Single‐cell high‐content imaging parameters predict functional phenotype of cultured human bone marrow stromal stem cells

Abstract Cultured human bone marrow stromal (mesenchymal) stem cells (hBM‐MSCs) are heterogenous cell populations exhibiting variable biological properties. Quantitative high‐content imaging technology allows identification of morphological markers at a single cell resolution that are determinant for cellular functions. We determined the morphological characteristics of cultured primary hBM‐MSCs and examined their predictive value for hBM‐MSC functionality. BM‐MSCs were isolated from 56 donors and characterized for their proliferative and differentiation potential. We correlated these data with cellular and nuclear morphological features determined by Operetta; a high‐content imaging system. Cell area, cell geometry, and nucleus geometry of cultured hBM‐MSCs exhibited significant correlation with expression of hBM‐MSC membrane markers: ALP, CD146, and CD271. Proliferation capacity correlated negatively with cell and nucleus area and positively with cytoskeleton texture features. In addition, in vitro differentiation to osteoblasts as well as in vivo heterotopic bone formation was associated with decreased ratio of nucleus width to length. Multivariable analysis applying a stability selection procedure identified nuclear geometry and texture as predictors for hBM‐MSCs differentiation potential to osteoblasts or adipocytes. Our data demonstrate that by employing a limited number of cell morphological characteristics, it is possible to predict the functional phenotype of cultured hBM‐MSCs and thus can be used as a screening test for “quality” of hBM‐MSCs prior their use in clinical protocols.


| INTRODUCTION
Human bone marrow-derived stromal (mesenchymal) stem cells (hBM-MSCs) are non-hematopoietic, self-renewing, plastic adherent cells with the ability to differentiate into several mesodermal lineages including osteoblasts (OB) and adipocytes (AD). 1 The efficacy of hBM-MSCs transplantation to enhance skeletal and non-skeletal tissue regeneration, for example, following bone fracture, cartilage injury, as well as cardiovascular and immune diseases, is being examined in a large number of clinical trials. The rationale is that hBM-MSCs can differentiate into functional bone forming osteoblastic cells (skeletal applications) or produce a large number of cytokines and growth factors (so-called cytokine factory) that improve tissue regeneration (non-skeletal applications). [2][3][4] Although MSCs-like cells have been isolated from different tissues such as adipose tissues, placenta, and Wharton's jelly of umbilical cord, bone marrow is considered the standard source for MSCs with in vitro and in vivo bone-forming ability 1,5,6 and is the basis for their clinical use for enhancing bone tissue formation.
The reported clinical outcome of hBM-MSC-based therapies for treating non-healed bone fractures or bone defects has revealed inconsistent results with respect to the efficacy of in vivo bone formation, which may be explained by variations in the phenotype of the transplanted cells. 1,7 In vitro cultured hBM-MSCs exhibit cellular heterogeneity with respect to their potential for osteoblast differentiation and bone formation. [8][9][10] Furthermore, cultured hBM-MSCs demonstrate inter donor variations related to donor age, sex, or disease state. [11][12][13] Comparing the outcome of clinical trials for bone regeneration that has used hBM-MSCs is often hampered by the lack of "common" cellular biomarkers that define the "quality" of transplanted cells. Identification of biomarkers that facilitate the selection of clinically relevant hBM-MSC populations is needed in order to improve the outcome and consistency of hBM-MSC-based therapies. Traditionally, hBM-MSCs have been defined by a limited number of CD markers, for example, CD44, CD73, CD90. 1,5,6 These markers are sensitive but not predictive of the differentiation capacity of the cells. 9,14 Several studies have identified a molecular signature for bone-forming hBM-MSCs by applying global analysis of gene expression, 9,15,16 miRNA expression, 17,18 or proteome analysis. 19,20 Although these approaches contribute significantly to understanding the biology and functions of hBM-MSCs, they are labor-intensive, use a large number of cells, and may not be easy to implement in a clinical setting.
Changes of cell morphology have been observed during cell proliferation 21,22 and differentiation 23,24 and thus are determinant of the biological functions of the cells. The relationship between cell morphology and biological functions have been studied in respect to changes in cytoskeletal fibers (actin and tubulin) that mediate cellular adaptation to microenvironmental stimuli and facilitate intracellular signal transduction. For example, the actin cytoskeleton plays a role in hBM-MSC lineage commitment, 25,26 and microtubules are key players in cell proliferation 27 and have been reported to contribute to osteoblast differentiation. 28 Alterations in cell shape affects actin stress fibers, which through intracellular signaling pathways initiate cell lineage commitment. 23,29 Moreover, nucleus shape can be modulated by cytoskeletal fibers in response to extracellular forces that affect cellular differentiation. 30 Liu et al showed that culturing cells on surface with different topography could affect nuclear morphology and cell differentiation in rat BM-MSCs. Cells cultured on high micropillars exhibited altered nucleus shape that favored osteoblastic differentiation, whereas cells grown on smooth or on low micropillars exhibited less deformed nuclei which were associated with adipocyte differentiation. 24 The availability of quantitative high-content imaging technologies at a single-cell resolution has allowed studies on the role of cell morphology as a predictor of cell differentiation potency. [31][32][33] However, these studies did not investigate the morphology of

| Donors and materials
Bone marrow aspirates were collected from lower extremities of 56 adult donors: 26 males (age 18-81 years) and 30 females (26-97 years), undergoing surgeries at the Department of Orthopedics and Traumatology, Odense University Hospital. The samples of bone marrow were considered as "waste material" from routine operations, and, therefore, the material collection was without extra risk for patients. All subjects received oral and written information and signed a consent. The project was approved by Scientific Ethics Committee of Southern Denmark (project ID: S-20160084).
hBM-MSCs were cultured from mononuclear cell population following gradient centrifugation using lymphoprep of the bone marrow, through plastic adherence as described previously. 13 The cells were cultured in MEM media supplemented with 10% fetal bovine serum (FBS; the same lot number 42F0266K was used for all samples) and 1% penicillin/streptomycin (P/S), in standard culture conditions (37 C in humidified 5% CO 2 incubator). After a week, when the first adherent cells were visible, media was switched to MEM media supplemented with 10% FBS, 1% P/S, 1% GlutaMAX, 1% sodium

Significance statement
Clinical trials employing cultured human bone marrow mesenchymal stem cells (hBM-MSCs) demonstrate the presence of a large inter-donor variability in biological functions and clinical efficacy. By applying high-content imaging methodology to cultured BM-MSCs obtained from a large cohort of donors, morphological features including cell area and nucleus geometry as predictors for cell differentiation and proliferation were identified. These parameters can serve as selection criteria for BM-MSC populations to be used in clinical trials of enhancing bone regeneration. pyruvate, and 1% nonessential amino acids (S-MEM growing medium).
When the cells reached around 80% of confluence, they were trypsinized and used for analysis and for further cell expansion.

| Study workflow
The study workflow is illustrated in Figure S1. The cells were trypsinized (1) and prepared for osteoblastic and adipogenic differentiation, cell proliferation, cell membrane marker expression, and morphology analysis.
Cell morphology was studied on whole populations of primary hBM-MSCs obtained from each donor and at single-cell resolution (2). After culturing hBM-MSCs in standard medium for 48 hours, the cells were fixed, stained for cytoskeletal fibers, and visualized using automated fluorescence microscope (Operetta HCS) (Perkin Elmer). After nuclear and cellular segmentation (3), the basic cell and nucleus morphology parameters as well as cytoskeletal texture patterns were analyzed (4). The table in Figure 1 lists cell morphology features observed in the representative photomicrograph (5) and provides mean values and coefficient of F I G U R E 1 Variations in cell and nucleus morphology of cultured hBM-MSCs. A, Representative photomicrographs of cultured hBM-MSCs with contrasting cell morphology. Left photomicrograph: cultured cells from donor #1 exhibit more fusiform and smaller size compared with donor #2 cells that are large cuboidal cells (right photomicrograph). Scale bar is 200 μm. B, Representative photomicrographs illustrating heterogeneity of nucleus morphology. Donor #1 (left photomicrograph) cultured cells exhibit more oval shaped nuclei compared with donor #2 cells that exhibit more rounded shaped nuclei (right photomicrograph). Scale bar is 200 μm. C, Quantitative frequency distribution of cell areas of cultured hBM-MSCs derived from donor #1 and donor #2. D, Quantitative frequency distribution of nucleus shapes expressed as nucleus width to length ratio of cultured hBM-MSCs derived from donor #1 and donor #2. Quantitative frequency distribution of mean cell areas (E) and nucleus shape (F) of cultured hBM-MSCs derived from all donors (n = 56). hBM-MSCs, Human bone marrow-derived stromal (mesenchymal) stem cells variations and standard deviation (SD) of each parameter. All the quantified cellular and nuclear morphological features were correlated with corresponding in vitro assays described in the following paragraphs.

| Cell proliferation
Short-term proliferation was performed in cultured hBM-MSCs at first passage. The cells were seeded in a 6-well plate (1,000 cells per well) in triplicates and cultured under standard conditions. At days 1, 3, 6, 9, 12, and 15, the cells were trypsinized and counted in a hemocytometer.
The proliferation ability of the cells from each donor was quantified as the area under the curve (AUC) and expressed as arbitrary units (AU). In addition, we calculated population doubling time (PDT) in hours between days 1 and 6 using the following formula PDT = 120 hours×log(2)/(log  10 minutes with rotation in room temperature (RT). After that, the cells were washed with PBS for several minutes to remove nonspecific binding. Alizarin red staining was quantified as red intensity using ImageJ software and expressed as AU.

| Alkaline phosphatase activity
The cells were washed, fixed with 3.7% formaldehyde-90% ethanol, and incubated with p-nitrophenyl phosphate (1 mg/mL) in 50 mM NaHCO 3 and 1 mM MgCl 2 , pH 9.6 for 20 minutes at 37 C. The reaction was stopped by adding 3 M NaOH. Absorbance was measured at 405 nm and corrected per cell number. The cell number was determined by CellTiter-Blue reagent (cell viability reagent) for 1 hour at 37 C. The fluorescent intensity of the reagent was measured in FLUOstar Omega plate reader (560ex/590em).

| Adipocytic differentiation
Human BM-MSCs at the first passage were plated at a density of 30,000 cells/cm 2 in a 4-well plate for 24 hours. At near full confluence, the media were replaced with adipocytic induction media containing DMEM supplemented with 10% FBS, 1% P/S, 5% horse serum, 1 μM BRL 49653, 3 μg/mL insulin, 100 nM dexamethasone, and 225 μM IBMX. Media were changed every 2-3 days for 2 weeks.

| Image analysis
Images of the hBM-MSCs cultured from each individual donor (15 different areas per well in 9 wells) were acquired using the Operetta high-content screening system (PerkinElmer, objective ×10 magnification, N.A 0.3). Cell and nucleus morphology parameters including area, roundness, width, length, and ratio of width to length were analyzed using Harmony 3.1 software by employing a predefined protocol based on the building blocks method, which includes nuclear and cellular segmentation, as shown in Figure S1. In addition to cell morphol-  Table S1.

| Bivariant analysis
Statistical analyses of the correlation between variables were performed using the Spearman two-tailed correlation test (r s = Spearman correlation coefficient). For the correlation analysis, outliers were identified and removed using the ROUT method, which detects outliers from nonlinear regression, based on the maximum false discovery rate Q = 1%. The number of independent donors (n) in each correlation analysis is described in the Results section and in each figure. Differences between groups were tested by unpaired two-tailed Student's ttest. All analyses were performed using GraphPad Prism 7.1 software.
Statistical significance was considered when P < .05.

| Multivariable analysis
A stable multivariable linear regression model was created for predicting the osteogenic and log transformed adipogenic differentiation outcome and performed for the whole data set of 56 donors. Due to the large number of potential morphological predictors, we applied the variable selection procedure introduced by Meinshausen. 34 This procedure allows to control the expected number (PFER) of selected variables that represent uninformative predictors, that is, one controls the number of false discoveries. The central building block of this procedure is a regression modeling approach that allows optimal selection of a predetermined number, q, of explanatory variables. In this case, the penalized regression "Lasso" method 35 Figure 2C). Interestingly, we observed that the hBM-MSCs that are ALP+, CD146+, and CD271+ exhibited significant negative correlations with nucleus width to length ratio: ALP+ (r s = −0.28, P < .05, n = 56), CD146+ (r s = −0.5, P < .0001, n = 56), and CD271+ (r s = −0.36, P < .01, n = 53; Figure 2D).

| Relationship of cell morphology and biological functions of hBM-MSCs
We examined the correlation between morphology of hBM-MSCs In contrast to cell proliferation findings, adipocyte differentiation capacity (estimated by the lipid droplets area of mature AD) or OB differentiation (based on mineralized matrix formation) did not correlate with cell area (P = .78 and P = .48, respectively, Figure 3B, C) or cell shape (P = .36 and P = .29, Figure 3E, F). Similar findings were obtained when cell morphology was determined from F-actin staining ( Figure S3).

| Relationship of cytoskeletal texture and biological functions of hBM-MSCs
High-content image analysis provides information regarding cellular texture parameters that reflect the cytoskeletal status of the cells.  suggesting that cultures enriched in cells with smaller nuclear area exhibited higher cell proliferation, whereas cells with larger nuclei were more prone to osteoblast differentiation. We did not detect significant correlation between nucleus area and ALP activity ( Figure 5C) or between nucleus area and adipocyte differentiation ( Figure 5D). Interestingly, nucleus shape (width to length ratio) was negatively correlated with osteoblast differentiation and demonstrated that cultures enriched in cells with oval-shaped nuclei exhibited enhanced osteoblastic differentiation (r s = −0.48, P < .001, n = 56) ( Figure 5F). Furthermore, a negative correlation between ALP activity and nuclear geometry was observed (r s = −0.44, P < .001, n = 55) ( Figure 5G). We did not detect significant correlation between nucleus shape and cell proliferation or adipocyte differentiation ( Figure 5E, H).

| Relationship between nucleus morphology and biological functions of hBM-MSCs
To further analyze the importance of nucleus geometry, we classified all donors based on the ability of their cells to differentiate into either OB or AD. We compared population mean values of nucleus shape between the two groups with contrasting differentiation potential: high OB group (high OB, low AD) and high AD group (high AD, low OB). The photomicrographs in Figure 6 illustrate nucleus shape of representative populations of hBM-MSCs classified as high OB ( Figure 6A) or high AD ( Figure 6B) groups.
As illustrated in Figure 6C, OB differentiation is enhanced in cultures enriched in cells with oval-shaped nuclei, whereas enhanced AD differentiation was observed in cultures enriched in more roundshaped nuclei. In addition, we calculated the value of variable prediction as determined by the estimated AUC = 0.84; 0.68 to 1.00 (97.5% confidence interval). This suggests that the predictive power is modest on average but significantly higher than expected by random selection (AUC = 0.5 for random data). To corroborate the relevance of our findings, we examined changes in nuclear morphology in two clonal hBM-MSC populations previously defined in our laboratory as exhibiting enhanced or reduced capacity for in vivo bone formation. 9 As seen in Figure 6D, hBM-MSCs with high in vivo bone forming capacity were enriched in oval-shaped nuclei.

| Multivariable analysis to identify predictors of hBM-MSC differentiation potential
To identify the morphological predictors for hBM-MSC differentiation potential, we performed a multivariable analysis that included all measured morphological and texture features of cells and nuclei. Using a stability selection procedure, we identified parameters with explanatory power of the primary outcome of osteoblastic ( Figure 7A) and F I G U R E 4 Relationship between the tubulin texture and hBM-MSC functions. The cytoskeletal fiber texture was determined following tubulin staining. The ridge pattern of cytoskeletal texture exhibited significant positive correlation with the proliferation activity (A) and the percentage of CD271+ cells (B). Each dot represents the average value of cultured cells derived from a single donor and n indicates the number of donors tested. The proliferation was quantified as area under the curve (AUC). hBM-MSCs, Human bone marrow-derived stromal (mesenchymal) stem cells adipocytic ( Figure 7B) differentiation. We found that nucleus geometry is the morphological feature that is highly stable predictor for OB differentiation potential of the cells (θ > 0.75) and negatively associated with osteoblast differentiation.
For adipocyte differentiation, the nuclear "Hole" texture pattern was selected as a highly predictive variable (θ > 0.75) with negative influence on adipocytic differentiation potential. However, "Hole" texture pattern, did not reach statistical significance in bivariant analysis, ( Figure S4). Moreover, two cell morphology features were identified as variables of moderate explanatory power (θ > 0.5) associated with AD formation namely nucleus width to length ratio (positively associated) and cell width to length ratio (negatively associated).
The relative performance of the model was measured by AIC that determines the predictive power of the model when applying it to a new data set. We performed the analysis comparing various models (m0, m1, m2) for osteogenic and adipogenic outcomes including different combination of covariates. For osteogenic outcome, models were as follow: m0 (intercept only, prediction only based on the mean value), nucleus geometry (m1) and combination of identified top four valuables including nucleus geometry, cell geometry (α-tubulin), nucleus length, and cell texture Spot (m2). Similarly, for adipogenic outcome, we included: m0 (intercept only), nucleus texture Hole (m1), and top four identified variables: nucleus texture Hole, nucleus geometry, cell geometry (actin), and cell texture Spot (m2). Our data demonstrated (Table S2) that by adding the F I G U R E 5 Relationship between nucleus morphology and hBM-MSC functions. Nucleus area was significantly correlated with (A) proliferation ability of hBM-MSCs and (B) mature osteoblast formation evidenced by alizarin red staining of mineralized matrix. C, ALP activity following osteoblastic differentiation induction or (D) adipocytic differentiation potential did not exhibit significant correlation with nucleus size. Nucleus geometry (ratio of nucleus width to length) showed a significant negative correlation with mature osteoblast formation (F) and with ALP activity (G). No significant correlation was observed between nucleus geometry and other cell functions including (E) proliferation and (H) mature adipocyte formation. ALP, alkaline phosphatase; hBM-MSCs, Human bone marrow-derived stromal (mesenchymal) stem cells highly predictive variable (θ > 0.75) identified by the multivariable analysis, the prediction (m1) of both osteogenic and adipogenic differentiation outcomes was improved. For osteogenic outcome, including additional variables did not improve the prediction (m2), in contrast to the adipogenic outcome, where including the top four variables gave the best prediction model (m2).

| DISCUSSION
Cultured hBM-MSCs represent a heterogenous population of cells with respect to their in vivo differentiation potential which may lead to inconsistent and variable results when being used in clinical trials. 5 The aim of this study was to identify cellular morphological characteristics as "cell F I G U R E 6 Relationship between nucleus geometry and hBM-MSC functions. Photomicrographs illustrating nuclear geometry of (A) cultured hBM-MSCs with enhanced osteoblastic differentiation and decreased adipocyte differentiation (high OB) and (B) a population with enhanced adipocytic differentiation and decreased osteoblast differentiation (high AD). Arrows show examples of round nuclei (high width to length ratio) or oval nuclei (low width to length ratio). C, Cultures of hBM-MSCs established from independent donors were grouped based on their differentiation capacities as either high OB or high AD and compared with respect to mean nucleus geometry. D, The nucleus geometry of two clonal populations of hBM-MSCs with either high or low in vivo bone formation potential. 9 Nuclei were stained with DAPI. Data are shown as mean ± SD with **P < .01; ****P < .0001. hBM-MSCs, Human bone marrow-derived stromal (mesenchymal) stem cells  21 The relationship between small cell size and proliferative potential has also been reported in other cell types. 52,53 It is plausible that cell size and nuclear size can be used as surrogate measures of replicative potential of cultures.
Osteoblast differentiation is the default differentiation pathway of cultured hBM-MSCs 54 and is the basis for their use in treatment of bone injuries. An inverse relationship has been observed between hBM-MSC capacity for differentiation into osteoblastic versus adipocytic cells. 55 In our study, the capacity for differentiation into AD was considered an "undesired" outcome of hBM-MSC differentiation. Interestingly, in multivariate analysis, nuclear geometry was the These changes could either be used to alter the differentiation outcome or predict the BM-MSC biology. 23,29,32,[56][57][58] The novelty of the current study lays in the use of naive hBM-MSC morphology.
The association of cellular shape and functional outcome may be explained by changes in actin dynamics. We have previously demonstrated that changes in cell shape induced by alternations in the actin cytoskeleton structure determine differentiation outcome of hBM-MSCs with cellular changes associated with actin de-polymerization led to enhanced adipocytic differentiation, whereas inhibition of actin depolymerization increased osteoblast formation of hBM-MSCs. 25,26 Furthermore, one study also shows that stimulation of a immunomodulating pathway can lead to a morphological response in BM-MSCs. 57,58 Thus, culturing hBM-MSCs on biomaterials with specific microstructure 59,60 or stimulation of certain pathways 57 could be used to obtained clinically relevant cell population with high OB differentiation potential or immunomodulatory properties suitable for clinical transplantation.
Multivariable analysis identified nucleus geometry as the most stable predictor factor of the differentiation capacity of the cells. The cell nucleus has been proposed to function as a mechano-sensor, 30,61 where cytoskeleton and nucleoskeleton linkers transmit extracellular and cytoplasmic forces that alter nuclear shape and thus affecting chromatin organization and transcriptional activity 30,62 as well as cellular differentiation. 63 In support of this notion, a recent study demonstrated that forcing hBM-MSCs to alter nuclear geometry by culturing on a micropatterned surface was associated with enhanced histone 3 acetylation, 64 a factor that is associated with increased osteoblast differentiation. 65 In addition, changes in nucleus geometry have been reported in rat MSCs to be associated with increase of osteogenic and decrease of adipogenic gene-marker expression. 24 Our study has some limitations. We used in vitro differentiation assay as a surrogate marker for the in vivo bone-forming capacity. Several markers of in vitro differentiation to OB are not predictive for bone forming capacity of the cells. 9 However, to corroborate our findings, we demonstrated that in vivo bone forming capacity is positively associated with the presence of elongated nuclei (low width to length ratio) based on data on a hBM-MSC cell line with known high heterotopic bone forming capacity ( Figure 6D). Also, standard culture conditions were used for expansion and differentiation of hBM-MSCs. Confirmation studies are needed to test clinical grade hBM-MSCs cultured under GMP conditions.
Routine clinical use of hBM-MSCs in therapy requires development of easy and noninvasive assays for determining "cell quality." We have previously demonstrated that Raman spectroscopy can be used to confirm the normal, nontransformed phenotype of hBM-MSCs prior to clinical transplantation. 66 In the current study, we demonstrate that using a limited number of morphological characteristics, it is possible to predict the proliferative capacity and the differentiation potential of hBM-MSCs. The clinical efficacy of using these criteria as quality parameters for transplanted hBM-MSCs remains to be determined in prospective clinical studies.

ACKNOWLEDGMENTS
The study was carried out as a part of Bonebank project (project number: 16-1.0-15) supported by Interreg 5a Germany-Denmark with funds from the European Regional Development. We would like also to thank The NovoNordisk foundation NNF 180C0054118 and the Lundbeck foundation R266-2017-4250 for funding. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.