (1 × TBE buffer, run for 40 min at 120 V). Complementary DNA (cDNA) was synthesized from 1 μg total RNA using the RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific) with oligo (dT)18 primers according to the manufacturer's protocol. Quantitative real-time PCR was performed with TB Green Premix Ex Taq II (Tli RNase H Plus, Takara) using a CFX96 Touch Real-Time PCR Detection System (Bio-Rad, Hercules, CA, USA).

Primer sequences [39]:

• m Enpp1 forward ACCCTCAGTGGCAACTTGCGTT

• m Enpp1 reverse TGCTTGAAGGCAGGTCCATAGC

• m Abcc6 forward CACACGATGCAGCTACCAGTGA

• m Abcc6 reverse GGTCATCCAGAGAAGCCTCCAT

• m Ank forward CGTGGACTCATGCTGGCATTCT

• m Ank reverse GTTCTCGGCATTCCAGGTGACT

• m Alpl forward CCAGAAAGACACCTTGACTGTGG

• m Alpl reverse TCTTGTCCGTGTCGCTCACCAT

• m Rps29 forward CGGTCTGATCCGCAAATAC

• m Rps29 reverse GGTCGCTTAGTCCAACTTAAT

• m Actb forward GCCTTCCTTCTTGGGTATG

• m Actb reverse GCATAGAGGTCTTTACGGATG

Cycling conditions were the following: Hot start 95 °C 30 s followed by 40 cycles of 95 °C 5 s, 59 °C for 10 s, 60 °C for 20 s, followed by melting curve analysis from 65 °C to 90 °C by 0.5 °C increments every 5 s. Analysis was performed using the CFX Maestro software v.4.1 (BioRad, Hercules, CA, USA). Relative gene expression was quantified using the ΔCt method, normalized to Rps29 alone or in combination with Actb as reference genes.

2.4.3. Blood collection and processing in mice

To assess the kinetics and correlations of the AP-PPi/Pi axis in wild type and Abcc6 deficient (Abcc6^−/−) mice we used the same phlebotomy samples to prepare plasma and sera. 500–1000 μl blood was gently drawn by cardiac puncture from isoflurane anaesthetized mice via a 23G heparinized needle and divided into Li–H300 tube (20.1309, Sarstedt) and serum-300 tubes with clotting activator (20.1308.100, Sarstedt). Blood was immediately gently mixed with the anti-coagulant or clotting activator by inverting the tubes 5 times. Tubes were centrifuged at 1000 g for 10 min at room temperature (RT) and plasma or sera were collected. Plasma was subsequently centrifuged again at 1000 g for 10 min at RT. All samples were aliquoted and stored at −80 °C until further use.

2.4.4. Serum phosphate and alkaline phosphatase activity in mice

Serum concentration of inorganic phosphate (Pi) was determined using a commercial kit (MAK307-1 KT, Merck Life Science) as described earlier [40]. Serum alkaline phosphatase activity was measured in an endpoint assay in a 96-well microtiter plate. In brief, 5 μL of sample was added to a total volume of 200 μl solution per well into 96-well plates (Costar 3912). The solution contained 30 mM TRIS-Glicin buffer (pH10.4), 12.5 mM MgCl2 and 3 mM 4-Nitrophenyl phosphate disodium (pNPP) salt hexahydrate (Phosphatase substrate, S0942 Merck Life Science). The plate was sealed, and the mixture was incubated for 60 min at RT. The absorbance at 405 nm (reflecting 4-nitrophenol production) was determined on a Enspire Multimode Plate reader (PerkinElmer). Difference in absorbance between samples and blank was calculated and alkaline phosphatase activity was quantified using calibration of serial dilutions of 4-nitrophenol (pNP) sodium salt dihydrate standard (36,612 Merck Life Science) as calibration curve (0,2–12,5 mM PnP calibration).

2.4.5. Plasma inorganic pyrophosphate in mice

The PPi content of the mouse plasma was determined enzymatically, using the ATPS bioluminescent method [41], with slight modifications. Briefly, PPi was converted into ATP in an assay containing 80 μM MgCl2, 50 mM HEPES pH 7.4, 32mU/ml ATP Sulfurylase (M1017B, New England Biolabs, Ipswich, MA, USA) and 8 μM adenosine 5′-phosphosulfate (A5508, Sigma–Aldrich, Saint Louis, MO, USA). Samples were incubated at 37 °C for 30 min, followed by enzyme inactivation at 90 °C for 10 min. In the next step, ATP levels were evaluated utilizing the BacTiterGlo (Promega, Madison, WI, USA) bioluminescent assay, prepared according to the manufacturer's instructions, and analyzed in a multimode plate reader (PerkinElmer). Plasma PPi concentrations were calculated using analytical calibration standards (Biothema, Sweden) and corrected for initial plasma ATP concentrations.

2.5. Human study

2.5.1. Patient cohorts

2.5.1.1. Zero cardiac calcification cohort

Cardiac computed tomography angiography was performed with dedicated cardiac CT scanner (CardioGraphe, GE Healthcare, Chicago, IL, USA) or a third-generation dual source CT scanner (SOMATOM Force, Siemens Healthineers, Forchheim, Germany) with ECG triggered axial acquisition mode at the Heart and Vascular Centre, Semmelweis University, Budapest, Hungary [42]. Coronary, aortic valve and mitral apparatus calcium scores were measured on non-contrast images. Individuals with zero total cardiac calcification score, combining calcium loads of the three regions, were selected from prospectively enrolled patients, with exclusion criteria including bisphosphonate or vitamin K antagonist therapy, calcium, phosphate or liver metabolism disorders, diabetes, osteoporosis and impaired renal function (estimated glomerular filtration rate (eGFR) < 60 ml/min/1.73 m²). All patients gave written informed consent for CT scanning, blood sampling, and statistical data analysis. The investigation was approved by the Scientific and Research Ethics Committee of the Hungarian National Medical Scientific Council (BMEÜ/3473–1/2022/EKU). The study protocol complied with the declaration of Helsinki.

2.5.1.2. PXE cohort

Seventy-one PXE patients were included from the PXE Reference Center of the Center of Medical Genetics (CMGG) of the Ghent University Hospital. In all PXE patients, diagnosis of PXE was histologically confirmed by performing Von Kossa staining on a lesional skin biopsy to identify mid-dermal elastic fiber calcification and fragmentation as well as by identifying bi-allelic pathogenic variants in the ABCC6 gene. All patients gave informed consent, and the study was approved by the Ethical Committee of the Ghent University Hospital, Ghent, Belgium. The study protocol complied with the declaration of Helsinki.

2.5.2. Blood sampling and analysis in humans

Blood from patients included in the non-calcified cohort in the Heart and Vascular Centre, Semmelweis University, was drawn into serum separation vacuum tubes for the assessment of Pi and AP activity, and to K₃EDTA vacuum tubes for plasma preparation for PPi concentration assessment as described previously [42]. Serum samples were analyzed following the routine procedures of the Heart and Vascular Centre. The plasma fraction from the K₃EDTA tube was isolated via two consecutive centrifugations (1000 g for 5 min at room temperature, repeatedly) and transferred into platelet separation tubes (Vivaspin Filtrate, 300 kDa, 13279-E, Sartorius, Göttingen, Germany). The platelet-free plasma for PPi assessment was obtained by further centrifugation at 2200g for 30 min at room temperature. Samples were stored at −80 °C until PPi measurement. In the Belgian PXE patient cohort blood samples were collected into vacuum citrate tubes, the plasma fraction was isolated via centrifugation (1000 g for 10 min at at 4 °C). Pi levels and AP activity were determined according to the routine procedures of the Ghent University Hospital. For PPi determination plasma was transferred into platelet separation tubes (Vivaspin Filtrate, 300 kDa, 13279-E, Sartorius, Göttingen, Germany), and platelet-free plasma was obtained by further centrifugation at 2300g for 35 min at 4 °C and stored at −80 °C until measurement.

2.5.3. Plasma inorganic pyrophosphate in humans

The PPi content of the zero calcification individuals was determined as described above for mice [41]. The PPi content of the PXE patients was determined similarly using the two-step ATPase luminescence assay [25]. Briefly, PPi was converted into ATP in an assay containing 104 μM MgCl2, 31 mM HEPES pH 7.4, 825mU ATP Sulfurylase (M0394L, Bioké NV, Leiden, The Netherlands) and 64 μM adenosine 5′-phosphosulfate (SC-214506, Bio-Connect, Huissen, The Netherlands). Samples were incubated at 37 °C for 30 min, followed by enzyme inactivation at 90 °C for 10 min. In the next step, ATP levels were evaluated utilizing the BacTiterGlo (G8230, Promega, Leiden, The Netherlands) bioluminescent assay, prepared according to the manufacturer's instructions, and analyzed in a multimode plate reader (Glomax). Plasma PPi concentrations were calculated using an internal calibration curve and corrected for initial plasma ATP concentrations.

2.6. Statistical analysis and mathematical modeling

Variables were tested for normality using the Shapiro–Wilk test. Normally distributed variables are presented as mean ± SEM, whereas non-normally distributed variables as median [interquartile ranges]. Variables with normal distribution are compared using unpaired two-tailed Student's T-test, those with non-normal distribution are compared using the two-tailed Mann–Whitney U test. Pearson coefficients are reported for normally distributed variables, while Spearman's rank correlation coefficients are used for non-normally distributed variables. Univariate and multivariate linear regressions were performed in R (glm, MASSpolr). One Abcc6^−/− mouse was excluded as an outlier and altogether 30 mice due to missing values, leaving 91 Abcc6^−/− and 50 wild-type mice for analysis; 66 adult PXE patients were also included. Independence and multicollinearity were assessed (Durbin–Watson test; VIF <5), and all models met criteria. Linearity, normality, and goodness-of-fit were evaluated using residual and Q–Q plots. Statistical significance was determined at the 0.95 confidence level (p < 0.05). Analyses were performed in GraphPad Prism 10.5.0 or in R (v4.3.3).

2.7. Computational analysis

2.7.1. Mouse snRNA-seq dataset

Single-nucleus RNA sequencing data were obtained from Yin et al. [43], which profiled hepatocyte polyploidy and ageing in mouse liver. The publicly available pre-processed and annotated AnnData object was filtered to retain only wild-type C57 B l/6 J animals, spanning both young (3 months) and old (22 months) age groups, with cell type labels and UMAP coordinates as provided by the publication. No additional preprocessing, normalization, or batch correction was applied, as quality metrics were within standardized ranges.

2.7.2. Mouse and human liver cell atlas datasets

Raw count matrices and cell-level metadata for mouse and human liver were obtained from Guilliams et al. [44]. Data were loaded using SciPy (v 1.15.2) and assembled into AnnData objects in Scanpy (v 1.11.5). The datasets contain cells profiled by multiple technologies (scRNA-seq and CITE-seq); technology-appropriate mitochondrial content thresholds were applied during quality control (<5% for scRNA-seq and CITE-seq; <1% for snRNA-seq). Cells exceeding 20,000 total counts and genes detected in fewer than 3 cells were excluded. Counts were normalized to 10,000 per cell and log1p-transformed and scaled. Pre-existing UMAP coordinates and cell annotation from the original atlas were retained.

2.7.3. Expression and transcription factor co-expression analysis

Expression of Enpp1/ENPP1, Ank/ANK, Abcc6/ABCC6, and Alpl/ALPL was visualized across annotated cell types using dot plots (fraction expressing: dot size; mean expression: color intensity). Cells were classified as hepatocytes or non-hepatocytes to contrast parenchymal versus non-parenchymal expression. For the Yin et al. dataset, expression analysis was performed on young animals, with old animals included exclusively for the age-stratified comparison of hepatocyte and non-hepatocyte expression shown in Figure 3A (right, bottom). To assess transcriptional co-regulation, Pearson correlation coefficients were computed between the four genes and the hepatocyte-enriched transcription factors Nr4a1, Esr1, Rxra, Srebf1, Foxa1, Hnf4a, Cebpb, and Nr1h3, using cell-type-averaged normalized expression values across all annotated populations. Results were visualized as annotated heatmaps using seaborn with a coolwarm colormap centered at zero. All analyses were performed in Python (v3.10.18) using Scanpy (v 1.11.5), pandas (v 2.3.3), NumPy (v 2.2.6), seaborn (v 0.13.2), and matplotlib (v 3.9.4).

Figure 3.

Hepatocyte-specific expression of Enpp1, Ank, Abcc6, and Alpl across liver single-nucleus and single-cell RNA sequencing datasets and transcription factor co-expression analysis.

(A–C) For each dataset, panels show (left) UMAP embedding colored by cell type, (center-left) stacked bar chart of cell type proportions, (center-right) dot plot of Enpp1/ENPP1, Ank/ANK, Abcc6/ABCC6, and Alpl/ALPL expression across cell types (dot size, fraction expressing; color intensity, mean expression), and (right) heatmap of mean expression in hepatocytes (Hep) versus non-hepatocytes (Non-Hep) with feature UMAP showing combined PPi pathway gene set score (B–C only). (A) SMART-seq-based snRNA-seq of mouse liver (Yin et al., 2024) from young (Y – 3 months) animals. Violin plots (bottom right) show per-cell expression distributions stratified by age (Y – 3 months and O – 22 months). (B) Mouse liver cell atlas (Guilliams et al., 2022). (C) Human liver cell atlas (Guilliams et al., 2022). (D) Heatmaps of Pearson correlation coefficients between the four genes and hepatocyte-enriched transcription factors (NR4A1, ESR1, RXRA, SREBF1, FOXA1, HNF4A, CEBPB, NR1H3) in human (left) and mouse (right), computed on cell-type-averaged normalized expression. Tile values indicate correlation coefficients; red indicates positive co-expression.

Y, young; O, old; TF, Transcription factor; Hep, Hepatocytes; Non-Hep, Non-hepatocytes; HB, Hepatobiliary cells; Endo, Endothelial cells; KC & DC, Kupfer cells and Dendritic cells; SC, Stellate cells; P. B cells, Plasma B cells; NK, Natural Killer cells; ILC1s, Innate lymphoid cells type 1; HsPCs, Haematopoietic stem and progenitor cells; pDCs, Plasmacytoid dendritic cells; cDC1s/cDC2s, Conventional dendritic cells type 1/2; Mig. cDCs, Migratory conventional dendritic cells; MD cells, Monocytes & Monocyte-derived cells; Macs, Macrophages; Cho, Cholangiocytes; PC, Plasma cells.

3. Results

3.1. Conservation of transcription factor binding sites suggest coordinated hepatic expression of PPi homeostatic genes

We hypothesized that PPi homeostasis is regulated through evolutionarily conserved transcriptional gene expression. Therefore, we first analyzed transcription factor (TF) binding sites in the regulatory regions (−50 kb/+1 kb from the transcription start site (TSS)) of Abcc6, Alpl, Ank, and Enpp1 and their human orthologues using the bioinformatics tool HOMER [29]. Since ALPL/Alpl have tissue-specific TSS, we focused on the liver-specific data unless specified otherwise. HOMER identified several shared motifs across orthologous promoters, with NR4A1, ESR1, NR1H3/LXRa, RXRA, and SREBF1 present in both species, while FOXA1, HNF4A and CEBP/B detected only in human promoters (Figure 1A). To validate these results, we applied the independent JASPAR tool [31]. Four TFs: NR4A1, ESR1, RXRA, and SREBF1, were consistently detected in both species by JASPAR as well. FOXA1, CEBP/B and HNF4A were reproducibly identified in human by both tools. Similarly to HOMER, JASPAR was unable to detect FOXA1 and CEBP/B motifs in mice, while NR1H3/LXRa was detected in both murine and human only by HOMER.

Figure 1.

Shared transcription factor motifs and co-expression of PPi homeostatic genes in mice and human.

(A) Promoter motif analysis of murine Abcc6, Alpl, Ank, and Enpp1 and their human orthologues using HOMER. Heatmap shows significant per-gene enrichments with color intensity representing −log₁₀ p-values. “#” indicates motifs supported by both HOMER and JASPAR; “&” indicates motifs identified only by JASPAR. (B) ATAC-seq results from summarized data obtained from publicly available mouse datasets (GSM8257061, GSM8257062, GSM8257063). Regions of ±50 kb around TSS of the PPi homeostasis genes are shown. The box below the ATAC-seq data indicates the locations of predicted TF binding sites shown on panel A, without significance threshold.

CEBPB, CCAAT/enhancer-binding protein beta; Enpp1/ENPP1, ectonucleotide pyrophosphatase/phosphodiesterase 1; ESR1, estrogen receptor 1; FOXA1, Forkhead box protein A1; HNF4A, hepatocyte nuclear factor 4 alpha; NR1H3/LXRα, nuclear receptor subfamily 1 group H member 3 (liver X receptor alpha); NR4A1, nuclear receptor subfamily 4 group A member 1; RXRA, retinoid X receptor alpha; SREBF1, sterol regulatory element binding transcription factor 1.

To further confirm our findings, we used the multi-tissue multiomics-based signaling pathway project webserver (https://www.signalingpathways.org) [45], which generally reports data from 2500bp distance to the TSS. The ChIP-seq based analysis of the PPi homeostasis genes confirmed the binding of NR4A1, ESR1, RXRA, SREBF1, HNF4a, and CEBP/B, but only in mice. According to this analysis NR1H3/LXRa was only detected in the mouse Ank and the human ABCC6 gene (Supplementary Fig. 1). The human TF binding sites were surprisingly much less retrievable by the web server; therefore, we decided to continue our analysis using publicly available bulk ATAC-seq datasets. In mouse and human liver ATAC-seq data we identified open chromatin regions in a 50 kbp window from the TSS of each of the four genes. We created bins of 1 kbp length and determined whether they contained ATAC-seq peaks. We next identified whether the bins having an ATAC-seq peak also contain any of the previously predicted binding sites. As shown in representative peaks and predicted TF binding sites in Figure 1B, we identified colocalization of binding sites of almost all previously predicted sites. However, we were unable to detect the colocalization of NR1H3/LXR1 in the ATAC-seq peaks in mouse and human liver samples for any of the four genes. ESR1 binding sites were not predicted in the proximal promoters either in human or in mice.

Together, these analyses demonstrate conserved enrichment of nuclear receptor–related factors, with SREBF1, FOXA1 and HNF4A appearing systematically in all regulatory regions.

3.2. Correlated expression of the PPi homeostasis genes in mouse liver

To validate our findings on the predicted co-regulation of the four PPi homeostasis genes we first analyzed publicly available datasets. We generated a unified dataset composed of 63 bulk RNA-seq samples from 8-week-old male mouse livers [[32], [33], [34], [35], [36], [37]]. In this unified murine dataset, we found significant high pairwise Spearman correlation (ρ > 0.80, p ≤ 3.4 × 10⁻⁶) between the expression levels of all the four PPi homeostasis genes (Figure 2A).

Figure 2.

Hepatic expression and correlation of PPi homeostatic genes in mice.

(A) Significant pairwise expression correlations of PPi homeostatic genes in murine RNA-seq datasets (8-week-old males, n = 63). Heatmap depicts −log₁₀ p-values by color intensity, with circle size proportional to Spearman correlation coefficient (ρ), relative to ρ = 1 along the diagonal. (B) Expression of PPi homeostatic genes in livers of 60-day-old wild-type and Abcc6^−/− littermates (n = 11 per group), normalized to Rps29. Data are shown as mean ± SEM, except Abcc6 (median ± IQR). Statistical comparisons: Mann–Whitney U test for Abcc6, unpaired two-tailed Student's t-test for other genes. (C) Hepatic expression of PPi homeostatic genes in 60-day-old littermates (n = 22). Pairwise relationships with Pearson correlation coefficients (r) and linear regression with corresponding two-tailed p-values.

Abcc6, ATP binding cassette subfamily C member 6; Alpl, alkaline phosphatase, liver/bone/kidney; Ank, progressive ankylosis protein; Enpp1, ectonucleotide pyrophosphatase/phosphodiesterase.

To experimentally validate in silico results we next analyzed in vivo the gene expression of 60-day old mice by real-time quantitative polymerase chain reaction (RT-qPCR). We first investigated if the expression of the four genes (Abcc6, Alpl, Ank and Enpp1) is different in the liver or the kidney of Abcc6 wild type and Abcc6^−/− littermates from Abcc6 ⁺/⁻ breeding pairs. We did not detect significant difference in the gene expression of the four genes of wild type and Abcc6^−/− littermates (Figure 2B). Importantly, the disrupted Abcc6 allele was similarly expressed to that of the wild type at the mRNA level, although the knock-out transcript cannot be translated to a functional ABCC6 protein [38]. As we found no significant genotype-specific difference in the qPCR experiments, we plotted the wild type and Abcc6^−/− mice together and analyzed the Pearson correlations between the expression of the four genes. In the liver, the four PPi homeostatic genes displayed significant pairwise correlations (n = 22, Pearson r > 0.5; Figure 2C). The strongest associations were observed between Enpp1 and Alpl or Ank (r > 0.8, p < 0.0001), followed by Ank with Abcc6 or Alpl (r > 0.7, p < 0.0001). The weakest, though still robust, correlations were between Abcc6 and Alpl or Ank (r > 0.56, p < 0.006). Similarly, in the kidney we detected strong correlations between the four genes (n = 23, Spearman ρ > 0.7, p < 0.0001) as shown in Supplementary Fig. 2A. Interestingly, although Alpl expression was approximately 10-fold higher in bone than in the liver, both remained stable across maturation and aging in both wild-type and Abcc6 knockout mice in both examined tissues (Supplementary Fig. 2B).

3.3. Hepatocytes exhibit cell-type-specific expression of PPi regulatory genes

The correlated expression of PPi homeostasis genes in bulk liver tissue, and its stability across maturation and ageing, raised the question of which hepatic cell type drives this expression. To address this, we analyzed a publicly available full-length single-nucleus RNA sequencing (snRNA-seq) dataset from young and old C57BL/6 J wild-type mice, which allowed us to resolve gene expression at cell-type resolution (Yin et al., 2024; Figure 3A). Across all major liver cell populations, Enpp1, Ank, Abcc6, and Alpl expression was strongly enriched in hepatocytes, with minimal signal detected in non-parenchymal cell types including endothelial cells, Kupfer cells, stellate cells, B and T cells (Figure 3A, dot plot). This hepatocyte-dominant expression pattern was broadly maintained across both young and old animals; however, stratification by age revealed a notable reduction in Abcc6 expression in old compared to young hepatocytes, while the remaining genes followed similar trends between age groups (Figure 3A, violin plots). This suggests that although the hepatocyte remains the primary site of PPi regulation throughout life, Abcc6 expression may be selectively vulnerable to age-related decline.

Because non-parenchymal cells (NPCs) are underrepresented in standard snRNA-seq and require specialized cell isolation protocols for adequate single-cell profiling, we complemented our analysis with integrated liver cell atlases that combine multiple sequencing technologies and enrichment strategies. To assess whether the observed pattern represents a general feature of liver biology rather than being specific to our dataset or model, we examined the mouse and human liver cell atlases (Guilliams et al., 2022), which provide broader and more balanced cell type representation across species. Similarly, all four genes involved in PPi homeostasis were predominantly expressed in hepatocytes, with the dot plot and feature UMAP confirming co-enrichment within the parenchymal compartment and near-absence in non-parenchymal populations (Figure 3B). The robustness of this finding across two independent mouse datasets, generated with different sequencing approaches and representing different biological contexts, strongly supports hepatocytes as the principal cell type governing PPi homeostasis in the liver.

Finally, to determine whether this hepatocyte-specific expression is conserved in humans, we interrogated the corresponding human liver cell atlas (Guilliams et al., 2022; Figure 3C). ENPP1, ANK, ABCC6, and ALPL displayed a strikingly similar expression pattern, with expression concentrated in hepatocytes and the combined gene set score mapping mostly to the hepatocyte cluster in the feature UMAP. The conservation of this cell-type-specific co-expression across species reinforces the conclusion that the hepatocyte is the central executor of extracellular PPi homeostasis. These findings are further supported by transcription factor co-expression analysis, in which cell-type-averaged expression of all four genes correlated positively with the same set of hepatocyte-enriched transcriptional regulators in both species (Figure 3D). In human liver, ENPP1, ABCC6, and ALPL showed particularly strong correlations with ESR1 (r = 0.94, 0.93, and 0.96, respectively), HNF4a (r = 0.92, 0.96, and 0.79), and FOXA1 (r = 0.84, 0.88, and 0.67). ANK displayed a broadly similar pattern, with the strongest associations observed for NR1H3 (r = 0.94), FOXA1 (r = 0.84), and HNF4a (r = 0.87). In mouse liver, correlations were generally more moderate but directionally consistent: Abcc6 and Alpl showed the strongest associations with SREBF1 (r = 0.90 and 0.91) and HNF4a (r = 0.89 and 0.87), while Enpp1 correlated most strongly with RXRA (r = 0.65) and NR1H3 (r = 0.69). Ank showed a more distributed pattern, with moderate correlations across SREBF1 (r = 0.68), FOXA1 (r = 0.58), and HNF4a (r = 0.60). Taken together, the convergence on overlapping TF sets across both species is consistent with shared transcriptional control of PPi homeostasis within the hepatocyte compartment.

3.4. Hnf4a deletion leads to downregulation of key PPi homeostasis genes

To independently validate the predicted transcriptional regulators of the PPi gene network, we surveyed publicly available liver transcriptomic datasets from TF knockout mice. Suitable datasets were scarce, limiting the number of regulators that could be evaluated. No consistent differential expression of PPi homeostasis genes was detected in the available Srebf1c (GSE100827) [46], Esr1 (GSE174535) [37], or Rxra (GSE317638) knockout datasets. We did not find a publicly available FoxA1 single knockout transcriptome dataset, however, in the FoxA1/A2 double KO mice (GSE140423) [47], Alpl expression was significantly reduced (log2fold-change (log2FC) = −1.75, q = 1.5 × 10⁻¹²), whereas Abcc6, Enpp1, and Ank expression was not altered. Furthermore, in the liver-specific Hnf4a knockout dataset (GSE210842) [48], three of the four PPi genes were significantly downregulated, including Abcc6 (log2FC = −5.43, q = 1.8 × 10⁻⁴), Enpp1 (log2FC = −1.94, q = 1.8 × 10⁻⁴), and Alpl (log2FC = −0.65, q = 3.5 × 10⁻⁴), albeit Ank was unchanged. Together, these independent knockout datasets provide additional functional support for the transcriptional network inferred from TF binding site analyses and co-expression studies. In particular, the marked downregulation of Abcc6, Enpp1 and Alpl following Hnf4a deletion identifies HNF4A as a major upstream regulator of hepatic PPi homeostasis.

3.5. Kinetics and correlation of circulatory PPi homeostatic factors in wild type and Abcc6^−/− mice

Since PPi homeostatic genes exhibited significantly correlated expression, we next examined how this coordinated regulation translates into circulating mineralization products. We analyzed serum AP activity and plasma PPi concentration in wild-type and Abcc6^−/− mice. Between weaning at postnatal day 21 (PND21) and 8 weeks of age, body mass increased 2.1–2.5-fold in a near-linear manner, respectively, coinciding with bone growth reflected by a quasi-linear gain in body length that plateaued at early adulthood (Supplementary Fig. 3A). We observed a significant decline in serum AP activity from weaning to 8 weeks of age in both wild-type and Abcc6^−/− mice, while plasma PPi levels increased substantially (∼1.5–2-fold, respectively) (Figure 4A), along with body weight (Supplementary Fig. 3A). Although AP activity did not differ between genotypes with age (Supplementary Fig. 3B), plasma PPi in Abcc6^−/− mice remained approximately half of wild-type levels throughout, consistent with the absence of functional ABCC6 (Supplementary Fig. 3C). Nevertheless, remarkably, PPi in Abcc6^−/− mice increased twofold by 8 weeks, similar to the >1.5-fold rise in wild-type mice (Figure 4A). No sex-specific differences were detected (Supplementary Fig. 3B and C).

Figure 4.

Relationships between alkaline phosphatase activity and plasma PPi concentration in young wild-type and Abcc6^−/− mice.

(A) Serum AP activity and plasma PPi levels in 3- to 8-week-old wild-type and Abcc6^−/− mice. Box plots show medians with interquartile ranges, minimum and maximum values, and individual data points. Significancy was tested using Mann–Whitney U test (p < 0.05). (B) Correlation of plasma PPi concentration with AP activity in young wild-type (n = 41) and Abcc6^−/− (n = 31) mice between postnatal day (PND) 21 and PND 62. Linear regression with 95% confidence intervals is indicated. Pearson correlation coefficients (r) and two-tailed p-values are indicated. Teal diamonds represent wild-type and purple circles Abcc6^−/− mice; transparency allows visualization of overlapping points.

Abcc6, ATP binding cassette subfamily C member 6; AP, alkaline phosphatase; PPi, pyrophosphate.

Given the concerted expression of PPi homeostatic genes, we investigated the correlations between circulatory mineralization factors, and found that in young Abcc6^−/− mice, PPi levels strongly and inversely correlate with AP activity (Spearman ρ ≈ −0.5, p = 0.005, n = 31), whereas wild-type mice show a weaker, yet significant, negative correlation (ρ ≈ −0.3, p = 0.045, n = 41) (Figure 4B). Extending the analysis to adult wild-type mice (up to PND200, n = 72), serum AP activity declined sharply during maturation but remained stable thereafter (Figure 5A). A two-phase exponential model fits AP activity, with strong negative correlation with age (ρ ≈ −0.85, p < 0.0001). Plasma PPi and the Pi/PPi ratio remain unchanged with age, while AP activity shows a weak positive correlation with Pi/PPi (ρ ≈ 0.25, p = 0.040) but not with PPi (Figure 5B). Univariate linear regression analysis identified age as the only significant, negative contributor to the Pi/PPi ratio (p = 0.029), but this effect was lost in the multivariate linear regression model including sex and serum AP activity.

Figure 5.

Age-dependent changes and correlations between serum AP activity, plasma PPi, and the Pi/PPi ratio in wild-type mice.

(A) Age dependence of serum alkaline phosphatase (AP) activity and Pi/PPi ratio. (B) Correlation between serum AP activity and PPi or the Pi/PPi ratio. Linear regression with 95% confidence intervals is shown. Spearman correlation coefficients (ρ) and corresponding two-tailed p-values are indicated at significant correlations. Teal diamonds represent individual wild-type mice (n = 77); transparency allows visualization of overlapping points.

AP, alkaline phosphatase; PPi, pyrophosphate; Pi/PPi, serum phosphate/plasma pyrophosphate ratio.

In Abcc6^−/− mice up to 200 days, AP activity declined markedly with age, however, unlike wild-type mice, PPi levels positively, while the Pi/PPi ratio negatively correlated with age. Extending the analysis to 600 days (n = 101) confirmed that AP activity fits a two-phase exponential curve (ρ ≈ −0.83, p < 0.0001) (Figure 6A) similar to wild type. However, unlike the wild-type mice, Abcc6^−/− mice show a persistent age-dependent increase in PPi (ρ ≈ 0.54, p < 0.0001) and a robust decline in the Pi/PPi ratio (ρ ≈ −0.6, p < 0.0001) throughout the lifespan (Figure 6A). Furthermore, AP activity correlates inversely with PPi (ρ ≈ −0.4, p < 0.0001) and positively with the PPi/Pi ratio (ρ ≈ 0.39, p < 0.0001) (Figure 6B). No sex-specific differences were identified (Supplementary Fig. 4), although females had slightly higher PPi level throughout. Univariate linear regression models revealed an inverse correlation of the Pi/PPi ratio with age (p < 0.0001) and a positive correlation with AP activity (p < 0.0001). In multivariate linear regression adjusted for age and sex (p = 1.3 × 10⁻⁷, adjusted R² = 0.305), AP activity remained a significant predictor of the Pi/PPi ratio (p < 0.0001) explaining over 30% of variance in Pi/PPi ratio, whereas age no longer contributed.

Figure 6.

Age-dependent changes and correlations between serum AP activity, plasma PPi, and the Pi/PPi ratio in Abcc6^−/− mice.

(A) Age dependence of serum alkaline phosphatase (AP) activity, PPi concentration and Pi/PPi ratio. (B) Correlation between serum AP activity and PPi or the Pi/PPi ratio. Linear regression with 95% confidence intervals is shown. Spearman correlation coefficients (ρ) and corresponding two-tailed p-values are indicated. Purple circles represent individual Abcc6^−/− mice (n = 101); transparency allows visualization of overlapping points.

Abcc6, ATP binding cassette subfamily C member 6; AP, alkaline phosphatase; PPi, pyrophosphate; Pi/PPi, serum phosphate/plasma pyrophosphate ratio.

3.6. Correlation between PPi homeostatic factors in human

Next, we investigated the factors of the PPi homeostasis in humans. To generate a cohort with individuals without cardiac ectopic calcification, we prospectively recruited patients referred for cardiac CT and included them into our study if they had zero cardiac calcification. In total we included 26 patients, including 18 males, with a broad age distribution (median age = 55 years; IQ range 49.5, 55.25 years). In these patients, we found no significant correlation of serum AP activity, plasma PPi or serum Pi levels with age. However, plasma PPi concentration exhibited a strong inverse correlation with AP activity (Spearman ρ ≈ −0.47, p = 0.014), while the Pi/PPi ratio positively correlated with AP activity (Pearson r ≈ 0.45, p = 0.020). In addition, in these individuals the plasma PPi level positively correlated with serum Pi concentration (Spearman ρ ≈ 0.61, p = 0.0008) (Figure 7A). Given the small cohort size (n = 26), no mathematical models were generated.

Figure 7.

Correlations of AP activity, PPi, and Pi/PPi ratio in human.

(A) Correlation between serum AP activity and PPi or the Pi/PPi ratio, or PPi and Pi levels in adult patients with zero cardiac calcification score (n = 26). (B) Correlation between serum AP activity and PPi or the Pi/PPi ratio, or PPi and Pi levels in adult PXE patients (n = 66). Blue upward triangles represent individual patients without detectable cardiac calcification; red downward triangles represent PXE patients. Transparency allows visualization of overlapping points. Linear regression with 95% confidence intervals is shown. Spearman or Pearson correlation coefficients (ρ or r) and corresponding two-tailed p-values are indicated at significant correlations.

AP, serum alkaline phosphatase; PPi, pyrophosphate; Pi, serum phosphate; Pi/PPi, serum phosphate/plasma pyrophosphate ratio; PXE, pseudoxanthoma elasticum.

We were also interested in the above correlations in ABCC6 deficient pseudoxanthoma elasticum patients. The Belgian PXE cohort comprised of 71 individuals including five children, whom we excluded from the analysis. The study cohort comprising 66 adult PXE patients including 25 males and 41 females had a median age of 47.5 years (IQ range 32.5, 54.25 years). Similarly to the patient cohort with zero cardiac calcification, in the PXE cohort no significant correlations were found between age and AP activity, PPi (Supplementary Fig. 5), or Pi levels neither in males nor in females. Furthermore, we only found a significant but weak negative association between AP activity and PPi levels (Pearson r ≈ −0.26, p = 0.038) and a somewhat stronger positive correlation between AP activity and Pi/PPi levels (Pearson r = 0.32, p = 0.009) (Figure 7B), present in both males and females (Supplementary Fig. 5). In PXE patients, univariate analysis identified AP activity as a significant contributor to the Pi/PPi ratio (p = 0.009), unlike age or sex. The association of the Pi/PPi ratio with AP activity remained significant (p = 0.018) in multivariate linear regression adjusted for age and sex. Although the model explained a modest proportion of variance (adjusted R² = 0.077), this finding highlights AP activity as an independent factor influencing the Pi/PPi ratio in PXE.

4. Discussion

4.1. Conserved and correlated transcriptional regulation of PPi homeostasis genes

Genes in key physiological pathways are often co-regulated by shared transcription factors (TFs). In skeletal tissues concerted regulation of PPi by Alpl, Enpp1, and Ank has been observed in knockout murine models [17] similarly to co-expression of Type I collagen and Tnap [49]. We hypothesized that PPi homeostasis genes are coordinately regulated hence share evolutionarily conserved TF binding sites in regulatory regions. Via motif analysis ESR1, NR4A1, RXRA, and SREBF1 were consistently identified in mice and human across analytical tools, supporting their role as core conserved regulators of Abcc6/ABCC6, Alpl/ALPL, Ank/ANKH, and Enpp1/ENPP1. FOXA1 and CEBP/B binding sites, however, were absent and HNF4A only sporadically detected in mouse promoters, whereas all three were robustly identified in human datasets, consistent with previously published experimental evidence indicating CEBP/B, HNF4A, and RXRa-mediated regulation of the human and murine Abcc6 promoters [[50], [51], [52], [53], [54], [55]]. Notably, we also identified a FOXO1 binding motif (enrichment 3.52, p < 10⁻⁶⁰) in the promoter region of murine Alpl, in agreement with previous studies reporting FOXO1-dependent ALPL expression contributing to ectopic calcification [56,57]. However, the restriction of FOXO1 to Alpl in our dataset indicates gene-specific regulatory inputs superimposed on shared transcriptional control.

To assess the potential functional relevance of the predicted TF binding sites, we integrated motif analysis with chromatin accessibility data. Bulk ATAC-seq analysis suggested the transcriptional regulation of PPi genes through a shared core set of TFs localizing to open chromatin regions in liver. However, lack of detectable colocalization for NR4A1 in ABCC6, ALPL, and ENPP1, as well as limited support for NR1H3/LXRa binding (restricted to human ALPL and ENPP1), suggests that these factors might act in distinct cell states or context-dependent manner not captured in bulk datasets. The integrated findings are consistent with the established roles of nuclear receptors in metabolic regulation, and previous reports providing direct evidence for ESR1 binding to ALPL enhancers [58] and RXRA to ABCC6 promoters [54] linking ESR1 and RXRA to vascular calcification [59,60] and bone metabolism [61].

Importantly, in agreement with the predicted shared transcriptional regulation inferred from TF binding site enrichment and open chromatin profiles, publicly available bulk RNA-seq datasets revealed co-expression of Abcc6, Alpl, Ank, and Enpp1 in mouse liver. Furthermore, according to single-cell transcriptomes, all four genes are enriched in hepatocytes, and correlate with a common set of TFs across both mice and humans, corroborating the shared regulatory control. This coordinated expression pattern was independently validated by RT-qPCR in both liver and kidney of wild-type and Abcc6^−/− mice, where all four genes showed robust pairwise correlations irrespective of the genotype. Analysis of independent TF knockout datasets provided additional experimental support. The strongest evidence was observed for HNF4A, whose deletion resulted in marked downregulation of Abcc6, Enpp1, and Alpl supporting HNF4A as an upstream regulator of the hepatic PPi gene program. The particularly strong reduction of Abcc6 further agrees with previous studies identifying HNF4A as a major regulator of hepatocyte-specific gene expression [[51], [52], [53],55].

The observed substantial co-expression of Enpp1 and Alpl may initially appear paradoxical given the opposing enzymatic activities of ENPP1 (PPi generation) and TNAP (PPi hydrolysis). Of note, we observed a significantly lower Alpl expression relative to bones. The transcriptional coupling of Enpp1 and Alpl in the liver may reflect their shared evolutionary origin within the alkaline phosphatase superfamily, as ENPPs and APs derive from a common ancestral protein and retain conserved catalytic features [62]. However, such coordination suggests a fine-tuned regulatory system controlling local and systemic PPi levels. A plausible explanation is that the simultaneous expression of ENPP1 and TNAP enables tight buffering of extracellular PPi concentrations, preventing both systemic deficiency and local accumulation that could promote calcium pyrophosphate precipitation. In this context, TNAP may act to spatially and/or temporally restrict PPi produced by ENPP1, maintaining homeostasis within a narrow physiological range via for instance additional regulation through FOXO1. Subcellular compartmentalization may add further explanation/complexity. Previous studies have shown that TNAP localizes predominantly to hepatocyte and cholangiocyte membranes facing bile canaliculi and ducts, as well as to endothelial cells in the portal triad [63,64] suggesting complementary roles in bile excretion and in the processing of circulating molecules [65]. This spatial distribution permits the notion that ENPP1 and TNAP may partly operate in distinct microenvironments within the liver, thereby functionally separating PPi production and hydrolysis partially. Importantly, the high level of Ank expression in the liver also showing correlated expression with Enpp1 hints for the first time, that the 30–40 % of total plasma PPi level attributed to ANK function [10,66] may substantially originate from this organ.

Together, these findings support a model in which a conserved TF network coordinates PPi metabolism genes in a context- and cell type-dependent manner. Consistent with the central role of the liver in systemic PPi homeostasis, hepatocytes emerge as the primary contributors, while additional tissues such as the kidney may have complementary role. The highly similar expression patterns observed between mouse and human further indicate strong evolutionary conservation.

4.2. Age-dependent coordination of AP activity and PPi levels in mice

Plasma PPi kinetics have not been investigated systematically in young mice, although TNAP is known to be highly expressed in early life, resulting in elevated AP activity [67,68]. Given that PPi is a direct TNAP substrate, we hypothesized that plasma PPi availability may be influenced during this developmental window, a possibility not previously addressed. Supporting this rationale, in lambs we previously reported low postnatal PPi levels that rose sharply to reach adult concentrations by weaning [69]. In our murine models AP activity exhibited a biphasic decline with age, stabilizing in adults, in line with data from 90 to 135-day old mice [70], while plasma PPi levels also plateaued. These dynamics are consistent with TNAP-mediated hydrolysis of PPi. Another novel finding of the present work is the strong reciprocal relationship between plasma PPi concentration and serum AP activity during post-weaning maturation, with a rapid decline in AP activity paralleled by a doubling of plasma PPi levels. Of note, given that circulating AP in mice is predominantly bone-derived [63], the marked decline in circulating AP activity during aging cannot be explained by transcriptional changes in Alpl, as its expression remains stable in both liver and bone. Instead, post-transcriptional or post-translational mechanisms are likely involved, including age-dependent differences in TNAP shedding or release, as well as changes in enzyme stability, turnover, activation [71] or tissue-specific processing [15], which warrant further investigation.

Nevertheless, our work suggests that developmental regulation of AP activity critically shapes extracellular PPi availability in early life. Lambs showing a similar postnatal increase in PPi raises the possibility that such dynamics represent a conserved developmental program, which support our notion on the concerted transcriptional regulation of the PPi homeostatic genes. Taking into consideration the potency of prenatally supplemented PPi to halt ectopic calcification in Enpp1 knock-out offspring [72], our results highlight early age as a critical therapeutic window to combat later-in-life-manifesting ectopic calcification.

4.3. Divergent regulation in wild-type and Abcc6^−/− mice

Although wild-type and Abcc6^−/− animals followed broadly similar developmental trends, the knockout mice consistently showed ∼50% lower plasma PPi and, importantly, a distinct pattern of regulation. In wild-type mice, univariate regression identified age as the only significant contributor to the Pi/PPi ratio, but this effect was lost after inclusion of sex and serum AP activity in a multivariate model. This pattern suggests that the apparent influence of age on the Pi/PPi ratio in wild-type animals is at least partly explained by covariation with AP activity (and/or sex), rather than reflecting an independent effect of age per se.

By contrast, Abcc6^−/− mice show a clearer, more robust regulatory signal. Univariate analyses revealed a strong inverse relationship between Pi/PPi and age and a strong positive relationship between Pi/PPi and AP activity. In a multivariate model adjusted for age and sex the model remained significant and explained over 30% of the variance in Pi/PPi; serum AP activity remained an independent, strong predictor while age no longer contributed. Thus, in the Abcc6^−/− background AP activity accounts for a substantial fraction of the variation in the Pi/PPi balance, whereas the effect of age appears to be mediated through or confounded by AP activity. Biologically, these results are consistent with partial compensation for loss of ABCC6 (for example via ANK or other PPi-regulating pathways) and indicate that AP activity is a key modulator of the Pi/PPi ratio in the knockout animals. Finally, whereas the multivariate model for Abcc6^−/− mice achieves significant explanatory power, the much weaker adjusted R² in wild-type mice implies that additional, unmeasured processes contribute to Pi/PPi regulation in the intact background, an important caveat and an avenue for future mechanistic studies.

4.4. Human validation and translational relevance

In humans, plasma PPi levels inversely correlated with AP activity, consistent with our murine findings. In individuals with zero cardiac calcification, PPi also correlated positively with serum Pi, while AP activity directly associated with the Pi/PPi ratio, a clinically relevant marker of calcification risk. This highlights the tight coupling of PPi and Pi metabolism and reinforce AP as a central physiological regulator, while raising the possibility that coordinated transcriptional control of PPi homeostasis genes contributes to this balance. In the disease setting, Pi/PPi ratio was recently shown to be a significant determinant of the aortic valve calcification in cardiovascular patients [42], also exhibiting a negative association between PPi and AP activity, similarly to patients with chronic kidney disease [73], in which PPi not only showed an inverse correlation with AP activity but a positive correlation with Pi levels. An earlier study however found a weak baseline correlation of plasma PPi with age and serum phosphate in CKD [74]. In our PXE patient cohort the correlations were weaker compared to the zero-calcification cohort, with only a modest inverse association between AP activity and PPi levels. Nevertheless, univariate regression identified AP activity as a significant contributor to the Pi/PPi ratio which remained significant in a multivariate model adjusted for age and sex. Although the model explained only a modest proportion of variance, these findings demonstrate that AP activity is an independent determinant of the Pi/PPi ratio in PXE, underscoring its role as a key modifier of extracellular PPi availability even in the disease setting. A striking difference emerged between Abcc6-deficient mice and the PXE patients: while PPi levels increased progressively with aging in knockout mice in contrast to wild-type mice, it remained stable with age in the PXE cohort, both in males and females. Of note, previous studies in PXE cohorts reported a positive correlation between plasma PPi levels and age [26], consistent with findings in heterozygous carriers and PXE patients but not in controls [25]. An age-related increase in PPi was further observed specifically in female PXE patients, supported by longitudinal data from seven patients, whereas this correlation was absent in males and healthy controls [75]. The controversial data warrant further investigation. However, identifying the mechanisms driving the strong ABCC6 dependent increase in PPi levels with age in mice may uncover therapeutic targets to modulate disease onset and progression in humans.

4.5. Limitations

Several limitations warrant consideration. Although our promoter analyses are supported by chromatin accessibility data and consistent co-expression patterns at both bulk, single-cell and single-nucleus levels, direct functional validation of TF binding and regulatory activity (e.g., ChIP-seq, reporter assays, currently available in detail only for HNF4A, CEBP/B, and RXRA) [[50], [51], [52], [53], [54], [55], [56], [57], [58], [59], [60], [61]] remains necessary. While the observed correlations are robust, they do not establish causality, and further mechanistic studies are required to define how TF networks and AP activity integrate to regulate systemic PPi. Finally, our human cohorts were relatively modest in size, and differences in preanalytical and analytical procedures between centers may have contributed to variability. Larger, longitudinal studies with harmonized protocols are needed, although the consistency of findings across datasets supports the robustness of our conclusions.

4.6. Implications and future directions

Our findings highlight several key implications. First, PPi homeostasis is under a coordinated shared transcriptional control, with conserved TFs acting as candidate regulators, with subtle gene-specific differences. This opens opportunities for therapeutic modulation of PPi levels via transcription factor–targeted interventions. Second, developmental regulation of AP activity emerges as a critical determinant of PPi availability, suggesting that early life is particularly vulnerable to disruptions in PPi homeostasis, which therefore may serve as a potential therapeutic window. Third, the shared regulation we observed between PPi, AP activity, and the Pi/PPi ratio across species underscore the translational relevance of murine models. Finally, further studies dissecting the TF regulatory landscape and the mechanisms underlying the age-related compensatory rise in PPi levels in mice may reveal additional, potentially druggable targets to combat ectopic calcification.

CRediT authorship contribution statement

Virgil Tamatey: Writing – review & editing, Methodology, Investigation, Data curation. Martin Várhegyi: Writing – review & editing, Methodology, Investigation, Data curation. Bence Blaha: Visualization, Software, Formal analysis. Judith Van Wynsberghe: Writing – review & editing, Investigation. Lona Lion: Writing – review & editing, Visualization, Methodology, Formal analysis. Zahira Ahmadi: Writing – review & editing, Investigation. Dénes Juhász: Writing – review & editing, Investigation, Data curation. Emese Bata: Writing – review & editing, Formal analysis. Dániel Márton Tóth: Writing – review & editing, Investigation. Muhyiddeen Muazu: Writing – review & editing, Investigation. Anikó Ilona Nagy: Writing – review & editing, Validation, Supervision, Resources, Funding acquisition. Celia P. Martinez-Jimenez: Writing – review & editing, Supervision, Resources, Project administration, Methodology, Funding acquisition. Olivier Vanakker: Writing – review & editing, Supervision, Resources, Methodology, Funding acquisition. Tamás Arányi: Writing – review & editing, Supervision, Resources, Project administration, Funding acquisition, Data curation, Conceptualization. Flóra Szeri: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Resources, Project administration, Methodology, Funding acquisition, Data curation, Conceptualization.

Funding

This study was funded by the National Research Development and Innovation Office of Hungary (FK131946 to FSz, K132695 and 152689Advanced to TA, K146732 and RRF-2.3.1-21-2022-00003 to AIN). Further financial support: FWO Junior Fundamental Research Project (grant G061521N) of the FWO Research Foundation Flanders, Belgium to OV and FSz. FSz, TA and OV are participants of the International Network on Ectopic Calcification (INTEC), and the International Scien. CP.M.-J. was supported by the Generalitat Valenciana through the Plan GenT (Project-ID CIDEXG/2023/30), the Agencia Estatal de Investigación through the ATRAE program (Project-ID ATR2024-154,995), and the Spanish Ministry of Science and Innovation (Project-ID PID2024-157730OB-I00). L.L. and MV were supported by pre-doctoral grants from the Spanish Ministry of Science and Innovation (PRE2024-003299), and Semmelweis University, respectively. VT, ZA and MM were supported by a grant from Stipendium Hungaricum. MM received SE 250+ 2025/2026/1 Excellence Grant from Semmelw.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following is the Supplementary data to this article.

Data availability

Data will be made available on request.