Batch Effects during Human Bone Marrow Stromal Cell Propagation Prevail Donor Variation and Culture Duration: Impact on Genotype, Phenotype and Function

Donor variation is a prominent critical issue limiting the applicability of cell-based therapies. We hypothesized that batch effects during propagation of bone marrow stromal cells (BMSCs) in human platelet lysate (hPL), replacing fetal bovine serum (FBS), can affect phenotypic and functional variability. We therefore investigated the impact of donor variation, hPL- vs. FBS-driven propagation and exhaustive proliferation, on BMSC epigenome, transcriptome, phenotype, coagulation risk and osteochondral regenerative function. Notably, propagation in hPL significantly increased BMSC proliferation, created significantly different gene expression trajectories and distinct surface marker signatures, already after just one passage. We confirmed significantly declining proliferative potential in FBS-expanded BMSC after proliferative challenge. Flow cytometry verified the canonical fibroblastic phenotype in culture-expanded BMSCs. We observed limited effects on DNA methylation, preferentially in FBS-driven cultures, irrespective of culture duration. The clotting risk increased over culture time. Moreover, expansion in xenogenic serum resulted in significant loss of function during 3D cartilage disk formation and significantly increased clotting risk. Superior chondrogenic function under hPL-conditions was maintained over culture. The platelet blood group and isoagglutinins had minor impact on BMSC function. These data demonstrate pronounced batch effects on BMSC transcriptome, phenotype and function due to serum factors, partly outcompeting donor variation after just one culture passage.


Introduction
Important successes in skeletal regenerative medicine must not obscure the fact that a panacea for nonunion fractures has not yet been found. Durable cartilage regeneration also remains elusive [1]. Sophisticated protocols for skeletal stem-cell activation will help in this regard [2]. BMSCs, representing the prototype mesenchymal stromal cells, are a versatile source of cells for skeletal regeneration [3]. However, culture conditions to direct reproducible bone and cartilage differentiation of BMSC are still not fully defined [4].
Organoid-like 3D BMSC differentiation models may provide more realistic readouts for studying osteochondral differentiation [5,6]. We have previously shown that cell culture-expanded BMSCs, but not stromal cells from white adipose tissue or umbilical cord, preserve their potential to create 3D organoid-like cartilage templates in vitro and fully re-establish bone and the hematopoietic marrow niche in vivo, in the presence of human platelet lysate (hPL) [7,8]. BMSC-derived scaffold-free cartilage organoid-like disk transplants could completely regenerate critically sized femur defects in a humanized mouse model. The skeletal regeneration competence was found to be predetermined by a discerning epigenetic landscape enabling common transcription factors to act on genes involved in ossification [9].
Batch effects in omics data are considered to occur when differences in experiment time, handling or reagent lots, to name some, result in a different outcome of the analysis [10]. In cell therapy, donor variation is generally considered to play an important role [11][12][13], but batch effects are barely considered. Pleiotropic raw materials for cell manufacture, such as animal-derived or human serum and undisclosed proprietary components of defined serum-free media, comprise a risk of deviating cell quality and quantity. More importantly, serum lot, source, or other unknown variables, can precipitate batch effects changing gene expression or any other outcome of interest in cell products [14].
Application of hPL as an efficient substitute for fetal bovine serum (FBS) in stromal cell culture has become common practice since the 2000s, following recommendations of the European Medicines Agency to avoid animal-derived raw materials for manufacturing cell therapeutics [15]. Notably, hPL production from so-called buffy-coats, derived from whole-blood donations as a side product of clinical red blood cell unit production, or from platelet-rich plasma collected by single-donor platelet apheresis, is highly divergent among producers [16] and efforts for standardization are still ongoing [17]. Besides the number of pooled donations and the mode of platelet processing influencing hPL quality, ABO blood group antigens of platelets and corresponding isoagglutinins in donor plasma may also be critical for cultured cells [18]. Therefore, we have compared the effect of hPL pooled either from fresh platelet concentrates of blood group O platelets (without AB antigens) and AB plasma (without isoagglutinins, hPL O/AB), with hPL pooled from expired platelet concentrates of mixed ABO blood groups (hPL mBG), and FBS, on stromal cell phenotype, genotype and function, respectively.
The intravenous application of stromal cells also bears the risk of an instant bloodmediated inflammatory reaction (IBMIR) leading to thromboembolic side effects during cell therapy [19]. Tissue factor cell surface expression seems to be a key factor, activating coagulation factor VII and finally inducing thrombin formation [20,21]. Previously, by rotational thromboelastometry (ROTEM), we found that BMSCs were less procoagulant in human plasma than stromal cells derived from white adipose tissue or umbilical cord [22]. After intravenous application, tissue factor high-expressing cord-derived stromal cells showed excessive pulmonary thromboembolism in a rat model. Such embolism risk was not observed after infusion of BMSCs with low tissue factor expression, confirming our in vitro data [22].
The aim of this study was to test the hypothesis that batch effects due to cell culture raw material precipitate culture-related trajectories in BMSC gene expression, phenotype or function. Here we tested the influence of hPL vs. FBS on BMSC transcriptomics, cell surface proteomics and function. Organoid-like cartilage disk formation was tested as a surrogate for osteo-chondral regenerative competence. The procoagulant activity of short vs. long-term cultured BMSCs, under the aegis of FBS vs. two types of hPL, respectively, was analyzed in addition by ROTEM as a marker of cell therapy hemocompatibility.

Bone Marrow Samples and Cell Culture
For human bone marrow mononuclear cell isolation, approval was obtained from the Institutional Review Board of the Medical University of Graz (protocol . Samples were collected in accordance with the Declaration of Helsinki after written informed consent from healthy volunteers. Cells were cultured in alpha-modified minimum essentials Eagle's medium, α-MEM (Sigma-Aldrich, St. Louis, MO, USA) supplemented with 5 mM N(2)-L-alanyl-L-glutamin (Dipeptiven, Fresenius Kabi, Austria), 100 U/mL penicillin and 0.1 mg/mL streptomycin (both Sigma-Aldrich). We supplemented this BMSC expansion medium additionally with 10% of either pooled hPL O/AB [23] or pooled hPL mixed blood group (mBG) in the presence of 2 IU/mL preservative-free heparin (Biochrom, Germany) [18] as indicated. An optimized concentration of 16.5% heat inactivated FBS (FBS Superior, Sigma-Aldrich) was used.
Bone marrow aspirations were divided 50/50 and stromal cells were isolated and expanded either under animal serum-free conditions with BMSC expansion media containing hPL O/AB or with media containing FBS as previously described [24] (Figure 1). We cryo-preserved BMSC passage 0 (P • , at the end of primary culture) for later analysis. For large-scale expansion, we seeded the cells directly into in four-layered cell factories (CF-4) in BMSC expansion medium supplemented with either hPL O/AB or FBS upon thawing. Expanded BMSCs from passage 1 (P1) were either analyzed or further cultured in these media until passage 4-5 (P4-5). An aliquot of BMSCs cultured in P • in hPL O/AB was transferred to BMSC expansion medium containing hPL mBG and also cultured until P4. BMSCs from P1 and P4 were further analyzed as indicated. In selected experiments, due to different cell growth of the different samples we used samples P4 +/− one passage. BMSC population doublings (PD) for all passages were calculated using the formula LN (fold increase cell number)/LN (2). Theoretical cumulative cell numbers were calculated, by assuming that all cells were further expanded, by multiplying the totally harvested cells of P1 with the fold-increase of the following passages P2-3 in smaller culture vessels. To quantify cell morphology, cells from donors one and two were seeded at a density of 10,000 cells/cm 2 overnight, stained with Calcein (LIVE/DEAD™ Viability/Cytotoxicity Kit Thermo Fisher Scientific, Waltham, MA, USA) for 20 min. Images were taken at 10x magnification as four times four fields of view with an Eclipse Ti inverted microscope (Nikon, Tokyo, Japan) equipped with an automated stage (Märzhäuser Wetzlar, Wetzlar, Germany) in phase contrast and GFP channel. Individual GFP + cells were detected after thresholding using NIS AR software. Object data containing area and length were extracted for 1208-1428 individual cells and used for statistical analysis.

RNA-Seq, MethylCap-Seq and Bioinformatics
For both RNA sequencing (RNA-seq) and Methyl-CpG binding domain-based capture and sequencing (MethylCap-seq), libraries were sequenced on a Hiseq 4000 (Illumina, San Diego, CA, USA) with 50 bp single-end reads. We conducted quality control using FASTQC and removed residual adapter sequences and low-quality reads using Trimmomatic. We mapped reads to the Ensembl GRCh38 human genome using Bowtie 2 as described previously [9]. We submitted raw reads to the GEO database (NUMBER: GSE194303).
To analyze RNA-seq data, we calculated the number of mapped reads/gene counts using HTseq. We annotated genes using the Ensembl version 97. Expression values of protein coding genes were normalized using Deseq2 package. Principal component (PC) analysis and hierarchical clustering analysis using Euclidean distance were conducted on the whole normalized dataset to determine how samples cluster together. Differential expression analysis between different groups was conducted using Deseq2. Genes with an adjusted p-value < 0.05 (Benjamini-Hochberg multiple testing correction) were considered significantly differentially transcribed. Gene ontology (GO) enrichment analysis was conducted using "clusterProfiler" package in R. Benjamini-Hochberg correction was used to adjust raw p-values for multiple testing (adjusted p-value < 0.05 was considered significant). For the MethylCap-seq data, we used the QSEA R package to identify the differentially methylated regions and to build the PC analysis.

Phenotypic Analysis
Identity, purity, viability and morphology of BMSCs cultured under hPL (O/AB and mBG) or FBS conditions was characterized by flow cytometry using a Gallios™ (Beckman Coulter, Brea, CA, USA) as previously described [25,26].
Extended flow cytometric cell surface marker screening was done using a LEGEND Screen Human phycoerythrin (PE) Kit (BioLegend, San Diego, CA, USA) according to the manufacturer's instructions containing 347 (old kit version) or 361 (new kit version) target antibodies, respectively. All BMSCs were stained with a backbone panel containing CD14, CD19, CD31, CD34, and CD45 as lineage (Lin)-negative markers, and CD29, CD44, CD73, CD90, CD105 as positive markers, beforehand. To analyze the expression of targets, single, viable LIN − /CD29 + /CD44 + and CD73 + /CD90 + /CD105 + cells were gated as depicted in Figure A1. Respective isotype controls were used to determine PE-labeled target antibody reactivity compared to 10 different isotypes' control reactivity. Data were analyzed using R.

BMSC Function: Coagulation Analysis of Early and Late BMSCs by Rotational Thromboelastometry
The procoagulant activity of BMSCs from 4 donors in plasma was analyzed by rotational thromboelastometry (ROTEM ® , Tem ® International, Munich, Germany) as described previously [22]. In brief, 5 × 10 5 BMSCs cultured in FBS and hPL O/AB (P1 and P4), and in hPL mBG (P4), respectively, were resuspended in 300 µL of citrated blood group AB plasma pooled from 10 healthy blood donors after receipt of written informed consent. After addition of 20 µL calcium chloride solution reagent (Tem ® International GmbH, Munich, Germany) the coagulation process was monitored for 60 min. AB plasma without cells was used as reference. The clotting time, clot formation time, maximum clot firmness and α-angle were measured in triplicate following manufacturer's instructions. Results were analyzed according to published standards as described previously [22].

Statistical Analysis
Statistical analysis was done using GraphPad Prism (GraphPad Software, San Diego, CA, USA). Data are shown as mean ± standard deviation (SD). Data were analyzed for normal distribution using Shapiro-Wilk or D'Agostino and Pearson tests. Paired t-test, repeated measures, one-way ANOVA with multiple comparisons (Tukey) were used for disk weights, clotting time and PDs. Friedman tests with multiple comparisons (Dunn) were conducted for clot formation time. To test the impact of hPL and FBS on cumulative cell numbers and cumulative PDs, linear regression analysis was conducted using R. Significant differences were depicted as indicated in the respective figures. A p-value < 0.05 was considered to be significant. Statistical analysis conducted for bioinformatics was carried out using R software and as described in detail in Sections 2.3 and 2.4, respectively.

BMSC Expansion
We selected four healthy donor-derived BM aspirates as starting material for this study. Heparinized BM aspiration aliquots were divided in two equal parts and equilibrated directly in 500 mL BMSC expansion medium supplemented with either FBS or hPL O/AB, to initiate P • , in four-layered cell factories (CF-4) on 2528 cm 2 growth area as described in [24,28]. Subsequent expansion was done either in FBS or in two types of hPL (O/AB vs. mBG) (Figure 1). Bone marrow (BM) from four healthy donors was divided 50/50 (e.g., 2 × 2.5 mL and resuspended directly in 500 mL culture medium supplemented with either fetal bovine serum (FBS) or pooled human platelet lysate (hPL) to initiate primary culture (P°) [23]. P° and passage (P1) cultures were performed with a selected lot of FBS or with hPL from blood group O dono platelets, lysed in blood group AB plasma, lacking isoagglutinins anti-A and anti-B (hPL O/AB), in four-layered cell factories (CF4), to establish a working cell bank. In subsequent cultures, hPL de rived from expired platelet concentrates irrespective of the blood group (mixed blood group, mBG was used for comparison as indicated. The batch effect of these different culture conditions on BMSC genotype, phenotype and function was studied as indicated. ROTEM, rotational thromboelastome try.

Transcriptome and Methylome Characterization
Gene expression profiling in BMSC samples from four healthy donors was per formed by RNA-seq at early (P1) and late time points (P4) after culture in either hPL-o FBS-supplemented media, respectively. Culture conditions, i.e., hPL vs. FBS, were the ma jor driver of variation. An impressive batch effect of culture conditions was observed al Bone marrow (BM) from four healthy donors was divided 50/50 (e.g., 2 × 2.5 mL) and resuspended directly in 500 mL culture medium supplemented with either fetal bovine serum (FBS) or pooled human platelet lysate (hPL) to initiate primary culture (P • ) [23]. P • and passage 1 (P1) cultures were performed with a selected lot of FBS or with hPL from blood group O donor platelets, lysed in blood group AB plasma, lacking isoagglutinins anti-A and anti-B (hPL O/AB), in four-layered cell factories (CF4), to establish a working cell bank. In subsequent cultures, hPL derived from expired platelet concentrates irrespective of the blood group (mixed blood group, mBG) was used for comparison as indicated. The batch effect of these different culture conditions on BMSC genotype, phenotype and function was studied as indicated. ROTEM, rotational thromboelastometry.
We confirmed previous observations that hPL-expanded BMSCs displayed slimmer cell shape [24], reached a higher cell density ( Figure A2) and a significantly enhanced proliferation resulting in a mean 3.3-fold higher cell number already after first passage (P1; Figure A3b) [25]. Microscopy measurements confirmed a significantly lower mean cell length and significantly lower mean cell area of hPL-expanded cells ( Figure A2). Beginning at P2, BMSCs were seeded at each passage exactly at 100 cells/cm 2 in cell culture flasks to create standardized high proliferation exhaustive culture conditions [29,30]. The low seeding density culture strategy resulted in a significantly reduced population doubling time of as short as 24 h in hPL-supplemented cultures compared to mean 40 h in FBS cultures. Maximum population doublings were observed at P2 and P3 for all culture batches in hPL and FBS, respectively. A significant decline in proliferation, reminiscent of proliferative senescence, was evident at P4 for FBS cultures ( Figure A3a), confirming previous observations [30]. Calculated cumulative mean cell numbers were 1.98 ± 0.81 × 10 15 and 4.06 ± 1.60 × 10 15 for the two independent hPL-driven cultures, corresponding to 29.8 ± 0.6 as well as 30.8 ± 0.5 population doublings within only 37 ± 3 days in hPL mBG and hPL O/AB, respectively. FBS cultures yielded 29.6-fold less cells with a calculated maximum of 0.10 ± 0.07 × 10 15 cells, corresponding to 25.2 ± 1.2 population doublings within 70 ± 9 days (all numbers mean ± SD; Figure A3c,d). We summarized the experimental strategy and the color code used throughout the study in Figure 1.

Transcriptome and Methylome Characterization
Gene expression profiling in BMSC samples from four healthy donors was performed by RNA-seq at early (P1) and late time points (P4) after culture in either hPL-or FBSsupplemented media, respectively. Culture conditions, i.e., hPL vs. FBS, were the major driver of variation. An impressive batch effect of culture conditions was observed already after P1 separating hPL-vs. FBS-driven cultures into trajectories due to 41% of the gene expression variance in principle component 1 (PC1). This separating batch effect of culture conditions persisted over time of culture. Differences in PC2 explaining 17% of the total variance mainly corresponded to donor variation and culture duration ( Figure 2a).
Hierarchical clustering of the 100 most variable genes in a heatmap showed that samples within hPL vs. FBS batches cluster together regardless of culture duration (Figures 2b and A2). Whole genome methylation analysis by MethylCap-seq revealed that both donor variation and hPL vs. FBS serum component differences during propagation contribute to variation in the DNA methylation pattern. Concerning the two most prominent components, represented by PC1 and PC2, the four donor-derived methylation datasets clustered together in PC1 more than PC2. The hPL-vs. FBS-driven culture conditions separated methylation patterns for all four donors, mainly in PC2. Minor differences were detectable comparing hPL O/AB with mixed blood group hPL ( Figure 2c). We found most significant DNA methylation changes on chromosomes 1, 8, 10 and 12 ( Figure 2d). We observed an overall gain of methylation in FBS compared to hPL with more than 21% of CpG islands being significantly hypermethylated in FBS while 12.5% were hypomethylated. By investigating the overlap between RNA-seq and MethylCap-seq, we identified 100 genes significantly downregulated and hypermethylated, and 38 genes significantly upregulated and hypomethylated, in hPL vs. FBS. GO term enrichment analysis conducted for the 100 genes downregulated and hypermethylated in hPL showed enrichment of GO categories including connective tissue development and striated muscle cell differentiation. This included reduced expression of genes negatively affecting chondrogenesis or genes involved in more pleiotropic functions such as ADAMTS7 and RFLNB (GO "negative regulation of chondrocyte differentiation"). No GO terms were found significantly enriched in the 38 genes upregulated and hypomethylated in hPL vs. FBS. ( Figure A5). Hierarchical clustering of the 100 most variable genes in a heatmap showed that samples within hPL vs. FBS batches cluster together regardless of culture duration (Figures  2b and A2). Whole genome methylation analysis by MethylCap-seq revealed that both donor variation and hPL vs. FBS serum component differences during propagation contribute to variation in the DNA methylation pattern. Concerning the two most prominent components, represented by PC1 and PC2, the four donor-derived methylation datasets clustered together in PC1 more than PC2. The hPL-vs. FBS-driven culture conditions separated methylation patterns for all four donors, mainly in PC2. Minor differences were The GO enrichment results of commonly differentially expressed genes during BMSC culture draw a clearly accentuated picture (Figure 3a). Comparing gene expression of BMSCs expanded in FBS vs. hPL displayed 3,625 significantly differentially regulated genes after P1 and 6615 differentially expressed genes after P4 suggesting progressive differences in response to extended cell culture. Interestingly, 1578 genes were detected commonly differentially transcribed in FBS vs. hPL-driven cultures irrespective of culture  (Figure 3b). Among these 1578 genes, 1540 showed similar profiles (UP or DOWN in both passages), including 993 genes significantly upregulated in FBS compared to 547 in hPL. This difference is also reflected in the enrichment analysis where a higher number of 124 GO categories were enriched in FBS compared to 54 in hPL. We found strong enrichment of genes categorized as interferon-response-related (n = 20) and antiviralresponse-related (n = 54) as well as energy metabolism responses (n = 33) in hPL-driven cultures (Supplementary Table S1). enriched in the 38 genes upregulated and hypomethylated in hPL vs. FBS. (Figure A5).
The GO enrichment results of commonly differentially expressed genes during BMSC culture draw a clearly accentuated picture (Figure 3a). Comparing gene expression of BMSCs expanded in FBS vs. hPL displayed 3,625 significantly differentially regulated genes after P1 and 6615 differentially expressed genes after P4 suggesting progressive differences in response to extended cell culture. Interestingly, 1578 genes were detected commonly differentially transcribed in FBS vs. hPL-driven cultures irrespective of culture duration ( Figure 3b). Among these 1578 genes, 1540 showed similar profiles (UP or DOWN in both passages), including 993 genes significantly upregulated in FBS compared to 547 in hPL. This difference is also reflected in the enrichment analysis where a higher number of 124 GO categories were enriched in FBS compared to 54 in hPL. We found strong enrichment of genes categorized as interferon-response-related (n = 20) and antiviral-response-related (n = 54) as well as energy metabolism responses (n = 33) in hPL-driven cultures (Supplementary Table S1).  Besides being upregulated in hPL, the immune-response-related genes, CD40, CD74, APOL4 and HLA-DRB1, also showed a strong positive correlation with cartilage disk weights (see also Figure A7b

Polychromatic Flow Cytometry
After confirming identity (<5% hematopoietic contamination CD14/19/34/45) and purity (>95% CD73/90/105/29) of the expanded BMSCs [31] after P1, using conventional multicolor flow cytometry ( Figure A6), we performed a >300 immune-related cell marker LEGEND screen with stringent gating strategy ( Figure A1). Heatmap analysis, providing an overview of multiplexed data, allowed the display of clear marker expression trajectories. In this display, BMSCs expanded in hPL O/AB clustered together as did those expanded in FBS. Marker profiles in hPL O/AB cultures at early (P1) and late time points (P4) were more adjacent indicating higher heterogeneity among FBS-expanded BMSCs. Surprisingly, three of four hPL mBG-expanded donors clustered together with their FBS-expanded counterparts at P4. Large cohorts of clearly differentially expressed markers were identified (Figure 4a). In a detailed analysis, we identified numerous immunity-related marker molecules to be differentially expressed by hPL O/AB-vs. FBS-expanded BMSCs further supporting the hypothesis that batch effects precipitate culture-related gene expression trajectories (Figure 4b).

Organoid-like 3D Cartilage Disk Formation
To validate skeletal regenerative function of the different early and late cell products we selected organoid-like 3D cartilage disk formation as one of the most stringent readouts for endochondral differentiation [7,9]. All hPL-and three of four FBS-driven P1 cell products displayed competence to form 3D cartilage discs under appropriate culture conditions. BMSCs derived from hPL-driven cultures were significantly more potent than cells from FBS cultures, producing more than twice as heavy cartilage disks. All P4 hPLexpanded cell products, made of cells beyond 29 PDs, showed significantly reduced weight of the end-products after 3D chondrogenesis. The hPL O/AB-and hPL mBG-expanded late cell products generated mean 20.3 ± 3.3 and 19.1 ± 5.9 mg weight disks, respectively. The FBS-expanded P4 cell products lost their capacity to form cartilage disks but just gathered into 2.6 ± 2.3 mg (mean ± SD) condensed aggregates (Figure 5a,b). Bern scoring of in vitro engineered cartilage [27] showed significant loss of cartilage quality in

Organoid-like 3D Cartilage Disk Formation
To validate skeletal regenerative function of the different early and late cell products we selected organoid-like 3D cartilage disk formation as one of the most stringent readouts for endochondral differentiation [7,9]. All hPL-and three of four FBS-driven P1 cell products displayed competence to form 3D cartilage discs under appropriate culture conditions. BMSCs derived from hPL-driven cultures were significantly more potent than cells from FBS cultures, producing more than twice as heavy cartilage disks. All P4 hPL-expanded cell products, made of cells beyond 29 PDs, showed significantly reduced weight of the end-products after 3D chondrogenesis. The hPL O/AB-and hPL mBG-expanded late cell products generated mean 20.3 ± 3.3 and 19.1 ± 5.9 mg weight disks, respectively. The FBS-expanded P4 cell products lost their capacity to form cartilage disks but just gathered into 2.6 ± 2.3 mg (mean ± SD) condensed aggregates (Figure 5a,b). Bern scoring of in vitro engineered cartilage [27] showed significant loss of cartilage quality in P4 FBS-expanded cultures (Figure 5c) with a highly significant correlation between Bern score and cartilage disk weight (R 2 = 0.8447; Figure A7). Safranin O plus Fast Green staining of sectioned disks or cell aggregates confirmed these findings (Figure 5d).

Procoagulant Activity of BMSCs
Thromboembolism is a feared risk of cell-based therapies, particularly triggered by extrahematopoietic cells after extended culture [22]. In a second set of functional experiments, we therefore measured the procoagulant capacity of early and late BMSC culture products expanded in FBS, hPL O/AB or hPL mBG, by ROTEM. BMSCs were resuspended in citrated AB plasma. After addition of CaCl2, the clotting reaction was monitored for 60 min. The time until clot formation initiation (amplitude of 2 mm, 'clotting time'), the time from initiation of clot formation until a 20 mm amplitude of clotting ('clot formation time'), the peak of the amplitude ('maximum clot firmness') and the kinetics of the clotting

Procoagulant Activity of BMSCs
Thromboembolism is a feared risk of cell-based therapies, particularly triggered by extrahematopoietic cells after extended culture [22]. In a second set of functional experiments, we therefore measured the procoagulant capacity of early and late BMSC culture products expanded in FBS, hPL O/AB or hPL mBG, by ROTEM. BMSCs were resuspended in citrated AB plasma. After addition of CaCl 2 , the clotting reaction was monitored for 60 min. The time until clot formation initiation (amplitude of 2 mm, 'clotting time'), the time from initiation of clot formation until a 20 mm amplitude of clotting ('clot formation time'), the peak of the amplitude ('maximum clot firmness') and the kinetics of the clotting represented as 'α-angle' were compared. The clotting time and the α-angle of P1 BMSCs did not differ significantly, irrespective of culture in FBS or hPL O/AB. Extended culture (P4) resulted in significantly shortened clotting time with changed clotting kinetics (Figure 6a,d).  After extended in vitro culture, measured at the end of P4, BMSCs showed a significantly shortened clotting time and increased α-angle reflecting a higher procoagulant activity acquired in all media conditions compared to P1. Notably, the clotting risk was most pronounced after FBS-compared to hPL-driven propagation. Minor variations in the clotformation time and maximum clot firmness were donor-dependent rather than serumdependent (Figure 6b,c). After extended in vitro culture, measured at the end of P4, BMSCs showed a significantly shortened clotting time and increased α-angle reflecting a higher procoagulant activity acquired in all media conditions compared to P1. Notably, the clotting risk was most pronounced after FBS-compared to hPL-driven propagation. Minor variations in the clot-formation time and maximum clot firmness were donor-dependent rather than serum-dependent (Figure 6b,c).

Discussion
Batch effects are evident in most complex analytic assay formats but are mostly neglected in cell manufacture. Here, we demonstrate profound batch effects on BMSC gene expression, phenotype and function, detected already after just one passage of large-scale preclinical cell propagation. We confirmed our historical results showing significantly superior BMSC multiplication in the presence of hPL compared to FBS [24,28,30,32,33].
Transcriptomics comparing FBS-vs. hPL-expanded BMSCs displayed distinct gene expression trajectories due to 3625 significantly differentially expressed genes in P1. Progressive differences in 6,615 differentially expressed genes were observed after four culture passages at low seeding density, i.e., augmented proliferation of up to 30 PDs, representing approximately 1 billion-fold expansion in hPL. The culture conditions played a greater role for the variance compared to donor variation. Interestingly, BMSCs in FBS showed an enrichment for certain GOs related to cartilage development. Chondrogenic markers COMP and ACAN were found to be upregulated in FBS. Key chondrogenic markers such as RUNX2 (osteochondro progenitor transcription factor), COL10A1, MMP13 (both hypertrophic chondrogenesis markers), BGLAP and CHI3L1 were found to be significantly upregulated in hPL. A similar observation was made previously [34] where ACAN expression was found to be strongly upregulated in FBS while SOX9, RUNX2 and ALP were found to be upregulated in hPL. We found higher values of SOX9 and ALP in hPL as well, but these were not significant due to high variability. Published data indicate that hPL maintained cells at an earlier stage of differentiation in terms of chondrogenesis, compared to FBS driving a more committed differentiation state [34].
BMSCs expanded in hPL showed an induction of genes related to interferon gamma (IFN-γ) and immune response gene categories in our study. While it is not yet clear how such response may influence cartilage development, a previous study indicated that IFN-γ could enhance chondrogenesis and anti-inflammatory activity [35] IFN-γ was also shown to be involved in the immune response of diseased cartilage [36,37].
We also found deviations in methylomics separating samples by donor origin and culture conditions in hPL vs. FBS, but barely by culture duration. It has also been shown that hPL-expanded cells displayed lower levels of senescence compared to cells expanded in FBS [38]. During senescence, epigenetic alterations and notably DNA methylation play an important role in shaping the expression of proliferation-related genes [39]. DNA methylation changes in long-term cultured cell lines are well established [40,41]. Methylation changes during long-term stromal cell culture were recently discovered to be mainly due to epigenetic drift, i.e., not directly regulated by a targeted mechanism [42]. Moreover, DNA methylation is a key epigenetic modification for the establishment and maintenance of cellular identity [43]. We observed an overall gain of methylation in FBS-expanded BMSCs suggesting a different aging process between the different cultures. We found limited overlap between the RNA-seq and the MethylCap-seq affecting 138 genes. This may suggest that other mechanisms, histone modification or nonepigenetic changes, may play a role. Interestingly, 100 of these overlapping genes were downregulated and hypermethylated in hPL compared to FBS and showed a clear enrichment for genes related to skeletal development. Strikingly, part of these genes belongs to GO category "negative regulation of chondrocyte differentiation" including RFLNB and ADAMTS7. A previous study showed that forced expression of ADAMTS7 and ADAMTS12 can suppress differentiation of uncommitted mesenchymal cells to the chondrocyte lineage [44]. Another gene, FGFR3 can act as negative regulator of chondrogenesis [45].
Cartilage regeneration represents one of the prime pursued applications for BM-SCs [61]. Cell expansion is a prerequisite for most stromal cell therapies due to the limited availability of primary cells [62]. In this study, we extended previous observations related to hPL vs. FBS serum supplements [7,63] in a head-to-head comparison of paired oligoclonal BMSC preparations from the same donor. We found that, already after one culture passage, hPL-expanded BMSC were significantly more potent than FBS-expanded BMSCs from the same donor in generating significantly larger 3D cartilage disks despite equal starting cell number. Bern scoring [27] indicated a significantly improved engineered cartilage quality. More impressively, FBS-expanded BMSCs lost their chondrogenic potential after excessive expansion, i.e., >20 population doublings. The hPL-expanded BMSC maintained chondrogenesis despite reduced cartilage disk weight but still displayed intact disk shape and engineered cartilage quality. These results argue in favor of using hPL for BMSC propagation for skeletal regeneration. Interestingly, we also showed that donor variation may have an impact on cartilage formation irrespective of culture in both FBS and hPL media. We were still able to produce cartilage disks with hPL-but not FBS-expanded BMSCs from one of four donors. We may speculate how the early transcriptomic and/or phenotypic batch effects translate into the subsequent significant differences in 3D chondrogenesis. We identified a number of genes showing strong positive correlation with 3D cartilage disk weight, including immune-related (HLA-DRB1 and invariant chain, CD74) metabolism-related (selenoprotein P, SELENOP; sortin-related receptor, SORCS2), morphogens (cysteine rich secretory protein, CRISPLD2, bone morphogenic protein-inducible protein and chitinase 3-like 1 chondrocyte protein, CHI3L1), apolipoprotein A4 (APOL4), delta and notch-like epidermal growth factor-related receptor (DNER) as well as RUNX2 ( Figure A8a,b). HLA-DRB1, CD74 and APOL4 were also found previously to be induced by IFN-γ in osteoarthritis chondrocytes [64]. DNER was shown to be negatively correlated with chondrogenic potential [65] Additional research is required to better understand the contribution of individual differentially regulated genes and the resulting signatures on BMSC function also in relation to the batch effects observed.
We selected clotting risk analysis as a second functional readout because intravascular application of tissue factor expressing cells, i.e., virtually all extrahematopoietic cell products, can result in an instant blood-mediated inflammatory response (IBMIR) [66]. It is not clear yet if unintended blood contact after local injection can also result in clotting-related side effects [67,68]. ROTEM results revealed a culture-induced increase of the procoagulant activity of BMSCs in AB plasma, reflected by significantly shortened clotting time values after extended culture (P4). Higher procoagulant activity of all P4 BMSCs with a statistically significant increase of α-angles was observed compared to early cells, indicating negative impact of extended culture and exhaustive propagation. The limited clotting risk of short-term expanded BMSCs confirmed earlier results [22,69]. Another important aspect is the capacity of the generated cartilage to adhere to the disease site. Although we have not conducted adhesion strength assays in our study, a systematic review suggested that hPL was as effective as FBS in promoting adhesion [70].
In the course of this study, we observed an unexpectedly distinctive serum-related batch effect resulting in pronounced differences in gene expression and phenotype as well as corresponding differences in BMSC chondrogenic function. This has broader implications, which are not restricted to skeletal regeneration. The distinct immune-related phenotypic changes may help to select most potent manufacturing conditions for immunomodulatory stromal cells. Marker profiling as indicated in this study and corresponding functional assays will allow the development of predictive potency assays for cell-based therapeutics [71]. Future studies need to address whether batch effects are restricted to serum components or if additional factors, particularly complex undisclosed proprietary ingredients in defined media, also elicit batch effects. There is an additional critical need for developing strategies for nonhealing bone fractures. In this work, we provided evidence that hPL provides a better environment for BMSCs regarding cartilage formation. Based on the endochondral bone formation capacity of BMSCs [7] these cartilage disks might also represent templates for endochondral bone formation.

Conclusions
Batch effects occur during cell therapy manufacture. Signatures identified in this study will help to develop sensitive potency assays to guarantee stringent release criteria for cell therapeutics.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/cells11060946/s1, Table S1: List of genes in alphabetical order related to viral response, metabolism or interferon response found significantly upregulated in hPL compared to FBS.