Introduction

Mesenchymal stem cells (MSCs) are pluripotent that can be obtained from various tissues, umbilical cord, amniotic fluid, and placenta1,2,3,4. It is capable of expanding and differentiating in vitro into a variety of mesodermal-type lineages, including bone, adipose, cartilage, muscle, tendon, and stroma, supporting hematopoiesis and the vascular system, and undergoes migration and paracrine signaling, which is not only crucial in embryonic development but also plays an important role in tissue homeostasis and repair throughout the life of an organism5,6. MSCs from different sources exhibit different differentiation tendencies, and studies have shown that BMSCs exhibit greater APL activity, calcium deposition, and transformation capacity at an earlier age,compared with adipose-derived MSCs7. Therefore, BMSCs were chosen for studying the molecular mechanisms of osteogenic differentiation.

Key transcription factors for osteoblast differentiation are required for the expression of osteoblast-specific genes, including Runt-associated transcription factor 2 (Runx2) and osteocalcin (Ocn), alkaline phosphatase (ALP), type I collagen, bone bridging proteins, and bone sialic acid protein8,9,10,11,12,13. Bone morphogenetic proteins (BMPs), especially BMP-2, BMP-4, and BMP-7, activate the Smad signaling pathway, which in turn leads to the expression of Runx2 and Ocn and inhibits adipogenic differentiation14,15,16. In addition to BMP, the Wnt/\(\beta \)-catenin signaling pathway plays a key role in osteoblast differentiation17. Activation of this pathway leads to the accumulation of \(\beta \)-catenin in the cytoplasm, which then translocates to the nucleus, activates the transcription of osteoblast-specific genes, and inhibits adipogenesis. On the other hand, adipocyte differentiation is regulated by peroxisome proliferator-activated receptor \(\gamma \) (PPAR\(\gamma \))18. PPAR\(\gamma \) is considered a major regulator of adipogenesis because it controls the expression of many adipocyte-specific genes. PPAR\(\gamma \) activation not only promotes adipogenesis but also inhibits osteogenesis19. The balance between osteoblast and adipocyte differentiation is a tightly regulated process20. Therefore, understanding the mechanisms that regulate this balance is essential for developing therapeutic approaches for diseases such as osteoporosis and obesity, which are characterized by an imbalance between bone and fat formation.

In mesenchymal stem cells, cellular matrix components play a key regulatory role in the directed differentiation of mesoderm into osteoblasts or adipogenic cells21. The extracellular matrix is a complex network of many molecules, including collagens, glycoproteins (GPs), and ECM-associated proteins22. Specific components of the ECM can influence cell fate decisions; for instance, collagen promotes osteoblast differentiation, whereas fibronectin may promote adipogenic cell differentiation23,24. Cell adhesion molecules and cell signaling molecules in the cell matrix are important regulators of the differentiation process, transmitting signals and modulating cell behavior by interacting with receptors on the cell surface25,26. For example, integrins increase the ability of cells to differentiate osteoblastically, maintain bone homeostasis, and regulate bone mass while inhibiting the adipogenic differentiation of BMSCs27,28,29. Taken together, the study of cellular matrix components contributes to our understanding of the mechanisms of germ layer-directed differentiation.

BMSCs exhibit the most pronounced difference between days 14 and 17 of differentiation, and the difference between days 17 and 21 of differentiation is not significant7. Therefore, our study combined tri-lineage differentiation with bulk RNA-seq analysis and single-cell RNA-seq (scRNA-seq) analysis to determine which factors simultaneously play key roles in osteogenic, adipogenic, and chondrogenic terminal phase differentiation, and cell staining and RT-qPCR were used to verify the changes in target genes during differentiation, aiming to provide a new basis for MSC therapies.

Results

hBMSCs osteogenic, adipogenic, and chondrogenic differentiation 14-day transcriptome data cross-identify 31 DEGs

A preliminary comparison of the six datasets (GSE36923, GSE44303, GSE109503, GSE140861, GSE28205, GSE37558) was performed before the differential gene analysis. Through the box-and-line plot of the distribution of the 6 datasets, we concluded that there was no batch effect within the datasets (Fig. 1a–c). According to the PCA plot of the distribution of the 6 datasets, the differences between different types of data were significant, and the analysis was feasible (Fig. 1d–i). Differential gene analysis of different types of data was performed separately, and DEGs coexpressed in similar datasets were selected to obtain 347 upregulated genes and 263 downregulated genes related to osteogenic differentiation, 658 upregulated genes and 704 downregulated genes related to adipogenic differentiation, and 106 upregulated genes and 48 downregulated genes related to cartilaginous differentiation (Fig. 2a–i). There was an antagonistic relationship between the three classifications. To identify the genes that were coexpressed in each category based on the Venn diagram, the genes coexpressed in osteogenic, adipogenic, and chondrogenic differentiation were compared two by two with opposite differentiation trends, and a total of 31 DEGs were obtained (Fig. 2j–o).

Figure 1
figure 1

Quality of RNA-seq data of bone marrow-derived MSCs at the late stage of differentiation. Box plot: (a) osteogenic differentiation dataset (GSE28205, GSE37558); (b) adipogenic differentiation dataset (GSE36923, GSE44303); (c) chondrogenic differentiation dataset (GSE109503, GSE140861); PCA plot; (d) GSE37558; (e) GSE36923; (f) GSE140861; (g) GSE28205; (h) GSE44303; (i) GSE109503.

Full size image
Figure 2
figure 2

Target gene identification and enrichment analysis of RNA-seq data from bone marrow-derived MSCs at the late stage of differentiation (a,b) volcano plots of differential gene expression in osteogenic differentiation datasets (GSE28205 and GSE37558); (c,d) volcano plots of differential gene expression in adipogenic differentiation datasets (GSE36923 and GSE44303); (e,f) volcano plots of differential gene expression in cartilageogenic differentiation datasets (GSE109503 and GSE140861) volcano plots of differential gene expression; (g) heatmap of differential gene expression of osteogenic differentiation datasets (GSE28205 and GSE37558) after integration by the RRA algorithm; (h) heatmap of differential gene expression of adipogenic differentiation datasets (GSE36923 and GSE44303) after integration of by the RRA algorithm; (i) heatmap of differential gene expression of chondrogenic differentiation dataset (GSE109503 and GSE140861) heatmap of differential gene expression after integration by the RRA algorithm. Venn diagrams of DEGs with opposite differentiation trends in the three differentiation directions. A total of 31 DEGs were obtained. (j) Intersection of osteogenic upregulated differential genes and adipogenic downregulated differential genes; (k) intersection of osteogenic upregulated differential genes and chondrogenic downregulated differential genes; (l) intersection of chondrogenic upregulated differential genes and adipogenic downregulated differential genes; (m) intersection of adipogenic upregulated differential genes and chondrogenic downregulated differential genes; (n) intersection of adipogenic upregulated differential genes and chondrogenic downregulated differential genes; (o) intersection of adipogenic upregulated differential genes and osteogenic downregulated differential genes; (p) differential gene enrichment analysis (GO) bar graph; (q) differential gene enrichment analysis (KEGG) network graph. AD adipogenesis, OS osteogenesis, CH chondrogenesis.

Full size image

Enrichment analysis revealed that DEGs were concentrated in the extracellular matrix and related to ECM-receptor interactions

Gene Ontology (GO) analysis revealed that the differentially expressed genes (DEGs) associated with an adjusted p-value (adj. P) < 0.05 were primarily enriched in biological processes and molecular functions related to collagen-containing extracellular matrix, hexosaminidase activity, binding to laminin, structural components of the extracellular matrix that confer tensile strength, and integrin binding (Fig. 2p). At KEGG pathway analysis revealed that DEGs with adj.P < 0.05 were associated with the ECM-receptor interaction, local adhesion, and glycolipid biosynthesis-ganglio series pathways (Fig. 2q).

ceRNA network construction

Using the String website, six target genes enriched in extracellular mesenchymal tissues were identified, namely, Hapln1, Col4a1, Lamc1, Itga10, Col10a14, and Tln2, and two genes enriched in the target genes involved in the biosynthesis of glycosphingolipid-ganglio series pathways were identified, namely, Hexb and St3gal5 (Fig. 3a,b). The gene symbols, abbreviations and functions are shown in Table 1. The two PPI networks of the target genes (60 nodes, 115 edges; 19 nodes, 20 edges) were obtained through the MCODE module of Cytoscape, with retention greater than or equal to 2, K-Core taken as 2, and Max. depth taken as 100 nodes. mRNAs are pink, miRNAs are purple, and lncRNAs are orange; the greater the connectivity of the nodes is, the greater the number of nodes. Lamc1 and Col4a1 are the core of the mRNA-miRNA-lncRNA interaction network and are closely related to noncoding RNAs that have been shown to play important roles in osteogenesis or adipogenesis, such as MALAT1, H19, XIST, and NEAT1.

Figure 3
figure 3

Expression of target genes in the database (a) mRNA–miRNA–lncRNA interaction network map composed of Hapln1, Col4a1, Lamc1, Itga10, Col10a1, and Tln2; (b) mRNA–miRNA–lncRNA interaction network map composed of Hexb and St3gal5. Expression in the HPA database; (c) histogram of differential gene tissue expression levels; (d) histogram of differential gene gene expression levels; expression in the Bgee database; (e) histogram of differential gene tissue expression levels.

Full size image
Table 1 Eight hub genes and their functions.
Full size table

Target gene expression was higher at both the adipose gene level and tissue level than at bone tissue

Analysis of the Human Protein Atlas (HPA) database revealed that the expression levels of Col4a1 and Lamc1 were significantly higher in adipose tissue compared to bone marrow tissue, both at the gene and tissue levels. Similarly, Tln2 exhibited slightly elevated gene expression in adipose tissue relative to bone marrow (Fig. 3c,d). Interestingly, while Hexb and St3gal5 displayed moderately higher gene expression in adipose tissue than in bone marrow, their tissue-level expression patterns were reversed, with higher levels observed in bone marrow compared to adipose tissue. The gene expression scores of the hub genes in abdominal adipose tissue, omental fat pads, subcutaneous adipose tissue, synovial joints, cartilage tissue, trabecular bone tissue, tibia, bone marrow, and myeloid cells were compared using the data obtained from the BGEE database. Col4a1 and Lamc1 had the highest expression in abdominal adipose tissue, omental fat pads, and subcutaneous adipose tissue, and Hexb had the highest expression in synovial joints, trabecular bone tissues, and bone marrow cells. Hapln1 was most highly expressed in cartilage tissues, and only Tln2 and Itga10 were expressed in bone marrow (Fig. 3e).

BMSCs are categorized into 11 clusters at single-cell resolution

We used t-distributed stochastic neighbor embedding (tSNE) to investigate cellular heterogeneity within clusters, and after the integration of the two datasets (GSE128423, GSE156635) (n = 44053), we identified 21 cell clusters and 11 cell types-osteoblasts (n = 2411), chondrocytes (n = 8251), adipogenic cells (n = 5396), endothelial cells (n = 7682), fibroblasts (n= 6949), neutrophils (n = 912), lymphocytes (n = 3957), megakaryocytes (n= 682), erythroblasts (n= 735), stem cells (n = 4908) and blastocytes (n = 2170) -and the biomarker expression of each subgroup is given (Fig. 4a,b). Erythrocytes were not removed since our focus was on the directed differentiation of stem cells to adipocytes, osteoblasts, and chondrocytes. The distribution of the target genes is shown in Fig. 4c.

Figure 4
figure 4

Subcluster identification of bone marrow-derived MSC single-cell sequencing data. (a) The TSNE algorithm was applied to the first 17 PCAs for dimensionality reduction, and 22 cell clusters were successfully classified; (b) cell clusters were manually annotated with CellMarker 2.0 according to the composition of marker genes and were successfully annotated into 11 subpopulations; (c) the expression levels of the target genes (Col4a1,Lamc1,Itga10,Hapln1,Hexb,Col10a1,Tln2,St3gal5)in the subpopulations; (d) the marker gene bubble map. Reliability of subcluster annotation of single-cell data.

Full size image

Scoring of scRNA-seq annotation reliability proves cluster annotation reliability

We chose three methods to evaluate the reliability of the single-cell annotations. First, we used the WNT pathway and the ADIPOGENESIS pathway to sort the cells (Fig. 5a,b). The WNT pathway was mainly distributed in osteoblasts, adipocytes, stromal cells, and chondroblasts, and the ADIPOGENESIS pathway was mainly distributed in adipocytes, stromal cells, and megakaryocytes, which was in line with our biological a priori experience. Then using the local similarity between cells, the stemness and differentiation potential of cells were assessed according to CytoTrace, and the degree of differentiation was in the following order: stem cells < chondrocytes < osteoblasts < adipoblasts. Col4a1 and Lamc1 were mainly distributed in adipocytes and osteoblasts; Tln2 and Hexb were expressed in chondrocytes, osteoblasts, and adipocytes; Col10a1 was distributed in a small number of chondrocytes, Itga10 and Hapln1 were distributed in a large number of chondrocytes, and St3gal5 was distributed in a large number of stem cells (Fig. 5c). Using Monocle2 to perform a time-matching analysis of the four types of cells, it was concluded that the BMSCs can be divided into two different types of sources, both of which can be differentiated into chondroblasts, and that the later stage is accompanied by the differentiation of the adipoblasts with the osteoblasts with a large number of adipoblasts (Fig. 5d). Finally, cell cycle maps for all the subpopulations were generated and we concluded that stem cells, lymphocytes, and neutrophils were in the early stage of differentiation, while the remaining subpopulations were in the late stage of differentiation (Fig. 5e).

Figure 5
figure 5

Proof of subgroup annotation reliability (a) the WNT pathway and the adult ADIPOGENESIS pathway were selected to score the cell subclusters; (b) the differentiation of the four classes of cells was assessed according to Cytotrace; (c) the expression levels of the target genes (Col4a1,Lamc1,Itga10,Hapln1,Hexb,Col10a1,Tln2,St3gal5)according to Cytotrace; (d) the differentiation of the four classes of cells was assessed based on Monocle2; (e)the differentiation of the 11 classes of cells was assessed based on Tricycle.

Full size image

Cell cycle analysis of target genes

Based on the scatter plots, Col4a1, Lamc1, Hapln1, Itga10, Tln2, Hexb, were expressed at all times in osteoblasts, Col10a1 and St3gal5 were expressed to a lesser extent, and Itga10 with Lamc1 became more abundantly expressed at mitotic phase (M). Scatter plots of Runx2 and Bglap expression cycles are given as reference (Fig. 6). In adipocytes, Col4a1 and Lamc1 were significantly more expressed than in osteoblasts, being abundantly expressed during mitosis (M), and Itga10, St3gal5, and Col10a1 were expressed during DNA synthesis (S), which is consistent with our previous prediction. Scatter plots of Ppar\(\gamma \) and Fabp4 are given as reference (Fig. 7). In chondrocytes, Hapln1 and Itga10 expression is reduced and Tln2 expression is elevated in mitosis (M). Scatter plots of Sox9 and Col2a1 are given as reference (Fig. 8).

Figure 6
figure 6

Expression period of target genes of Lamc1 interaction in osteoblasts (a) Col4a1; (b) Lamc1; (c) Itga10; (d) Hapln1; (e) Hexb; (f) Col10a1; (g) Tln2; (h) St3gal5;(i) Bglap; (j) Runx2.

Full size image
Figure 7
figure 7

Expression period of target genes of Lamc1 interaction in in adipocytes (a) Col4a1; (b) Lamc1; (c) Itga10; (d) Hapln1; (e) Hexb; (f) Col10a1; (g) Tln2; (h) St3gal5;(i) Adipoq; (j) Fabp4.

Full size image
Figure 8
figure 8

Expression period of target genes of Lamc1 interaction in chondrocy (a) Col4a1; (b) Lamc1; (c) Itga10; (d) Hapln1; (e) Hexb; (f) Col10a1; (g) Tln2; (h) St3gal5; (i) Col2a1; (j) Sox9.

Full size image

Intercellular interactions in adipogenic, osteogenic differentiation target two pairs of ligands that interact with Lamc1

In BMSCs, osteoblasts communicate with adipogenic cells when adipogenic cells are used as receptors, and all three types of cells communicate with stem cells when stem cells are used as receptors (Fig. 9a). Considering that the pathway enriched for target genes is the ECM-receptor interaction pathway, to determine the mechanism of interactions in the stroma, we investigated the most significant ligand–receptor interactions in the LAMININ pathway (Fig. 9b). Lamc1-CD44 and Lamc1-Dag1 exhibited extreme differentiation, followed by interactions at the level of individual ligands, and the ligands that interact with Lamc1 were identified. LAMC1-CD44 interaction may affect stem cell differentiation to bone or cartilage and adipose, and LAMC1-DAG1 interaction may affect osteogenic, adipogenic, and chondrogenic differentiation relationships (Fig. 9c–e).

Figure 9
figure 9

The molecular mechanism of Lamc1 interaction (a) hierarchical plot of cellular communication; (b) the most important ligand-receptor interactions in the laminin pathway (top 20); (c) LAMC1-Dag1 interaction chord diagram; (d) LAMC1-CD44 interaction chord diagram; (e) Demonstrate the specific ligand–receptor interaction mechanisms of stem cells, osteoblasts, and adipogenic and chondrogenic cells in the Laminin pathway.

Full size image

Target genes promote osteogenic differentiation and inhibit adipogenic differentiation

The target genes Col10a1 and Hapln are predicted to play important roles in chondrogenic differentiation, and we have confirmed this in past reports. Similarly, Col4a1 and Itga10 have been shown to play important roles in osteogenic differentiation, and therefore, no subsequent validation will be performed. During osteogenic differentiation, the alizarin red staining results revealed gradual reddening with increasing days of induction, with MC3T3-E1 cells reaching the reddest at 7 days (Fig. 10a,b). We detected the expression of target genes and marker genes at 3 days, 5 days, and 7 days. Lamc1, Tln2, Hexb, and St3gal5 tended to be upregulated, and the expression of the marker gene Runx2 gradually increased to a maximum at 7 days (Fig. 10c–e). During adipogenic differentiation, as the duration of induction increased, the red intensity of the 3T3-L1 cells gradually decreased, and on day 8, the red intensity of the 3T3-L1 cells decreased (Fig. 10f,g). We detected the expression of target genes and marker genes after 4 days, 6 days, and 8 days. Lamc1, Tln2, Hexb, and St3gal5 tended to be downregulated, and the expression of the marker gene Ppar\(\gamma \) gradually increased to a maximum on 8 days (Fig. 10h–j). Laminin can increase the expression of calcium and increase the concentration of osteocalcin. After Lamc1 was knocked down in MC3T3-E1 and 3T3-L1 cells, the gene expression levels of Runx2 and Ocn were significantly downregulated, and those of Ppar\(\gamma \) and Fabp4 were also downregulated (Fig. 11c–e,h–j).Similarly, the red color of the knocked-down cells was significantly lighter than that of the controls (Fig. 11a,b,f,g). Similarly, the previous experiment was repeated with the housekeeping gene GAPDH and yielded similar results (Supplementary Fig. S1).

Figure 10
figure 10

Results of target gene expression experiments. (a) Alizarin red staining of osteoblasts induced for 3, 5 and 7 days; (b) capture images with camera,100\(\times \) microscope,scale bar \(=500\;\upmu \hbox {m}\); qPCR results of target gene expression in osteoblasts: (c) 3 days; (d) 5 days; (e) 7 days. (f) Oil Red O staining of adipocytes induced for 4, 6 and 8 days; (g) capture images with camera, 100\(\times \) microscope, scale bar \(=\) \(500\;\upmu \hbox {m}\); qPCR results of target gene expression in adipocyte: (h) 4 days; (i) 6 days; (j) 8 days.

Full size image
Figure 11
figure 11

Results of target gene expression experiments after knockdown of Lamc1. (a) Alizarin red staining of osteoblasts induced for 3, 5 and 7 daysafter knockdown of Lamc1; (b) capture images with camera, 100\(\times \) microscope,scale bar \(=\) \(500\;\upmu \hbox {m}\); qPCR results of target gene expression after knockdown of Lamc1 in osteoblasts: (c) 3 days; (d) 5 days; (e) 7 days. (f) Oil Red O staining of adipocytes induced for 4, 6 and 8 days after knockdown of Lamc1; (g) capture images with camera, 100 \(\times \) microscope, scale bar \(=\) \(500\;\upmu \hbox {m}\); qPCR results of target gene expression after knockdown of Lamc1 in adipocyte: (h) 4 days; (i) 6 days; (j) 8 days.

Full size image

Methods

Data collection and processing

Gene expression profiling data (GSE36923, GSE44303, GSE109503, GSE140861, GSE28205 and GSE37558) were obtained from the Gene Expression Omnibus (GEO). GSE109503 contained RNA-seq data, and the remaining five gene sets were obtained from bulk RNA-seq data. GSE109503 and GSE140861 are chondrogenic differentiation datasets; GSE109503 has four subsets (two controls and two experimental groups); GSE140861 has six subsets (three controls and three experimental groups); GSE28205 and GSE37558 are osteogenic differentiation datasets; GSE28205 has six subsets (three control groups and three experimental groups); GSE37558 has six subsets (three control groups and three experimental groups); GSE36923 and GSE44303 are adipogenic differentiation datasets; GSE36923 has seven subsets (four control groups and three experimental groups); and GSE44303 has seven subsets (four controls and three experimental groups).

Single-cell RNA-seq data (GSE128423 and GSE156635) were obtained from the Gene Expression Omnibus (GEO) based on a 10\(\times \) Genomics assay with eight normal BMSC tissue samples. The summary results of the gene expression profiling dataset information are provided in Table 2.

Table 2 Summary information on gene expression profiling datasets.
Full size table

Microarray data were normalized ,and variance was analyzed using the Limma package in the R/Bioconductor software for matrix data from each GEO dataset, and RNA-seq data were normalized ,as was variance analyzed using the edgeR package. DEGs of the same cell type identified from each of the six datasets were integrated based on different differentiation categories using the RobustRankAggreg package. |log2FC| \(\ge \) 1 and P \(\le \) 0.05 were considered to indicate statistical significance.

PCA plots for each dataset were plotted using the FactoMineR package and the factoextra package in the R/Bioconductor software. pheatmap package was used to plot heat maps of the genes after data merging. ggplot2 package was used to plot volcano plots after data merging. Venn plots were generated using TBtools software (version x64 v1.09876). GO function enrichment and KEGG pathway enrichment were performed for 31 differential genes (DEGs) using the clusterProfiler package in R software. adj. P \(\le \) 0.05 was considered statistically significant. Bar and bubble plots were drawn using the ggplot2 package, and network plots were drawn using the enrichplot package.

scRNA-seq data were preprocessed using the Seurat package in R software for matrix data from each GEO dataset, the PercentageFeatureset function was used to determine the proportion of mitochondrial genes, and correlation analyses were used to investigate the relationships between sequencing depth and mitochondrial gene sequences and total intracellular sequences. Each gene was expressed in at least 3 cells. The gene expression in each cell was greater than 500 and less than 6000, the mitochondrial content was less than 30%, and the UMI in each cell was at least greater than 200. The scRNA-seq data were normalized by the NormalizeData method after data filtering, and highly variable genes were identified by the FindVariableFeatures method to compute the mean expression and dispersion and converted to logarithmic format. Next, the data from the two scRNA-seq datasets were integrated using the Harmony package, and then we performed principal component analysis (PCA) and reduced the data to the first 17 PCA components (the number of components selected based on the standard deviation of the principal components—in the platform area of the Elbow Graph). To visualize the data, we used the Seurat package for t-distribution random neighbor embedding (tSNE) to project the cells in two-dimensional space. We visualized clusters on 2D maps generated using t-distributed random neighbor embedding (t-SNE). Clusters were annotated according to CellMarker 2.0 (http://yikedaxue.slwshop.cn/), and the annotated clusters were scored using the AUCell package, selecting the WNT pathway-a key pathway for osteogenesis-and the ADIPOGENESIS pathway-a key pathway for adipogenesis. The cell cycle of the annotated clusters was inferred using the tricycle package, and the cell cycle in which the target genes are located was determined. The Cytotrace package was used to determine the stemness and differentiation potential of stem cells, chondrocytes, osteoblasts, and adipocytes, and based on the results of Cytotrace, Monocle2 was used to determine the direction of differentiation of the four types of cells and to determine the site of expression of the target gene.

The CellChat package was used to query the role of intercellular communication, and the ECM-Receptor database was selected for subsequent analysis based on the KEGG enrichment results. The identify OverExpressedGenes function of the CellChat package was used to identify highly expressed genes, the identify OverExpressedInteractions function was used to identify highly expressed pathways, the project Data function was used to project the relationship between the highly expressed genes and the pathways to the PPI network, the computeCommunProb function was used to calculate the probability of cell-to-cell communication, data with the minimum number of cells less than 10 were removed, the computeCommunProbPathway function was used to calculate the cell-to-cell communication at the level of signaling pathways, the aggregateNet function was used to calculate the aggregated cell-to-cell communication network, and the “LAMININ” signaling pathway was extracted.

The mechanism map was drawn via Figdraw (https://www.figdraw.com/static/index.html).

PPI network construction

The String website (https://cn.string-db.org/) analyzed protein functional interactions and selected interactions with a composite score>0.4. The RNAInter database (http://www.rnainter.org/) is a complete source of RNA interaction data for studying the relationships between mRNAs, lncRNAs, and microRNAs. The RNAInter database was used to identify miRNAs that interact with mRNAs, as well as lncRNAs that interact with miRNAs, and interactions with a score > 0.5 was selected. Cytoscape (version v3.9.1) is an open-source bioinformatics software platform for visualizing molecular interaction networks. The relationship between target genes and non-coding RNAs was plotted using Cytoscape.

Tissue distribution and gene expression identification

Protein and gene expression levels in bone marrow and adipose tissue were compared with those in the HPA database using The Human Protein Atlas (https://www.proteinatlas.org/) as a validation tool to determine whether there are differences in the expression of these genes in osteogenic and adipogenic differentiation. Expression scores in abdominal adipose tissue, omental fat pads, subcutaneous adipose tissue, tibia, synovial tissue, synovial joints, cartilage tissue, trabecular bone tissue, and bone marrow tissue were used to validate whether there are differences in the expression of these genes in osteogenic and adipogenic as well as cartilaginous differentiation using the BGEE database (https://bgee.org/). Bar graphs were drawn using GraphPad Prism 9.

Cell culture and differentiation

MC3T3-E1 cells (Procell Life, Science & Technology, China) were cultured in MEM\(\alpha \) medium supplemented with 10% fetal bovine serum (FBS), 100 units/ml penicillin and streptomycin at \(37\;^\circ C\) in a 5% \(\hbox {CO}_2\) incubator. The medium was changed every 2–3 days and the cells were used for experiments in which the cells were grown to the logarithmic growth phase. For osteogenic differentiation, MC3T3-E1 cells were cultured in osteogenic medium supplemented with 10 mM \(\beta \)-glycerophosphate and 50 \(\upmu \)g/mL L ascorbic acid osteogenic inducer. The cells were induced for 3, 5, or 7 days. 3T3-L1 cells (Procell Life, Science & Technology, China) were cultured in DMEM medium supplemented with 10% fetal bovine serum (FBS), 100 units/ml penicillin and streptomycin at \(37\;^\circ C \) in a 5% \(\hbox {CO}_2\) incubator.The medium was changed every 2-3 days, and the cells were used for experiments in which the cells were grown to the logarithmic growth phase. For adipogenic differentiation, 3T3-L1 cells were cultured in adipogenic induction medium supplemented with 0.5 \(\upmu \)M 3-isobutyl-1-methyl xanthine , 1 \(\upmu \)M dexamethasone and 10 \(\upmu \)g/mL insulin adipogenic inducer for 2 days, and the medium was replaced with adipogenic maintenance medium supplemented with 10 \(\upmu \)g/mL insulin for 2 days. The medium was finally replaced with DMEM, and the culture was continued for 2–4 days.

Alizarin red staining method and Oil Red O staining

After fixation with 80% ethanol for 30 min, cells were incubated with fresh alizarin red solution (Cyagen) for 10 min. After fixation with 4% neutral formaldehyde for 30 min, cells were incubated with saturated Oil Red O staining solution (Beyotime Biotechnology; Shanghai; China) for 10 min.

Quantitative reverse transcription PCR

Total RNA was extracted using triazole lysate lysis (Sangon Biotech; Shanghai; China), isopropanol, chloroform, and 75% ethanol solutions. Total RNA reverse transcription was induced using the HiScirpt Reverse Transcriptase kit (Novozymes Bio; Nanjing; China) The genes and primers used in the osteogenesis and adipogenesis process are listed in Table 3. Performed with a QuantStudio 5 real-time fluorescent quantitative PCR system (Applied Biosystems; Shanghai; China). Each sample was performed in triplicate. The sequences of the mRNAs were calculated according to the \(\Delta \Delta {\hbox {C}_{\hbox {T}}}\) relative stacking method. mRNAs were sequenced as Table 3.

Table 3 qPCR primer sequences.
Full size table

Lamc1 siRNA interference

Lamc1 siRNA was obtained from GenePharma Corporation. siRNA was transfected into MC3T3-El cells using the gp transfection mate (GenePharma Corporation) according to the manufacturer’s instructions. 24 h later, induction was initiated, and cells were collected 3–7 days later for alizarin red staining, and real-time quantitative RT PCR analysis. The siRNA was transfected into 3T3-L1 cells using gp transfection mate (GenePharma Corporation) according to the manufacturer’s instructions. After 90% confluence, induction was initiated, and the cells were collected 4–8 days later for oil red o staining, and real-time quantitative qPCR analysis experiments. The sequence of the siRNA was as follows. NC:5\(^\prime \)-UUCUCCGAACGUGUCACGUTT-3\(^\prime \) and 5\(^\prime \)-ACGUGACACGUUCGGAGAATT-3\(^\prime \); Lamc1-siRNA: 5\(^\prime \)-GCCGUAAUCUCAGACAGUUTT-3\(^\prime \) and 5\(^\prime \)- AACUGUCUGAGAUUACGGCTT-3\(^\prime \).

Molecular docking analysis

The Hdock website (http://hdock.phys.hust.edu.cn/) was used to perform simulated docking studies of Laminin-111 (Entry ID:5mc9)with DAG1 (Entry ID: 5llk) or CD44 (Entry ID:4pz3), the PymoL software (version v3.0.0) was used to perform pre-docking treatments (water molecules from the structure and co-crystallized ligands were removed, H atoms were added) and Pymol was used for the visual presentation of the docking complexes, with the results with the highest docking scores being selected.

Discussion

To investigate the molecular mechanisms of the multidirectional differentiation of BMSCs, especially osteogenesis, adipogenesis and chondrogenesis, BMSCs exhibited significant cellular heterogeneity, and an integrated analysis of the multiscale transcriptome was performed in the present study. These DEGs were analyzed by bulk RNA-seq and scRNA-seq, and it was found that the late stage of differentiation was the most significant in terms of interactions with the ECM receptor. It is well known that the ECM enhances cell recruitment through cell surface receptors, which determine cellular interactions with the ECM and trigger specific cellular functions such as adhesion, migration, proliferation, and differentiation30. Several studies have reported the critical role of the matrix in regulating the differentiation of MSCs toward the bone or adipose lineage30,31.Recent studies have shown that stem cells can mechanically sense the hardness of their microenvironment, and the direction of differentiation of MSCs may change with changes in the matrix composition32,33. For example, the extracellular matrix supplemented with glucan sulfate promotes the osteogenic differentiation of mesenchymal stem cells34, and type I collagen inhibits adipogenesis and differentiation by activating yes-associated proteins35. The latest research by Yangzi Jiang et al. showed that ECM produced by adult stem cells from different tissues has different cartilage induction abilities36. Osteoblasts are the main bone-forming cells, and they produce extracellular proteins that constitute the major components of bone. During bone formation, BMSCs differentiate into osteoblasts, which produce bone by synthesizing an extracellular matrix composed of various proteins. The initial deposition of the extracellular matrix is called the osteoid, which is then mineralized into bone tissue by the accumulation of hydroxyapatite (\(\hbox {Ca}_10(\hbox {PO}_4)_6(\hbox {OH})_2\)) as calcium phosphate37. As osteoblasts mature, they transform into osteocytes, and revealing the detailed mechanisms that regulate the later stages of osteoblast differentiation and function is important for clinical applications38. However, studies on the mechanisms underlying the differentiation of bone marrow-derived MSCs in the late stages of differentiation still need to be performed. We therefore selected an mRNA microarray dataset at the end of differentiation to identify genes that showed significant heterogeneity in the differentiation of BMSCs into adipocytes, osteoblasts, and chondrocytes. lamc1 tended to increase osteoblastic differentiation and decrease adipocytes, and Col4a1 had the opposite effect on differentiation. Highly expressed Lamc1 and Col4a1 play key roles in the protein interaction network, and their interactions with the key molecules involved in osteogenic, adipogenic, and chondrogenic differentiation, H19, XIST, NEAT1, and MALAT1, have been identified. According to our results, Hui-jian Chen et al. reported that Lama4 and Col4a1 are expressed in white adipose tissue at different sites and are upregulated during adipocyte differentiation39. Interestingly, the mRNA expression of Col4a1 was significantly increased after adipogenic induction of MSCs in a recent study, which is consistent with our expectations40. More importantly, laminin in the cell matrix contains three subunits: \(\alpha \) , \(\beta \), and \(\gamma \). Lamc1 is the most commonly used \(\gamma \) chain and is present in most laminin molecules41. During the process of osteogenesis, Laminin activates downstream signaling pathways such as Rho GTPase and MAPK by interacting with Integrin, which in turn regulates the processes of cellular bone morphogenesis and bone matrix deposition, and participates in the directional differentiation of stem cells to osteoblasts42,43.Lamc1 is present in LN-111 and LN-511, it has been shown to promote osteogenic differentiation by binding to integrins41,which are key factors for the development of osteoblasts. The HPA database, Bgee database, and single-cell transcriptome data analysis revealed that Lamc1 expression was significantly greater in adipocytes than in osteoblasts, but Lamc1 tended to promote bone formation and inhibit adipogenic, and this tendency became more obvious with increasing induction time. Further studies showed that after Lamc1 knockdown, the expression of osteogenic and adipogenic markers was significantly decreased, and differentiation was inhibited. It is therefore reasonable to assume that Lamc1 plays an important role in regulating the balance of bone adipogenesis44.

The distribution of target genes in bone marrow-derived MSCs, the cell cycle, and ligand–receptor interactions were determined by single-cell transcriptome analysis. Among these genes, Lamc1, which is expressed mainly during the G2/M phase in adipogenic cells and may affect intracellular signaling through interactions with Dag1, was used as a focus for determining the balance between osteogenic and adipogenic differentiation. Dag1 (also known as \(\alpha \)-glycoprotein) is involved in the adhesion between the extracellular matrix and cells, helps to maintain cell aggregation and structural stability, and participates in the assembly process of laminin and the basement membrane45. Recent studies have shown that Dag1 plays a role in cartilage formation and osteoblast differentiation in vivo46,47.Another finding is that the interaction of Dag1 with laminin has been reported48.To prove our prediction, molecular docking experiments were performed on laminin and Dag1. The experimental results showed that laminin and Dag1 formed a total of 11 hydrogen bonds (Fig. 12a). Our data support the conclusion that the interaction of Dag1 and Lamc1 may influence the balance between the adipogenic and osteogenic differentiation of MSCs through cell adhesion, signaling, and gene regulation (Fig. 12c). Similarly, a total of four hydrogen bonds were formed after molecular docking experiments were performed on laminin and CD44, initially validating the interaction between Lamc1 and CD44 (Fig. 12b).

Figure 12
figure 12

Molecular docking and mechanism maps (a) Results of molecular docking of laminin to DAG1; (b) Results of molecular docking of laminin to CD44; (c) mechanism maps.

Full size image

In conclusion, this study aimed to identify DEGs that may be involved in osteogenic differentiation, adipogenic differentiation, and the onset and progression of chondrogenic differentiation of bone marrow mesenchymal stem cells (BMSCs). A total of eight hub genes were identified, four of which were discovered and proposed for the first time in this study. Although these findings are preliminary, they suggest the existence of previously unknown regulatory relationships governing the differentiation process of BMSCs. Unfortunately, our experiments did not reveal new regulatory relationships between chondrogenesis and adipogenesis or osteogenesis, which may be related to our choice of a more conservative data merging approach.However, this conservative strategy ensured the accuracy of bioinformatic predictions. Despite these limitations, the identified hub genes and their potential interactions provide valuable insights and lay the foundation for further investigation into the molecular mechanisms underlying the multi-lineage differentiation of BMSCs.

Conclusion

This study aimed to identify the key biological processes involved in the late stages of differentiation of BMSCs. The following conclusions were demonstrated: eight target genes were identified in the context of trilineage differentiation antagonism, which is concentrated in the extracellular matrix and is associated with ECM-receptor interactions.Notably, Lamc1 expression was higher in adipose tissue than in bone tissue. Furthermore, Lamc1 exhibited maximal expression during the mitotic phase of the cell cycle. Bioinformatic analysis suggested that Lamc1 interacts with CD44 and may influence the differentiation of stem cells towards osteoblastic, chondrogenic, or adipogenic lineages. Additionally, the interaction between Lamc1 and Dag1 may affect the balance of trilineage differentiation in BMSCs. Experimental validation confirmed that Lamc1 was upregulated during osteogenic differentiation and downregulated during adipogenic differentiation, indicating its essential role in regulating the differentiation process.