TWIST1 and TSG6 are coordinately regulated and function as potency biomarkers in human MSCs

Abstract

Mesenchymal stem/stromal cells (MSCs) have been evaluated in >1500 clinical trials, but outcomes remain suboptimal because of knowledge gaps in quality attributes that confer potency. We show that TWIST1 directly represses TSG6 expression that TWIST1 and TSG6 are inversely correlated across bone marrow–derived MSC (BM-MSC) donor cohorts and predict interdonor differences in their proangiogenic, anti-inflammatory, and immune suppressive activity in vitro and in sterile inflammation and autoimmune type 1 diabetes preclinical models. Transcript profiling of TWIST1^HiTSG6^Low versus TWIST^LowTSG6^Hi BM-MSCs revealed previously unidentified roles for TWIST1/TSG6 in regulating cellular oxidative stress and TGF-β2 in modulating TSG6 expression and anti-inflammatory activity. TWIST1 and TSG6 levels also correlate to donor stature and predict differences in iPSC-derived MSC quality attributes. These results validate TWIST1 and TSG6 as biomarkers that predict interdonor differences in potency across laboratories and assay platforms, thereby providing a means to manufacture MSC products tailored to specific diseases.

TWIST1 and TSG6 are biomarkers that predict inter-donor differences in BM-MSC potency across laboratories and assay platforms.

INTRODUCTION

Mesenchymal stem/stromal cells (MSCs) represent a high-value resource for regenerative medicine applications due to their ability to support hematopoiesis (1–3), generate osseous tissue (4), and release paracrine factors (5, 6) and extracellular vesicles (7) in response to environmental cues that promote therapeutic effects. However, while MSC products have been tested in humans for various indications, outcomes have been largely suboptimal, and no MSC-based therapies are currently approved for use in the United States (8). A major limitation in development of efficacious MSC-based products with predictable therapeutic indexes in humans is that methods used to isolate cells yield functionally heterogeneous populations (5, 9–11), and no industry-wide standards exist for donor selection and manufacturing processes (12, 13). Moreover, in the United States, each MSC investigational new drug application is evaluated on the basis of its individual attributes and not compared to other existing MSC-based products (8), which confounds efforts to compare the potency and efficacy of different products in patients. The latter may explain why results from independent clinical trials are largely incongruent (14, 15). This problem is exacerbated by the paucity of metrics available to assess product heterogeneity in terms of both biological function and potency, which hampers efforts to select the most suitable “batch” of MSCs for treating a given disease indication or patient population. Consequently, many clinical studies produce MSCs using similar donor selection and manufacturing processes to treat diseases of unrelated pathogenesis.

To address these limitations, efforts have focused on identifying biomarkers that stratify patients into responders versus nonresponders (16) and define MSC products on the basis of potency rather than phenotype or composition of matter. While low-throughput assays that relate potency to quantifiable differences in cell morphology, gene expression, or activity in cell-based assays have been reported, these approaches are limited in scope because they seek to predict the potency or improve efficacy for one class of disease indication. We recently identified TWIST1 as a biomarker that predicts interdonor differences in stem/progenitor and paracrine activities of bone marrow (BM)–derived MSCs (BM-MSCs) (17). TWIST1 is a member of the basic helix-loop-helix family of transcription factors (18) and plays important roles in angiogenesis (19–22), maintenance of hematopoietic stem cells (23, 24) and skeletal muscle progenitors (25, 26), regulation of myeloid lineage determination (27), and cytokine signaling (28). We showed that TWIST1 mRNA levels predict interdonor differences in the growth, viability, colony-forming unit–fibroblast (CFU-F) activity, and trilineage differentiation potential of BM-MSCs; that silencing of TWIST1 down-regulated transcripts encoding proteins with proangiogenic activity and induced those with anti-inflammatory and immunomodulatory activity; and that manipulating TWIST1 levels predictably altered the capacity of BM-MSCs to support endothelial tubule formation and suppress activated CD3⁺ T cell proliferation in cell-based assays (29). BM-MSCs with disparate TWIST1 levels were also shown to exhibit predictable differences in anti-inflammatory activity in a mouse model of acute lung injury (29). Together, these results indicate that stem/progenitor and paracrine functions of BM-MSCs are mechanistically linked via the action of TWIST1, which we exploited to develop a Clinical Indications Prediction (CLIP) scale that discriminates BM-MSCs on the basis of proangiogenic versus anti-inflammatory/immunomodulatory activity (17, 29). A growing number of independent studies have demonstrated that proangiogenic activity exists at the expense of anti-inflammatory and immune suppressive activities as predicted by our CLIP scale (30). We and others have also identified tumor necrosis factor (TNF)–stimulated gene/protein 6 (TSG6) as an important mediator of MSC anti-inflammatory activity. For example, MSCs were shown to attenuate myocardial infarction–induced cardiac injury (31), peritoneal injury (32), acute pancreatitis (33), and zymosan-induced mouse peritonitis (34); reduce colitis (35, 36); and suppress corneal inflammatory lymph angiogenesis (37) in a TSG6-dependent manner. Lee et al. (38) also demonstrated that TSG6 expression levels exhibit significant interdonor heterogeneity and positively correlate with BM-MSC anti-inflammatory activity in mouse models of sterile inflammation and acute lung injury and negatively correlate with osteogenic potential (38). However, whether TWIST1 and TSG6 function independently or are mechanistically linked in BM-MSCs and whether cooperative gene effects enhance the predictive power of each biomarker are indeterminate.

Here, we demonstrate that TWIST1 represses TSG6 via direct promoter binding, that TWIST1 and TSG6 expression are inversely correlated in BM-MSC donor cohorts obtained from different sources and expanded across laboratories, and that these mRNAs predict interdonor differences in BM-MSC potency across assay platforms and preclinical disease models. Transcript profiling of TWIST1^HiTSG6^Low versus TWIST1^LowTSG6^Hi BM-MSCs implicated TWIST1 and TSG6 in regulating cellular oxidative stress responses and demonstrated that transforming growth factor–β2 (TGF-β2) modulates cell size and anti-inflammatory activity by up-regulating TSG6 expression. Last, we show that TWIST1 and TSG6 levels are positively and negatively correlated, respectively, with BM-MSC donor stature. Together, these data validate TWIST1 and TSG6 as robust potency biomarkers, confirm a mechanistic link between stem/progenitor and paracrine functions of BM-MSCs, and challenge the paradigm that MSC potency is engendered solely by interaction with the host microenvironment. Our results further highlight the importance of donor screening in clinical trial design, which, to date, has been prejudiced toward optimizing cell yields and provide a rational platform in the form of the CLIP scale to manufacture MSC products tailored to specific disease indications.

RESULTS

TWIST1 and TSG6 predict interdonor differences in BM-MSC growth, viability, and CFU-F activity

We previously reported that population-averaged expression levels of TWIST1 and FGFR2IIIc positively correlate with growth/viability and CFU-F activity of BM-MSCs sourced from different human donors (29). Analysis of a separate donor cohort (RD, n = 8) from a commercial vendor revealed significant interdonor differences in population-averaged growth/viability (F_7,24 = 37.46, P < 0.0001) and CFU-F activity (F_7,40 = 32.32, P < 0.0001; Fig. 1A) and normalized TWIST1 (F_7,38 = 17.76, P < 0.0001) and FGFR2IIIc mRNA levels (F_7,38 = 59.9, P < 0.0001; Fig. 1B). Regression analysis demonstrated that TWIST1 and FGFR2IIIc mRNA levels positively correlate with cell growth/viability (Fig. 1C) and CFU-F activity (Fig. 1D) and with each other in these donors (Fig. 1E). Analysis of a third donor cohort (BMH, n = 5) expanded in human platelet lysate (hPL) supplemented media also exhibited significant interdonor differences in growth/viability (F_4,15 = 6.33, P = 0.0034), CFU-F activity (F_4,25 = 26.12, P < 0.0001), and TWIST1 mRNA levels (F_4,25 = 15.27, P < 0.0001), and the latter significantly correlated with CFU-F activity in these donors (Fig. 1F). Pooling data from all three donor cohorts evaluated to date revealed a significant positive correlation between TWIST1 and FGFR2IIIc versus CFU-F activity (Fig. 1G), demonstrating the robustness of these relationship across donors. In addition, we found that normalized TSG6 levels also exhibited significant interdonor variability (F_8,46 = 148.6, P < 0.0001; Fig. 1H), consistent with previous reports (38), and negatively correlated with TWIST1 mRNA levels, cell growth/viability, and CFU-F activity (Fig. 1I).

Fig. 1. TWIST1 and TSG6 are inversely correlated and predict interdonor differences in BM-MSC attributes.

(A and B) Population-averaged differences in growth/viability and CFU-F activity (A) and TWIST1and FGFR2IIc mRNA levels (B) in BM-MSCs (n = 8). (C and D) Scatter plots of TWIST1and FGFR2IIIc levels versus cell growth/viability (MTT assay) (C) and CFU-F activity (D) in BM-MSCs from (A). (E) Scatter plot of TWIST1 versus FGFR2IIc mRNA levels in BM-MSCs from (A). (F) Scatter plot of TWIST1 versus CFU-F activity in BM-MSCs from a separate donor cohort (BHM) expanded in hPL. (G) Scatter plots of TWIST1 and FGFR2IIIc versus CFU-F activity based on data pooled from n = 3 donor cohorts [7000: n = 8, blue; RD from (A): n = 8, red; BHM from (F): n = 5, green]. (H) Population-averaged differences in TSG6 mRNA levels in BM-MSCs from (A). (I) Scatter plots of TSG6 versus TWIST1, cell growth/viability (MTT), and CFU-F activity in BM-MSCs from (H). Data in (A), (B), and (H) are means ± SD from three to six technical replicates per sample with *P < 0.05, **P < 0.01, and ***P < 0.005 versus RDO1 by one-way analysis of variance (ANOVA) with Tukey post hoc test. Data in (C) to (G) and (I) are means ± SD from four to six technical replicates per sample, and r is the Pearson’s product moment correlation coefficient with corresponding P values determined by linear regression. OD₅₇₀, optical density at 570 nm.

TWIST1 represses TSG6 expression via direct promoter binding

To explain the observed correlation between TWIST1 and TSG6 expressions, we analyzed global chromatin immunoprecipitation sequencing (ChIP-seq) data from BM-MSCs engineered to express a hemagglutinin (HA)–tagged full-length TWIST1 cDNA (HATW1; GSE220905), which identified HATW1-enriched regions at the TSG6 gene locus (Fig. 2A). Inspection of these peaks identified four consensus enhancer box (Ebox) motifs within the TSG6 proximal promoter (Fig. 2B), and ChIP analysis confirmed binding of HATW1 to this region (Fig. 2C). Mutagenesis studies demonstrated that expression of a luciferase reporter cloned downstream of the TSG6 proximal promoter was significantly up-regulated when Ebox 2 was mutated but not Eboxes 1, 3, or 4 (Fig. 2D). Moreover, small interfering RNA (siRNA)–mediated silencing of TWIST1 up-regulated TSG6 expression in BM-MSCs to a significantly greater extent as compared to a scrambled siRNA (Fig. 2E). Treatment of BM-MSCs with the small-molecule harmine (39), which inhibited TWIST1 protein expression in a dose-dependent manner (Fig. 2F) without altering TWIST1 mRNA levels, significantly up-regulated TSG6 and IDO1, a known TWIST1 target, compared to vehicle-treated cells (Fig. 2G). Inhibition of TWIST1 by harmine treatment also enhanced stimulus-directed osteogenic differentiation (Fig. 2H), consistent with studies showing that TWIST1 restrains osteogenic potential (40–43). Therefore, TWIST1 directly represses TSG6 expression, thereby providing a mechanistic basis for their orthogonal relationship in all BM-MSCs tested to date.

Fig. 2. TWIST1 directly represses TSG6 expression in BM-MSCs.

(A) Chromosomal location of the TSG6 gene (top) (red line) and coverage of HATW1-enriched regions at the TSG6 locus by ChIP-seq analysis (bottom) (blue peaks). (B) Sequence of the TSG6 proximal promoter where Ebox motifs are indicated by rectangles and their sequences highlighted in red. Arrows are primers used for ChIP-PCR (polymerase chain reaction) analysis in (C). (C) Uncropped immunoblot of ChIP-PCR analysis showing pulldown of HATW1 using an anti-HA antibody at the TSG6 locus in BM-MSCs. (D) Normalized luminescence of luciferase reporter driven by the TSG6 proximal promoter or promoters where the indicated Eboxes were mutated. Data are means ± SD from single experiments with four technical replicates. In Prism, ***P < 0.005 versus TSG6 by one-way ANOVA with Tukey post hoc test. (E) Quantitative reverse transcription PCR (qRT-PCR) of TWIST1 and TSG6 levels in native BM-MSCs and those transfected with a scrambled or TWIST1-specific siRNA. Data are means ± SD from biological replicates with four technical replicates per experiment. In Prism, P values are by one-way ANOVA with Tukey post hoc test. (F) Cropped immunoblot of cell extracts from BM-MSCs pretreated with the indicated concentrations of harmine for 24 hours. (G) qRT-PCR of TWIST1, TSG6, and IDO1 levels in BM-MSCs from (F). Data are means ± SD from four technical replicates, and P values are by one-way ANOVA with Tukey post hoc test. (H) Osteogenesis of BM-MSCs pretreated with vehicle or harmine (40 μM). Data are means ± SD from four technical replicates, and the P value is by Student’s t test. IgG, immunoglobulin G; RLU, relative light unit; GAPDH, glyceraldehyde-3-phosphate dehydrogenase; DAPI, 4′,6-diamidino-2-phenylindole.

TWIST1 and TSG6 predict BM-MSC potency in cell-based and preclinical disease models

To evaluate TWIST1 and TSG6 as potency biomarkers, we showed that tubule formation by human umbilical vein endothelial cells (HUVECs) was more highly induced by conditioned medium (CM) from TWIST1^Hi (n = 3) as compared to TSG6^Hi (which are TWIST1^Low) BM-MSCs (n = 3) based on morphometric analysis (Fig. 3A). CM from a TWIST1^Hi donor (7052) also stimulated HUVEC invasion into a collagen matrix to a greater extent than a TSG6^Hi donor (6015; Fig. 3B). This result is consistent with previous findings showing TWIST1 induces and represses, respectively, proteins with proangiogenic and angiostatic activity (29). Interdonor differences in the ability of BM-MSCs (n = 13) to suppress CD3⁺ T cell proliferation following antibody-mediated activation of peripheral blood mononuclear cells (PBMNCs) (Fig. 3C) also negatively correlated with TWIST1 and FGFR2IIIc and positively correlated with TSG6 levels (Fig. 3D). To extend these findings to in vivo models, we retrospectively quantified TWIST1 levels in populations obtained from multiple donors (n = 7) whose potency in a sterile inflammation model (Fig. 3E) was shown to be positively correlated with TSG6 levels (38). In these BM-MSCs, TWIST1 levels were significantly negatively correlated with TSG6 mRNA levels (Fig. 3F) and anti-inflammatory activity quantified by reduction of neutrophil levels within the cornea at 24 hours after injury, which predicts corneal opacity at 7 days after injury (Fig. 3G). Using an adoptive transfer model of type 1 diabetes mellitus (T1D), we further assayed the ability of BM-MSCs (n = 7) to prevent T1D onset as defined by two consecutive glycemic measurements of ≥250 mg/dl (Fig. 3H). We confirmed that TSG6^Hi (6015, 7012, and 7013) BM-MSCs were more effective in preventing T1D onset in mice compared to those characterized as TWIST1^Hi (7052, 7064, and 7074; Fig. 3I). As expected, TWIST1 expression was negatively correlated with TSG6 in these BM-MSCs (Fig. 3J), and their potency was negatively correlated with TWIST1 levels (Fig. 3K) and positively correlated with TSG6 levels (Fig. 3L).

Fig. 3. TWIST1 and TSG6 function as potency biomarkers.

(A) Photomicrographs (top) of HUVECs cultured in CM from the indicated BM-MSCs (n = 3 per group) and morphometric analysis (lower) of tubule structures (scale bar, 500 μm). The bar graph shows total tubule length. Data are means ± SD, and P values are by Student’s t test. (B) Photomicrographs of collagen invasion assay using HUVECs treated with CM from BM-MSCs 6015 and 7052 (scale bar, 100 μm). The bar graph shows nuclear invasion distance. Data are means ± SD, and P values are by Student’s t test. (C) Ratio of CD3⁺Ki67⁺/CD3⁺ PBMNCs in cocultures with the indicated BM-MSCs (n = 13) at 4 days after stimulation with anti-CD3/CD28 antibodies. (D) Scatter plots of TWIST1, FGFR2IIIc, and TSG6 versus CD3⁺Ki67⁺/CD3⁺ cells quantified by flow cytometry from assays in (C). Data are means ± SD from biological replicates run in duplicate. (E) Schematic of corneal sterile inflammation model wherein myeloperoxidase (MPO) levels measured at 1 day after injury provide a surrogate measure of neutrophil levels and predict corneal opacity at 7 days after injury. (F and G) Scatter plots of TWIST1 versus TSG6 in BM-MSCs (F) (n = 7) and reduction in neutrophil levels (G) based on MPO levels in the mouse sterile inflammation model. (H) Schematic of the adoptive transfer model of T1D. (I) Kaplan-Meier curves of T1D mice administered BM-MSCs [1 × 10⁶ cells, intravenously (IV)] from the indicated donors. The log-rank (Mantel-Cox) test was used to determine P values (n = 8 mice per group). (J to L) Scatter plots of TWIST1 versus TSG6 in BM-MSCs administered to mice in (I) and TWIST1 (K) and TSG6 (L) versus T1D incidence. r is the Pearson’s product moment correlation coefficient with corresponding P values determined by linear regression. N.S., not significant.

Lee et al. (38) previously reported that TSG6 levels negatively correlate with the height and weight of BM-MSC donors. Reanalysis of these donors showed that TWIST1 levels positively correlate with donor height (Fig. 4A) and trended higher in male versus female donors (Fig. 4B). To validate this relationship, we generated BM-MSCs from a tall (72 inches) and short (65 inches) male donor of the same race and similar age (25 to 35 years old; Fig. 4C). Analysis of these donor populations revealed that normalized TWIST1 mRNA levels were significantly higher in BM-MSCs from the tall versus short donor, while TSG6 levels were significantly higher in BM-MSCs from the short versus tall donor (Fig. 4D). In addition, BM-MSCs from the tall donor were found to exhibit significantly higher growth rates (Fig. 4E) and viability (Fig. 4F) when measured after a single passage and higher CFU-F activity when measured at the end of two consecutive passages (Fig. 4G). These data are consistent with previous results demonstrating that TWIST1 reliably predicts interdonor differences in BM-MSC fitness (29). They also link donor stature to BM-MSC potency as defined by the CLIP scale and hence provide a simple means to prescreen donors to produce products of defined potency.

Fig. 4. TWIST1 correlates with BM-MSC donor stature.

(A and B) Scatter plot of TWIST1/GAPDH levels in BM-MSCs versus donor height (A) and bar graph of TWIST1/GAPDH levels versus donor sex (B). Data are means ± SD, and P values are by Student’s t test (n = 7). (C) Schematic showing isolation of BM-MSCs from a tall (72 inches) and short (65 inches) male donor. (D) Bar graphs of normalized TWIST1 and TSG6 mRNA levels in male donors from (C). Data are means ± SD of three technical replicates, and P values are by Student’s t test. (E and F) Bar graphs of total cell yield (E) and cell viability (MTT assay) of BM-MSCs from male donors in (C) after a single passage. (G) Bar graph of CFU-F activity of BM-MSCs from (C) quantified after each indicated passage. Data in (F) and (G) are means ± SD from four technical replicates, and P values are by Student’s t test.

TWIST1^HiTSG6^Low and TWIST1^LowTSG6^Hi BM-MSCs have unique gene signatures

To further interrogate TWIST1 and TSG6 function, we compared the transcriptomes of TWIST1^HiTSG6^Low (group 1; donors 7052, 7064, and 7075) and TWIST1^LowTSG6^Hi (group 2; donors 6015, 7012, and 7013) BM-MSCs by microarray analysis. Unsupervised hierarchical clustering of the top 240 most highly differentially expressed genes (DEGs) segregated populations on their TWIST1/TSG6 status, although one donor from group 1 (7064; G1-21,22) more closely aligned with those from group 2 than its respective group (fig. S1A). Principal components analysis demonstrated that 43.83% and 31.76% of the variation between populations were accounted for by PC1 and PC2, respectively (fig. S1B). This analysis further identified a total of 109 DEGs in TWIST1^HiTSG6^Low versus TWIST1^LowTSG6^Hi BM-MSCs that were altered by ≥log₂ fold change (fig. S1C), of which 36 and 73 were up-regulated in groups 1 and 2, respectively (fig. S1D). On the basis of gene set enrichment analysis (GSEA), these up-regulated DEGs mapped to the cellular component gene ontology (GO) terms “extracellular matrix,” “extracellular region,” and “extracellular space”; molecular function GO terms involving heparin, carbohydrate, glycosaminoglycan, and polysaccharide binding; and the biological process GO terms “skeletal system development,” “positive regulation of T cell activation,” “hematopoietic or lymphoid organ development,” “positive regulation of leukocyte activation,” “oxidation reduction,” and Wnt and TGF-β receptor signaling pathways (fig. S1E). We confirmed that transcripts encoding proteins localized to the extracellular space including TSG6, CD24, MGP, CXCL6, FGF7, TGF-β2, and OXTR were expressed at significantly higher levels in TWIST1^LowTSG6^Hi (group 2) versus TWIST1^HiTSG6^Low (group 1) BM-MSCs (fig. S1F). These transcripts are known to be up-regulated in response to inflammatory stimuli and encode proteins involved in extracellular remodeling at inflammatory sites, suppression of immune responses to danger-associated molecular patterns, and regulatory T cell induction, thereby reflecting the potent immunomodulatory activity of TWIST1^LowTSG6^Hi BM-MSCs (group 2). Last, we analyzed both BM-MSC cohorts by flow cytometry, which revealed that TWIST1^LowTSG6^Hi MSCs (group 2) had significantly larger forward scatter, indicative of larger cell size as compared to TWIST1^HiTSG6^Low BM-MSCs (group 1; fig. S1G). Group 2 BM-MSCs also exhibited larger side scatter, indicative of increased cell complexity, but this difference did not reach statistical significance. Together, these findings demonstrate that TWIST1^Hi versus TSG6^Hi BM-MSCs have distinct gene signatures, which mirror differences in their potency in both cell-based and preclinical models (Fig. 3).

TWIST1 and TSG6 modulate oxidative stress responses in BM-MSCs

TWIST1 is known to exhibit antioxidant properties in cancer cells (44), and GSEA of up-regulated DEGs in TWIST1^LowTSG6^Hi MSCs returned genes associated with oxidation/reduction and response to oxygen levels. To determine whether BM-MSCs from groups 1 and 2 evaluated by microarray exhibit differences in cellular oxidative stress, we subjected representative populations from each group to flow cytometric analysis, which revealed that TWIST1^LowTSG6^Hi (6015, G2-11,12) BM-MSCs retained higher intracellular reactive oxygen species (ROS) levels compared to TWIST1^Hi TSG6^Low (7052, G1-11,12) BM-MSCs (Fig. 5A). Moreover, while siRNA-mediated silencing of TWIST1 induced significant increases in total ROS levels in both populations, levels were highest in TWIST1^LowTSG6^Hi (6015) BM-MSCs (Fig. 5B), which also exhibited measurable increases in cell size and cellular granularity within 24 hours after treatment (fig. S2A). Silencing of TWIST1 also induced expression of TSG6 mRNA in both populations, whereas TGF-β2 expression was significantly up-regulated only in TWIST1^LowTSG6^Hi (6015) BM-MSCs (Fig. 5C). To further assess cellular oxidative stress responses, we cultured these BM-MSCs in varying concentrations of hydrogen peroxide for 24 hours, which revealed clear differences in sensitivity. For example, viability measurements indicated that TWIST1^HiTSG6^Low (7052) BM-MSCs were more resistant to low doses and less resistant to high doses of peroxide as compared to their TWIST1^LowTSG6^Hi (6015) counterpart (Fig. 5D). Moreover, low doses of peroxide up- regulated TWIST1, and high doses up-regulated TSG6 and TGF-β2 expression in TWIST1^LowTSG6^Hi (6015) BM-MSCs, whereas TWIST1 and TGF-β2 expression were largely unaffected by peroxide treatment, and TSG6 expression was only modestly induced by high peroxide doses in TWIST1^HiTSG6^Low (7052) BM-MSCs (Fig. 5E). TGF-β2 is induced in response to oxidative stress (45) and protects cells from free radical–mediated injury (46), which may explain the increased resistance of TWIST1^Low TSG6^Hi BM-MSCs to oxidative stress. To determine whether TSG6 affords similar protection, we silenced its expression in TWIST1^Low TSG6^Hi (6015) BM-MSCs, which resulted in significant increases in cell size based on morphology (fig. S2B) and forward and side scatter based on flow cytometric analysis (Fig. 5F). It also induced significant increases in TGF-β2, but not TWIST1 (Fig. 5F). Other genes that track with TGF-β2 including CXCL6 and OXTR were also induced to significant levels (fig. S2B). Alternatively, overexpression of TSG6 in TWIST1^LowTSG6^Hi reverted cells to a TWIST1^HiTSG6^Low-like morphology (fig. S2C), induced measurable decreases in cell size and cellular granularity, and down- regulated expression of TGF-β2 but did not significantly alter expression of TWIST1 (Fig. 5G). OXTR levels were also significantly down-regulated as expected (fig. S2C). To confirm these observations, we analyzed two additional BM-MSC donors from the microarray studies, which confirmed that TWIST1^LowTSG6^Hi (7012) BM-MSCs retained higher cellular ROS levels compared to their TWIST1^HiTSG6^Low (7074) counterpart (fig. S2D). Silencing of TWIST1 in these donors also yielded similar impacts on TSG6 and TGF-β2 expression levels (fig. S2E) and induced measurable increases in cell size and granularity (fig. S2F). Last, silencing and overexpression of TSG6 in TWIST1^LowTSG6^Hi (7012) BM-MSCs increased and reduced cell size, respectively, on the basis of morphological and flow cytometric analysis as expected (fig. S2, G and H). These results are consistent with our previous observations that modest overexpression of TSG6 in TWIST1^HITSG6^low enhanced their anti-inflammatory activity in a murine model of corneal injury (38), but superphysiological levels failed to improve therapeutic potency in this model (fig. S3). They also suggest that the ability to transiently up-regulate TSG6 and TGF-β2 affords TWIST1^LowTSG6^Hi BM-MSCs greater protection against oxidative stress.

Fig. 5. Impacts of TWIST1, TSG6, and TGF-β2 on cellular ROS levels and oxidative stress responses in BM-MSCs.

(A) Histogram of cellular ROS levels in TWIST1^HiTSG6^Low (7052) and TWIST1^LowTSG6^Hi (6015) BM-MSCs quantified by flow cytometric analysis. (B) Dot plots from flow cytometric analysis of ROS levels in native and TWIST1-specific siRNA-transfected BM-MSCs from (A). (C) qRT-PCR quantification of normalized TWIST1, TSG6, and TGF-β2 levels in BM-MSCs from (B). Data are means ± SD from triplicates. *P < 0.05, **P < 0.01, and ***P < 0.005 versus control by one-way ANOVA with Dunnett’s multiple comparisons test. (D) Cell viability of TWIST1^HiTSG6^Low (7052) and TWIST1^LowTSG6^Hi (6015) BM-MSCs at 24 hours after exposure to the indicated doses of hydrogen peroxide. Data are means ± SD of triplicates. **P < 0.01 by one-way ANOVA with Dunnett’s multiple comparisons test. (E) qRT-PCR quantification of normalized TWIST1, TSG6, and TGF-β2 levels in BM-MSCs from (D). Data are means ± SD from triplicates. *P < 0.05, **P < 0.01, and ***P < 0.005 versus untreated cells by one-way ANOVA with Dunnett’s multiple comparisons test. (F) Representative dot plots from flow cytometric analysis (top) and qRT-PCR quantification of TSG6, TGF-β2, and TWIST1 levels (bottom) in TWIST1^LowTSG6^Hi BM-MSCs (6015) at 24 hours after transfection with a scrambled or TSG6-specific siRNA (10 mM). Data are means ± SD from triplicates. **P < 0.01 and ****P < 0.001 by Student’s t test. (G) Representative dot plots from flow cytometric analysis (top) and qRT-PCR quantification of TSG6, TGF-β2, and TWIST1 (bottom) levels in TWIST1^LowTSG6^Hi in BM-MSCs (6015) at 24 hours after transfection with a control or TSG6 plasmid (0.1 to 0.5 mg/cm²). Data are means ± SD from triplicates. **P < 0.01 and ****P < 0.001 versus control by one-way ANOVA with Dunnett’s multiple comparisons test.

TGF-β2 alters BM-MSC anti-inflammatory activity in vivo

TGF-β is a profibrotic cytokine (47) that modulates expression of extracellular matrix proteins (48). Our transcript profiling data indicate that TGF-β2 expression tracks with TSG6 (fig. S1F) and both transcripts are induced in BM-MSCs by cellular oxidative stress (Fig. 5E). TGF-β2 has been reported to both enhance (49, 50) or impair (51) the anti-inflammatory activity of murine BM-MSCs. We showed that TWIST1^LowTSG6^Hi (6015) BM-MSCs secrete significantly higher levels of TGF-β2 compared to TWIST1^HiTSG6^Low (7052) cells, which we confirmed by extending this analysis to several additional BM-MSC populations used in the microarray studies (Fig. 6A). We further demonstrated that siRNA-mediated silencing of TGF-β2 in TWIST1^LowTSG6^Hi (6015) BM-MSCs significantly decreased TSG6 mRNA levels (Fig. 6B) and augmented osteogenic differentiation (fig. S4A). Treatment of TWIST1^Low TSG6^Hi (6015) BM-MSCs with the TGFBR1 inhibitor SB431542 (52) decreased TGF-β2, TSG6, and OXTR mRNA levels while increasing TWIST1 levels (Fig. 6C). Alternatively, treatment of TWIST1^HiTSG6^Low BM-MSCs (7052) with 50 or 200 pg/ml of recombinant human TGF-β2 resulted in significant increases in forward and side scatter based on flow cytometric analysis (Fig. 6D) and up-regulated expression of TSG6, OXTR, and CXCL6 mRNAs in a dose-dependent manner (Fig. 6E). However, TWIST1 expression was significantly induced by 50 pg/ml but not 200 pg/ml of TGF-β2 in this donor (Fig. 6E). Treatment of a second TWIST1^HiTSG6^Low (7074) population with TGF-β2 resulted in similar impacts on cell size (Fig. 6F) and TSG6 and OXTR expression (Fig. 6G), but in this donor, TWIST1 levels were decreased (Fig. 6G). Last, we found that diabetes incidence was significantly reduced in mice administered TWIST1^HiTSG6^Low BM-MSCs (7052) pretreated with TGF-β2 (200 pg/ml) as compared to mice administered naïve, nontreated cells (Fig. 6H). TGF-β2 also decreased the osteogenic potential of TWIST1^HiTSG6^Low BM-MSCs (7052; fig. S4B). Together, these data indicate that TGF-β2 induces TSG6 and alters the size, granularity, osteogenic potential, and anti-inflammatory activity of BM-MSCs in a manner antagonistic to TWIST1.

Fig. 6. TGF-β2 modulates BM-MSC cell size, osteogenic capacity, and anti-inflammatory activity.

(A) Enzyme-linked immunosorbent assay (ELISA) of TGF-β2 in CM from TWIST1^LowTSG6^Hi (left, 6015; right, 7012) and TWIST1^Hi TSG6^Low (left, 7052; right, 7074 and 7068) BM-MSCs. Data are means ± SD from triplicates, and **P < 0.01 is by Student’s t test. (B) qRT-PCR quantification of TGF-β2 and TSG6 levels in TWIST1^LowTSG6^Hi (6015) BM-MSCs at 48 hours after transfection with a control or TGF-β2-specific siRNA. Data are means ± SD from triplicates, and *P < 0.05 is by Student t test. (C) qRT-PCR quantification of TGF-β2, TSG6, OXTR, and TWIST1 levels in TWIST1^LowTSG6^Hi BM-MSCs (6015) cultured with or without the indicated concentrations of SB431542 for 4 days. Data are means ± SD from triplicates. *P < 0.05, **P < 0.01, ***P < 0.005, and ****P < 0.001 by one-way ANOVA versus Cont with Dunnett’s multiple comparisons test. (D) Bar graphs showing average mean forward scatter (FSC) and side scatter (SSC) of TWIST1^HiTSG6^Low BM-MSCs (7052) cultured in the absence or presence of recombinant human TGF-β2 (50 or 200 pg/ml) for 4 days. Data are means ± SD from triplicates. *P < 0.05, **P < 0.01, and ***P < 0.005 versus untreated by one-way ANOVA with Tukey post hoc test. (E) qRT-PCR quantification of TSG6, OXTR, CXCL6, and TWIST1 in BM-MSCs from (D). Data are means ± SD from triplicates. *P < 0.05 and ***P < 0.005 versus Cont by one-way ANOVA with Dunnett’s multiple comparisons test. (F and G) Flow cytometry dot plots (F) and qRT-PCR quantification of TSG6, OXTR, and TWIST1 levels (G) in TWIST1^HiTSG6^Low BM-MSCs (7074) treated as in (D). Data are means ± SD from triplicates. *P < 0.05, **P < 0.01, and ***P < 0.005 versus untreated by one-way ANOVA with Tukey post hoc test. (H) Diabetic incidence in nonobese diabetic (NOD)/scid mice after transfer of diabetogenic splenocytes and/or TWIST1^HiTSG6^Low BM-MSCs (7052) (1 × 10⁶ cells per mouse) pretreated for 4 days with or without TGF-β2 (200 pg/ml). The log-rank (Mantel-Cox) test was used to determine P values (n = 8 mice per group).

TWIST1 and TSG6 expression predicts functional attributes of induced pluripotent stem cell–derived MSCs

Several studies have shown that MSCs can be generated from induced pluripotent stem cells (iPSCs) by treatment with the TGF-β inhibitor SB431542 (53–55) and that these iPSC-derived MSCs (iPSC-MSCs) exhibit greater osteogenic capacity compared to BM-MSCs (56). Consistent with these findings, we found that these iPSC-MSCs expressed ~8-fold higher TWIST1 levels compared to TWIST1^HiTSG6^Low (7052) BM-MSCs and ~10-fold lower TSG6 levels compared to TWIST1^LowTSG6^Hi (6015) BM-MSCs (Fig. 7A). Flow cytometric analysis (Fig. 7B) further demonstrated that iPSC-MSCs exhibit lower forward and side scatter than TWIST1^HiTSG6^Low (7052) BM-MSCs (Fig. 7C), which is consistent with their higher growth rate. Previously, we demonstrated that TSG6^Hi but not TSG6^Low BM-MSCs effectively suppressed the proinflammatory mediators TNF-α, CXCL1, and CXCL2 in peritoneal lavage of mice challenged with Zymosan A (38). iPSC-MSCs (1.5 × 10⁶ cells) administered to mice challenged with Zymosan A were also ineffective in reducing expression of TNF-α, CXCL1, and CXCL2 in peritoneal lavage, whereas dexamethasone administration, which served as a positive control, significantly attenuated Zymosan A–induced inflammation in mice (Fig. 7D). Recently, we observed that microcarrier culture conditions improved the anti-inflammatory activity of iPSC-MSCs (personal communication). Consistent with this observation, we found that these culture conditions increase TSG6 and decrease TWIST1 expression levels in cells (Fig. 7E). In addition, when iPSC-MSCs were expanded in a hollow-fiber bioreactor system, they expressed higher levels of TSG6 than TWIST1^HiTSG6^Low BM-MSCs (Fig. 7F), and similar to BM-MSCs, these iPSC-MSCs significantly suppressed sterile inflammation responses and reduced corneal opacity in a TSG6-dependent manner (57). Last, we found that treatment of iPSC-MSCs with recombinant human TGF-β2 significantly decreased TWIST1 levels and increased TSG6 levels in a dose-dependent manner (Fig. 7G) while decreasing osteogenic potential (Fig. 7, H and I). Collectively, these data indicate that TWIST1 and TSG6 expression levels also predict differences in the potency of iPSC-MSCs, further expanding and validating their utility as potency biomarkers.

Fig. 7. TWIST1 and TSG6 expression predicts functional attributes of iPSC-MSCs.

(A) qRT-PCR quantification of TWIST1 and TSG6 levels in BM-MSCs (7052 or 6015) and iPSC-MSCs at the indicated passage. Data are means ± SD from triplicates. (B and C) Representative dot plots (B) from flow cytometry analysis of BM-MSCs (7052) and iPSC-MSCs and bar graphs (C) of mean FSC and SSC values. Data are means ± SE from flow cytometry analysis, and P values are by Student’s t test. (D) ELISA of TNF-α, CXCL1, and CXCL2 levels in peritoneal lavage from mice (n = 4 to 6 per group) at 4 hours after challenge with Zymosan A [1 mg/mouse, intraperitoneally (IP)]. Mice were administered phosphate-buffered saline (PBS; positive control; PC), dexamethasone (Dex; 1.5 mg/kg, IP), or iPSC-MSCs (1.5 × 10⁶ per mouse, IP) at 15 min after challenge. NC, negative control. Data are means ± SD, and *P < 0.05 is by one-way ANOVA with Tukey post hoc test. (E) qRT-PCR quantification of TWIST1 and TSG6 levels in iPSC-MSCs grown in monolayer (ML) cultures or on microcarriers (MC) for 8 days. Data are means ± SD from triplicates. (F) qRT-PCR quantification of TSG6 levels in TWIST1^HiTSG6^Low BM-MSCs (7075) grown in ML cultures and iPSC-MSCs grown in a hollow-fiber bioreactor system (Quantum system, Terumo). (G) qRT-PCR quantification of TWIST1 and TSG6 levels in iPSC-MSCs at 6 days after treatment with the indicated concentrations of TGF-β2. *P < 0.05, **P < 0.01, ***P < 0.005, and ****P < 0.001 versus untreated cells by one-way ANOVA with Dunnett’s multiple comparisons test. (H) Alizarin Red S staining of iPSC-MSCs pretreated for 6 days with or without TGF-β2 then cultured for 11 days in osteogenic induction media. (I) Quantification of staining data from (H). **P < 0.01, ***P < 0.005, and ****P < 0.001 versus untreated cells by one-way ANOVA with Dunnett’s multiple comparisons test.

DISCUSSION

Implicit in the design of MSC-based clinical therapies is the need to generate large numbers of cells to achieve recommended patient doses. Consequently, most clinical manufacturing protocols select for MSC populations that exhibit robust growth or augment growth by culturing cells in proprietary medium supplements. As a result, it is commonplace to use similar manufacturing regimens to produce MSC products for treating medically unrelated disease indications. These well-established practices have deemphasized the need to stratify MSC populations on the basis of donor characteristics, and while efforts to develop assays that predict product efficacy remains a key focus for the field, these are not linked to donor selection criteria. In this study, we identify key attributes of TWIST1 and TSG6 that validate their utility as biomarkers to predict interdonor differences in MSC potency across platforms and disease indications. First, we show that TWIST1 directly represses TSG6 expression and that expression of these mRNAs are inversely correlated across multiple BM-MSC human donor cohorts sourced from different providers and expanded in different laboratories (Fig. 8). Therefore, this correlation is highly robust. Second, we show that population-averaged, normalized levels of TWIST1 and TSG6 predict interdonor differences in BM-MSC potency in both cell-based assays and preclinical disease models (Fig. 8). Because TWIST1 levels also correlate with CFU-F activity, this relationship provides an internal “standard” to stratify populations based on potency using our previously described CLIP scale (29). Third, outcomes from both cell-based assays and animal models reveal clear interdonor differences in BM-MSC potency and hence challenge the existing paradigm that potency is determined by interaction with the host microenvironment in vivo. While our studies do not rule out a role for the host microenvironment, they clearly demonstrate that donor-dependent differences in potency predict therapeutic efficacy in preclinical disease models independent of the host microenvironment. Fourth, we show that TWIST1 and TSG6 correlate with donor stature and that BM-MSCs from tall- versus short-statured donors exhibit intrinsic differences in behavior (Fig. 8). These results are consistent with observed interdonor differences in potency and suggest that genetic or epigenetic factors may contribute to such differences. Several genetic alterations including single-nucleotide polymorphisms have been identified at the TWIST1 locus that are associated with susceptibility to breast cancer (58), prognosis in gastric cancer (59), and osteoporosis (60). Fifth, while our studies focused on BM-MSCs, we show that these relationships extend to iPSC-MSCs, which were found to express significantly higher and lower TWIST1 and TSG6 levels, respectively, as compared to BM-MSCs, and exhibit weak anti-inflammatory activity in vivo when expanded in monolayers as predicted by the CLIP scale. Moreover, culture-dependent (microcarrier or bioreactor) alterations in TWIST1 and TSG6 levels correlated with changes in iPSC-MSC anti-inflammatory activity in vivo. In this regard, we have observed that bioreactor-induced sheer stress serves as a potent inducer of TSG6 in BM-MSCs similar to oxidative stress and that this induction occurs, in part, because of down-regulation of TWIST1.

Fig. 8. TWIST1 and TSG6 function as potency biomarkers and modulators of cellular oxidative stress in BM-MSCs.

TWIST1 and TSG6 expression levels are inversely correlated in BM-MSC populations and predict interdonor differences in proangiogenic versus anti-inflammatory/immune suppressive activities in cell-based assays and preclinical disease models. TWIST1 also positively correlates with the height and sex of BM-MSC donors, while TSG6 negatively correlates with these attributes. Profiling studies further indicate that TWIST1^HiTSG6^Low and TWIST1^LowTSG6^Hi BM-MSCs exhibit varying sensitivities to oxidative stress and that the latter population up-regulates TGF-β2, TSG6, extracellular matrix (ECM) proteins, and other inflammatory modulators in response to high levels of such stress, which confers resistance and enhanced anti-inflammatory activity in vivo. Created with BioRender.com.

Our data also define a unique TWIST1/TGF-β2/TSG6 signaling axis in BM-MSCs. Specifically, we found that TWIST1^HiTSG6^Low and TWIST1^LowTSG6^Hi populations exhibit unique gene signatures, express significantly different intracellular ROS levels, and have different sensitivities to oxidative stress induced by peroxide exposure (Fig. 8). Therefore, TWIST1 and TSG6 predict not only interdonor differences in potency but also differences in physical (cell size/granularity) and metabolic (ROS generation) parameters. In the latter case, increased resistance of TWIST1^LowTSG6^Hi BM-MSCs to oxidative stress was linked to up-regulation of TGF-β2 and TSG6 expression in response to such stress, which protect cells from free radical–induced injury and oxidative stress–induced apoptosis (Fig. 8). Stress-induced up-regulation of other extracellular matrix and inflammatory modulators are consistent with the stromal-like phenotype of these cells and their enhanced anti-inflammatory activity in vivo. Alternatively, TWIST1^Hi TSG6^Low BM-MSCs maintain low ROS levels, which itself may act as a second messenger to promote stem/progenitor self-maintenance including rapid growth, high CFU-F activity, and multipotency. Continued proliferation of these cells under conditions of oxidative stress likely enhances oxidative damage, which manifests as increased sensitivity to high peroxide doses. These results are consistent with previous studies showing that BM-MSCs cultured in atmospheric oxygen exhibit stress-induced impairments in mitochondrial function (61). Last, we previously reported that TWIST1 confers osteogenic potential while restraining osteogenic differentiation in BM-MSCs and that its silencing enhances the latter (29). Here, we show that TGF-β2 suppresses osteogenic differentiation of BM-MSCs, particularly at high doses, while simultaneously repressing TWIST1 expression in BM-MSCs (Fig. 6) and iPSC-MSCs (Fig. 7). Therefore, TGF-β2 at low doses may stimulate osteogenic differentiation via TWIST1 down-regulation and, at higher doses, inhibit differentiation independent of TWIST1.

In summary, our findings have important implications regarding the clinical manufacture of BM-MSCs. Our data indicate that populations selected on the basis of rapid growth will be characterized as TWIST1^HiTSG6^Low and exhibit a proangiogenic phenotype. On the basis of our CLIP scale (17, 29), these cells are appropriate for treating ischemic-related diseases and suboptimal for acute/chronic inflammatory and immune-mediated disorders. Generally, this prediction is consistent with studies demonstrating a clinical benefit of MSC-based therapies in patients with heart disease (62–65) and a lack of benefit in those with acute inflammatory disorders, such as acute respiratory distress syndrome (66–68). Le Blanc et al. (69) demonstrated that BM-MSCs minimally expanded by plating at high density (4000 cells/cm²) afforded a clinical benefit in patients with steroid refractory graft-versus-host disease (GvHD), and subsequent studies have shown that MSCs are most effective in treating pediatric GvHD patients (70, 71). However, patients with acute or chronic GvHD show lower response rates to treatment with MSCs expanded in human platelet lysate as compared to fetal bovine serum (72, 73). These data are consistent with our observation that BM-MSCs expanded in hPL express high TWIST1 levels (Fig. 1F). Similarly, McKinnirey et al. (74) reported that female versus male MSCs were more potent in suppressing PBMNC proliferation in vitro because of their higher expression of Indoleamine 2,3-dioxygenase 1 (IDO1), Interleukin 1 receptor antagonist (IL1RN), and prostaglandin E₂. These findings are consistent with our studies showing that MSCs derived from short-statured or female donors express higher basal and induced levels of TSG6, exhibit more potent anti-inflammatory activity in a sterile inflammation model (38), and express low levels of TWIST1. Numerous studies have also reported that preconditioning MSCs with cytokines, such as TNF and INFG, enhances their anti-inflammatory and immune suppressive activities (75, 76), which is also consistent with our studies showing that TWIST1 is a target of INFG action in BM-MSCs (29) and that pretreatment with TGF-β2 licenses the immune suppressive activity of TWIST1^Hi BM-MSCs in vivo (Fig. 6G). In summary, results reported herein validate TWIST1 and TSG6 as biomarkers for quantifying interdonor differences in BM-MSC potency across various assay platforms and disease indications. They also underscore the value of using the CLIP scale to monitor impacts of donor selection and large-scale cell manufacturing on product potency.

MATERIALS AND METHODS

Cell isolation and culture

Primary BM-MSCs were isolated from iliac crest marrow aspirates (Lonza, USA) as previously described (11). Briefly, nucleated cells were isolated using a Ficoll-Paque density gradient (Cytiva), suspended in 25 ml of complete culture medium [CCM; α–minimum essential medium supplemented with 2 mM l-glutamine, 17% fetal bovine serum (HyClone), penicillin (100 U/ml), and streptomycin (100 μg/ml)] and expanded in a 15-cm culture dish (Nunclon Delta) at 37°C in 5% CO₂ in a humidified chamber. Media and supplements were purchased from Gibco unless indicated otherwise. Nonadherent cells were discarded by decanting after 24 hours, and the adherent monolayer was washed 2× with phosphate-buffered saline (PBS) and recharged with fresh CCM. Medium changes were performed every 3 to 4 days thereafter, and BM-MSCs were harvested at ~70% confluency with 0.25% trypsin-EDTA, washed in CCM, and plated in 60-mm dishes (500 to 1000 cells/cm²). The iPSC-derived MSCs were generated as described previously and expanded in a vertical wheel bioreactor (PBS Biotech) while attached to polystyrene microcarriers (55, 77). BM-MSCs transduced with a control retroviral vector MSCV-PIG (Addgene) or one engineered to express a full-length human TWIST1 ORF tagged with HA at its N terminus (HATW1) were used to conduct ChIP analysis (29). CFU-Fs were quantified as described previously (29). Alternatively, frozen vials of passage 1 BM-MSCs were obtained from the Institute for Regenerative Medicine, Texas A&M University School of Medicine, and the cells were cultured as previously described (31). Where indicated, BM-MSCs were plated at 500 cells/cm² and cultured with the CCM containing various concentrations of recombinant human TGF-β2 (R&D Systems) or SB431542 (Sigma-Aldrich) for 4 days with medium changes on day 2. Where indicated, scrambled or TWIST1-, TSG6-, and TGF-β2-specific siRNAs (10 nM; Santa Cruz Biotechnology Inc.) or TSG6 or control plasmids (0.1 to 0.5 μg per well in a six-well plate) (38) were transfected into BM-MSC using Lipofectamine RNAiMAX or 2000 (Life Technologies), respectively, according to the manufacturer’s protocol, and after 24 to 48 hours, cells were harvested and used for analysis.

TSG6 promoter cloning and mutagenesis

The promoter region of TSG6 was amplified via polymerase chain reaction (PCR) using Q5 High-Fidelity DNA polymerase (New England Biolabs) from genomic DNA extracted from BM-MSCs using the Zymo Quick-DNA kit (Zymo Research). The pGL4.10[luc2] vector (Promega) was subjected to digestion with Eco RV and Hind III. The amplified PCR product was digested with Hind III and ligated into the pGL4.10[luc2] vector to produce TSG6-pGL4.10[luc2]. The TSG6 promoter site has four canonical Eboxes, which were individually mutated via inverse site-directed mutagenesis. This was achieved by creating forward primers that contained the mutated Ebox site, along with an end-to-end reverse primer to PCR-amplify a mutated vector directly from TSG6-pGL4.10[luc2]. Dpn I was used to cleave the parental, unmutated TSG6-pGL4.10[luc2], and following self-ligation of the PCR product, DH10b bacteria (Life Technologies) were transformed with the resulting DNA to generate Ebox-mutated TSG6-pGL4.10[luc2] clones. These clones were used for subsequent luciferase assays. Primers used were as follows: TSG6 promoter forward, 5′ACAAAGGAAGAGGAAAGTTTGG3′ and reverse, 5′AAGCTTATCGTCAGTTGTAGTGAAGTAA3′; TSG6 Ebox1 mutant forward, 5′GTTGCCAATGAAAATTAGTCAGAGGATTTG3′ and reverse, 5′CAAAACTAAGGAGATAGAATGATC3′; TSG6 Ebox2 mutant forward, 5′GCTGACCTCAAAAATTAAAAAAACTATTTAAAAGATGTGG3′ and reverse, 5′AAGTACTCTCCAATGGCATTTTC3′; TSG6 Ebox3 mutant forward, 5′ATGTGGAAACAAAATTTTTCAGTCACATTTC3′ and reverse, 3′CTTTTAAATAGTTTTTTTCAACTGTG3′; and TSG6 Ebox4 mutant forward, 5′TACTTCACTAAAAATTACGATATCAAGCTTGGCAATC3′ and reverse, 5′ACAGCCTGTTAGGGGCTG3′.

TSG6 promoter luciferase assays

To evaluate the inhibitory effect of TWIST1 on TSG6, human embryonic kidney–293 cells were initially transfected with either TSG6-pGL4.10[luc2] or one of four Ebox-mutated TSG6-pGL4.10[luc2] vectors. Subsequently, the cells were divided into two groups and transfected with either pMSCV-Pig-HA-TWIST1 or a pMSCV-Pig empty vector control. After a 48-hour incubation, luciferase expression was determined using the Dual-Glo Luciferase Reporter System (Promega). The cells were lysed using the Dual-Glo assay reagent, and after a 5-min incubation at room temperature, the luminescence was measured via plate reader (BioTek). The level of luminescence was quantified as a reduction in luminescence in the TWIST1 transfected sample compared to the empty vector control.

Chromatin immunoprecipitation

ChIP was performed according to the manufacturer’s protocol using the ChIP-IT Express kit (Active Motif). Briefly, ~5 × 10⁶ HATW1 or MSCV-Pig cells were fixed in a monolayer with 1% paraformaldehyde, quenched with glycine, then gently lifted by scraping, and homogenized with a Dounce homogenizer in manufacturer-provided lysis solution on ice to release the fixed chromatin. Fixed chromatin samples were concentrated in two volumes of ~300-μl shearing buffer, sonicated with a 1/8-inch probe (Qsonica) on ice for five pulses of 20 s at 25% power at 30-s intervals. Shearing was empirically optimized to produce ~500bae pair fragments. Immunoprecipitation reactions (200 μl) were performed overnight on chromatin samples (15 μg) using the manufacturer-provided protein A beads and anti-HA tag antibody HA-7X (sc-7932X, Santa Cruz Biotechnology). Using kit-provided buffers, Protein A beads were washed on a magnetic stand, and the captured chromatin was subsequently eluted. Samples were cleaned up with Proteinase K followed by phenol chloroform extraction. Protein lysates were prepared from HATW1 or MSCV-Pig cell monolayers using NP-40 lysis buffer, and protein concentrations were determined using the Pierce BCA Protein Assay Kit (Thermo Fisher Scientific). Protein samples (20 μg) were prepared in SDS sample buffer containing 5% β-mercaptoethanol, denatured at 95°C for 10 min and electrophoresed on 10% SDS–polyacrylamide gel electrophoresis gels using SDS Running Buffer (Bio-Rad). Transfer papers and 0.45-μm polyvinylidene difluoride (PVDF) membranes were washed with tris-buffered saline (TBS) containing 10% methanol, and the same buffer was used for transfers. Transfer was conducted for 30 min using a semidry rig. Membranes were washed with TBS, and transfer efficiency was assessed with Ponceau Red, which was washed off with TBST. The membrane was blocked for 2 hours at room temperature using nonfat dry milk reconstituted in TBST. Proteins were probed with an anti–HA-peroxidase high affinity antibody (3F10) at a 1:500 dilution for 1 hour at room temperature. Labeled bands were visualized using the SuperSignal West Pico PLUS Chemiluminescent Substrate (Thermo Fisher Scientific, 34577) to expose film.

Quantitative reverse transcription PCR

Total RNA was isolated using the Zymo Quick-RNA Miniprep Kit (R1054, Zymo), converted to cDNA using the qScript cDNA synthesis kit (Quantabio), and amplified by PCR using the POWER SYBR Green PCR Master Mix (Thermo Fisher Scientific) and TWIST1, FGFR2IIIc, IDO1, or GAPDH primer sets according to the manufacturer’s instructions. In addition, the following TaqMan probe/primer sets (Thermo Fisher Scientific) were used for PCR amplification using the TaqMan Fast Master Mix (Thermo Fisher Scientific): human TSG6, CD24, MGP, CXCL6, FGF7, TGF-β2, OXTR, CRISPLD2, PTX3, IL-18, and GAPDH. Reactions were performed on a 7900 HT sequence detector (Applied Biosystems), and transcript levels were quantified using the relative Ct method by using GAPDH as an internal control.

Western blot

MSCs were washed 2× with ice-cold PBS and collected in radioimmunoprecipitation assay buffer [50 mM tris-Cl (pH 7.4), 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, and 0.1% SDS] by gently scraping. Lysates were cleared by centrifugation at 12,000g for 20 min, and their protein content was measured using the Bradford method (Bio-Rad). Protein samples were suspended in 4× NuPAGE LDS Sample Buffer, heated for 3 min at 95°C, and separated by electrophoresis on NuPAGE 4 to 12% bis-tris polyacrylamide gels (Invitrogen). Proteins were transferred to PVDF membranes, incubated with TBST [10 mM tris (pH 8.0), 150 mM NaCl, and 0.1% Tween 20] and 5% nonfat dry milk (Bio-Rad) for 1 hour at room temperature, rinsed in TBST, and incubated overnight at 4°C in TBST containing 5% bovine serum albumin with a TWIST1 antibody (E7E2G, Cell Signaling Technology) and glyceraldehyde-3-phosphate dehydrogenase (GAPDH; sc-32233, Santa Cruz Biotechnology).

Adoptive transfer T1D mouse model

Female nonobese diabetic (NOD)/LtJ (12 weeks old) and NOD/scid mice (7 weeks old) were used for an adoptive transfer model of T1D. All mice were purchased from the Jackson Laboratory (Bar Harbor, ME) and cared for at the Texas A&M University Comparative Medicine Program under a protocol approved by the Institutional Animal Care and Use Committee. To induce an adoptive transfer in the T1D model, 10⁷ splenocytes from prediabetic 12-week-old female NOD mice were intravenously injected into 7-week-old female NOD/scid mice. BM-MSCs (1 × 10⁶) or vehicle control were injected intravenously twice at 15 min after splenocyte transfer and on day 4 thereafter. Blood glucose levels were measured twice a week by tail bleeding according to National Institutes of Health guidelines, and diabetes in mice was defined as having the two consecutive glycemic values above 250 mg/dl.

Microarray

RNA was isolated from BM-MSCs with RNeasy Mini Kit (QIAGEN) and assayed on human microarray (HOA 7.1; OneArray, San Diego, CA).

Endothelial tubule formation and invasion assays

HUVECs (30,000 cells/well) were seeded on top of growth factor reduced Matrigel (BD Biosciences) in a 96-well plate with Medium 200PRF (Invitrogen) supplemented with CM from BM-MSCs and incubated for 6 hours. Tube formation was analyzed and quantified using the WIMASIS software (Onimagin Technologies SCA, Spain). For invasion assays, HUVECs (30,000 cells) were seeded on type 1 collagen matrices (2.5 mg/ml; Pel-Freez Biologicals, AK), which was supplemented with BM-MSC CM (×1) and 1 μM sphingosine-1-phosphate (Avanti Polar Lipids, AL). Cells were allowed to invade into type 1 collagen matrices in the presence of HUVEC media containing M199 medium (Gibco, MA) supplemented with 1× RSII, ascorbic acid, and basic fibroblast growth factor (40 ng/ml) and vascular endothelial growth factor (40 ng/ml; R&D Systems, MN). Following overnight invasion, the cells were fixed in 3% glutaraldehyde in PBS overnight, and gels were stained with 10 μM 4′,6-diamidino-2-phenylindole, and confocal images were captured with a Nikon Eclipse TI. Nuclear invasion distance was quantified using NIS Elements AR software of images rendered in the volume view with Alpha depth coding selected. Distance measurements were taken from the monolayer to the bottom of each cell nucleus.

ROS assay

BM-MSCs were plated at 200 cells/cm² and cultured for 5 days with CCM, and the total ROS levels in cells was measured by flow cytometry with a commercial kit (Abcam, Waltham, MA).

Osteogenic differentiation

BM-MSCs were cultured in CCM supplemented with 10 nM dexamethasone, 10 mM β-glycerophosphate (Sigma-Aldrich), and 50 μM ascorbate-2-phosphate (Sigma-Aldrich) for 11 to 14 days with medium changes every 2 to 3 days. Cell monolayers were then fixed with 10% formalin for 30 min and stained with 2% Alizarin Red S. For calcium quantification, an Alizarin Red–based assay was used as described previously (38).

Mouse model of peritonitis and measurements of inflammation

To induce inflammation in 8-week-old male C57BL/6 mice (the Jackson Laboratory), 1 ml of Zymosan A solution (1 mg/ml; Sigma-Aldrich) was administered by intraperitoneal injection. After 15 min, mice were administered PBS, dexamethasone (1.5 mg/kg; Sigma-Aldrich), or iPSC-MSCs (1.5 × 10⁶ cells per mouse) by intraperitoneal injection. Inflammatory exudates were collected by peritoneal lavage at 4 hours after treatment, and the cell-free supernatant was used to measure levels of the proinflammatory mediators by ELISA (mTNF-α, mCXCL1, and mCXCL2; R&D Systems).

Statistical analysis

All data are reported as means ± SD. The statistical significance between two independent experimental groups was assessed using a two-tailed, unpaired Student’s t test. The statistical significance among more than two groups was assessed using a one-way analysis of variance (ANOVA) with multiple comparisons evaluated by the Tukey’s or Dunnett’s test. The log-rank (Mantel-Cox) test was used to determine differences in the survival experience of mice between experimental groups. Linear correlations are reported as the Pearson’s r and corresponding P values determined by linear regression. Significance level was set at P ≤ 0.05.

Supplementary Materials

This PDF file includes:

Figs. S1 to S4