Sulfate biology genes are associated with mineralization in mouse and human

Abstract

Sulfate plays a critical role in bone health and development. More than 90 sulfate-related genes are highly conserved across mammalian species, but very few of these genes have been linked to adverse bone phenotypes in humans. To extend our knowledge of sulfate-related gene expression dynamics during mineralization, this study leveraged 6 publicly available transcriptomic datasets, covering human osteosarcoma cell line Saos-2 mineralization, 2 mouse calvarial osteoblast mineralization models, vascular smooth muscle cell (VSMC) calcification, and 2 neurogenic heterotopic ossification datasets. We focused on a total of 12 sulfate-related genes that were upregulated during mineralization of Saos-2 cells. Six of these genes (Slc26a11, Sgsh, Sqor, Sult1a1, Tpst1, and Ust) were also consistently upregulated during mouse osteoblast and VSMC mineralization. Additionally, 3 genes (Cth, Got1, and Sulf1) were upregulated in Saos-2 mineralization but downregulated in mouse primary osteoblasts. Cbs, Chst3, and Chst13 were unchanged in the mouse primary cell models. Cbs, Chst13, Sgsh, Sulf1, and Ust also increased in models of heterotopic ossification. We have now identified several genes (CHST13, TPST1, UST, SULF1, GOT1, SLC26A11, and SULT1A1) that have not previously been considered for adverse bone conditions in humans, suggesting that additional sulfate biology genes may be linked with human skeletal conditions. Network analysis showed large co-expression clusters of genes, including sulfate biology and bone genes, that were upregulated across the calcification time courses. Gene ontology term enrichment analysis demonstrated significant enrichment in terms associated with mineralization, including ossification, bone mineralization, cartilage development, and extracellular matrix organization for these clusters of genes. This study provides a collated list of sulfate-related genes and networks that are associated with mineralization, which will facilitate future functional studies of sulfation pathways associated with bone pathology.

Graphical Abstract

Graphical Abstract.

Introduction

The extracellular matrix that supports the development and maintenance of bone consists of organic and inorganic components. Collagen is the major organic component. It is highly crosslinked to form a meshwork onto which the inorganic components (primarily a calcium- and hydroxyl-deficient apatite with unique CO₃²⁻ and HPO₄²⁻ substitutions and some amorphous calcium phosphate¹) are deposited. The Phylobone database lists 28 collagenous and 227 non-collagenous proteins that have been found in the ECM of bone,² with the collagenous fraction comprising approximately 90% of the organic component. Bioactive components of the non-collagenous organic fraction include the small leucine-rich proteoglycans biglycan and decorin and the proteoglycan perlecan (HSPG2).^3,4

Proteoglycans consist of a core protein covalently attached to one or more glycosaminoglycans, which are long, unbranched polysaccharides containing repeating disaccharide units,^5,6 each consisting of an N-sulfated or N-acetylated amino sugar and a uronic acid. Common glycosaminoglycans are heparan sulfate, chondroitin sulfate, keratan sulfate, and dermatan sulfate. Many of these glycosaminoglycans are important in the formation and maintenance of bone. For example, chondroitin sulfate is the major glycosaminoglycan of bone⁷ and is crucial for forming and maintaining cartilage and bone,⁸ as demonstrated through various chondrodysplastic mouse models.^9–11 Heparan sulfate regulates key factors of ossification and bone mass.¹² Keratan sulfate is involved in the organization of the extracellular matrix in bone, cartilage, and the cornea, and can also act as a signaling molecule.¹³ Since these sulfated glycosaminoglycans make an essential contribution to the structure and function of bone, sulfate is an obligatory mineral in bone biology.¹⁴

Ninety-one genes have previously been associated with sulfate biology.¹⁵ Proteins encoded by these genes are involved in transport of sulfate across the plasma membrane, sulfonation of key targets, removal of sulfate from molecules, generation of sulfate, and synthesis and transport of the universal sulfonate donor 3′-phosphoadenosine-5′-phosphosulfate.⁵ Functional sulfated glycosaminoglycans are formed and degraded through the activity of a number of these sulfate biology enzymes and transporters.

Sulfotransferases, including carbohydrate sulfotransferases (CHST family) and uronyl 2-sulfotransferase (UST), are responsible for transferring sulfate to glycosaminoglycans.^16,17 In contrast, several enzymes, notably N-sulfoglucosamine sulfohydrolase (SGSH), catalyze the hydrolysis of N-sulfated glucosamine residues in heparan sulfate, facilitating its breakdown and recycling within the lysosome.¹⁸ Sulfatase 1 (SULF1) removes sulfate groups from heparan sulfate.¹⁹ It is expressed in cartilage and bone and during endochondral ossification.²⁰ This activity modulates the effects of heparan sulfate by altering the binding sites for signaling molecules, including Noggin, BMP, WNT, and FGF.^21–23

The SLC26 family of transporters mediates the transport of various anions across cellular membranes, facilitating the exchange of sulfate and playing a crucial role in maintaining cellular ion homeostasis and pH balance.²⁴ There are also several enzymes involved in the generation of sulfate, including glutamic-oxaloacetic transaminase 1 (GOT1) and cystathionine beta synthase (CBS), which are involved in the catabolism of homocysteine and cysteine, releasing sulfate. Cystathionine gamma-lyase (CTH) is a crucial enzyme in the trans-sulfuration pathway, responsible for converting cystathionine into cysteine, producing hydrogen sulfide (H₂S). Cystathionine gamma-lyase-mediated H₂S production enhances osteoblast activity and promotes bone fracture healing by inducing sulfhydration of RUNX2, a key transcription factor in bone formation.²⁵ Cystathionine gamma-lyase thus contributes to bone regeneration and remodeling. Sulfide quinone oxidoreductase (SQOR, formerly SQRDL) is a mitochondrial enzyme that is involved in the first step of H₂S oxidation, catalyzing the transfer of electrons from H₂S to ubiquinone in the respiratory chain. Sulfotransferase 1A1 (SULT1A1) is involved in the conjugation of sulfate to hormones, neurotransmitters, drugs, and xenobiotic compounds. It is important in the metabolism of catecholamines, thyroid hormone, and estrogenic compounds, which are involved in the maintenance of bone integrity and turnover.^26–28 SULT1A1 activity has been detected in human osteoblast-like cells.^29,30 The tyrosylprotein sulfotransferases (TPST1 and TPST2) catalyze the post-translational sulfation of tyrosine residues in various proteins to form tyrosine O-sulfate esters,³¹ a modification essential for the proper function of many secreted and membrane-bound proteins.

Of the 91 identified sulfate biology genes,¹⁵ 17 have a skeletal phenotype (Table 1). The list includes 9 genes that are mutated in the mucopolysaccharidoses, a group of lysosomal storage diseases caused by accumulation of glycosaminoglycans in the lysosome. Ten of the 17 genes are included in at least 1 current skeletal dysplasia panel (eg, https://blueprintgenetics.com/tests/panels/malformations/skeletal-dysplasias-core-panel/, https://www.invitae.com/us/providers/test-catalog/test-89100, https://www.fulgentgenetics.com/skeletal-dysplasias). These panels also include 3 genes for core proteoglycan proteins, aggrecan (ACAN gene; mainly involved in cartilage biology), biglycan (BGN), and perlecan (HSPG2). Other sulfate biology genes, which also contribute to maintaining the required sulfate content of glycosaminoglycans, are therefore likely candidates for skeletal dysplasias of unknown genetic origin.

Table 1.

Sulfate biology genes linked to bone phenotypes in humans.

Gene	OMIM	Human phenotypes
SLC4A1	179800	Rickets/osteomalacia
SLC26A1	610130	Reduced bone mineral density
SLC26A2	222600, 256050	Diastrophic dysplasia
PAPSS2	612847	Skeletal dysplasia
CHST3	143095	Spondyloepiphyseal dysplasia
CHST11	618167	Osteochondrodysplasia, mainly affecting limbs
CHST14	601776	Connective tissue disorder. Distinct joint contractures and craniofacial abnormalities
HS2ST1	619194	Neurofacioskeletal syndrome
ARSB	253200	Mucopolysaccharidosis type VI. Short stature, coarse facial features, joint contractures, claw hands
ARSE/ARSL	302950	Chondrodysplasia punctata, X-linked recessive. Aberrant bone mineralization, severe underdevelopment of nasal cartilage, distal phalangeal hypoplasia
ARSK	619698	Mucopolysaccharidosis type X. Short-trunk, short stature, mild scoliosis, genu valgus
GALNS	253000	Mucopolysaccharidosis type IVA. Skeletal dysplasia with incomplete ossification and successive imbalance of growth
GNS	252940	Mucopolysaccharidosis type IIID. Mild dysostosis multiplex, ovoid vertebrae
IDS	309900	Mucopolysaccharidosis type II. Skeletal deformation, coarse facial features
SGSH	252900	Mucopolysaccharidosis type IIIA. Skeletal dysplasia, short stature
SUMF1	272200	Neurodegenerative disorders, scoliosis, hip dislocation, osteopenia
SULF1	600383	Mesomelia-synostoses syndrome. Decreased ossified bone volume

Genes associated with lysosomal storage diseases are shown in bold. Genes included on at least one skeletal dysplasia panel are underlined.

OMIM, Online Mendelian Inheritance in Man (https://www.omim.org/).

In this study, we used 6 public databases to explore the association of sulfate biology gene expression with mineralization. The Saos-2 human osteosarcoma cell line, first established in 1975,³² synthesizes extracellular matrix factors comparable to a typical osteoblastic phenotype.³³ These mature osteoblast-like cells have the ability to mineralize in vitro and in vivo.^33–35 Saos-2 cells achieve an osteocyte-like phenotype during long-term culture under mineralizing conditions, and provide a useful model for studying human osteocyte differentiation and responsiveness in vitro.³⁶ Thus, utilizing the Saos-2 culture system is highly applicable to investigations of the synthesis and deposition of factors that surround bone cells in the extracellular matrix.^33,37,38

We also analyzed 2 public databases of mouse calvarial osteoblasts undergoing calcification in vitro.^39,40 These cells are employed for studying mineralization due to their ability to mimic in vivo osteoblast behavior, including the synthesis and mineralization of bone matrix.⁴¹ They allow for detailed examination of the entire differentiation process, including alkaline phosphatase activity, calcium deposition, and gene expression changes. The primary cells more accurately represent the heterogeneity and functional activity of osteoblasts than cell lines.

We have included 2 models of calcification of tissues other than bone (ectopic calcification), a clinically significant problem. First, we examined transcriptional changes during calcification of mouse vascular smooth muscle cells (VSMCs) in vitro, which models the calcification of blood vessels and heart valves that occurs during human aging.⁴² In addition, we studied 2 transcriptomic datasets of muscle calcification in mouse in vivo, modeling human neurogenic heterotopic ossification (NHO).^43,44 Neurogenic heterotopic ossification is an abnormal development of bone in extra-skeletal tissues after damage to the central nervous system, including stroke, traumatic brain injury, and spinal cord injury, in the presence of localized muscular inflammation.^45,46 Muscle damage plus a glucocorticoid spike can cause the same process in a mouse model.⁴⁴ Neurogenic heterotopic ossification often leads to severe functional impairments, posing significant clinical challenges.

These 6 established models of mineralization in bone or soft tissues were investigated to highlight the significance of sulfate biology. We identified known and novel associations between sulfate biology genes and mineralization, suggesting additional genes that could be investigated in skeletal dysplasias and as therapeutic targets.

Materials and methods

Datasets used in the study

Gene expression values for the human cell line Saos-2 were derived from the FANTOM5 project^47,48 (https://fantom.gsc.riken.jp/5/). The experiment is described in detail in the supplementary material of the previous paper.⁴⁸ In summary, 3 separate cultures of the Saos-2 cell line were induced to mineralize, using ascorbic acid and β-glycerophosphate. Quantification of mRNA expression used genome scale 5′-end profiling (cap analysis of gene expression^47,49) with results presented as tags per million (TPM). Gene-based expression values were downloaded from https://fantom.gsc.riken.jp/5/datafiles/phase2.6/extra/gene_level_expression/. Results from prior to treatment and days 1, 4, 7, 14, 21, and 28 (designated D0, D1, D4, D7, D14, D21, D28, respectively) were used in the present study. A sample from an untreated control culture taken at 28 d (D28C) was also included.

Mouse calvarial osteoblast data were from the BioGPS mouse MOE430_2.0 gene atlas^39,50 (http://biogps.org), available as GSE11339 (PRJNA106547) through NCBI GEO Datasets (https://www.ncbi.nlm.nih.gov/gds). Two replicate samples were available for 5, 14, and 21 d (D5, D14, and D21) after initiation of mineralization. This dataset is referred to as OB-1 in the text. We also examined a similar experiment (OB-2), available from the EMBL-EBI Array Express repository, accession number E-MTAB-1391.⁴⁰ Three or four replicate samples were available for 0, 9, and 27 d (D0, D9, and D27) after initiation of mineralization. In both studies, osteoblasts were isolated from calvaria of neonatal C57BL/6 mice. Mineralizing medium (ascorbic acid and β-glycerophosphate) was added from day 7 of culture (D0 of the experiment).⁴⁰ The Affymetrix .cel files downloaded from the repositories were normalized using the Transcriptome Analysis Console (ThermoFisher). Normalized results were unlogged prior to analysis and are presented as arbitrary fluorescence units (AU).

Vascular smooth muscle calcification was studied using Affymetrix microarray data from a mouse model,⁴² available from the EMBL-EBI Array Express repository, accession number E-MTAB-1680. Vascular smooth muscle cells were isolated from the aortas of 6-wk-old C57BL/6 mice and cultured under mineralizing conditions as above.^40,42,51 Samples were taken at D0, D9 and D27. There were 3 or 4 replicates for each time point. The .cel files downloaded from the repository were processed using the Transcriptome Analysis Console. Microarray data for a mouse model of NHO were from NCBI GEO Datasets, accession number GSE165062 (PRJNA693157).⁴³ Neurogenic heterotopic ossification was induced in mouse muscle by transection of the spinal cord at the level of thoracic vertebrae 11-13, followed by intramuscular cardiotoxin injection,⁴³ and the mice were sacrificed 2 d after treatment. RNA was extracted from hamstring muscle. Untreated control samples, spinal cord injury-only controls and cardiotoxin-only controls were included (3 or 4 replicates per condition). RNA was hybridized to Illumina MouseRef-8 v.2.0 Gene Expression BeadChips. The normalized data file for this dataset (file GSE165062_matrix_normalized_final.txt.gz) was downloaded. This study is referred to as NHO-1 in the text. A second NHO experiment (NHO-2) was also examined (GSE218699; PRJNA 904909). In this model, NHO was induced without spinal cord injury, by injecting cardiotoxin in conjunction with dexamethasone treatment.⁴⁴ Mice were sacrificed at 4 d, and RNA was extracted from affected hamstring muscles and subjected to RNA sequencing. There were 4 replicates of the control condition (muscle injury only, no NHO) and 9 of the dexamethasone-treated condition (developed NHO by 4 d). Normalized and filtered data were kindly provided by the authors. Results are given as counts per million (CPM).

Analysis of sulfate biology genes in models of calcification

The results for individual sulfate biology genes were retrieved from the datasets (Table S1). Where there were multiple probesets for one gene, the highest expressing probeset is presented. Significance values for expression differences were calculated using GraphPad Prism, V10.3.1 (https://www.graphpad.com/). Tests used and significance levels are indicated in figure legends.

Network analysis of gene expression during calcification

To explore the genes that vary during the mineralization time courses, we used the network analysis tool BioLayout (http://biolayout.org)^52,53 to create clusters of co-expressed genes. Genes that cluster together in a network analysis are likely to have similar functions.⁵⁴ Prior to network analysis, data were filtered to remove low-expression genes or probesets. For the Saos-2 data, genes where all samples showed <1 TPM expression and genes where fewer than 4 samples had non-zero expression were removed. For the normalized Affymetrix microarray datasets OB-1, OB-2 and VSMC, we determined an expression level considered to be above background, based on the distribution of expression values shown during the normalization process. This value corresponded to a relative fluorescence level of 35 for OB-1, and of 50 for OB-2 and the VSMC data. All probesets where no sample reached this level were removed prior to network analysis. Network analysis was performed using the Pearson correlation coefficient. Thresholds are indicated in the results and were chosen to ensure that approximately the same number of genes was included in each network. Clustering used the Markov Clustering Algorithm with an inflation value of 2.0 for all analyses.⁵⁵ A chi-squared test with 1° of freedom and Yates correction was used to assess over- or under-representation of sulfate biology genes in the clusters based on the expected number if the genes were distributed throughout the clusters in proportion to cluster size.

Enrichment of gene ontology (GO) terms for the genes within a cluster was assessed using the PANTHER Overrepresentation Test (released 2024-08-07) (https://geneontology.org/) and GO Ontology database (released 2024-06-17; DOI: 10.5281/zenodo.12173881). The reference list was Homo sapiens or Mus musculus as appropriate. Fisher’s Exact test was used to calculate probabilities and the Bonferroni correction was applied to adjust for multiple testing. A list of genes involved in bone formation and maintenance was curated based on genes listed in skeletal dysplasia panels (https://www.invitae.com/us/providers/test-catalog/test-89100; https://blueprintgenetics.com/tests/panels/malformations/skeletal-dysplasias-core-panel/; https://www.fulgentgenetics.com/skeletal-dysplasias; https://panelapp.genomicsengland.co.uk/panels/309/) plus genes known to have a role in bone maturation^56,57 (Table S2).

Results

Most sulfate biology genes were found in the datasets

No dataset contained all of the 91 sulfate biology genes previously identified.¹⁵ Six were not annotated in the Saos-2 data (ARSF, ARSH, HS6ST3, SULT1A3, SULT1A4, and SULT1C3), giving a total of 85. These genes have now been identified in the human genome (GRCh38.p14) but were not found in the version used to derive the FANTOM5 data (GRCh37).^47,48 Thirteen of the 91 were not found on the Affymetrix expression microarray used for the mouse primary cells (MOE430_2.0), leaving a total of 78 sulfate biology genes (137 probesets) that could be examined. Ten of the missing genes (Arsd, Arsf, Arsh, Arsl (formerly Arse), Chst6, Sult1a2, Sult1a3, Sult1a4, Sult1c3, and Sult1c4) are not currently identified in the mouse genome (GRCm39) and three (Gal3st3, Slc26a9, and Sult2a1) are now mapped to the genome but were not annotated on the array. Nineteen sulfate biology genes were not found in the filtered gene list from NHO-1, based on the Illumina array. These included eleven of those listed above for the Affymetrix array (Arsd, Arsf, Arsh, Arsl, Chst6, Sult1a2, Sult1a3, Sult1a4, Sult1c3, Sult1c4, and Sult2a1) as well as Chst13, Chst15, Got1, Hst3st5, Hs3st4, Ndst2, Sulf2, and Sult6b1. However, a result was found for Gal3st3 and Slc26a9. The inconsistency between genes found with the Affymetrix and Illumina arrays reflects the differing genes represented by annotated probesets on the arrays and the filtering applied to the NHO-1 data.⁴³ There were 73 genes (98 probesets) that could be studied for NHO-1. For NHO-2 there were 58 sulfate biology genes in the final analysis after filtering for lowly expressed genes where log₂(CPM) < 0.5 and genes expressed in only one sample.⁴⁴ Expression levels for all sulfate biology genes found in these datasets are available in Table S1.

Key sulfate biology genes had increased expression during mineralization of Saos-2 human osteosarcoma cells

We initially analyzed the 85 annotated sulfate biology genes¹⁵ in the FANTOM5 time course of Saos-2 osteosarcoma mineralization.⁴⁸ We identified sulfate gene expression trends at key time points: (D0, D1, D4, D7, D14, D21, D28 and D28C). Fifty-one of the 85 sulfate biology genes showed expression ≥1 TPM in at least four samples (Table S1). Eleven genes with maximum expression >15 TPM exhibited increased expression over the time course, generally starting from D4, peaking and then decreasing or plateauing by D28 of treatment (Figure 1). The untreated control sample at D28 generally had expression similar to (CTH, GOT1, TPST1, UST) or lower than (CBS, CHST3, CHST13, SGSH, SLC26A11, SQOR) the treated sample at the same time point, although for SULF1 the control was substantially higher than the D28 treated sample. The highest expression was seen for TPST1, and the greatest proportional increase with treatment was seen for CHST13. Six other sulfate biology genes (GALNS, GNS, IDS, PAPSS1, SLC35B2, and SUMF2) showed high (>50 TPM) expression without a consistent pattern across time points or replicates (Table S1). One class of sulfate biology genes, the cytosolic sulfotransferases, had very low (<1 TPM) to no expression in the Saos-2 time course. Only SULT1A1 was expressed in multiple samples (19 of 24 samples), with a maximum value of 0.9 TPM (Figure 1).

Figure 1.

Sulfate biology genes showing regulated expression during mineralization of the Saos-2 human osteosarcoma cell line. Expression levels for 11 genes with increasing expression are shown. Last panel shows low expression gene SULT1A1, presented for comparison with mouse results (Figures 2-4). Final column in each graph shows the results for an untreated control after 28 d in culture (D28C). Y axis—expression level in tags per million (TPM); X axis—time points. Details of the experiment are available in Arner et al.,⁴⁸ and the full results can be viewed on the FANTOM5 Zenbu browser (https://fantom.gsc.riken.jp/zenbu/). Statistical analysis used ordinary one-way ANOVA with Tukey’s multiple comparison test. The most significant p-value between time points for each gene is shown: ^*p-adjusted < .05; ^**p-adjusted < .01; ^***p-adjusted < .001; ^****p-adjusted < .0001.

Sulfate biology genes showed varying patterns of expression in mouse models of osteoblast mineralization

Saos-2 is a human osteosarcoma cell line that has undergone genetic alterations both as a tumor and as a result of long-term cell culture. To compare the results using this cell line, we examined data from mouse primary calvarial osteoblasts in culture. Expression levels of the annotated sulfate biology genes in the two models (OB-1 and OB-2) are provided in Table S1.

The BioGPS database (OB-1; http://biogps.org/) contains microarray expression data for mouse calvarial osteoblasts at 3 time points, D5, D14, and D21 after commencement of mineralization treatment.^39,50 There is no pretreatment sample in this dataset and the first time point at D5 likely occurs after expression of many sulfate biology genes has already increased, as seen in the Saos-2 time course (Figure 1). There were distinct expression levels among the 11 sulfate biology genes identified as being regulated during mineralization in Saos-2 cells (Figure 2A). A high-expression group (peak expression ≥500 AU) comprised Sulf1, Got1, Tpst1, and Sqor. Cth, Ust, Sgsh, and Slc26a11 exhibited moderate expression levels (peak expression between 99 and 250 AU), while Cbs, Chst3, and Chst13 showed the lowest expression (maximum <60 AU). Only Sqor, Sult1a1, and Tpst1 increased significantly across the time course, while Got1 and Cth declined (Figure 2A).

Figure 2.

Expression levels of sulfate biology genes in mouse calvarial osteoblasts. (A) Results for OB-1 dataset. (B) Results for OB-2 dataset. Statistical analysis used ordinary one-way ANOVA with Tukey’s multiple comparison test. X axis—time points; Y axis—expression in AU. Significant p-values are shown: ^*p-adjusted < .05; ^**p-adjusted < .01; ^***p-adjusted < .001; ^****p-adjusted < .0001. Where a gene had more than one probeset, the highest expressing probeset is presented.

To validate the results from the BioGPS data, we accessed a second dataset of mouse calvarial osteoblasts undergoing mineralization (OB-2).⁴⁰ In this study, RNA was extracted at D0, D9, and D27 after commencement of mineralization treatment, providing a zero time control and also a later time point than the BioGPS data. The results replicate those from OB-1 (Figure 2A). Three of the genes that varied in the Saos-2 time course (Chst13, Sgsh, and Ust) were expressed at a substantial level but did not change extensively over the OB-2 time course (Figure 2B). In contrast, Tpst1 and Sqor increased about 2-fold over 27 d and Slc26a11 increased about 1.5-fold, while Sult1a1 expression increased from a negligible level at D0 to approximately 400 AU by D27. Tpst1 showed the highest expression upon mineralization (1500-2000 AU) while Sulf1 was highest prior to mineralization (2500-3000 AU). Got1 and Sulf1 decreased 4- and 2-fold respectively and Cth declined to a very low level after the initiation of mineralization (Figure 2B).

Sulfate biology genes showed varying patterns of expression during ectopic calcification in mouse models

Ectopic calcification is a clinically relevant condition where soft tissues mineralize and take on features of bone. We accessed 2 models of this problem, calcifying primary VSMC and skeletal muscle undergoing NHO.

Like calvarial osteoblasts, VSMC can mineralize in vitro, simulating human vascular calcification.⁵⁸ We used a public dataset available on Array Express.⁴² Vascular smooth muscle cells were mineralized and RNA was extracted prior to treatment (D0) and at D9 and D27 after the initiation of treatment. The patterns of expression of sulfate biology genes were almost identical to those of the calvarial osteoblasts from the same research group (OB-2) (Figure 3). Cth, Got1 and Sulf1 declined following treatment while Sqor and Tpst1 increased, although notably not as much as in the osteoblasts. Sult1a1 displayed a similar expression pattern to osteoblasts, showing a 15-fold increase over the 27-d time course (Figure 3).

Figure 3.

Expression levels of sulfate biology genes in mouse vascular smooth muscle cells (VSMCs) undergoing mineralization. Statistical analysis used ordinary one-way ANOVA with Tukey’s multiple comparison test. X axis—time points; Y axis—expression in AU. Significant p-values are shown: ^*p-adjusted < .05; ^**p-adjusted < .01; ^***p-adjusted < .001; ^****p-adjusted < .0001. Where a gene had more than one probeset, the highest expressing probeset is presented.

We analyzed microarray data from a previously published mouse NHO model.^43,59 In this model (NHO-1), NHO was induced by combined spinal cord and muscle injury. Tissue was collected 2 d post-injury. Among the sulfate biology genes identified as showing high and variable expression in Saos-2 cells and/or the mouse primary cells that were found in this dataset, five showed expression difference between the treated (NHO) and saline control groups (Figure 4A). For Sgsh and Chst3, this change was largely attributable to the double treatment that resulted in the calcification process, since the single treatment controls were not significantly different from the saline control. Sulf1 and Tpst1 increased in response to cardiotoxin treatment and there was no additional increase in the combined treatment. Sqor increased in response to spinal cord injury with no additional impact of the double treatment. Additionally, unlike the findings in primary osteoblast and VSMC models, Sult1a1 showed no difference among any of the groups. The time point assessed in this experiment potentially represents an earlier stage in mineralization than the previous mouse experiments. In the Saos-2 data, most expression increases occurred at 4 d and beyond, and it is not clear where this 2-d time point in the NHO mice might align with the time courses in cultured cells.

Figure 4.

Expression levels of sulfate biology genes in mouse muscle undergoing NHO. (A) Results for NHO-1 dataset. Results shown are from the normalized data file for this dataset (file GSE165062_matrix_normalized_final.txt.gz). The AVG_Signal results from this file are plotted; error bars show the ARRAY_STDEV. There were 4 samples in the saline, SCI and CDTX groups and 3 in the SCI+CDTX (NHO) group. Chst13 and Got1 were not present in this filtered dataset; Cbs, Cth, Slc26a11, and Ust had negligible expression. X axis: saline—saline injection only; SCI—spinal cord injury only; CDTX—cardiotoxin injection into muscle only; NHO—NHO generated by spinal cord injury plus injection of cardiotoxin into muscle. Y axis: expression in AU. Statistical analysis used one-way ANOVA. Significant p-values are shown: ^*p-adjusted < .05; ^**p-adjusted < .01; ^***p-adjusted < .001; ^****p-adjusted < .0001. Where a gene had more than one probeset, the highest expressing probeset is presented. (B) Results for NHO-2 dataset. RNA sequencing results are shown for 4 (control) or 9 (NHO) replicates. Cth was not present in the filtered dataset. X axis: control—single injection of cardiotoxin into muscle, daily saline injection; Dex—single injection of cardiotoxin into muscle, daily dexamethasone injection, developed NHO by D4. Statistical analysis used Student’s t-test with Welch’s correction for unequal variances. Significant p-values are shown: ^*p-adjusted < .05; ^**p-adjusted < .01; ^***p-adjusted < .001; ^****p-adjusted < .0001.

Another NHO dataset was available (NHO-2). These samples were collected four days after initiation of treatment with cardiotoxin and dexamethasone, to induce NHO without spinal cord injury.⁴⁴ There were few significant differences between the control and NHO groups. Tpst1, Sqor and Sgsh were unchanged while Sulf1 and Chst13 marginally increased in the NHO (dexamethasone treated) group (Figure 4B). Ust, which was unchanged at D2 in NHO-1 showed a highly increased level at D4 in NHO-2. Unlike NHO-1, Cbs was significantly increased in the NHO group although the level was low. These differences may reflect the four-day timepoint in the second experiment.

Network analysis identified patterns of increasing or stable sulfate biology gene expression in Saos-2 cells undergoing mineralization

Network analysis can aid in interpretation of transcriptomic data by rendering the results of gene expression studies as a network, consisting of nodes representing transcripts connected by edges representing the similarity of their expression profiles across multiple conditions,^52,53,55 in this case time points following initiation of mineralization.

To gain insights into the function of sulfate biology genes in mineralization and extracellular matrix organization, we performed network analysis on the Saos-2 dataset, at r ≥ 0.75 and inflation value of 2.0. The analysis included 14 955 genes (of which 47 were sulfate biology genes). There were three large clusters with distinctive expression patterns related to the time course. Saos-2 Cluster0001 (3299 nodes) was driven by high expression in the control sample, Saos-2 Cluster0002 (1465 nodes) showed an increase in average expression of genes in the cluster over the treatment time course with D28C generally lower than D28 and Saos-2 Cluster0003 (636 nodes) showed a decline in expression (Figure 5A).

Figure 5.

Expression profiles of the major up- and downregulated clusters from network analysis. (A) Saos-2 clusters 0001, 0002, and 0003. (B) OB-1 clusters 0001 and 0002. (C) OB-2 clusters 0001 and 0002. (D) VSMC clusters 0001 and 0002. Clusters in which gene expression reduced during mineralization are shown on the right; clusters in which gene expression increased during mineralization are shown on the left. The X axis shows the samples; each column represents one sample. The Y axis shows the average expression level of cluster genes in the sample.

Four sulfate biology genes were present in Saos-2 Cluster0001 (listed in Figure 5A). As shown in Figure 1, SULF1 had low expression at D0 and increased over time, with a higher value in the control (D28C). CHST1 had low expression and the other genes showed no trend with time except that the D28C sample was high. This cluster was enriched for GO terms relating to DNA, RNA and protein synthesis, probably reflecting the more proliferative state of the untreated cells at 28 d of culture.

Nineteen sulfate biology genes were present in Saos-2 Cluster0002 (Figure 5A). There was a significant excess of sulfate biology genes in this cluster (5 would be expected; p-value = 0). Two key bone proteoglycan genes, DCN and BGN, were also found in Saos-2 Cluster0002. GO biological process term enrichment analysis indicated significant involvement in heparan sulfate proteoglycan catabolic process (12-fold enrichment), collagen metabolic process (4.06) and glycosaminoglycan metabolic process (3.5).

There were no sulfate biology genes in Saos-2 Cluster0003 where average expression declined over the time course. GO biological process terms enriched in this cluster related to biosynthetic processes, some mitochondrial functions and protein metabolism. The highest expressing gene, TPST1, was not clustered. The list of genes in all clusters and summary of the GO term enrichment analysis with p-values is provided in Table S3 and the average expression pattern of the clusters is shown in Figure 5A.

Network analysis identified common patterns of increasing and decreasing sulfate biology gene expression during primary cell mineralization in mouse

BioLayout network analysis was then performed for the OB-1 dataset, at a correlation threshold of r ≥ 0.85, encompassing 24 858 probesets including 81 probesets for sulfate biology genes (55 unique genes). Two main clusters were identified (Figure 5B). OB-1 Cluster0001 contained 8162 probesets. Gene expression in this cluster decreased after D5. Ten sulfate biology genes were identified (16 probesets) (Figure 5B). This is a small depletion of sulfate biology genes (p-value = 0.02). The GO analysis for OB-1 Cluster0001 revealed the highest enrichment in processes related to replication and transcription, including positive regulation of rRNA processing (4.6-fold enrichment), co-transcriptional RNA 3′-end processing (4.6), nuclear pore organization (4.32), and DNA replication initiation (4.22). Notably, these processes are primarily associated with cellular and DNA functions, with no terms directly related to mineralization or bone ossification.

OB-1 Cluster0002 comprised 5968 probesets, with expression levels increasing after D5. Twenty-six sulfate biology genes (33 probesets) were identified (Figure 5B). This is a significant enrichment (p-value = 0.006). GO analysis for OB-2 Cluste0002 identified several critical mineralization-related GO terms, including positive regulation of bone mineralization (2.32-fold enrichment), cartilage development (2.02) and ossification (1.87). The term sulfur compound metabolic process (enrichment of 1.6) was also found for this cluster. There was no term relating to production of glycosaminoglycans but these findings highlight the potential role of sulfate biology genes during the mineralization process, particularly from mid to late stages. The GO term analysis and cluster lists are presented in Table S3.

Using the second murine calvarial osteoblast dataset (OB-2), the network analysis performed at a correlation co-efficient threshold of 0.85 included 24 954 probesets (77 probesets for sulfate biology genes; 53 unique genes). Again, there were two main clusters (Figure 5C and Table S3). OB-2 Cluster0001 comprised 7372 probesets. Expression declined after D0. Thirteen sulfate biology genes were included (listed in Figure 5C). Gene ontology functional analysis revealed that this cluster was predominantly enriched in processes related to cellular functions, including mitochondrial tRNA modification (4.89-fold enrichment), tRNA surveillance (4.89), and positive regulation of protein localization to the telomere (4.89). As for Saos-2 Cluster0003 and OB-1 Cluster0001, these findings suggest that the genes within OB-2 Cluster0001 are primarily involved in cellular maintenance and replication processes.

In contrast, OB-2 Cluster0002, which included 6427 probesets, exhibited an increase in expression from D9 to D27. There was a total of 32 probesets for 26 sulfate biology genes (Figure 5C). This was a significant (p-value = 0.006) excess. Key enriched processes for genes in OB-2 Cluster0002 included bone mineralization (2.82-fold enrichment), extracellular matrix organization (2.44), bone remodeling (2.64), collagen metabolic process (2.59), ossification (2.25), bone development (2.27), and cartilage development (1.95). There was no term relating to production of glycosaminoglycans.

To assess whether ectopic mineralization resulted in the same co-expression patterns, network analysis was also performed on the VSMC dataset at r ≥ 0.86, resulting in the inclusion of 26 091 gene probesets, including 74 probesets for sulfate biology genes (51 unique genes). Two main clusters were identified from this analysis (Figure 5D and Table S3).

Gene expression within VSMC Cluster0001 (6743 probesets) decreased from D9 to D27. Fifteen sulfate biology genes (20 probesets) were identified, shown in Figure 5D. The GO analysis of genes in this cluster revealed significant enrichment in several biological processes, including positive regulation of protein localization to the telomere (fold enrichment 1.46), DNA replication initiation (5.0), purine nucleobase biosynthetic process (5.46), and DNA strand elongation (4.24). These findings suggest that the genes within this cluster are primarily involved in cellular processes related to DNA metabolism and cellular proliferation. VSMC Cluster0002 included 5939 genes. The expression trend in this cluster increased from D9 to D27. This cluster contained 19 sulfate biology genes, listed in Figure 5D. This represents a small excess (p-value = .04). The GO enrichment analysis of the genes in VSMC Cluster0002 revealed significant enrichment in several biological processes directly related to mineralization. Notable enriched GO terms included sulfur compound catabolic process (2.95 fold enrichment), ossification (2.03), extracellular matrix organization (2), cartilage development (1.83), and skeletal system development (1.6). The enrichment of these related processes underscores the common features of physiological and ectopic calcification and the potential role of sulfate biology genes in the context of VSMC ectopic calcification.

Sulfate biology gene expression correlated with bone biology gene expression in upregulated clusters

Each upregulated cluster was examined for the presence of a selection of genes from the commercial panels for bone dysplasias plus other genes known to be involved in bone biology (Table S2). Several genes were represented in the upregulated Cluster0002 of all models (BMP2, COL1A2, DCN, PTH1R, SPARC, TGFB1, WDR19) while some were in Saos-2 Cluster0001 (upregulated in treated samples, but higher in the untreated sample D28C) and Cluster0002 of other datasets (DYNC2H1, IBSP, PHEX, PLS3). Others of these genes were in Cluster0002 of more than one dataset. These results suggest that the sulfate biology genes are co-regulated with a range of bone-related genes, consistent with a role for sulfate in bone formation. In contrast, some bone-related genes were consistently in the downregulated clusters (Saos-2 Cluster0003; Cluster0001 of the mouse models) where there was a deficiency of sulfate biology genes. These included ACTA2, FLNB, LBR, and ORC6.

Discussion

Summary of the findings

Overall, this analysis of sulfate biology genes across diverse datasets, including the human osteosarcoma cell line Saos-2, murine calvarial osteoblasts, VSMC, and the NHO models, has yielded significant insights into their association with mineralization and calcified extracellular matrix formation. There was consistent expression of many sulfate biology genes across the osteosarcoma and osteoblast models, which most closely represent physiological bone formation in vivo. The models of ectopic calcification were less similar, presumably reflecting the different origin of the cells and suggesting that calcification does not completely override the original transcriptomic phenotype. Network analysis identified a distinct Cluster0002 characterized by increased gene expression over time, with significant overrepresentation of sulfate biology and bone-related genes, and enrichment of GO terms relating to bone biology, extracellular matrix assembly, and cartilage metabolic processes. The findings highlight the potential of sulfate biology genes to facilitate critical processes related to bone mineralization.

A summary of the findings for the 12 genes of interest in human and mouse in the 6 model systems is presented in Table 2, and phenotypes associated with genetic variants are shown in Table 3. Further discussion of the functions of the proteins encoded by these genes is presented below.

Table 2.

Summary of results for 12 sulfate biology genes.

Gene	Saos-2	OB-1	OB-2	VSMC	NHO-1	NHO-2
CBS/Cbs	High, no trend	Mod, no change	Low, no change	Low, no change	Low, increase in SCI and SCI+CDTX	Low, increase in Dex
CHST3/Chst3	Low, peak D14-D21	Low, no change	Low, no change	Low, no change	Mod, increase in SCI+CDTX	Mod, decrease in Dex
CHST13/Chst13	Mod, peak D14-D21	Low, no change	Low, no change	Low, slight decrease at D27	Not found	Low, slight increase in NHO
CTH/Cth	Mod, no trend	Mod at D5, decrease by D14	High at D0, decrease by D9	Mod at D0, decrease by D9	Low, no difference	Low, no difference
GOT1/Got1	Mod, plateau after D4	High at D5, decrease by D14	High at D0, decrease by D9	High, decrease by D9	Not found	High, slight decrease in NHO
SGSH/Sgsh	Mod, peak D21	Mod, no change	Mod, slight increase at D27	Mod, slight increase at D27	Mod, increase in CDTX, greatest in CDTX+SCI	Mod, no difference
SLC26A11/Slc26/a11	Low, peak D21	Mod, no change	Mod, slight increase with time	Mod, slight increase at D27	Low, no difference	Mod, no difference
SQOR/Sqor	Low, plateau from D4	High, increase at D21	High, increase at D27	High, slight increase at D27	High, increase in SCI and SCI+CDTX	Mod, no difference
SULF1/Sulf1	Mod, plateau after D14	High, no change	High, decrease by D9	High, decrease by D9	Mod, increase in CDTX and SCI+CDTX	Mod, slight increase in NHO
SULT1A1/Sult1a1	Negligible, no trend	Mod, small increase at D21	High, low at D0, increase with time	High, increase at D27	High, greatest in SCI	Mod, no difference
TPST1/Tpst1	High, plateau after D1	High, increase by D14	High, increase by D9	High, increase at D27	High, increase in CDTX and SCI+CDTX	Mod, no difference
UST/Ust	Mod, plateau from D4	Mod, no change	Mod, slight increase at D27	Mod, no change	Low, slight increase in SCI+CDTX	Mod, increase in NHO

Expression levels refer to the highest value seen for the gene. For Saos-2, low <20 TPM, moderate (mod) between 20 and 50 TPM, high >50 TPM. For OB-1, low <40 AU, mod 40-500 AU, high >500 AU. For OB-2 and VSMC, low <100 AU, mod 100-500 AU, high >500 AU. For NHO-1, low <20 AU, mod 20-150 AU, high >150 AU. For NHO-2, low <5 CPM, mod 10-80 CPM, high >200 CPM.

“No trend” indicates that there was no reproducible change with time across replicates. “No change” indicates that there was no significant difference among time points.

Abbreviations: Saos-2, Fantom5 project dataset (https://fantom.gsc.riken.jp/5/); OB-1, calvarial osteoblast microarray dataset GSE11339; OB-2, calvarial osteoblast microarray dataset E-MTAB-1391; VSMC, vascular smooth muscle microarray dataset E-MTAB-1680; NHO-1, neurogenic heterotopic ossification (NHO) microarray dataset GSE165062; NHO-2, RNA sequence dataset GSE218699.

Table 3.

Phenotypes associated with genes that showed variable expression during Saos-2 mineralization.

Gene	OMIM	Human phenotypes	Animal phenotypes
CBS	613381	Homocystinuria with osteoporosis	Growth retardation (Mouse) Decreased bone mineral density (Mouse) Impaired endochondral ossification (Mouse)
CHST3	143095	Spondyloepiphyseal dysplasia with congenital joint dislocations	Abnormal bone marrow morphology (Mouse)
CHST13	610124	Potential Kashin–Beck disease: chondrocyte necrosis, apoptosis, cartilage degeneration	No report
CTH	607657	Cystathioninuria Occasional talipes equinovarus	Skeletal muscle atrophy on cysteine-deficient diet (Mouse)
GOT1	138180	No report	No report
SGSH	252900	Mucopolysaccharidosis type IIIA Joint contractures Carpal tunnel syndrome	Mucopolysaccharidosis IIIA (Dachshund, Wire-Haired; Huntaway dog) Increased bone mineral content/density. Short tibia (Mouse)
SLC26A11	610117	No report	No report
SQOR	617658	Reduced bone mineral density	No report
SULF1	600383	Mesomelia-synostoses syndrome Decreased ossified bone volume	Reduced bone length, premature ossification of vertebrae, and fusion of sternebrae and tail vertebrae (Mouse)
SULT1A1	171150	No report	No report
TPST1	603125	Fetal loss and reduced body weight	No report
UST	610752	No report	No report

Information is from Online Mendelian Inheritance in Man (OMIM; https://omim.org) for human and Online Mendelian Inheritance in Animals (OMIA; https://omia.org) or Mouse Genome Informatics (MGI; https://www.informatics.jax.org) for animals, and references therein.

Sulfate biology in mineralization

Two carbohydrate sulfotransferase genes, CHST3 and CHST13 showed an increase between D0 and D28 in the Saos-2 mineralization time course. Chst3 and Chst13 were unchanged in the mouse osteoblasts. Mutations in CHST3 are associated with spondyloepiphyseal dysplasia with congenital joint dislocations (OMIM 143095). Features include short stature, spinal abnormalities, and abnormal ossification. CHST3 overexpression increased proliferation of cartilage endplate derived stem cells in bone marrow co-culture.⁶⁰ Although no bone phenotype has been associated with CHST13 mutations (OMIM Database; https://www.omim.org), there is a potential association with Kashin–Beck disease (KBD), which is characterized by chondrocyte necrosis, apoptosis, cartilage degeneration, and extracellular matrix degradation. Kashin–Beck disease is generally thought to be environmental.⁶¹ There were reduced levels of CHST3, CHST12, CHST13, and UST in metacarpophalangeal joint cartilage from KBD patients.^62,63 The level of Chst13 in the primary mouse cells undergoing mineralization was quite low. CHST13 may play a role in bone mineralization, a perspective that has yet to be thoroughly investigated.

No human conditions or mouse bone phenotypes have been associated with UST, another glycosaminoglycan sulfation enzyme. The role in bone formation is not clear, but the consistent increased expression in models of mineralization suggests that this enzyme may be important.

TPST1 is also involved in sulfate transfer, contributing to ECM stability by incorporating sulfate into secreted proteins, including proteoglycans, such as dermatan sulfate proteoglycans, and collagen, enhancing matrix resilience.^64–68 To date, there is limited evidence directly linking TPST1 mutations to bone phenotypes in humans or animal models. TPST1 KO (Tpst1−/−) mice exhibit early post-natal pulmonary failure, primary hypothyroidism and reduced body weight but no bone phenotypes.^66,69 In contrast, TPST2 KO (Tpst2−/−) mice show a moderate pubertal growth delay and infertility in males, with certain proteins being under-sulfated, indicating the importance of tyrosine sulfation.⁶⁹  TPST2/Tpst2 showed low expression in the Saos-2 time course and relatively high expression with little change over time or treatment in the mouse models. Further investigation of the role of TPST1 in bone biology is warranted by our results.

The increased expression of SULF1 and SGSH, encoding proteins that remove sulfate groups from heparan sulfate, likely supports extracellular matrix reorganization and the regulation of binding sites for growth factors, such as WNT and BMP family members, which are essential for osteoblast maturation.^70–72 The differences between human and mouse for both SULF1/Sulf1 and SGSH/Sgsh may reflect the differing requirements for heparan sulfate turnover in a cell line compared with primary cells or tissues. Sulf1 KO mice have defects in bone development⁷³ and SULF1 overexpression in Chinese hamster ovary cells released the inhibition of BMPs by the antagonist Noggin.⁷⁴ In quail, SULF1 was involved in early differentiation of chondrocytes.⁷⁵ In humans, mesomelia-synostoses syndrome (OMIM 600383) is associated with microdeletions within chromosome 8q13, always including SULF1 and SLCO5A1. Patients have skeletal abnormalities, including mesomelic limb shortening, acral synostoses, and specific craniofacial dysmorphisms. A potential synergistic effect between SLCO5A1 and SULF1 has been suggested.⁷⁶ These findings implicate SULF1 in the early stages of endochondral and intramembranous ossification. In the present study, SULF1 expression was 2-fold higher in the Saos-2 control sample (D28C) than in the treated sample (D28). It increased in treated Saos-2 samples but declined over the time course of mouse calvarial osteoblast and VSMC calcification. These observations suggest that the removal of sulfate from heparan sulfate is important early in mineralization but declines as mineralization progresses. The NHO models indicate a response to muscle damage rather than calcification.

Sulfate homeostasis is supported by several of the proteins encoded by genes that were regulated during mineralization. CTH, GOT1, and CBS all varied in at least one of the models. Modulation of CTH activity influences bone regeneration and repair; in traumatic occlusion mouse models, CTH aggravates periodontal damage, reducing alveolar bone height and producing morphological disorders in the periodontal ligament.⁷⁷  CTH was expressed by the Saos-2 cells and increased slightly over time. In the mouse osteoblast and VSMC models Cth expression reduced significantly after the first time point, suggesting that its role is required early in mineralization. Mutations in CBS result in homocystinuria and hyperhomcysteinemia (OMIM236200). Skeletal manifestations include osteoporosis, scoliosis, and long limbs and digits, possibly due to inhibition of collagen crosslinking by homocysteine. Homozygous null mice suffered from growth retardation and early mortality.⁷⁸ Cystathionine beta synthase is upregulated during the differentiation of mesenchymal stromal cells to osteoblasts.⁷⁹

GOT1 and SLC26A11 variants have not been associated with bone phenotypes to date⁸⁰ (https://www.informatics.jax.org/marker/phenotypes/MGI:95791; https://www.informatics.jax.org/marker/MGI:2444589). The substantial level of expression and the differing changes during mineralization of GOT1/Got1 suggest that this enzyme may be important in bone. Similarly, the constitutive or regulated level of SLC26A11/Slc26a11 mRNA in our studies suggests that this protein may be involved in mineralization.

The cytosolic sulfotransferases (SULT family) were found to have low to no expression in these models, which is consistent with the major roles of these sulfotransferases being Phase II metabolism of xenobiotic molecules in other tissues, such as the liver.⁸¹ The exception was Sult1a1, which increased over time in calcifying mouse primary cell models. SULT1A1 primarily mediates the sulfonation of phenolic compounds but can also sulfonate catecholamines, which leads to inactivation of catecholamine signaling.^82,83 This may be relevant to the known catabolic effects of catecholamines on bone.^26–28 SULT1A1 activity has been detected in some human osteoblast-like cells,^29,30 although the mRNA level in Saos-2 cells was very low at all time points. Known variations in copy number and single nucleotide polymorphisms in the human SULT1A1 gene could indirectly influence bone mineralization processes and skeletal integrity by altering hormone metabolism.^84,85

Strengths and limitations of the study

In this study, publicly available databases were used to assess expression of sulfate biology genes in various models of cellular mineralization. The use of a cell line, primary cells, and in vivo models of calcification is a strength of this study. While there are limitations to this approach, there was remarkable concordance among the studies.

The data came from 3 different modes of assessing gene expression, cap analysis of gene expression 5′-end profiling, RNA sequencing, and microarrays. Despite the emergence of RNA sequencing, microarrays (a fundamental method for gene expression profiling over many years) remain a valid tool in transcriptomic research, and there are many existing microarray datasets that can be mined for information, saving unnecessary human and animal experimentation. Microarray gene expression levels achieve good correlation with values from RNA sequencing.^86,87 However, RNA sequencing offers several advantages over microarrays, as it captures the entire transcriptome without the background noise associated with hybridization in microarrays.^88,89 Genome scale 5′-end profiling offers an alternative approach that can quantify expression based on transcript start sites and has the potential to identify alternative or regulated promotors.^47,49 The similarity of the results suggests that the findings of the present study are not platform-related.

The human Saos-2 model is a cancer cell line that has been in culture for many years. These cells exhibit osteoblast characteristics, including expression of osteocalcin (BGLAP gene) and alkaline phosphatase (ALPL gene). The line is considered to have a more mature osteoblastic profile than other osteosarcoma-derived cell lines, U-2 OS and MG-63.⁹⁰ Long-term cultured cancer cell lines have altered karyotype and gene expression patterns due to malignant transformation and prolonged culture.^91–93 The differences between the mouse and human cell results may be explained by the changes in the Saos-2 genome that have allowed the cells to escape some regulatory processes.

Primary osteoblasts in culture are closer than a long-term cancer cell line to the cells in vivo. In bone, macrophages are intercalated throughout bone tissues and actively contribute to remodeling and regeneration processes.^50,94,95 High expression of sulfate genes SULT1A1, SGSH, and SLC26A11 was found in human monocytes and macrophages (FANTOM5 dataset⁴⁷). Macrophages are present in calvarial osteoblasts cultures.⁵⁰ Indeed, co-culture of calvarial osteoblasts with macrophages was necessary for mineralization to occur in vitro.⁵⁰ In animal models in vivo, depletion of osteal tissue macrophages leads to impaired bone formation and decreased bone mineral density. Following transformation and many years in culture, the Saos-2 cell line has apparently escaped the requirement for macrophage input, but the primary cell and mouse tissue models contained macrophages (and their mRNA) as well as the primary osteoblast, vascular smooth muscle, or skeletal muscle cells.⁵⁰ This is evidenced by the increasing expression of the gene for the macrophage-specific colony stimulating factor 1 receptor (Csf1r gene) in the OB-1 and OB-2 datasets and a stable level in the VSMC dataset (Figure 6) and by the association of macrophages with NHO.^96,97 Some of the differences between the Saos-2 results and the mouse primary cell results may be due to changes in the Saos-2 genome or transcriptome reflecting this loss of the requirement for macrophages.

Figure 6.

Expression of Csf1r (a macrophage-specific gene) in mouse primary cells undergoing mineralization. Statistical analysis used ordinary one-way ANOVA with Tukey’s multiple comparison test. X axis—time points; Y axis—expression in AU. Significant p-values are shown: ^*p-adjusted < .05; ^**p-adjusted < .01; ^***p-adjusted < .001; ^****p-adjusted < .0001. The highest expressing probeset is presented.

The Saos-2 dataset included a long-term untreated control. This revealed that at least for this cell line, mineralization occurs even in the absence of the calcifying treatment, as evidenced by alizarin red staining of the cultures (shown in the appendix of the earlier paper⁴⁸). It is not clear whether the primary osteoblasts would also calcify under control culture conditions. Both the NHO-1 experiment and the Saos-2 Cluster0001 show that sulfate biology genes may have altered expression even in the absence of the mineralization trigger. The NHO-1 dataset revealed that some of the sulfate biology genes respond to muscle or spinal cord damage with no increase in expression in the doubly treated mineralizing samples. These findings do not preclude a role in mineralization but highlight the need for appropriate controls. In addition, the NHO datasets represent a single time point, which may miss important early or late changes in gene expression.

Conclusions

This analysis of sulfate biology gene expression during calcification in 6 different mouse and human models suggests that sulfate metabolism has an important association with formation and homeostasis of bone. The upward trend in expression of key sulfate biology genes during calcification indicates that these genes act synchronously to fortify the extracellular matrix, creating a suitable scaffold for inorganic mineral deposition. Although some of the genes (CHST13, SQOR, SULF1, and SGSH) have been associated with bone phenotypes in humans and/or animal models, we have now identified several genes (TPST1, UST, GOT1, SLC26A11, and SULT1A1) that have not previously been considered for adverse bone conditions. The bone phenotype of individuals carrying functional variants in these genes warrants further investigation, which may lead to novel treatments and increased understanding of bone development and maintenance. The network analysis revealed other sulfate biology genes that clustered with those examined here. These genes should also be studied further.

This expanded knowledge of sulfate pathways in mineralization introduces sulfate biology as a therapeutic focus for maintaining bone density and matrix integrity, presenting an alternative perspective on managing bone-related disorders. Targeting sulfate biology may help to stabilize extracellular matrix and promote mineralization, potentially preventing developmental bone deficiencies. For example, it may be relevant for managing osteopenia in preterm infants, who rapidly develop sulfate deficiency.⁹⁸ In addition, osteoporosis is an increasing cause of morbidity and mortality in the aging population, and new treatments are constantly being sought. Our study provides a collated list of sulfate-related genes and networks that are associated with mineralization, which lays the groundwork for further studies to elucidate the regulatory mechanisms of sulfate biology genes, potentially establishing them as therapeutic targets in bone health and disease management.

Author contributions

Kun-Di Lee (Conceptualization, Data curation, Methodology, Formal analysis, Investigation, Writing—original draft, Writing—review & editing, Visualization), Paul A. Dawson (Conceptualization, Writing—review & editing, Supervision, Funding acquisition), and Kim M. Summers (Conceptualization, Methodology, Formal analysis, Investigation, Data curation, Writing—original draft, Writing—review & editing, Visualization, Supervision)

Funding

This study was funded in part by a grant from the Australian National Health and Medical Research Council (grant number 2020999) to P.A.D. K.D.L. is funded by a University of Queensland Research Training Scholarship. P.A.D. is supported by a Mater Foundation Principal Research Fellowship. The authors are grateful for the core support from the Mater Foundation, Brisbane. The Translational Research Institute receives core funding from the Australian Government.

Conflicts of interest

The authors declare no conflict of interest.

Data availability

All the datasets used in this study are available in the public domain. Accession numbers are given in the text.