Esta referencia está disponible actualmente en inglés. La versión en español está en preparación.

Mesenchymal Stromal Cell Heterogeneity by Single-Cell RNA-seq

Por Benn Bluestein-Veyra, M.Sc. en Química Orgánica · Actualizado el 20 de octubre de 2026

MSC single-cell RNA-seq measures the transcriptome of each mesenchymal stromal cell in a culture, one cell at a time. Studies of cultured bone marrow, umbilical cord, adipose and dermal MSCs reported a shared core program with distinct subpopulations inside it.^[1][2][3]

This reference follows the question from clonal assays to droplet sequencing, describes how tissue source, passage and culture conditions shift the subpopulations, and maps the stromal cells of bone marrow in vivo.^[3][4][5]

Clonal assays first showed that MSC cultures are heterogeneous

Pittenger and colleagues isolated MSCs from the marrow of volunteer donors and induced them to differentiate into adipocytic, chondrocytic or osteocytic lineages. Individual cells expanded into colonies retained this multilineage potential.^[6]

Muraglia and colleagues tested 185 human bone marrow stromal clones. All clones but one formed bone, about one third were tripotent, and 60% to 80% showed osteochondrogenic potential.^[4]

Clones lost adipogenic and chondrogenic potential as cell doublings accumulated, which the authors read as a hierarchy in which tripotent progenitors give rise to osteochondrogenic and then osteogenic precursors.^[4]

Donors and colonies vary in growth and lineage potential

Phinney and colleagues expanded marrow stromal cells from 17 healthy donors aged 19 to 45. First-passage growth rates varied up to 12-fold between donors, and alkaline phosphatase activity varied up to 40-fold.^[7]

The authors concluded that each culture holds cells at various stages of differentiation, and that this mixture is propagated during expansion.^[7] Russell and colleagues built a 96-well clonal assay and detected all eight possible categories of trilineage potential in normal bone marrow MSCs.^[8]

In that assay, tripotent cells made up nearly half of the colony-forming cells and expanded more than unipotent clones, with about 2-fold higher CD146 fluorescence.^[8] Time-lapse imaging by Rennerfeldt and colleagues showed that diversity also emerges within single colonies, and that media exchange coincided with diversification.^[9][10]

Droplet and plate methods capture the transcriptome of each cell

A single-cell RNA-seq study begins with a suspension of single cells and a method that tags each transcript with the identity of its cell (Fig. 1).^[11][12] Two capture designs dominate MSC work: droplet systems that count transcript ends in thousands of cells, and plate methods that sequence full-length transcripts from fewer cells.^[12][13]

Zheng and colleagues described a droplet system that counts the 3′ ends of mRNAs in tens of thousands of cells per sample, with about 50% capture efficiency. They profiled about 250,000 cells across 29 samples and validated detection of rare populations with cell lines and synthetic RNAs.^[12]

Picelli and colleagues introduced Smart-seq2, which uses template switching and preamplification to raise the yield and length of cDNA from single cells. Its libraries improved detection, coverage, bias and accuracy over the original Smart-seq method.^[13]

The two designs trade cell number for depth. Droplet capture favors large surveys of subpopulations, and full-length plate methods favor coverage and detection within each cell.^[12][13]

Clustering, UMAP and trajectory analysis turn counts into subpopulations

Luecken and Theis set out the standard analysis: quality control, normalization, data correction, feature selection and dimensionality reduction, followed by cell-level and gene-level analysis. Their best-practice recommendations rest on independent comparison studies.^[11]

Clusters of cells with similar expression are then displayed in two dimensions. Becht and colleagues compared UMAP with five other tools on mass cytometry and single-cell RNA-seq data and found the fastest run times, highest reproducibility and most meaningful organization of clusters.^[14]

Trajectory methods order cells along a continuous process. Trapnell and colleagues introduced Monocle, which places cells in pseudotime and revealed switch-like changes in regulatory factors during differentiation of primary human myoblasts.^[15]

• Quality control removes low-quality cells and doublets before analysis.^[11]
• Cell cycle and batch effects dominated the variation in one Wharton’s jelly MSC dataset and were regressed out before clustering.^[2]
• Highly variable genes drive principal component analysis, graph-based clustering and the UMAP display.^[2][14]
• Pseudotime orders cells by transcriptional progress and supports inference of transcription factors along a path.^[15][16]

MSC single-cell RNA-seq of cultured cells reveals a structured population

Freeman and colleagues applied single-cell RNA-seq to mouse bone marrow MSCs. Genes associated with multipotency were expressed at high, consistent levels, while osteogenic, chondrogenic, adipogenic, neurogenic and smooth muscle genes varied widely between cells.^[17]

Certain immunomodulatory genes also varied from cell to cell, independently of proliferation cycles. The authors read the pattern as lineage priming, in which individual cells already express genes of different lineages.^[17]

Cell cycle state accounts for much of the variation in cord MSCs

Huang and colleagues profiled 361 umbilical cord MSCs from two cords at different passages, with and without IFN-gamma and TNF-alpha. Network analysis found limited heterogeneity across donors and passages, and cytokine exposure produced uniform changes.^[1]

The variation that remained tracked entry into G2/M phase. One featured gene, CD168, was expressed in a cell cycle-dependent way, and sorted CD168-high cells regenerated the same expression pattern in culture.^[1]

Wharton’s jelly MSCs separate into proliferative, matrix and chemokine subsets

Sun and colleagues sequenced Wharton’s jelly MSCs from three donors and found that cell cycle and batch were the dominant sources of variation. After regressing both out, graph-based clustering identified six candidate subpopulations (Fig. 2).^[2]

MKI67 was higher in two clusters, collagen genes in another and chemokine genes in a fourth, while classic MSC markers were expressed at similar levels across clusters. Sorted CD142+ and CD142- cells differed in proliferation and in a scratch migration assay.^[2]

The highly variable genes mapped to chemokine and cytokine signaling, integrin signaling and angiogenesis. Wharton’s jelly and adipose MSCs shared most of these genes despite their different tissue origins.^[2]

Tissue source, passage and culture history shift the subpopulations

Wang and colleagues built an atlas of more than 130,000 single-MSC transcriptomes from bone marrow, adipose tissue, umbilical cord and dermis. They identified five conserved subpopulations shared across tissues and seven tissue-specific ones (Fig. 3).^[3]

Extracellular matrix genes contributed most to heterogeneity in the atlas, and tissue-specific subpopulations differed in matrix-associated immune regulation, antigen processing and presentation, and senescence. An umbilical cord-specific subpopulation differed in immunosuppression-related gene expression.^[3]

Uncultured cord cells and their cultured descendants differ

Zhang and colleagues sequenced freshly isolated, uncultured umbilical cord MSCs and found two populations separated by genes including CD73. One group was enriched for immune regulatory and stemness genes, the other for extracellular matrix production and osteoblast and chondrocyte differentiation.^[18]

Wang and colleagues compared cord tissue with its cognate culture-expanded MSCs. They identified three subgroups in culture, inferred transcription factors for state transitions by trajectory analysis, and linked telomere-related pathways to expansion in vitro.^[16]

Serial passage adds pre-senescent and senescent states

Taherian Fard and colleagues followed serially passaged MSCs derived from human embryonic stem cells. Cells passed through pre-senescent states before entering three distinct senescent states, and gene regulatory networks lost connectivity along the way.^[19]

Donor, harvest and medium handling shape the starting mixture as well. Differences between donors persisted across repeat aspirates, and media exchange coincided with diversification within colonies.^[7][9]

Bone marrow stroma in vivo centers on CXCL12-abundant reticular and LEPR+ cells

CXCL12 is a small chemokine encoded on human chromosome 10q11.21.^[20] Murphy and colleagues solved CXCL12 with heparin disaccharides, finding one bound in the dimer interface by His25, Lys27 and Arg41 (Fig. 4).^[21]

Sugiyama and colleagues found most hematopoietic stem cells in contact with stromal cells expressing high CXCL12, which they named CXCL12-abundant reticular (CAR) cells. CAR cells surrounded sinusoidal endothelium or sat near the endosteum.^[22]

Omatsu and colleagues showed that CAR cells are adipo-osteogenic progenitors. Their short-term ablation in mice reduced the adipogenic and osteogenic potential of marrow cells and their production of SCF and CXCL12.^[23]

The authors proposed that glycosaminoglycans bind around CXCL12 dimers as they hold and present the chemokine to its receptor CXCR4.^[21]

LEPR marks the colony-forming stromal cells of adult marrow

Zhou and colleagues found that leptin receptor (LEPR) highly enriches mouse bone marrow MSCs. About 0.3% of marrow cells were LEPR+, and these cells accounted for 94% of the colony-forming fibroblast units.^[24] Human LEPR is encoded on chromosome 1p31.3.^[25]

LEPR+ cells formed bone, cartilage and adipocytes in culture and in vivo, and expressed high Cxcl12 and Scf reporters. Fate mapping showed that they arise after birth and give rise to most bone and adipocytes formed in adult marrow.^[24]

Single-cell and spatial maps resolve the marrow niche

Tikhonova and colleagues mapped mouse marrow vascular, perivascular and osteoblast populations at single-cell resolution. They found previously unappreciated heterogeneity and resolved which cells produce hematopoietic growth factors, chemokines and membrane-bound ligands.^[26]

The same study profiled the niche during stress-induced hematopoiesis and found considerable transcriptional remodeling, including an adipocytic skew of perivascular cells.^[26]

Baccin and colleagues combined single-cell and spatial transcriptomics and split CAR cells into Adipo-CAR and Osteo-CAR subsets. These localized to sinusoidal and arteriolar surfaces and acted as cytokine-secreting cells that form perivascular micro-niches.^[5]

Their RNA-Magnet algorithm inferred the three-dimensional organization of the marrow from single-cell transcriptome data, which the authors offer as a systematic approach to whole organs.^[5]

Wolock and colleagues sequenced non-hematopoietic marrow cells and combined a computed cell state hierarchy with transcription factor expression. They mapped differentiation paths to the osteocyte, chondrocyte and adipocyte lineages and validated them with reporter strains.^[27]

Single-cell data add resolution to MSC characterization

The ISCT minimal criteria define MSCs by plastic adherence, expression of CD105, CD73 and CD90 without hematopoietic markers or HLA-DR, and trilineage differentiation in vitro.^[28] Single-cell data show that cells meeting these criteria share the classic markers across subpopulations.^[2]

The subpopulations differ in cell cycle state, matrix genes, chemokine genes and lineage priming.^[1][2][17] Recording donor, tissue, passage and culture conditions alongside the sequencing data lets laboratories compare these features across studies.^[3][11]

Single-cell data also connect culture biology to tissue biology. Immune regulatory and matrix-forming groups appear in uncultured cord cells and in the multi-tissue culture atlas, placing cultured subpopulations in a tissue context.^[3][18]

Surface markers found this way can be tested directly. Sorting on CD168 or CD142 linked a transcriptional cluster to a measurable property in culture.^[1][2]

Where to go from here

The ISCT minimal criteria reference covers the marker panel that defines the cells sequenced in these studies, and the source tissues reference compares bone marrow, adipose tissue, umbilical cord and placenta. The licensing and preconditioning reference follows the cytokine exposures that shift MSC gene expression in culture.

The secretome reference describes what MSC cultures release into their medium, and the perinatal tissue donor eligibility reference explains the federal framework for cord and placental material. Verified practices can request lot documentation, and our team is glad to discuss the methods described on this page.