Morphological features of IFN-γ–stimulated mesenchymal stromal cells predict overall immunosuppressive capacity

Significance

Substantial evidence exists demonstrating the immunosuppressive function of mesenchymal stromal cells (MSCs), but inconsistent clinical results suggest that better understanding of MSC-mediated immunosuppression and identification of features predictive of immunosuppressive capacity would advance MSC-based therapeutics. In this work, we present a robust analytical approach to quantify the immunosuppressive capacity of MSCs by integrating high-dimensional flow cytometry data from multiple experimental conditions into a single measure of immunosuppressive capacity. Additionally, we identified morphological features of MSCs that predicted immunosuppressive capacity, as well as the magnitude of IFN-γ–mediated immunosuppression enhancement. These improved methods of MSC characterization could be used to identify MSC preparations with desired immunosuppressive capacity, as well as screen for pretreatments that enhance their immunosuppressive function.

Abstract

Human mesenchymal stromal cell (MSC) lines can vary significantly in their functional characteristics, and the effectiveness of MSC-based therapeutics may be realized by finding predictive features associated with MSC function. To identify features associated with immunosuppressive capacity in MSCs, we developed a robust in vitro assay that uses principal-component analysis to integrate multidimensional flow cytometry data into a single measurement of MSC-mediated inhibition of T-cell activation. We used this assay to correlate single-cell morphological data with overall immunosuppressive capacity in a cohort of MSC lines derived from different donors and manufacturing conditions. MSC morphology after IFN-γ stimulation significantly correlated with immunosuppressive capacity and accurately predicted the immunosuppressive capacity of MSC lines in a validation cohort. IFN-γ enhanced the immunosuppressive capacity of all MSC lines, and morphology predicted the magnitude of IFN-γ–enhanced immunosuppressive activity. Together, these data identify MSC morphology as a predictive feature of MSC immunosuppressive function.

Human mesenchymal stromal cells (MSCs) can potently suppress immune responses in vitro and in animal models of human disease (1, 2), but to date MSC-based therapies have produced mixed results in clinical trials for treatment of inflammatory and autoimmune diseases (3, 4). A major challenge in the development of consistently effective MSC-based immunosuppressive therapies is that MSC lines derived from different donors and manufacturing processes (i.e., cell expansion) can possess markedly dissimilar immunosuppressive function (3, 5, 6). Although methods exist to assess MSC immunosuppression in vitro, they are often based on only a few measured outcomes, assay culture conditions, and donor MSC samples (5, 7–9). To improve upon these methods, we developed an experimental and analytical approach to quantify MSC-mediated immune suppression using principal-component analysis (PCA) to integrate multiple measurements of T-cell activation assessed at a range of MSC densities. This approach allowed us to determine a single value for immunosuppressive capacity for MSC lines derived from two different manufacturing processes and 13 independent donors.

Another major challenge associated with MSC-based immune therapies is the lack of well-defined predictive markers to identify MSC lines with therapeutically relevant biological activities or manufacturing processes that produce more effective MSC-based products. Efforts have been made to identify MSC quality attributes associated with immunosuppression (6, 7), but the majority of clinical studies (10) rely upon the surface markers described by Dominici et al. (11). Having previously shown that morphology can predict MSC mineralization capacity (12), we hypothesized that morphological features associated with immunosuppression in MSCs could be identified and used to predict their performance in our quantitative immunosuppression assay.

Using our quantitative method for assessing the overall immunosuppressive capacity of MSCs, we identified and quantified differences in immunosuppression related to donor and passage. Additionally, we used this assay to demonstrate consistent enhanced immunosuppression for MSC lines that were stimulated with IFN-γ, as has been reported previously (8, 13, 14). Using high-content imaging of MSCs following IFN-γ stimulation, morphological features were identified that significantly correlated with immunosuppressive capacity and predicted the immunosuppressive capacity of other MSC and non-MSC lines, as well as the effects of IFN-γ pretreatment on immunosuppressive capacity. These findings highlight MSC morphology as an attribute associated with immunosuppressive capacity and provide additional evidence that morphological profiling can be used to predict MSC in vitro function (12, 15–17).

Results

PCA Permits Quantitative Assessment of Overall MSC Immunosuppressive Capacity.

To quantify the immunosuppressive capacity of MSC lines, 10⁵ peripheral blood mononuclear cells (PBMCs) from a heathy human donor (PBMC donor 1, “PBMC1”) were stimulated with T-cell–activating beads for 3 d in the presence of increasing concentrations of MSCs. PBMCs were then collected, and activation was assessed in CD4⁺ and CD8⁺ T cells by flow cytometry-based measurements of proliferation [carboxyfluorescein succinimidyl ester (CFSE) dilution], secretion of the proinflammatory cytokines IFN-γ and tumor necrosis factor-α (TNF-α), and surface expression of the activation marker CD25 (Fig. 1). Our initial studies were carried out using early-passage cells (passage 3) from six independent MSC lines (three each from two independent commercial sources; Table S1 for descriptions of all cell lines used). MSCs from all cell lines suppressed activation of CD4⁺ and CD8⁺ T cells in a dose-dependent manner, although the extent of suppression varied among MSC lines (Fig. 1).

Fig. 1.

The relative immunosuppressive capacity of MSC lines varies according to T-cell activation measurement and MSC concentration. A constant number of human PBMCs from a single donor (10⁵ per well) were stimulated with T-cell–activating beads in the presence of increasing concentrations of cells from six independent MSC lines. After 3 d, activation of CD4⁺ and CD8⁺ T cells was measured by proliferation (% CFSE diluted), CD25 expression (% CD25⁺), and cytokine secretion (% IFN-γ⁺ and % TNF-α⁺). Percent activation was calculated for each value relative to positive and negative control samples. The range of the y axes differs between graphs because low concentrations of some MSC lines enhanced activation above that observed in positive controls for some activation measurements. Dotted lines represent 100% activation. Data shown are the mean and SEM values of biological experiments performed with five replicates per condition.

Table S1.

Summary of hMSC line donor specifications for original (test) dataset using PBMC donor 1

Donor ID	Gender	Age	Passage	Vendor
127756	M	43	3, 5	Lonza
167696	F	22	3, 7	Lonza
8F3560	F	24	3, 7	Lonza
PCBM1632	M	24	3, 7	AllCells
PCBM1641	F	23	3, 7	AllCells
PCBM1662	F	31	3, 7	AllCells

We observed that the relative magnitude of immunosuppression mediated by MSC lines differed between T-cell activation measurements (Fig. 1). Whereas proliferation was maximally reduced by 80–90% in CD4⁺ and CD8⁺ cells by the highest concentrations of MSCs (Fig. 1, % CFSE Diluted), the frequency of CD25⁺ cells was only reduced by 20–40% for the same cell concentrations (Fig. 1, % CD25⁺). Additionally, the relative ranking of MSC lines in our cohort based upon magnitude of immunosuppression differed between T-cell activation measurements as well as between MSC concentrations. For example, MSC line PCBM1632 had the greatest observed suppression of T-cell proliferation for both CD4⁺ and CD8⁺ cells at the majority of cell densities (Fig. 1, % CFSE Diluted), but this was not the case for the percentage of IFN-γ–producing cells at multiple cell densities (Fig. 1, % IFN-γ⁺). Furthermore, cell density had a dramatic effect on the observed ranking of MSC immunosuppressive capacity based on percentage of TNF-α–producing CD4⁺ T cells where PCBM1632 had the most significant reduction at the lowest cell density, but at the highest cell density had one of the least significant reductions in TNF-α production (Fig. 1, % TNF-α⁺). Finally, we observed that low cell densities could enhance rather than inhibit some activation parameters above that of positive controls (>100% activation), especially IFN-γ secretion among CD4⁺ T cells (Fig. 1, % IFN-γ⁺). Therefore, using data from a single T-cell activation measure or from a single MSC concentration to quantify immunosuppressive capacity could lead to spurious conclusions.

To address these issues, we developed an analytical approach that incorporates various measures of T-cell activation and integrates data obtained from multiple MSC concentrations into a single value representing a given MSC line’s overall immunosuppressive capacity (Fig. 2A). The initial step in this approach uses unsupervised PCA to effectively reduce the dimensionality of the T-cell activation response and determine a composite variable for overall T-cell activation. A total of 16 activation variables [% CFSE diluted, % CD25⁺, % IFN-γ⁺, % TNF-α⁺, and associated median fluorescence intensity (MFI) values for each marker for both CD4⁺ and CD8⁺ T cells] were used to construct principal components (Fig. 2A). All replicates of both control and experimental conditions were used for PCA. When the overall experimental results were plotted in PC space (Fig. 2B), we observed that positive and negative T-cell activation controls clearly separated along principal component 1 (PC1), and experimental conditions fell between the positive and negative controls. The extent of T-cell activation observed in samples containing the lowest concentration of MSCs is nearly indistinguishable from positive control samples, but as the concentration of MSCs increases, samples cluster farther away from the positive controls and toward negative controls (Fig. 2B). PC1 accounted for 60–80% of the observed variance in all experiments using PBMCs from a single donor, suggesting that PC1 is a suitable measurement of overall T-cell activation (Fig. 2B).

Fig. 2.

Immunosuppressive capacity of MSC lines is quantified by integrating data from multiple MSC concentrations using a PCA-generated composite T-cell activation measurement. (A) Activation data are collected from both CD4⁺ and CD8⁺ T cells by flow cytometry after 3 d of stimulation in the presence or absence of MSCs. Histograms depict representative data from unstimulated T cells (red histograms) and activated T cells (blue histograms). Data from all experimental conditions were then analyzed by PCA. (B) Five concentrations of early-passage cells from six independent MSC lines (500, 1,000, 2,000, 5,000, and 10,000 MSCs per well) were cocultured with a constant number of PBMCs from a single donor (10⁵ per well) and T-cell–activating beads. PBMCs were activated for 3 d, and T-cell activation measurements (CFSE dilution, cytokine secretion, and CD25 expression) in CD4⁺ and CD8⁺ T cells were used to construct principal components. Each dot represents a single replicate (five replicates per cell concentration), and the size of each dot corresponds to the number of MSCs per well. The first principal component (PC1) was used as an unbiased composite measure for overall T-cell activation. (C) Activation data from T cells stimulated in the presence of a titration of early- and late-passage cells from six independent MSC lines were analyzed as in B. The first principal component (PC1) constructed from these data was then transformed relative to positive and negative control samples and plotted against MSC concentration. The shaded area beneath the line represents the area under curve (AUC). (D) AUC for early- and late-passage cells for each of the six MSC lines was calculated from data shown in C. AUC values were transformed relative to positive and negative controls so that an AUC value of 100 represents a lack of observed MSC-mediated suppression, and a value of 0 corresponds to complete MSC-mediated suppression at all cell concentrations. The AUC of early-passage MSC lines was significantly lower than that of late-passage MSC lines (P < 0.0001, two-way ANOVA). Data shown are the mean and SEM values of biological experiments performed with five replicates per condition.

Cell Line- and Passage-Dependent Differences in MSC Immunosuppression.

Using PC1 as a composite variable for T-cell activation, we developed an area-under-curve (AUC) analysis to incorporate T-cell activation data from multiple MSC concentrations into a single value representing a given MSC line’s overall immunosuppressive capacity. PCA using all available data were used to determine the composite T-cell activation value PC1, which was then transformed relative to control samples and plotted vs. MSCs per well (Fig. 2C). The AUC for each MSC line/passage group was calculated using the average transformed PC1 value for each concentration of MSCs (SI Materials and Methods). For ease of presentation and to facilitate interexperimental comparisons, AUC values were transformed such that a hypothetical MSC line that showed complete immunosuppression (comparable to negative control) at all cell concentrations would have an AUC of 0, whereas a hypothetical MSC line with no observed immunosuppression (comparable to positive control) would have an AUC value of 100. By convention, AUC values are representative of the observed level of overall T-cell activation, and therefore lower AUC values represent stronger immunosuppression and higher AUC values represent weaker immunosuppression. We used this combined PCA/AUC analysis to quantify the overall immunosuppressive capacity of early- and late-passage cells from each MSC line in our initial cohort (Fig. 2 C and D). We tested both early- and late-passage cells (p3 and p7 for all lines except 127756 where late passage represents p5) from all six of our initial cohort of MSC lines to increase the observed dynamic range of immunosuppression, as immunosuppressive capacity is reduced at higher passages (Fig. 2D, P < 0.0001, two-way ANOVA).

MSC Morphology Is Altered by Exposure to IFN-γ.

The immunosuppressive capacity of MSCs is enhanced by exposure to the proinflammatory cytokine IFN-γ (18, 19), and we hypothesized that MSCs would undergo characteristic morphological changes following exposure to IFN-γ that could be informative of their overall immunosuppressive capacity. To test this hypothesis, we used an experimental design in which morphological and immunosuppressive assays were performed in parallel with early- and late-passage MSCs (Fig. 3) to identify morphological features associated with immunosuppression.

Fig. 3.

Multiple MSC lines from independent donors are culture expanded and seeded for simultaneous morphological assessment and immunosuppressive capacity using a coculture assay with human PBMCs. For morphological assessment, MSCs are precultured for 24 h in control growth medium and then exposed to 0, 10, and 50 ng/mL IFN-γ medium for 24 h. After 24 h, cells are fixed and analyzed for high-dimensional cellular (FITC-maleimide, green) and nuclear (Hoechst, blue) morphological features, and a characteristic morphological signature is obtained for each MSC line sample. Lines demarcate cell and nuclear boundaries identified in image analysis. These morphological signatures are then correlated with AUC values to identify morphological features that can predict MSC immunosuppressive capacity.

The overall morphological signatures of six MSC lines under control and IFN-γ–stimulated conditions (10 and 50 ng/mL) were determined using both unsupervised (A) and supervised (B) PCA (Fig. 4 A and B). Unsupervised PCA used all 558 morphological features, whereas supervised PCA used a subset of 21 features (Fig. 4, top right table) found to be significantly different (P < 0.0001) between unstimulated (0 ng/mL IFN-γ) and stimulated groups for all cell lines. The mean of PC1 was significantly different (P < 0.0001) between both concentrations of IFN-γ and the unstimulated group for both unsupervised (Fig. 4C) and supervised (Fig. 4D) PCA. Moreover, there was no significant difference between the overall morphological profiles of MSCs cultured in 10 and 50 ng/mL IFN-γ. Representative cell images from a single MSC line for each condition illustrate the characteristic morphological changes observed upon IFN-γ stimulation, for example, increase in cell and nuclear aspect ratio/eccentricity, decrease in cell and nuclear form factor (Fig. 4E). Fig. 4F further highlights the distinct separation in the overall single-cell morphological profiles of unstimulated and stimulated MSCs using PC1, but also reveals the existence of single-cell heterogeneity within each population.

Fig. 4.

MSCs exhibit distinct morphological response upon stimulation with IFN-γ. Unsupervised (A) and supervised (B) principal-component analysis (PCA) performed on six MSC lines based on their morphological responses to unstimulated 0 ng/mL (black), 10 ng/mL (orange), and 50 ng/mL (red) IFN-γ . The table summarizes differences in cell- and nucleus-associated morphological features (used in supervised PCA) upon IFN-γ stimulation. C and D present the mean ± SD principal component 1 position (PC1 from A and B) for each condition using unsupervised and supervised morphological features, respectively (*P < 0.0001, significantly different from unstimulated). PC1 accounted for 47.1% (unsupervised) and 53.7% (supervised) of the data variance. Data in A–D represent six cell lines (biological replicates) from five independent experiments (n = 30 total points). (E) Separate PCA performed using single-cell morphological data (>1,000 cells presented in density plots of PC space) from one MSC line (PCBM1632) under unstimulated (black) and stimulated (10 ng/mL IFN-γ, orange) conditions. Representative cells were chosen randomly from the centroids and are shown for each condition. (F) Histogram representation of PC1 from single-cell data shown in E.

Morphological Features of MSCs After IFN-γ Stimulation Correlate with Immunosuppressive Capacity.

Having confirmed that MSCs alter their overall morphology upon IFN-γ stimulation, we sought to identify morphological features that correlated with our immunosuppression results and could be used to predict a given MSC line’s overall immunosuppressive capacity. In the absence of IFN-γ stimulation, no morphological features of MSCs correlated well with immunosuppressive capacity. In contrast, we found that multiple morphological features of IFN-γ–stimulated MSCs were significantly correlated with immunosuppressive capacity (Fig. 5). For both concentrations of IFN-γ, similar correlation trends were observed for five cellular features (perimeter, form factor, maximum feret diameter, nucleus/cytoplasm ratio, and Δminor axis length). Of particular note, the differential feature Δminor axis length_cell was positively correlated with AUC, as a larger Δminor axis length_cell for a given MSC sample was associated with higher AUC values and thus lower immunosuppressive capacity. Previously, we observed that this differential feature correlated similarly with osteogenic potential as larger Δminor axis length_cell was associated with the absence of mineralization (12).

Fig. 5.

Morphological response of MSCs to IFN-γ correlates with immunosuppression. (A) Correlation of individual morphological features from MSCs stimulated by 10 ng/mL IFN-γ (Top) and 50 ng/mL IFN-γ (Bottom) with quantitative AUC immunosuppression results. Each point represents a single MSC line/passage with its associated median morphological feature and AUC values. (B) Four-dimensional graphs showing correlation of the top three morphological features (perimeter, nucleus/cytoplasm ratio, maximum feret diameter) with AUC (color map shown below graphs) using morphological data from both 10 and 50 ng/mL IFN-γ conditions. Representative cells stimulated with 10 ng/mL IFN-γ from a high immunosuppressive (low AUC) MSC line (C) and a low immunosuppressive (high AUC) cell line (D). Data in A and B represent morphological data from five independent experiments using six MSC lines (n = 30 total points).

The three morphological features most significantly correlated with immunosuppressive capacity are presented in Fig. 5B in the form of 4D graphs. MSC lines with high immunosuppressive capacity (low AUC values) clustered in the region that corresponds to a morphological profile of low cell perimeter, low cell maximum feret diameter, and high nucleus/cytoplasm ratio after IFN-γ stimulation (Fig. 5B). A linear regression model constructed using these three features was significantly correlated with immunosuppression (Fig. S1, P < 0.0001, R = 0.78), whereas a model made using the same three features in unstimulated controls did not correlate with immunosuppressive response (Fig. S1, P > 0.02, R = 0.55). Representative cells from both high and low immunosuppressive MSC lines after stimulation with 10 ng/mL IFN-γ are shown in Fig. 5 C and D, respectively.

Fig. S1.

Morphological features from stimulated MSCs better predict immunosuppression response. (A) Four-dimensional graphs showing correlation of the top three morphological features (perimeter, nucleus/cytoplasm ratio, maximum feret diameter) with AUC (color map shown below graphs) using morphological data from both unstimulated and 10 ng/mL IFN-γ conditions. (B) Linear regression models constructed using the morphological features cell perimeter, nucleus/cytoplasm ratio, and cell maximum feret diameter in unstimulated (Left) and 10 ng/mL IFN-γ (Right) to compare experimental overall immunosuppression (AUC_exp) with predicted overall immunosuppression (AUC_pred). Data represent five independent experiments using six MSC lines (n = 30 total points). The linear regression model was constructed using JMP and used to determine predicted AUC values in Fig. 6.

MSC Morphology After IFN-γ Stimulation Predicts Overall Immunosuppressive Capacity.

We sought to corroborate our initial findings using new MSC lines and with PBMCs isolated from a different donor (Table S2). Because the susceptibility of T cells to MSC-mediated suppression can differ between individuals (8, 20), we tested several MSC lines with PBMCs from both donors and used these results to normalize AUC values obtained from experiments using PBMCs from our second donor (PBMC2). T cells from PBMC2 were more susceptible to MSC-mediated immune suppression, as AUC values for MSC lines tested with PBMC2 cells were consistently lower than AUC values obtained when the same MSC lines were tested with T cells from PBMC1 (Fig. S2).

Table S2.

Summary of hMSC line donor specifications for new (validation) dataset with PBMC donor 2 (PBMC2)

Previously tested with PBMC1	Donor ID	Sex	Age	Passage	Vendor
Yes	127756	M	43	5	Lonza
Yes	167696	F	22	3, 7	Lonza
Yes	8F3560	F	24	7	Lonza
Yes	PCBM1632	M	24	3, 7	AllCells
Yes	PCBM1641	F	23	7	AllCells
Yes	PCBM1662	F	31	7	AllCells
No	110877	M	22	3, 7	Lonza
No	1F3422	M	39	3	Lonza
No	BM3018	M	41	3	AllCells
No	BM2893	M	40	3	AllCells
No	NHDF271678	F	58	4	Lonza

Fig. S2.

Method for normalizing AUC values from validation experiment with PBMCs derived from a second donor (PBMC2). The ratio of AUC values from cell lines tested with both PBMC cell lines (PBMC1 and PBMC2) are calculated and averaged. This average AUC ratio is used to normalize AUC values from the validation experiments to allow for comparison (and prediction) of immunosuppression capacity of MSC lines. n/a, not assessed.

Following normalization of the AUC values between experimental datasets (Fig. S2), AUC values for the validation dataset also demonstrated high correlation (Fig. 6A) with the top three morphological features identified from the test dataset (cell perimeter, nucleus/cytoplasm ratio, max feret diameter). Predicted AUC values were calculated for the validation dataset using the model constructed from the test dataset (Fig. S1) and compared with the actual experimental AUC values. We found that AUC values predicted by the model were highly correlated with AUC values obtained experimentally (Fig. 6B, P < 0.001, R = 0.72 for AUC_pred vs. AUC_exp), and closely followed the line of unity. This correlation of predicted vs. experimental AUC values was evident across the full dynamic range of immunosuppressive capacity (Fig. 6C).

Fig. 6.

Morphological response of MSCs to IFN-γ predicts degree of immunosuppression. (A) Four-dimensional graph showing correlation of the top three morphological features (perimeter, nucleus/cytoplasm ratio, maximum feret diameter) with AUC (color map shown below) using morphological data from 10 ng/mL IFN-γ conditions for additional MSC lines (n = 16 biological replicates from one experiment) with a different PBMC donor (Table S2 for descriptions of cell lines). (B) Predictive model for AUC developed using linear regression on the top three highly correlated morphological features and AUC values from test set (perimeter, nucleus/cytoplasm ratio, and maximum feret diameter) and used to predict the AUC of validation set cell lines based on their early morphological signatures. Predicted AUC values (AUC_pred) vs. experimental AUC values (AUC_exp) are shown overall for the validation dataset (16 different cell line samples from one experiment) with linear fit (solid line) and reference AUC_pred = AUC_exp line for comparison. (C) Predicted (black) vs. experimental (white) AUC values shown in bar graph for each individual sample (ordered Left to Right from high to low immunosuppression). (D) The enhancing effect of IFN-γ stimulation on MSC immunosuppression was demonstrated using eight MSC lines, which consisted of the six cell lines in Table S1 and two RB MSC lines. Overall immunosuppression response (transformed relative to inactivated and activated PBMCs) presented as PC1 for each MSC line at a range of cell densities (x axis). Five replicates were used for each cell concentration. Shaded region for unstimulated (black) and 10 ng/mL IFN-γ stimulated (orange) used for AUC analysis of overall immunosuppressive capacity. (E) AUC immunosuppression scores for each MSC line for unstimulated (−IFN-γ, black) and stimulated (+IFN-γ, orange). Data presented as mean ± SEM (n = 5 per condition). Paired t test showed significant difference in AUC values for unstimulated and stimulated conditions for each cell line (P < 0.0001). (F) Quantification of the change in immunosuppression [−ΔAUC = −(AUC_stimulated – AUC_unstimulated], where a larger –ΔAUC indicates a greater improvement in immunosuppression due to stimulation by IFN-γ. Correlation of –ΔAUC values with early differential morphological features ΔMajor Axis Length_{Nucleus,10–0} (G) and ΔPerimeter_{Nucleus,10–0} (H), which were found to be predictive of the change in immunosuppression upon IFN-γ stimulation.

Although these morphological features were useful predictors for immunosuppression with MSC lines expanded using our standard conditions [α-MEM with 16.6% (vol/vol) lot-selected FBS], they did not effectively predict the immunosuppressive capacity of MSCs expanded by a different process [RoosterBio (RB) MSC lines, Table S3], and none of the predictive morphological features previously identified (cell perimeter, nucleus/cytoplasm ratio, and maximum feret diameter) correlated with AUC values in RB MSC lines (Fig. S3A). Instead, unique morphological features were found to correlate with immunosuppression in RB MSC lines: differential nuclear morphological features extent, form factor, and minimum feret diameter (Fig. S3B). Although the same characteristic morphological change upon IFN-γ stimulation was observed as with the other cell lines (increase in aspect ratio and decrease in nucleus/cytoplasm ratio), this finding suggests that predictive morphological features may differ depending on the manufacturing process.

Table S3.

Summary of RoosterBio (RB) MSC lines

Donor ID	Cell type	Gender	Age	Passage	Vendor
RB9	MSC	M	43	2, 5	RoosterBio
RB14	MSC	F	20	2, 5	RoosterBio
RB16	MSC	F	29	2, 5	RoosterBio

Fig. S3.

Unique morphological features correlate with immunosuppression for MSCs derived from different manufacturing process. (A) Correlation of AUC values for RoosterBio (RB) MSC lines with morphological features found to correlate with immunosuppression using original manufacturing process (Fig. 6). (B) Correlation of AUC values for RB cell lines with morphological features unique to this subset of MSCs. Four-dimensional graphs showing correlation of morphological features identified from original manufacturing process (C) and RB manufacturing process (D) with AUC values (color code shown below graphs). Bottom right table identifies morphological features that change upon IFN-γ stimulation for all RB MSC lines. Data represent five independent experiments using three RB MSC lines (n = 14 total points).

Increased Immunosuppression Following IFN-γ Pretreatment Can Be Predicted by MSC Morphology.

We next examined the effects of IFN-γ pretreatment on MSC immunosuppressive capacity by stimulating eight MSC lines with 10 ng/mL IFN-γ for 24 h before PBMC coculture. As expected, IFN-γ pretreatment significantly enhanced the immunosuppressive capacity of all MSC lines (Fig. 6 D and E, P < 0.0001, paired t test). The difference in AUC values obtained under control and IFN-γ–stimulated conditions was used to quantify the magnitude of this enhancement for each MSC line as –ΔAUC, where larger values indicate an increased effect of IFN-γ stimulation on immunosuppressive capacity (Fig. 6F). Although this enhancement in immunosuppression could be predicted by noting the difference in the panel of morphological features in Fig. 4, we sought to correlate the magnitude of this enhanced immunosuppression due to IFN-γ stimulation with MSC morphology. We found that Δmajor axis length_{nucleus,10–0} and Δperimeter_{nucleus,10–0} were highly correlated with IFN-γ–induced enhancement of immunosuppression and could potentially be used to predict the effects of other soluble factor pretreatments (Fig. 6 G and H). Of note, both features exhibited consistent correlative trends to other differential features in that a greater response to IFN-γ (i.e., increased Δmajor axis length_nucleus and Δperimeter_nucleus) correlated with a smaller increase in immunosuppression. These results suggest that morphological profiling can predict not only a given MSC line’s inherent immunosuppressive capacity, but also the magnitude of IFN-γ–mediated enhanced immunosuppression.

SI Materials and Methods

Cell Lines and MSC Manufacturing/Expansion.

Human bone marrow-derived stem cells were obtained from six different donors purchased from Lonza (127756, 167696, 8F3560) and AllCells (PCBM1632, PCBM1641, PCBM1662) at passage 2 (Table S1 for donor specifications). MSC culture conditions were chosen based on well-established protocols (55). Briefly, MSCs were expanded by plating at a density of 10,500 cells per T175 flask (60 cells per cm²) using standard MSC growth medium: 500 mL of α-MEM, 6 mL of 200 mM l-glutamine, 6 mL of 10,000 U/mL penicillin–streptomycin (Life Technologies), and 100 mL of lot-selected FBS (JM Bioscience). Upon reaching 80% confluence, cells were trypsinized and replated at 10,500 cells per T175 flask, designated as one “passage.” Based on the cell yield at 80% confluence, each passage consisted of approximately seven to nine population doublings. MSCs from each donor were continuously expanded (with no freezing/thawing occurring between passages) with fractions of cells being frozen at passages 3, 5, and 7 (p3, p5, and p7) for donors 167696, 8F3560, PCBM1632, PCBM1641, and PCBM1662. Donor 127756 was unable to expand beyond p5. MSCs were cryopreserved in freezing medium consisting of 30% (vol/vol) FBS, 5% (vol/vol) DMSO (Sigma), 1% l-glutamine, 1% penicillin–streptomycin in α-MEM. Throughout this work, the passage number refers to the number of times the MSCs were trypsinized before cryopreservation. Most of the MSC lines used in this study have been extensively characterized, and further information about the donors’ surface marker expression, genomic, epigenetic, and proteomic profiles, as well as performance in multiple bioassays for MSC function has been published previously (12, 38, 47, 48, 56–58).

Additional cell lines used for validation are described in Tables S2 and S3. Bone marrow-derived cell lines 110877, 1F3422, BM3018, and BM2893, and a fibroblast cell line were all cultured in the same FBS-containing growth medium as MSC lines listed in Table S1. Bone marrow-derived MSC lines purchased from RoosterBio (RB) were expanded in RB expansion medium at an initial plating density of 8.33 × 10⁵ MSCs per T225 flask (∼3,700 cells per cm²). Upon reaching 80% confluence, cells were trypsinized and replated at 8.33 × 10⁵ MSCs per T225 flask, designated as one passage. Based on the cell yield at 80% confluence, each passage consisted of approximately three to five population doublings. RB MSC lines from each donor were continuously expanded for several passages (with no freezing/thawing occurring between passages), with fractions of cells being frozen throughout the expansion process at passages 2 and 5. All MSC samples were obtained from patients with the proper informed consent. All cell lines used in this work possessed viability >95% (assessed using Trypan Blue) before plating for immunosuppression coculture and morphological assessment.

MSC/PBMC Immunosuppression Coculture Assay.

MSCs were harvested from culture flasks and five concentrations for each MSC line (500, 1,000, 2,000, 5,000, and 10,000 cells per well) were added to 96-well flat-bottomed plates in standard MSC growth medium and allowed to adhere for 48 h. Medium was then removed without disturbing adherent MSCs, and 10⁵ unseparated PBMCs were added to each well at a final concentration of 5 × 10⁵ PBMCs per mL in standard RPMI medium [1× RPMI with 10 mg/mL glycine, 100 U/mL penicillin, 100 U/mL streptomycin, and 10% (vol/vol) FBS]. Cryopreserved PBMCs were thawed the previous day and rested in standard RPMI medium overnight, and then stained with CFSE (Life Technologies; C34554) before being added to MSC cultures. PBMCs used in this study (AllCells; PB006F) were collected from healthy adult human donors. To activate T cells, 10⁵ beads coated with anti-CD3 and anti-CD28 (Life Technologies; 11131D) were added to PBMC cultures. At least five replicates for each condition were used in most experiments. After 3 d of activation in the presence of MSCs, unseparated PBMCs were harvested for analysis by flow cytometry. All PBMC samples were obtained from patients with the proper informed consent.

Flow Cytometry.

For surface staining, one-half of the collected cells were washed in washing buffer [HBSS containing 2% (vol/vol) FBS], and then stained with optimized concentrations of anti–CD4-PerCP (peridinin-chlorophyll) (BioLegend; 300528), anti–CD8α-APC (allophycocyanin) (BioLegend; 301049), and anti–CD25-Pacific Blue (BioLegend; 356130) for 30 min at room temperature. Stained cells were then washed once with washing buffer and fixed in HBSS containing 2% (wt/vol) formalin for 1 h at 4 °C. After fixation, cells were washed twice and resuspended in an appropriate volume of washing buffer (50–100 µL) and analyzed on a MACSQuant Analyzer flow cytometer (Miltenyi Biotec). For intracellular cytokine staining, cells not used for surface staining were stimulated for 4–6 h with 50 ng/mL PMA (Sigma-Aldrich; P1585) and 1 µg/mL ionomycin (Sigma-Aldrich; I9657) in the presence of brefeldin A (BD Biosciences; 555029). Cells were then washed once and stained with anti–CD4-PerCP and anti–CD8α-APC as described above before fixation with HBSS containing 2% (wt/vol) formalin for 1 h. Following fixation, cells were washed twice and left overnight at 4 °C in washing buffer after which cells were incubated in 1× permeabilization buffer (BD Biosciences; 554723) for 15 min, and then stained with anti–IFN-γ-PECy7 (phycoerythrin-cyanine 7) (BioLegend; 506518) and anti–TNF-α-Pacific Blue (BioLegend; 502920) for 30 min at room temperature. After washing twice with permeabilization buffer and once with washing buffer, PBMCs were then resuspended in an appropriate volume of washing buffer (50–100 µL) and analyzed on a MACSQuant analyzer flow cytometer (Miltenyi Biotec). Flow data were analyzed with FlowJo (Tree Star; version X 10.0.6) using compensation matrices generated by the software after analyzing single-color control samples. Scatter properties were used to determine a gate for PBMC cells that did not contain activating beads or MSCs, and then CD4⁺ and CD8⁺ cells were analyzed independently for CFSE dilution and staining for CD25, IFN-γ, and TNF-α. Gates identifying CFSE-diluted, CD25⁺, IFN-γ⁺, and TNF-α⁺ cells were identical for all samples analyzed within a given experiment. Histogram gates used to identify “CFSE-diluted” cells were placed so that <1% of T cells in unstimulated control wells (i.e., PBMCs cultured for 3 d in the absence of activating beads or MSCs) were included by the gate. All antibodies used for flow cytometry in this study are listed in Table S7.

Table S7.

Antibodies used for flow cytometry in this study

Antigen	Clone	Conjugation	Manufacturer/catalog no.	Validation profile
hCD4	RPA-T4	PerCP	BioLegend 300528	1DegreeBio
hCD8α	RPA-T8	APC	BioLegend 301049	1DegreeBio
hCD25 (IL2RA)	M-A251	Pacific Blue	BioLegend 356130	Antibodypedia
hIFN-γ	B27	PE/Cy7	BioLegend 506518	1DegreeBio
hTNF-α	MAb11	Pacific Blue	BioLegend 502920	1DegreeBio

PE/Cy7, phycoerythrin-cyanine 7.

Raw flow cytometry values (Y_EXPERIMENTAL) for all activation variables were transformed relative to the average values for both positive (Y_POS) and negative (Y_NEG) controls (Eq. S1). After transformation, positive control samples will have an average value of 100, whereas negative control samples will have an average of 0:

$$\%\mathrm{Activation}=100\left(\frac{Y_{\mathrm{EXPERIMENTAL}}-\overline{Y_{\mathrm{NEG}}}}{\overline{Y_{\mathrm{POS}}}-\overline{Y_{\mathrm{NEG}}}}\right).$$ [S1]

Quantitative Assessment of MSC Immunosuppressive Capacity.

Transformed data acquired by flow cytometry were used to determine a composite variable for T-cell activation. Sixteen variables were used for principal-components analysis (PCA)—eight variables each from CD4⁺ and CD8⁺ cells (% CFSE diluted, % CD25⁺, % IFN-γ⁺, % TNF-α⁺, and associated median fluorescence intensity values for each marker). All replicates of both control and experimental conditions were used for PCA. Principal components were generated from correlations estimated by restricted maximum likelihood with JMP 12.1.0 software, with the first principal component (PC1) capturing the highest percentage of data variance, and the second principal component (PC2) orthogonal to PC1 capturing the next highest percentage of data variance. In all experiments, PC1 accounted for ∼60–80% of the total observed variance and was therefore used as a composite variable for T-cell activation. Eigenvectors for all variables used in combination to produce the first and second principal components for each experiment are reported in Table S5. For clarity of presentation, as well as simplifying interexperimental comparisons and downstream analyses, PC1_EXPERIMENTAL values were transformed relative to controls into PC1 (Eq. S2):

$$\mathrm{PC}1=100\left(\frac{\mathrm{PC}{1}_{\mathrm{EXPERIMENTAL}}-\overline{\mathrm{PC}{1}_{\mathrm{NEG}}}}{\overline{\mathrm{PC}{1}_{\mathrm{POS}}}-\overline{\mathrm{PC}{1}_{\mathrm{NEG}}}}\right).$$ [S2]

Table S5.

Eigenvectors from PCA

			Fig. 2B		Fig. 6D
T cells	Activation measure		PC1	PC2	PC1	PC2
CD4	CFSE	%Diluted	0.27659	−0.21815	0.2718	0.00691
CD4	CD25	%+	0.28617	−0.11468	0.27062	−0.1492
CD4	IFN-γ	%+	0.28625	0.05704	0.26558	−0.00457
CD4	TNF-α	%+	0.29449	0.05274	0.24719	0.01995
CD4	CFSE	MFI	0.28003	−0.21065	0.26005	−0.15781
CD4	CD25	MFI	0.27582	−0.1577	0.22068	−0.45708
CD4	IFN-γ	MFI	0.25773	0.00111	0.25667	0.23656
CD4	TNF-α	MFI	0.22006	0.33217	0.23165	0.33473
CD8	CFSE	%Diluted	0.27627	−0.22543	0.26756	0.04775
CD8	CD25	%+	0.25977	0.03333	0.267	−0.08024
CD8	IFN-γ	%+	0.07805	0.43247	0.27158	−0.14014
CD8	TNF-α	%+	0.19144	0.47814	0.25855	0.03444
CD8	CFSE	MFI	0.27553	−0.18912	0.26644	−0.01742
CD8	CD25	MFI	0.27312	−0.07296	0.23177	−0.39721
CD8	IFN-γ	MFI	0.22008	0.11866	0.19671	0.45456
CD8	TNF-α	MFI	0.13455	0.48051	0.19621	0.42567

MFI, median fluorescent intensity.

Average PC1 values were plotted vs. MSC concentration, and the area under curve (AUC) for this graph was calculated as a measure of the overall immunosuppressive capacity for a given MSC line. MSC concentrations are designated X_1…5, and their experimentally determined PC1 values as Y_1…5 (see graphic below).

The area under the curve (AUC_EXPERIMENTAL) can be calculated as the sum of the areas of four adjacent trapezoids, each with the base dimension (D_1…4) defined as the difference between sequential MSC concentration values. The area of each trapezoid is equal to the base D_n multiplied by the average of its corresponding PC1 values (Y_n and Y_n+1). This equation can then be simplified as the sum of multiples of each PC1 value (Eq. S3):

$$\mathrm{AU}{\mathrm{C}}_{\mathrm{EXPERIMENTAL}}=Y_1\left(\frac{X_2-X_1}{2}\right)+Y_2\left(\frac{X_3-X_1}{2}\right)+Y_3\left(\frac{X_4-X_2}{2}\right)+Y_4\left(\frac{X_5-X_3}{2}\right)+Y_5\left(\frac{X_5-X_4}{2}\right).$$ [S3]

For ease of presentation and to facilitate interexperimental comparisons, AUC_EXPERIMENTAL values were transformed such that a hypothetical MSC line that showed complete immune suppression at all concentrations would have an AUC of 0, and an MSC line with no immunosuppressive capacity would have an AUC value of 100 (Eq. S4):

$$\mathrm{AUC}=\frac{\mathrm{AU}{\mathrm{C}}_{\mathrm{EXPERIMENTAL}}}{X_5-X_1}.$$ [S4]

AUC values are inversely proportional to immunosuppressive capacity, so lower AUC values correspond to higher levels of MSC-mediated immune suppression. Within-group variance of the AUC was generally consistent between MSC lines and experimental conditions, suggesting that parametric tests are appropriate when comparing AUC means between experimental groups. With five replicates per group, a representative experiment had 80% power to detect a between-group AUC mean difference of 6.94 when α = 0.05 (Table S6).

Table S6.

Post hoc power analysis (using JMP 11) for statistical methods performed throughout this work

Figure	Description	Sample size	SD	α	Power	Difference to detect	Observed difference
Fig. 2D	Early passage vs. late passage (two-way ANOVA)	10	3.248 (average)	0.05	0.8	6.94	5.97–26.71
Fig. 4C	10 ng/mL IFN-γ vs. unstimulated (two-tailed t test)	60	3.552	0.025	0.8	3.23	11.52
	50 ng/mL IFN-γ vs. unstimulated (two-tailed t test)	60	3.378	0.025	0.8	3.07	12.61
Fig. 4D	10 ng/mL IFN-γ vs. unstimulated (two-tailed t test)	60	1.171	0.025	0.8	1.06	6.03
	50 ng/mL IFN-γ vs. unstimulated (two-tailed t test)	60	1.01	0.025	0.8	0.92	6.82
Fig. 6E	+IFN-γ vs. –IFN-γ (two-tailed paired t test)	16	2.95 (average)	0.05	0.8	4.44	19.1

MSC Culture and Staining for Morphological Analysis.

For MSC morphological analysis, cells were thawed and cultured in growth medium for 48 h before seeding. Cells were seeded at a density of 1,000 cells per well (four total wells per experimental group) in 12-well plates (Corning) and cultured for 24 h in growth medium. Growth medium was replaced with growth medium containing 0, 10, or 50 ng/mL IFN-γ (Life Technologies) and cultured for an additional 24 h after which cells were fixed with 4% (wt/vol) paraformaldehyde for 15 min. Cell and nuclear morphology were assessed using FITC-maleimide (Life Technologies) and Hoechst (Sigma-Aldrich), respectively. Briefly, fixed samples were incubated with 20 µM FITC-maleimide for 30 min, washed with PBS, incubated with 1 mg/mL Hoechst for 5 min, and washed with PBS before imaging.

High-Content Imaging of MSC Morphology.

Samples for morphological analysis were imaged using an inverted Nikon Ti-U microscope with automated stage (Prior) and filters (Chroma Technology) compatible with FITC (cell morphology) and Hoechst (nuclear morphology). For each group, 160 random 10× images were acquired per well (n = 4 wells) for each IFN-γ condition (0, 10, or 50 ng/mL). At least 1,000 cells were assessed for each experimental group, with approximately equal numbers of cells analyzed from each of four replicate wells. Automated quantification of cellular and nuclear shape features was performed using CellProfiler (59) to obtain quantitative morphological data for each cell consisting of 46 cellular shape features and 46 nuclear shape features (Table S8). The CellProfiler algorithm (termed pipeline) used to analyze cell and nuclear morphology can be viewed in Table S9.

Table S8.

Single-cell and nuclear shape features measured by CellProfiler

Cell shape features		Nucleus shape features
Area	Zernike_4_2	Area	Zernike_4_2
Compactness	Zernike_4_4	Compactness	Zernike_4_4
Eccentricity	Zernike_5_1	Eccentricity	Zernike_5_1
Extent	Zernike_5_3	Extent	Zernike_5_3
FormFactor	Zernike_5_5	FormFactor	Zernike_5_5
MajorAxisLength	Zernike_6_0	MajorAxisLength	Zernike_6_0
MaximumFeretDiameter	Zernike_6_2	MaximumFeretDiameter	Zernike_6_2
MinorAxisLength	Zernike_6_4	MinorAxisLength	Zernike_6_4
MinimumFeretDiameter	Zernike_6_6	MinimumFeretDiameter	Zernike_6_6
MaximumRadius	Zernike_7_1	MaximumRadius	Zernike_7_1
MeanRadius	Zernike_7_3	MeanRadius	Zernike_7_3
MedianRadius	Zernike_7_5	MedianRadius	Zernike_7_5
AspectRatio	Zernike_7_7	AspectRatio	Zernike_7_7
Perimeter/Area Ratio	Zernike_8_0	Perimeter/Area Ratio	Zernike_8_0
Perimeter	Zernike_8_2	Perimeter	Zernike_8_2
Solidity	Zernike_8_4	Solidity	Zernike_8_4
Zernike_0_0	Zernike_8_6	Zernike_0_0	Zernike_8_6
Zernike_1_1	Zernike_8_8	Zernike_1_1	Zernike_8_8
Zernike_2_0	Zernike_9_1	Zernike_2_0	Zernike_9_1
Zernike_2_2	Zernike_9_3	Zernike_2_2	Zernike_9_3
Zernike_3_1	Zernike_9_5	Zernike_3_1	Zernike_9_5
Zernike_3_3	Zernike_9_7	Zernike_3_3	Zernike_9_7
Zernike_4_0	Zernike_9_9	Zernike_4_0	Zernike_9_9

Definition of each feature is available in the CellProfiler manual (cellprofiler.org/manuals.shtml) under “MeasureObjectSizeShape.”

Table S9.

CellProfiler pipeline used to automatically quantify cellular and nuclear morphological features

Module	Description
IdentifyPrimaryObjects	Individual nuclei analyzed as primary objects
	Parameters
	Discard objects touching border of image: yes
	Thresholding method: Otsu global
	Three class thresholding
	Minimize weighted variance
	Middle intensity class pixels assigned to foreground
	Threshold correction factor: 1.25
	Lower and upper bounds on threshold: 0.0025–1.0
	Method to distinguish clumped objects: shape
	Method to draw dividing lines between clumped objects: shape
	Automatically calculate size of smoothing filter: yes
	Automatically calculate minimum allowed distance between local maxima: yes
	Speed up by using lower resolution image to find local maximal: yes
	Retain outlines of the identified objects: yes
	Fill holes in identified objects: yes
IdentifySecondaryObjects	Cells identified as secondary objects associated with nuclei
	Parameters
	Method to identify secondary objects: propagation
	Thresholding method: background global
	No smoothing
	Threshold correction factor: 1.1
	Regularization factor: 0.05
	Fill holes in identified objects: yes
	Discard secondary objects that touch the edge of the image: yes
	Discard the associated primary objects: yes
	Retain outlines of the new primary objects: yes
	Retain outlines of the identified secondary objects: yes
GrayToColor	Creating composite green/blue image for thresholding evaluation
OverlayOutlines	Overlaying cell and nucleus outlines onto composite image to visually evaluate quality of thresholding process
MeasureObjectSizeShape	All features from Table S8 measured here for each cell/nucleus
ExportToSpreadsheet	Measurements exported to .csv file
DisplayDataOnImage	Unique numbering of individual cells to allow for identification of poorly thresheld cells (or debris) that could then be removed from analysis in the exported spreadsheet
SaveImages	Save outlined composite images (examples shown in Figs. 4 and 5)

Determination of Differential Morphological Features.

Differential morphological features were added to the overall morphological signature for each group by taking the difference in the median values for each IFN-γ culture condition as indicated in the following equations:

$$\mathrm{\Delta Featur}{\mathrm{e}}_{10-\text{0}}=\mathrm{Median}\left(\mathrm{Featur}{\mathrm{e}}_{\mathrm{10}\mathrm{ng}/\mathrm{mL}\hspace{0.17em}\mathrm{IF}{\mathrm{N}}_{\mathrm{\gamma }}}\right)-\mathrm{Median}\left(\mathrm{Featur}{\mathrm{e}}_{\mathrm{0}\mathrm{ng}/\mathrm{mL}\hspace{0.17em}\mathrm{IF}{\mathrm{N}}_{\mathrm{\gamma }}}\right),$$

$$\mathrm{\Delta Featur}{\mathrm{e}}_{50-\text{0}}=\mathrm{Median}\left(\mathrm{Featur}{\mathrm{e}}_{\mathrm{50}\mathrm{ng}/\mathrm{mL}\hspace{0.17em}\mathrm{IF}{\mathrm{N}}_{\mathrm{\gamma }}}\right)-\mathrm{Median}\left(\mathrm{Featur}{\mathrm{e}}_{\mathrm{0}\mathrm{ng}/\mathrm{mL}\hspace{0.17em}\mathrm{IF}{\mathrm{N}}_{\mathrm{\gamma }}}\right),$$

$$\mathrm{\Delta Featur}{\mathrm{e}}_{50-10}=\mathrm{Median}\left(\mathrm{Featur}{\mathrm{e}}_{\mathrm{50}\mathrm{ng}/\mathrm{mL}\hspace{0.17em}\mathrm{IF}{\mathrm{N}}_{\mathrm{\gamma }}}\right)-\mathrm{Median}\left(\mathrm{Featur}{\mathrm{e}}_{\mathrm{10}\mathrm{ng}/\mathrm{mL}\hspace{0.17em}\mathrm{IF}{\mathrm{N}}_{\mathrm{\gamma }}}\right).$$

Determination of an Overall MSC Morphological Signature.

Overall morphological signatures were constructed for each group by taking the median value of the 93 total cellular and nuclear features in each IFN-γ condition for a total of 279 morphological features. Differential morphological features were also calculated as outlined in SI Materials and Methods. The addition of differential morphological features resulted in an assessment of 558 total parameters for each group.

Identification of Representative MSC Morphological Phenotypes.

To identify representative cells for different conditions, PCA was performed on single-cell data using the 93 total cellular and nuclear features. The centroid for each condition was calculated using PC1 and PC2, and representative single cells were selected that were proximal to the centroid. Single cells proximal to the centroid were chosen based on PC1 and PC2 values being within a distance of 5% of the SD for the entire PCA. Representative cells were selected to illustrate qualitative overall differences in morphology observed between unstimulated (0 ng/mL IFN-γ) and stimulated (10 ng/mL IFN-γ) MSCs, as well as morphological differences between MSCs with high and low immunosuppressive capacity.

Data Analysis and Statistical Comparisons.

High-dimensional morphological and immunosuppressive data were analyzed using PCA with JMP (version 12). Overall differences in MSC morphology following IFN-γ stimulation were highlighted by performing PCA on the morphological signatures (93 total features) for each group at 0, 10, and 50 ng/mL IFN-γ. Supervised PCA was performed using a subset of morphological features found to be significantly different between 0 ng/mL IFN-γ and 10 ng/mL IFN-γ conditions for each group. A regression analysis of MSC morphology with immunosuppressive capacity was performed using JMP12, and correlation coefficients were determined between the overall immunosuppressive capacity (as measured by AUC) and each of the 558 morphological features. Morphological features were deemed significant if the P value of the correlation was less than the Bonferroni-corrected significance level (α = 0.05/558 ∼ 0.00009). Linear regression models using morphological features to predict immunosuppression were constructed using JMP12 software and applied to validation cell lines (Fig. S1). For both immunosuppression and morphological assay analyses, paired t test, ANOVA, and post hoc analysis were performed using GraphPad Prism 6. Retrospective analysis of our experimental sample sizes, sample variation, and significance levels revealed that our studies were adequately powered to detect statistical differences in our immunosuppression and morphological data (Table S6).

Data Availability.

All data presented throughout this work are available upon request to the authors.

Discussion

MSC-based cellular therapies continue to be the subject of intense research despite a lack of clearly demonstrated clinical effectiveness in treating inflammatory disorders (2, 4). A major contributor to these inconsistent clinical outcomes may be the inherent variability in the immunosuppressive capacity of MSC lines due to differing tissues of origin, manufacturing processes, and donor-specific differences (10, 21–23). Although this functional heterogeneity is well known, no reliable predictors of MSC immunosuppressive capacity have been identified (3, 24). Another significant challenge facing development of MSC-based therapies is the lack of well-defined, robust assays for assessing their immunosuppressive function. Current assays for assessing MSC immunosuppressive capacity exist; however, they often rely on quantification of only a few markers in a single culture condition (8, 9, 20). Based on these limitations, we developed a robust, quantitative method of assessing the immunosuppressive capacity of human bone marrow-derived MSCs and further demonstrated that morphological features of IFN-γ–stimulated MSCs can be used to predict their overall immunosuppressive capacity.

Criteria for a standardized immunosuppressive capacity assay for MSC-based products have been suggested (5, 6), but a specific protocol for quantification of immunosuppressive capacity has not yet gained wide acceptance. Measuring the activation of T cells stimulated in the presence or absence of MSCs in vitro is a commonly used assay for assessing the effects of stimulatory and inhibitory factors on MSC-mediated immune suppression; however, this approach is rarely used to compare the inherent immunosuppressive capacity of MSC lines, and interexperimental variability can make it difficult to reliably distinguish between MSC lines with similar immunosuppressive capacities (5). Although T-cell proliferation is often used to measure activation in MSC immunosuppression studies, other measures of activation such as cytokine secretion and up-regulation of growth factor receptors like CD25 more directly assess activated T-cell effector functions (25, 26). In this study, we assessed cytokine secretion (IFN-γ and TNF-α) and CD25 expression in addition to proliferation in both CD4⁺ and CD8⁺ T cells activated in the presence of MSCs (Fig. 1). We found that considering only a single activation measurement to compare immunosuppressive capacity could lead to erroneous conclusions, as the magnitude of suppression differs between activation measurements, and suppression of some activation measurements could be observed at lower MSC concentrations than others (Fig. 1).

To avoid some of these limitations, we developed a method for quantifying the overall immunosuppressive capacity of MSC lines that is robust to interexperimental variability. For each experiment, we generated a composite T-cell activation variable from multiple activation measurements, allowing information regarding effector functions and proliferation to be incorporated into a single measure of global T-cell activation (Fig. 2 A and B). Similar to other studies, we observed that the magnitude of T-cell responses to stimulation can differ substantially between independent PBMC donors (27, 28), and when multiple measurements of T-cell activation are considered, each independent PBMC donor shows a unique profile of T-cell activation. By using PCA to produce a composite activation variable specific to the PBMC donor in a given experiment, our method accounts for these unique profiles by considering each activation measurement in an unbiased manner and incorporating those with the most variation into the first principal component (PC1; Fig. 2A). The presence of these unique activation profiles can be observed when data collected from two individual PBMC donors are used together to construct principal components, as PC1 explains much of the variation due to MSC-mediated immune suppression, whereas the second principal component separates the results by PBMC donor (Fig. S4). Using an unbiased composite measurement of T-cell activation rather than a single measurement reduces the impact of PBMC donor-specific variability that can impede interexperimental comparisons.

Fig. S4.

PCA of MSC-mediated immune suppression with two independent PBMC donors. In parallel cultures, PBMC from two donors (PBMC1 and PBMC2) were activated in the presence of five different concentrations of cells from six independent MSC lines (n = 5 replicates for each concentration). After 3 d, T-cell activation was measured by flow cytometry (CFSE dilution, cytokine secretion, and CD25 expression) in both CD4⁺ and CD8⁺ T cells. These data were then used to construct principal components as shown. Data from PBMC donor 1 are depicted as solid markers, and data from PBMC donor 2 are depicted as faded markers. Marker size indicates the concentration of MSCs (larger dot, more MSCs per well). In this experiment, PCA2 reflects a difference in the activation profile of donor 1 and donor 2, which is largely explained by a higher frequency of IFN-γ⁺ and TNF-α⁺ cells among activated CD8⁺ T cells in PBMC from donor 2.

Our method for quantifying the immunosuppressive capacity of MSC lines also integrates data from multiple concentrations of MSCs using an AUC analysis. Incorporating data from a dynamic range of immunosuppression for each MSC line provides a robust and repeatable measurement of immunosuppressive capacity for a given MSC line and avoids any dubious conclusions about the relative immunosuppressive capacity of MSC lines that could be made when only one or a few MSC concentrations are considered. For some MSC lines, we observed an enhancement of T-cell activation at low MSC concentrations even though higher concentrations of MSCs from the same line showed clear immunosuppression (Fig. 2C). This enhancement has been observed by others as well, and its incorporation into the AUC gives a broader measure of the immunosuppressive capacity of a given MSC line that may have biological relevance in some in vivo scenarios where MSCs present in small numbers could potentially increase rather than decrease inflammatory symptoms (27, 29). The AUC for a given MSC line differed between experiments in which independent PBMC donors were used due to donor-specific differences in T-cell susceptibility to MSC-mediated immune suppression (Fig. S2). However, although the absolute AUC value for a given MSC preparation differed when independent PBMC donors were used, the relative value of MSC lines in our cohort was generally consistent regardless of the PBMC source (Fig. S2). These results suggest that the presence of a “reference” immunosuppressive cell line in each experiment would be beneficial to make direct comparisons of MSC lines tested in different experiments (8). A similar strategy could also facilitate comparing experimental results from different laboratories by using results obtained for shared “cell reference materials” to normalize results between laboratories (5, 30). Although we limited our experiments to MSC-mediated suppression of T cells in this study, this multivariate PCA/AUC approach could be adapted for the study of MSC-mediated suppression of other types of immune cells, such as B cells or monocytes.

MSC-mediated immunosuppression is commonly measured using only proliferation in CD4⁺ T cells, with many studies measuring suppression at only a few MSC:T-cell ratios. We therefore chose to compare our comprehensive AUC approach to the more commonly used approach assessing only proliferation in CD4⁺ T cells (Fig. S5). Using data combined from multiple experiments, we correlated AUC values with percentage suppression calculated using only CFSE dilution among CD4⁺ T cells from these same experiments. Although CFSE dilution from some cell densities correlated moderately with AUC (R² = 0.58 for 2,000 MSCs per well, R² = 0.70 for 5,000 MSCs per well; R² = 0.61 for 10,000 MSCs per well), CFSE dilution alone was unable to account for all of the variation in immunosuppressive capacity observed by the AUC metric (Fig. S5). Therefore, although less comprehensive measures of MSC-mediated immunosuppression are appropriate for detecting qualitative differences in immunosuppressive capacity among MSC lines, studies using such methods will have reduced power to detect quantitative differences or differentiate between MSC lines with comparable immunosuppressive capacities.

Fig. S5.

MSC-mediated immunosuppressive capacity as measured by PCA/AUC analysis is only partially captured by measuring CD4⁺ T-cell proliferation. PBMCs were stimulated for 3 d in the presence of MSCs obtained from eight independent human donors at various cell densities as in Fig. 1. T-cell activation measurements (CFSE dilution, cytokine secretion, and CD25 expression) in CD4⁺ and CD8⁺ T cells were then used to construct principal components as in Fig. 2, and the first principal component was used to calculate an AUC value measuring the overall immunosuppressive capacity for each MSC line (as described in Fig. 2 and SI Materials and Methods). These AUC values were then plotted against the percentage suppression of CD4⁺ T-cell activity as calculated using only percentage CFSE dilution in CD4⁺ T cells—1 of the 16 variables used to construct the first principal component used to calculate the AUC. Lines of best fit and R² values were determined for each cell density individually. Data from three experiments using early-passage, late-passage, and IFN-γ–stimulated MSC lines are shown to demonstrate the full dynamic range of MSC-mediated immune suppression. Although suppression of CD4⁺ T-cell proliferation is correlated with AUC at several cell densities, the comprehensive AUC value captures variation that is missed when only this single activation measurement is considered.

Our AUC results demonstrate that immunosuppressive capacity decreases with passage for each MSC line and provides a comprehensive and quantitative description of that difference (Fig. 2D). We observed that even the least-immunosuppressive early-passage MSC line showed better suppression than the most-suppressive late-passage MSC line, and the combined cohort of early- and late-passage lines provided a large dynamic range of immunosuppressive capacity. The immunosuppression results collected in this study encompass data on MSC lines produced from three different commercial sources and includes donor-matched suppression results from both early- and late-passage cells as well as unstimulated and IFN-γ–treated conditions. This study contains a substantial amount of information from characterization of in vitro MSC-mediated immunosuppression, which may be important for future correlational studies looking for other predictive factors associated with immunosuppressive capacity, including gene expression and secretome analyses.

Concomitant with the challenges of effectively characterizing the overall immunosuppressive capacity of MSCs is the inability to identify cell characteristics that can effectively predict this behavior in vitro and in vivo. We previously demonstrated that early morphological features can serve as predictors of MSC mineralization capacity (12) and sought to extend this predictive characterization to immunosuppressive capacity by assessing MSC morphology in culture conditions relevant to MSC-mediated immune suppression. We chose to investigate the morphological response of MSCs to the cytokine IFN-γ as it is abundant at sites of inflammation, enhances the immunosuppressive function of MSCs, and is required for optimal MSC-mediated immunosuppression in some animal disease models (13, 31, 32). We found that, upon IFN-γ stimulation, all MSC lines adopted morphological phenotypes distinct from unstimulated MSCs, with IFN-γ–stimulated MSCs displaying increased elongation (increased aspect ratio/eccentricity), increased complexity (decrease in form factor and solidity), and lower nucleus/cytoplasm ratios, all within 24-h stimulation.

IFN-γ receptors signal in part through the JAK–STAT pathway, which mediates transcriptional changes through translocation of STAT proteins into the nucleus (33). Evidence of IFN-γ–stimulated morphological changes has been found in the case of monocytes where enhanced actin polymerization was observed in the presence of IFN-γ (34). In other cell types (HepG2 and HeLa), differences in cell adhesion signaling (drug and matrix-induced) that led to adoption of distinct morphologies were discovered to significantly impact IFN-γ signaling as actin cytoskeletal arrangement was directly linked to STAT1 dephosphorylation (35). This suggests that morphology not only serves as a phenotypic readout for IFN-γ treatment, but could also play a direct role in the magnitude of the MSC immunosuppression response as IFN-γ signaling promotes production of immunomodulatory factors such as indoleamine 2,3-dioxygenase, prostaglandin E₂, and IL-6 (20, 36, 37).

MSCs in coculture with PBMCs are exposed to T-cell–derived IFN-γ (as flow cytometry confirmed), and their ability to suppress T-cell activation was highly correlated with their morphology after 24 h of IFN-γ treatment. This correlation was demonstrated with six independent MSC lines at both early and late passages, and prediction models constructed with morphological data from these experiments were able to corroborate and accurately predict the immunosuppressive capacity of additional untested MSC lines from multiple donors (and a control fibroblast cell line). Although increase in cell size with extended culture has been widely observed for MSCs (12, 38–40), overall cell area alone was a poor predictor of immunosuppression (P > 0.11) and assessment of more subtle cell morphological features upon IFN-γ stimulation (perimeter, maximum feret diameter, and form factor) better predicted overall immunosuppression (Fig. 5). This further underscores the importance of designing functionally relevant morphological profiling tools based on the context of the assay, that is, extracting information about MSC immunosuppression by observing morphology in response to a stimulus (IFN-γ) that enhances the immunosuppressive function of MSCs.

Using data compiled from publications using the same MSC lines used in this study (12, 38, 41), we directly compared the potential predictive power of several commonly used MSC markers to that of the morphological features identified in this study. We found that only the morphological features (upon 24-h IFN-γ stimulation) significantly correlated with immunosuppressive capacity, whereas the expression of traditional MSC surface markers and trilineage differentiation capacity were not significantly correlated with our AUC results (Table S4). These results emphasize the importance of using functionally relevant assays to more effectively predict the immunosuppressive capacity of an MSC population. Other researchers could potentially use this morphological profiling approach to better characterize their cell lines and predict the immunosuppressive capacity of different cell lines as well as the effect of changes in manufacturing conditions (i.e., culturing conditions) on immunosuppressive capacity by quantifying morphological responses of cells to IFN-γ.

Table S4.

Morphological predictors are more strongly associated with immunosuppressive capacity than other commonly measured MSC attributes

Assay predictor	P value	Significant?	Assay timeline, d
Nucleus/cytoplasm10 ng/mL	1.36 x 10−8	Yes	2
Perimetercell,10 ng/mL	1.54 x 10−7	Yes	2
Nucleus/cytoplasm50 ng/mL	2.56 x 10−7	Yes	2
Maximum feret diametercell,50 ng/mL	9.59 x 10−7	Yes	2
Maximum feret diametercell,10 ng/mL	1.09 x 10−6	Yes	2
Perimetercell,50 ng/mL	2.05 x 10−6	Yes	2
ΔMinor axis lengthcell,10–0	2.99 x 10−5	Yes	2
Form factorcell,10 ng/mL	1.35 x 10−4	No	2
Alkaline phosphatase (12)	5.16 x 10−4	No	14
Form factorcell,50 ng/mL	8.32 x 10−4	No	2
MFI CD73 (38)	1.74 x 10−3	No	1
ΔMinor axis lengthcell,50–0	2.59 x 10−3	No	2
% CD105 (38)	3.66 x 10−3	No	1
Proliferation (38)	6.42 x 10−3	No	4
Adipogenesis (38)	6.65 x 10−3	No	14
MFI CD44 (38)	7.17 x 10−3	No	1
% CD73 (38)	1.90 x 10−2	No	1
MFI CD90 (38)	2.03 x 10−2	No	1
MFI CD105 (38)	2.53 x 10−2	No	1
MFI CD166 (38)	3.11 x 10−2	No	1
MFI CD29 (38)	3.54 x 10−2	No	1
% CD29 (38)	4.62 x 10−2	No	1
% CD166 (38)	7.43 x 10−2	No	1
% CD44 (38)	9.28 x 10−2	No	1
Chondrogenesis (41)	1.63 x 10−1	No	21
Mineralization (12)	1.85 x 10−1	No	35
% CD90 (38)	3.39 x 10−1	No	1

Immunosuppressive capacity as measured in this study (AUC) was assessed for association with surface marker expression and other common MSC functional bioassays using MSC lots derived from seven donors (PCBM1632, PCBM1641, PCBM1662, 110877, 127756, 167696, and 8F3560) at low and high passages. Associations with morphological predictors identified in this study are shown in boldface type, whereas associations with previously published results are shown in regular type (data obtained with permission from refs. 12, 38, 41). Linear regression was used to test all assay predictors for association with AUC except chondrogenesis and mineralization, where an ANOVA was used to compare average AUC values between performance based groups (i.e., high vs. low mineralization). An adjusted significance threshold of P < 8.7 × 10⁻⁵ was required for an association to be considered significant to control for the number of tests performed [P < 0.05/(558 morphological predictors + 17 common MSC assays) = 0.05/575 ∼ 8.7 × 10⁻⁵]. Assay timeline estimates the total amount of time necessary to complete each assay. Chondrogenic response (41) and median fluorescent intensity (MFI) data (38) were determined from previously published experiments.

MSCs expanded with different manufacturing processes did not demonstrate the same correlation as our original dataset (Fig. S3). However, unique morphological features that correlated with immunosuppression were identified for these MSC lines. Although we tested fewer MSC lines expanded by this second manufacturing process, this result may indicate that morphological characterization must be performed with each manufacturing process to establish useful criteria for identifying MSC preparations with desired immunosuppressive potential. Similarly, MSCs from other tissue sources (adipose, umbilical cord, dental pulp) may possess tissue-specific morphological phenotypes and responses that will need to be characterized to identify potential morphological quality attributes unique to these cell types (42). As this study tested only three MSC lines produced by this second manufacturing process, further study will be required to confirm that unique morphological profiles are correlated with immunosuppressive capacity in bone marrow-derived MSC lines produced by varying sources or manufacturing processes.

This work also demonstrated that changes in morphology were not only associated with inherent immunosuppressive capacity, but could also be used to determine the effects of IFN-γ pretreatment on MSC immunosuppression (Fig. 6). Using our AUC analysis, we observed consistent enhancement of immunosuppression for eight MSC lines when samples were stimulated with IFN-γ for 24 h before coculture with PBMCs. The presence of this conserved trend of enhanced immunosuppression could be predicted based on the overall difference in morphological signatures (Fig. 4, control vs. stimulated). Furthermore, the magnitude of this enhancement highly correlated with the differential features perimeter_nucleus and major axis length_nucleus (Fig. 6 G and H). Nuclear morphology relates to stemness and differentiation (43, 44), and there is direct coupling of the nucleus to the cytoskeleton through proteins such as Lamin A/C (44, 45). Evidence suggests that nuclear morphological changes associated with a biological function may be mediated by epigenetic modifications (46–48). The effect of IFN-γ stimulation on permissive chromatin (at the IDO promoter) associated with MSC immunosuppressive function was demonstrated recently with several of the MSC lines used in this study and may be related to the change in nuclear morphology associated with IFN-γ treatment (47). This further supports the notion that nuclear morphological data can be informative of MSC immunosuppressive capacity and treatment conditions that enhance this capacity. Again, as was the case with our osteogenic morphological analysis (12), larger differential values (i.e., a greater morphological response to IFN-γ) correlated with a reduced functional response to stimulation. Using this technique, a high-throughput assessment of stimulatory regimens could be performed by analyzing the morphological response of MSCs, which could facilitate discovery of optimal conditions for preconditioning MSCs for immunotherapies. This approach could potentially identify “personalized” stimulatory regimens for a given patient or disease indication and result in improved clinical outcomes using autologous or allogeneic MSC-based products.

Although derived from high-content, single-cell measurements, our morphological correlations with immunosuppression still represent population-based techniques, which require further investigation and refinement to better address inherent characterization difficulties associated with MSC heterogeneity. Preliminary evidence exists that demonstrates morphological subpopulations with distinct functional capacities (proliferation and differentiation), and continued investigations into the emergence of heterogeneity will facilitate improved characterization and understanding of cell therapy products (49–51). Improvements on single-cell morphological analysis using high-dimensional single-cell classification techniques such as SPADE or viSNE could allow for identification of morphologically distinct subpopulations of MSCs that could be enriched for immunotherapies (52). These approaches could also provide insight into the progression and establishment of MSC heterogeneity as it relates to differences in donor (or tissue) source and in manufacturing (40, 53). Advancements in live-cell tracking and monitoring in vitro would further allow for direct correlation of early morphological phenotypes with assay outcomes (15, 54).

Improved understanding of morphological signatures in the context of stem cell heterogeneity and specific biological functions will supplement current MSC characterization methods used in clinical applications involving immunomodulation. Morphology has been identified in many studies as a predictive marker of osteogenic capacity (15, 16); however, limited research exists that demonstrates a correlation of morphology with other MSC functions such as immunosuppression. Further improvement on the methods for morphological assessment and applicability across multiple donors is necessary to fully demonstrate its possible utility as a predictor of not only in vitro but also in vivo MSC immunosuppressive capacity. This work also provides further insight into the establishment of analytical approaches making use of an MSC internal “reference cell line” (8), as we were able to determine the relative immunosuppressive capacity of validation MSC lines based on normalization of data acquired from MSCs and PBMCs derived from different donors. Our development of predictive models of in vitro immunosuppression based on morphological data followed by successful application to new MSC lines lays the foundation for further study into identifying morphological signatures that can predict MSC potency.

Materials and Methods

Cell Lines and MSC Manufacturing/Expansion.

Human bone marrow-derived stem cells were obtained from six different donors purchased from Lonza (127756, 167696, 8F3560) and AllCells (PCBM1632, PCBM1641, PCBM1662) at passage 2 (Table S1 for donor specifications). Additional cell lines used for validation are described in Tables S2 and S3. MSC culture conditions were chosen based on well-established protocols (55). Most of the MSC lines used in this study have been extensively characterized and further information about the donors’ surface marker expression, genomic, epigenetic, and proteomic profiles, as well as performance in multiple bioassays for MSC function has been published previously (12, 38, 47, 48, 56–58). Detailed MSC culturing and expansion methods are described in SI Materials and Methods. All cell lines used in this work possessed viability >95% (assessed using Trypan Blue) before plating for immunosuppression coculture and morphological assessment.

Quantitative Assessment of MSC Immunosuppressive Capacity.

PBMCs from healthy human donors (10⁵ per well) were stimulated with an equal number of T-cell–activating beads (Life Technologies; 11131D) in the presence of five concentrations of each MSC line (500, 1,000, 2,000, 5,000, and 10,000 cells per well) in 96-well plates. After 3 d, PBMCs were harvested and activation was assessed in both CD4⁺ and CD8⁺ T cells by proliferation (CFSE dilution), CD25 expression, and production of IFN-γ and TNF-α. PCA was used to generate a composite T-cell activation metric (PC1) using all available flow cytometry data for a given experiment. Transformed PC1 values were then plotted over MSC concentration for each MSC line, and the AUC calculated from these plots was used to quantify the immunosuppressive capacity of each MSC line. AUC values are inversely proportional to immunosuppressive capacity, and all AUC values were transformed for ease of presentation such that a hypothetical MSC line that showed complete immune suppression at all concentrations would have an AUC of 0, and an MSC line with no immunosuppressive capacity would have an AUC value of 100. More detailed descriptions of this experimental and analytical approach are found in SI Materials and Methods and Tables S5–S7. All MSC and PBMC samples were purchased from commercial sources with appropriate informed consent as indicated on the certificates of analysis. Research on commercially obtained human cell lines is exempt from FDA IRB (Research Involving Human Subjects Committee) oversight.

Determination of an Overall MSC Morphological Signature.

For MSC morphological analysis, cells were seeded at a density of 1,000 cells per well (four total wells per experimental group) in 12-well plates (Corning) and cultured for 24 h in growth medium. Growth medium was replaced with growth medium containing 0, 10, or 50 ng/mL IFN-γ (Life Technologies) and cultured for an additional 24 h and then fixed with 4% (wt/vol) paraformaldehyde for 15 min. Cell and nuclear morphology were assessed using fluorescein (FITC)-maleimide (Life Technologies) and Hoechst (Sigma-Aldrich), respectively. Briefly, fixed samples were incubated with 20 µM FITC-maleimide for 30 min, washed with PBS, incubated with 1 mg/mL Hoechst for 5 min, and washed with PBS before imaging. Samples for morphological analysis were imaged using an inverted Nikon Ti-U microscope with automated stage (Prior) and filters (Chroma Technology). At least 1,000 cells were assessed for each experimental group, with approximately equal numbers of cells analyzed from each of four replicate wells. Automated quantification of cellular and nuclear shape features was performed using CellProfiler (59) to obtain quantitative morphological data for each cell consisting of 46 cellular shape features and 46 nuclear shape features (Tables S8 and S9). Overall morphological signatures were constructed for each group by taking the median value of the 93 total cellular and nuclear features in each IFN-γ condition for a total of 279 morphological features. More detailed descriptions of this experimental and analytical approach are found in SI Materials and Methods and Table S6.