Subcutaneous human adipose tissue was obtained external to the knee joint in patients undergoing knee arthroscopy or from abdominal lipoaspirate (REB #18-5480 and #18-6345, see Table 1 for summary of donor characteristics). MSC(AT) were isolated and expanded using MesenCult™-ACF Plus xeno-free and antibiotic-free growth medium (StemCell Technologies, Vancouver, Canada) on standard tissue culture polystyrene flasks coated with animal component-free cell attachment substrate (StemCell Technologies) according to the manufacturer’s instructions. Cells were expanded in a standard incubator at 37˚C in 5% CO₂ under ambient air. For passaging, flasks at approximately 80% confluency were washed with PBS (Wisent, St-Bruno, Canada) and harvested using TrpLE (Gibco, Waltham, USA) prior to re-plating at 5,000 cells/cm². All experiments were performed using MSC(AT) between passage 3 and passage 5.

At approximately 70-80% confluence, MSC(AT) were transiently (16-20 h) cultured under varied CPP conditions, including 3D normoxic (3D-N), 2D hypoxic (2D-H), or 2D normoxic (2D-N) conditions. These culture steps were performed on separate flasks cultured in parallel using MesenCult™-ACF Plus medium supplemented with 1% (v/v) human serum album (HSA; Canadian Blood Services, Ottawa, Canada). 3D-N culture was performed by harvesting MSC(AT) from adherent flasks and plating cells on ultra-low attachment surfaces (Corning, Corning, USA) at 26,700 cells/cm² and 200,000 cells/mL in medium supplemented with 2 ng/mL IL-6 (Peprotech, Cranbury, USA; used to support cell viability in the 3D culture), based on previously reported methods that allow spontaneous aggregation of MSCs into cell clusters (39). Flasks from the same batch of cells were cultured in parallel under 2D-N (maintained in standard tissue culture incubator) or 2D-H conditions (38 mmHg O₂, i.e., 5% O₂ under standard atmospheric pressure) using a HypoxyLab workstation (Oxford Optronix, Milton, UK) for the same duration as 3D culture.

Prior to experiments, 3D cell aggregates were collected from ultra-low attachment flasks. Following multiple PBS washes of the flask and mixing, samples were removed for cell counting. Cell enumeration was performed after dissociating the aggregates using Accumax™ solution (Sigma, St. Louis, USA), according to previously published methods (40). For MSC(AT) cultured under 2D-N and 2D-H conditions, flasks were washed with PBS and incubated in TrpLE solution, followed by neutralization with complete medium and cell counting. All cell counts were performed using the Vi-Cell XR Cell Counter (Beckman Coulter, Brea, USA). Prior to plating cells for experiments, excess PBS was added to cell suspensions to dilute residual growth factors/cytokines before centrifugation (350 x g, 5 min) to pellet the cells.

Cell diameter and circularity was measured for single cell suspensions using the Vi-Cell XR Cell Counter (Beckman Coulter). 3D-N MSC(AT) were dissociated into single cell suspensions as described above. To analyze maximum feret diameters and circularity of whole intact 3D-N MSC(AT) aggregates, 10X phase-contrast images were captured using an EVOS XL Core Cell Imaging System (ThermoFisher, Waltham, USA). A semi-automated algorithm was developed in ImageJ (41) based on rolling ball subtraction to create binary images for particle analysis, and a minimum of 230 aggregate measurements were performed per donor and condition.

Surface marker expression of MSC(AT) was measured following previously established protocols (8) and in accordance with IFATS/ISCT guidelines (positive marker threshold: >80%, negative marker threshold: <2%) (42). The following PE-conjugated anti-human antibodies from BioLegend (San Diego, USA) were used: anti-CD90 (cat. 328109), anti-CD73 (cat. 344004), anti-CD44 (cat. 338807), anti-CD29 (cat. 303003), anti-CD13 (cat. 301703), anti-CD34 (cat. 343506), anti-CD31 (cat. 303105), anti-CD45 (cat. 304008), and anti-CD105 (cat. 323205). For staining, single cell suspensions of 2D-H and 2D-N MSC(AT) were obtained by TrpLE dissociation, while 3D-N aggregates were digested using Accumax™ solution as described above. To evaluate the effects of Accumax™ digestion on the surface marker profile of 2D MSC(AT), a subset of 2D-N MSC(AT) were digested using the same Accumax™ digestion protocol as used for 3D cell aggregates. Samples were characterized using the FC500 flow cytometer (Beckman Coulter) and analyzed using FlowJo version 10 software (Ashland, USA).

Western blot analysis was performed to confirm CD105 expression in MSC(AT) cultured under varying CPP conditions. Cells were lysed using a buffer containing 50 mM Tris–HCl (pH 7.5), 1mM EGTA, 1mM EDTA, 1% (w/v) Nonidet P40, 1mM sodium orthovanadate, 50 mM sodium fluoride, 5 mM sodium pyrophosphate, 0.27 M sucrose, and a protease inhibitor cocktail (Roche). Protein concentration of all samples was measured using the Pierce BCA protein assay (ThermoFisher). Protein samples were loaded in a 10% polyacrylamide gel (20 μg/well) for electrophoresis followed by transfer to nitrocellulose membranes. Membranes were then blocked with TBS-T containing 5% (w/v) BSA and they were immunoblotted in the same buffer overnight at 4°C with an anti-CD105 primary antibody (cat. 323205, BioLegend, 1:1,000 dilution), or for 2 h with an anti-β-actin antibody (Sigma, 1:10,000 dilution) used for the loading control. Washes were then performed with TBS-T and the blots were then incubated with secondary HRP-conjugated antibodies in 5% skimmed milk. The blots were washed in TBS-T and the signal was detected with the enhanced chemiluminescence reagent (ECL; GE Healthcare, Chicago, USA) and using a chemiluminescent imaging system (Bio-Rad, Hercules, USA).

After harvesting cells from 2D-N, 2D-H, and 3D-N conditions, cells were added to 24-well plates at 60,000 cells/well in MesenCult™-ACF Plus medium supplemented with 1% HSA (v/v) with or without addition of pro-inflammatory licensing cytokines. The licensing cytokines (all purchased from Peprotech) consisted of IFNγ (30 ng/mL), TNFα (10 ng/mL), and IL-1β (5 ng/mL). Cells harvested from 2D-N or 2D-H conditions were maintained under normoxic or hypoxic (38 mmHg O₂) conditions for 24 h, respectively, while cell aggregates from the 3D-N condition were plated on ultra-low attachment 24-well plates (Corning) without dissociation. After the culture period, conditioned medium was collected, centrifuged (1,000 x g, 5 min) and frozen at -80°C. The remaining cells were washed in PBS and RNA was extracted using a RNeasy Mini kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions. RNA concentration and purity was measured using a DS-11 Spectrophotometer (DeNovix, Wilmington, USA).

The nCounter platform (NanoString, Seattle, USA) was used as a highly sensitive tool that detects target mRNAs with high specificity and without amplification. Samples (100 ng RNA/sample) were run on the nCounter MAX Analysis system (St. Michael’s Genomics Molecular Biology Core facility) according to the manufacturer’s instructions using a custom CodeSet 58-gene panel. The data was processed using the nSolver version 4.0 software (NanoString) according to the manufacturer’s instructions to obtain mRNA counts normalized to the synthetic positive control probes and to reference genes. The following were measured as potential reference genes: ABCF1, GAPDH, GUSB, HPRT1, LDHA, RPL19, RPLP0, TUBB, POLR1B, and TBP. Reference gene stability was evaluated using geNorm analysis in the nSolver software; POLR1B and TBP were subsequently discarded as reference genes and not used for normalization. Normalized mRNA counts below 20 were assigned values of 1 if >33% of samples were within range. The following genes were undetectable (<33% of samples were within range) under both licensed and unlicensed conditions: ANGPT2, BDNF, BMP7, CCR7, CD200, CTLA4, CXCR4, IGF1, IL10, IL12A, NGF, PDGFB, PROK1, and CXCL12. Under licensed conditions, VASH1 and SOX9 were undetectable. Under unlicensed conditions, CD274, IDO1, NFKBIA, NOS2, and PDGFA were undetectable. We selected a broad spectrum of MSC(AT) transcripts initially but many of the undetectable genes are not commonly or are inconsistently expressed by MSCs. Other genes may have tissue of origin-dependent or context-dependent expression. All NanoString data has been deposited in NCBI’s Gene Expression Omnibus (43) accessible through GEO Series accession number GSE212368.

Conditioned medium samples were analyzed for soluble factors using a custom 10-analyte LEGENDplex immunoassay (BioLegend) according to the manufacturer’s instructions, and samples were run on a FACSCanto™ II flow cytometer (BD Biosciences, Franklin Lakes, USA). Values below the assay limit of detection were imputed as half of the lower limit of detection if >33% of samples were within range. IL-10, PIGF, and PD-L1 were undetectable in the samples tested. IL-1RA was undetectable in unlicensed samples only.

Pilot gene expression experiments were performed to i) examine gene expression in MSC(AT) cultured using the combination of 3D and hypoxic culture (3D-H), and ii) to evaluate effects of IL-6 treatment on MSC(AT) cultured under 2D-N conditions using the same IL-6 concentration and treatment duration as used for 3D-N conditions. Both sets of pilot experiments were performed under licensed conditions using the methods outlined above. After licensing, RNA was isolated from MSC(AT) by Trizol-chloroform extraction and cDNA was generated using SuperScript™ IV VILO™ Master Mix (Invitrogen, Waltham, USA). qPCR was run using custom primers ( Suppl. Table 1 , Invitrogen) and FastStart Universal SYBR Green Master Mix (Roche, Basel, Switzerland) on a QuantStudio™ 5 system (ThermoFisher). Results were normalized (ΔΔCT) against reference genes (B2M and RPL13A) and presented as fold-change values relative to 2D-N culture conditions.

Peripheral blood-derived CD14⁺ monocytes were isolated from a leukopak (StemCell Technologies) by Ficoll density gradient separation and selection with CD14⁺ magnetic beads as previously described (44). Cryopreserved monocytes were thawed and plated on 24-well plates at 100,000 cells/well and allowed to acclimate for 48 h in co-culture medium consisting of: 1 mM sodium pyruvate (Gibco), 1% penicillin-streptomycin (Gibco), 10% FBS (Wisent), and 10% low-glucose DMEM (Sigma) in RPMI medium (Gibco). Prior to co-culture, MSC(AT) were harvested from 2D-N, 2D-H, and 3D-N conditions, plated on 0.4 μm transwell inserts at 10,000 cells/insert in a separate 24-well plate, and allowed to attach for 2 h. The inserts were then transferred to MΦ wells and co-cultured for 20-24 h. Transwell inserts containing MSC(AT) were then removed, and lipopolysaccharide was spiked into wells at a final concentration of 2.5 ng/mL. After a 4 h incubation, MΦs (both adherent and in suspension) were collected and stored in Trizol (Roche) at -80°C. Conditioned medium was also collected and stored at -80°C.

Levels of TNFα in conditioned medium were measured by ELISA (R&D Systems, Minneapolis, USA) according to the manufacturer’s instructions. To analyze MΦ gene expression, RNA was isolated by Trizol-chloroform extraction and cDNA was generated using SuperScript™ IV VILO™ Master Mix (Invitrogen). qPCR was run using custom primers ( Suppl. Table 1 , Invitrogen) and FastStart Universal SYBR Green Master Mix (Roche) on a QuantStudio™ 5 system (ThermoFisher). Results were normalized (ΔΔCT) against reference genes (ACTB, B2M, and TBP) and presented as fold-change values relative to MΦ cultured without MSC(AT) (SOLO condition).

To prepare conditioned medium for the HUVEC tube formation assay, MSC(AT) were harvested from 2D-N, 2D-H, and 3D-N conditions, plated at 15,000 cells/cm² and 100,000 cells/mL of growth medium (no added HSA or exogenous cytokines), and incubated for 24 h in cell culture incubators. MSC(AT) harvested from all three CPP conditions were maintained under these conditions to prepare conditioned medium. To prepare unconditioned medium controls, growth medium was incubated on a cell culture plate for the same duration. Following the 24 h incubation period, the conditioned medium was collected, centrifuged (1,000 x g, 5 min), and the supernatant was frozen at -80°C for future use in the HUVEC tube formation assay.

Five to seven days prior to the tube formation assay, P4 HUVECs (Cat. CC2519, Lonza, Basel, Switzerland) were thawed and expanded in EGM™-2 (Lonza) according to the manufacturer’s instructions. The HUVEC tube formation assay was performed based on previously published methods (38). 96-well plates were pre-coated with Cultrex Reduced Growth Factor Basement Membrane Extract, PathClear (R&D Systems) according to the manufacturer’s instructions. HUVECs were harvested and plated on the coated plates at 42,500 cells/cm² in conditioned medium. The following medium formulations were used as controls: unconditioned medium, EGM™-2 medium (positive control), and basal medium without addition of growth factors/supplements (negative control). All wells were imaged 6 h after plating the cells using the EVOS XL Core Cell Imaging System (ThermoFisher) at 4X objective. The images were analyzed using the Angiogenesis Analyzer plugin in ImageJ (45).

Plots were created using GraphPad Prism 6.0 (La Jolla, USA) and JMP Pro 14 (Cary, USA) software. All statistical tests are specified in the figure captions. One- and two-way ANOVA, as well as simple linear regression was performed using GraphPad Prism software. Unbiased hierarchical clustering (Ward method), principal component (PC) analysis (default estimation method), and desirability profiling was performed using JMP software. Analysis of differential gene expression was performed in nSolver (NanoString) using single linear regressions for each covariate (i.e., gene) with false discovery rate (FDR)-corrected p values calculated using the Benjamini-Yekutieli method (46). Data were considered statistically significant based on a threshold of p<0.05. For functional in vitro assays (MΦ polarization and HUVEC tube formation) where PC analysis was performed on the assay readouts, the PC1 score was taken as a ‘composite functional score’, based on previously published methods (14). For linear regression, analysis between functional PC1 scores and MSC(AT) characteristics (genes, soluble factors, and morphometric features), statistically significant (p<0.05) correlations are reported along with correlations with 0.05<p<0.1 which were considered as near-significant. Normal residuals were checked to fulfill assumptions of linear regression, and parametric correlations were performed independently for each individual MSC(AT) characteristic that was measured.

Desirability profiling (47) was used as a tool for multiple response optimization that assigns individual responses a score from zero to one based on the range of data values, with zero representing an undesirable response, and one representing a highly desirable response. Minimization or maximization functions were assigned to each MSC(AT) characteristic (including genes, soluble factors, and morphological features) to indicate whether higher or lower values were desirable, based on outcomes from linear regression analyses with functional PC1 scores. Based on the regression analyses, all statistically significant (p<0.05) and near-significant (p<0.1) MSC(AT) characteristics were included in the desirability analysis. The R² values from the regression analyses were applied as weightings for each MSC(AT) characteristic.

A matrix of in vitro readouts was used to investigate the responses of putative CQAs to varying donors and CPPs that were specifically selected for their known ability to enhance MSC immunomodulatory and angiogenic properties. The matrix consisted of multivariate morphometric measurements, gene expression, soluble factor analysis, and functional immunomodulatory and angiogenic readouts ( Figure 1 ). The matrix of readouts demonstrated sensitivity to multiple sources of MSC(AT) variability, including variations in select CPPs, donor heterogeneity, and licensing with pro-inflammatory cytokines. Linear regression analyses were used to refine putative CQAs by evaluating correlations with anchor functional in vitro immunomodulatory and angiogenic outcomes. We further provided a range of values for assessing basal MSC(AT) fitness range according to the significant and near-significant putative CQAs identified from the regression analyses. Desirability analysis was then applied to analyze the profile of putative CQAs and to assign empirical rankings for donors and CPP conditions that result in desirable MSC(AT) immunomodulatory and/or angiogenic functionality.

The CPPs investigated included 2D-N, 2D-H, and 3D-N culture conditions. These CPPs were chosen based on previous literature demonstrating that 3D and 2D hypoxic culture can enhance the immunomodulatory and angiogenic potency of MSCs (28, 29, 31, 32, 35). MSC(AT) cultured under 2D-N, 2D-H and 3D-N conditions satisfied surface marker expression criteria (42) and were CD90⁺CD73⁺CD44⁺CD13⁺CD34^-CD31^-CD45^- ( Figure S1A ). Notably, CD105 appeared to be cleaved under enzymatic conditions required for dissociating 3D cell aggregates for flow cytometry analysis ( Figure S1B ). Western blot analysis verified CD105 expression by MSC(AT) cultured using 3D-N conditions, albeit at lower levels relative to 2D-N and 2D-H conditions ( Figure S1C ). MSC(AT) cultured under 2D-N and 2D-H conditions displayed a characteristic spindle-like morphology, while MSC(AT) cultured under 3D-N conditions formed cell aggregates of varying sizes ( Figure S2A ). Cell aggregates in the 3D-N condition had a median ferret diameter of 37.82 μm (range: 12.38 – 269.40 μm) and median circularity of 0.56 (range: 0.044 – 0.97). Morphometric measurements of MSC(AT) revealed differences in cell morphology with changes in CPP conditions. Analysis of single cell suspensions obtained from each culture condition demonstrated significantly reduced diameter and greater circularity of 3D-N MSC(AT) relative to 2D-N and 2D-H MSC(AT) ( Figure S2B, C ).

For gene expression and soluble factor analysis, 2D-N, 2D-H, and 3D-N MSC(AT) were subject to pro-inflammatory licensing conditions (with a cocktail of three pro-inflammatory cytokines, TNFα, IFNγ, and IL-1β to simulate a wide range of disease conditions) or cultured under unlicensed conditions in the absence of pro-inflammatory stimuli. The combination of 3D and hypoxic culture (3D-H) was investigated by gene expression analysis using an abbreviated panel of anti-inflammatory/angiogenic markers measured in two MSC(AT) donors ( Figure S3 ) and demonstrated no significant benefit of combining these culture conditions. Thus, 3D-N was used as the select CPP in all subsequent experiments. The effects of IL-6 (used to support cell viability under the 3D-N culture condition) were also evaluated on MSC(AT) cultured under 2D-N culture conditions using the same cytokine concentration as used for the 3D-N culture method. An abbreviated panel of genes was selected using markers that were significantly differentially expressed in the 3D-N culture condition relative to 2D-N culture. Gene expression analysis revealed no significant effect of IL-6 on 2D-N culture ( Figure S4 ), suggesting that the 3D geometry was the major factor that primed the cells rather than IL-6.

Gene expression analysis was performed using a curated panel of markers including predominantly immunomodulatory and angiogenic genes (data provided in Suppl. Tables 2, 3, 4, and 5 ). The panel was selected based on previous literature (11, 17, 48, 49) and our experience using MSC(M) in an osteoarthritis clinical trial (8). It was used to evaluate MSC(AT) fitness across multiple donors and while modulating select CPP conditions associated with enhanced immunomodulatory and angiogenic functionality. Unbiased hierarchical clustering analysis revealed that MSC(AT) gene expression profiles clustered according to both variations in CPP conditions and donor heterogeneity ( Figures 2A, B ). Under licensed conditions, CPP variations, specifically 3D-N configurations of culturing MSC(AT) clustered separately from 2D-N and 2D-H cultures for all but one donor (Donor 1) which clustered together regardless of CPP variations. For MSC(AT) cultured under 2D-N and 2D-H conditions, the gene expression profiles clustered together by donor rather than oxygen tension under licensed conditions. Interestingly, under unlicensed conditions, gene expression profiles clustered primarily by donor, regardless of variations in CPPs. Overall, variations in CPP conditions by changing culture geometry or oxygen tension and donor heterogeneity resulted in shifts in gene expression profiles dependent on pro-inflammatory licensing conditions or unlicensed conditions. For example, genes such as HGF (multifunctional growth factor) and TNFAIP6 (anti-inflammatory gene encoding TSG6), were upregulated by 3D-N culture conditions relative to 2D-N culture conditions, under licensed conditions ( Figure 2C ). NOS2 (encoding for inducible nitric oxide synthase, iNOS which may indicate cell stress or enhanced immunosuppressive functions for mouse, but not human MSCs (50)) and PRG4 (lubricating proteoglycan with potential immunomodulatory functions (51, 52)) were also upregulated under these conditions. ICAM1 (immunomodulatory marker), PRG4, and TNFAIP6 were upregulated by 3D-N culture conditions relative to 2D-N under unlicensed conditions ( Figure 2D ). Furthermore, 2D-H culture conditions induced augmented expression of the immunomodulatory marker PTGS2 relative to 2D-N culture conditions under unlicensed settings.

Levels of soluble factors in conditioned medium were queried using a curated sub-panel of immunomodulatory and angiogenic factors, based on significant results from the gene expression analysis (soluble factor data provided in Suppl. Tables 6, 7 ). Unbiased hierarchical clustering demonstrated that soluble factor profiles clustered based on both donor heterogeneity and variations in CPP conditions under licensed and unlicensed conditions ( Figures 3A, B ). Analysis of individual soluble factors was performed to evaluate the effects of CPP and donor heterogeneity on each soluble factor. Under licensed conditions, MSC(AT) cultured in 3D-N configurations expressed significantly higher levels of the multifunctional cytokine TGF-β relative to 2D-N, and significantly lower levels of the immunomodulatory factor soluble PD-L2 relative to 2D-H ( Figure 3C ). Furthermore, MSC(AT) cultured in 3D-N configurations expressed significantly higher levels of the growth factor HGF and the immunosuppressive factor IL-1RA relative to both 2D-N and 2D-H MSC(AT) under licensed conditions. Under unlicensed conditions, 2D-H and 3D-N MSC(AT) expressed significantly higher levels of TGF-β relative to 2D-N, while soluble PD-L2 was significantly upregulated by 2D-H relative to 3D-N MSC(AT) ( Figure 3D ). Expression of the angiogenic markers angiopoietin-1 (Ang-1) and VEGF showed statistically significant differences between donors, with Donor 3 expressing the highest levels of both factors under both licensed and unlicensed conditions ( Figures 3C, D ). Taken together, the curated panel of genes and soluble factors were differentially sensitive to donor heterogeneity and CPPs under licensed and unlicensed conditions. Interestingly, donor heterogeneity was masked under licensed conditions, and dominated under unlicensed conditions. Further, culturing MSC(AT) under 3D-N conditions rendered them with an elevated profile of anti-inflammatory/immunosuppressive genes (HGF, TNFAIP6, PTGES, PTGS2, TLR2, NFKBIA, TGFB1, PRG4, IDO, ICAM1, TLR4) and soluble factors (TGF-β, HGF, IL-1RA), corroborating previous reports that have investigated MSC culture in 3D formats (31, 32, 53).

To further probe immunomodulatory properties of MSC(AT), an indirect co-culture assay was performed to evaluate functional MΦ polarization ( Figure 4A ). Alterations to CPP conditions by changing MSC(AT) culture configuration (3D-N) or oxygen tension (2D-H) significantly reduced levels of pro-inflammatory TNFα in conditioned medium relative to MΦs alone ( Figure 4B ), suggesting that these culture conditions enhanced anti-inflammatory functions of MSC(AT). Gene expression analysis of MΦs revealed statistically significant increased expression of inflammation-resolving MΦ markers (CD206, HMOX1, and IL10) in co-cultures with MSC(AT) in 3D-N and 2D-H culture conditions relative to MΦ alone ( Figure S5A ). In addition, MSC(AT) cultured in 3D-N conditions significantly upregulated MΦ expression of CD86 (pro-inflammatory) and CD163 (inflammation-resolving), while 2D-N culture conditions significantly upregulated expression of HMOX1 only. PC analysis was applied as an unbiased dimension reduction tool to evaluate the full gene expression panel and TNFα protein levels ( Figure 4C , loading plots for PC1 and PC2 displayed in Figures S5B, C ). Increased expression of CD86, CD206, HMOX1, and STAB1 genes, and reduced expression of TNFα protein were the main contributors to higher scores along the PC1 axis (accounting for 35.7% of variation). Reduced expression of CD274 and HLADRA, and increased expression of CD163, IL12A, and TREM1 genes drove higher scores along the PC2 axis (accounting for 22.2% of variation).

To further probe the results of the PC analysis, we plotted individual PC1 and PC2 scores for each CPP and donor combination as these scores capture multivariate heterogeneity in a reduced, single dimension with PC1 capturing a larger proportion of the total existing variation in the dataset relative to PC2. Analysis of individual PC1 scores demonstrated that separation along the PC1 axis was driven by co-culture with MSC(AT) as indicated by statistically significant differences in PC1 scores for MSC(AT) cultured under 2D-N, 2D-H, and 3D-N conditions relative to solo MΦ (positive control) without MSC(AT) ( Figure 4D ). Given that the PC1 axis accounted for differences in MΦ phenotypic marker profiles relative to positive control, PC1 scores were considered as a ‘composite functional score’ (14) for MSC(AT) immunomodulatory function. With the exception of CD86 gene expression, higher scores along the PC1 axis were generally associated with greater inflammation-resolving MΦ polarization as evidenced by increased expression of CD206, HMOX1, and STAB1, along with reduced expression of TNFα protein. Using PC1 scores as an analytic, variations in CPP conditions resulted in differences in MSC(AT)-mediated MΦ polarization toward inflammation-resolving subtypes. Notably, MSC(AT) cultured in 3D-N configurations displayed the highest PC1 scores, and these were significantly different from 2D-N MSC(AT), suggesting that 3D-N MSC(AT) displayed a superior capacity to polarize MΦ phenotype toward inflammation-resolving subtypes. No significant differences in culture conditions were observed along the PC2 axis. Analysis of PC1 scores across individual donors showed no significant differences ( Figure 4E ); while not significant, Donor 1 displayed the highest PC1 scores suggesting that Donor 1 may have intrinsic improved immunomodulatory basal functionality, and Donor 3 displayed the lowest scores. In contrast, a significant effect of donor heterogeneity was observed along the PC2 axis. Given that PC2 scores represent changes in both pro-inflammatory and inflammation-resolving MΦ markers, these data suggest that donor heterogeneity dictates MSC(AT)-mediated polarization of MΦ toward mixed phenotypes.

To evaluate the angiogenic functions of MSC(AT), the effects of MSC(AT) conditioned medium on HUVEC tube formation was explored ( Figure 5A , Figure S6A ). Recognizing that there is no standard quantitation method for HUVEC tube formation and that different measurements can yield different insights into in vitro angiogenesis (54), we employed the ImageJ Angiogenesis Analyzer plugin which provides twenty different types of measurements (45). PC analysis was applied to analyze all twenty parameters that profile tube formation image analyses ( Figure 5B , loading plots for PC1 and PC2 displayed in Figure S6B, C ). An increased number of nodes, number of junctions, and the total master segment length were the main contributors to higher scores along the PC1 axis (accounting for 69.2% of variation), indicative of greater angiogenesis. Increased total branches length and number of branches, along with reduced total mesh area drove higher scores along the PC2 axis (accounting for 14.3% of variation). Analysis of individual PC1 scores demonstrated that the positive control group (HUVECs cultured in pro-angiogenic medium) displayed significantly higher PC1 scores relative to the negative control (HUVECs cultured in basal medium), and to HUVECs cultured in conditioned medium derived from 2D-N or 3D-N MSC(AT) culture conditions ( Figure 5C ). As above, the PC1 scores were considered as a ‘composite functional score’ for MSC(AT) angiogenic functions given that higher scores along this axis reflected greater HUVEC tube formation. Surprisingly, conditioned medium derived from MSC(AT) cultured under 2D-H or 3D-N conditions did not elicit significant increases in HUVEC PC1 scores, and there were no significant differences between PC1 scores across MSC(AT) by varying CPP conditions. PC1 scores were instead driven by MSC(AT) donor differences ( Figure 5D ). Donors 1, 2, 3, and 5 all displayed statistically significant higher scores relative to Donor 4 indicative of greater angiogenic function. Analysis of PC2 scores similarly showed no significant effect of the conditioned medium from the different groups representing variability in culture conditions, or the controls on HUVEC tube formation profiles ( Figure 5C ). Some variation along the PC2 axis was driven by donor with Donors 1 and 3 displaying significantly higher PC2 scores relative to Donor 4 ( Figure 5D ).

To further analyze effects mediated by variations in donor and experimental batches, PC analysis was performed on all biological and technical replicates ( Figure S7A ) and PC1 scores were investigated for each donor ( Figures S7B, C ). Under this analysis, conditioned medium derived from MSC(AT) cultured under 2D-H conditions had the highest mean HUVEC PC1 scores for four out of five donors (Donor 2, 3, 4, and 5) compared to scores for the 2D-N and 3D-N conditions. Taken together, the data suggests that donor heterogeneity dominated over effects mediated by variations in CPPs in our analysis of angiogenic readouts. The net effect of CPPs on HUVEC tube formation was variable, donor-dependent, and in part driven by CPP conditions that favour angiogenesis.

The full set of gene and soluble factor expression profiles as well as morphological features measured for MSC(AT) were examined for correlations to composite functional (PC1) scores (indicative of inflammation-resolving MΦ polarization) generated for MSC(AT) immunomodulatory fitness. Under licensed conditions, MSC(AT) expression of THBS1 (R² = 0.5481, p=0.0025, angiogenic gene), CCN2 (R² = 0.3020, p=0.0418, encoding for the multifunctional growth factor), and EDN1 (R² = 0.2854, p=0.0491, angiogenic gene) genes demonstrated significant inverse correlations with MΦ PC1 composite scores ( Figure 6A , Table 2 ), suggesting that lower expression of these genes correlated with improved immunomodulatory MSC(AT) functionality. Correlations of MΦ PC1 composite scores with ACTA2, PDCD1LG2, TNFAIP6, ANGPT1, and CXCL8 were near-significant. Under unlicensed conditions, MSC(AT) expression of six genes (CCN2, TSG101, THBS1, PDGFA, VEGFA, and EDIL3) were significantly inversely correlated with MΦ PC1 composite scores, with CCN2 (R² = 0.4934, p=0.0035), TSG101 (R² = 0.4229, p=0.0087, negative growth regulator and regulator of vesicular trafficking), THBS1 (R² = 0.3853, p=0.0135), and PDGFA (R² = 0.3754, p=0.0152) displaying the strongest correlations ( Figure 6B , Table 2 ). Correlations of MΦ PC1 composite scores with ACTA2, TGF-β protein, ANGPT1, and ANG were near-significant. Interestingly, many of the statistically significant and near-significant correlations were inverse correlations and they predominantly consisted of angiogenic-associated genes (THBS1, CCN2, EDN1, PDGFA, VEGFA, EDIL3, ANGPT1, and ANG).

Analysis of morphological features of MSC(AT) single cell suspensions derived under varying CPP conditions and/or donors revealed that cell diameter was significantly inversely correlated with inflammation-resolving MΦ PC1 composite scores ( Figure 6C , Table 2 ), suggesting that smaller cells exhibit greater immunomodulatory properties, concordant with observations by Klinker et al. (15). Correlation of MΦ PC1 composite scores with cell circularity was near-significant ( Table 2 ). Taken together, the significant and near-significant correlations between MSC(AT) gene expression and morphometric features with MΦ pro-resolving polarization constituted a matrix of multivariate readouts that were considered as putative CQAs for informing MSC(AT) immunomodulatory fitness.

Linear regression analyses were also performed to investigate correlations of MSC(AT) markers with functional angiogenic HUVEC PC1 composite scores. A statistically significant positive correlation was found between HUVEC PC1 composite scores and levels of the pro-angiogenic factor VEGF in MSC(AT) conditioned medium measured under licensed conditions (R² = 0.3048, p=0.0328) ( Figure 6D ), while VEGF levels measured under unlicensed conditions showed a positive near-significant correlation (R² = 0.2416, p=0.0743) ( Table 3 ). No significant or near-significant correlations to HUVEC PC1 scores were observed for the other measured soluble factors. Regression analyses of HUVEC PC1 scores with MSC(AT) genes revealed significant inverse correlations for SMAD7 (R² = 0.4237, p=0.0117, inhibitor of TGF-β signaling), HGF (R² = 0.3807, p=0.0187), and ANG (R² = 0.3404, p=0.0285, pro-angiogenic marker) measured under licensed conditions ( Figure 6E ). Correlations of HUVEC PC1 composite scores with expression of IDO1, TSG101, and CXCL8 were near-significant ( Table 3 ). Under unlicensed conditions, expression of the chondrogenic marker SOX9 (R² = 0. 2906, p=0.0381) was significantly inversely correlated with HUVEC PC1 scores ( Table 3 ). Inverse correlations of HUVEC PC1 composite scores with expression of TNFAIP6 were near-significant. No significant or near-significant correlations were observed between MSC(AT) cell diameter and circularity measurements with HUVEC PC1 composite scores. As above, the significant and near-significant correlations between expression levels of VEGF protein and the identified genes with HUVEC tube formation were considered as a matrix of multivariate readouts that served as putative CQAs for informing MSC(AT) angiogenic fitness.

Desirability profiling was applied as an analytical tool using the matrix of putative CQAs to empirically rank CPP conditions and MSC(AT) donors that favour immunomodulation or angiogenic fitness. Putative CQAs were selected using a broader p value threshold of p<0.1 (near-significance) based on the regression analyses presented above. This p value threshold was selected to filter the large initial panels of genes and soluble factors (curated based on literature) through a pipeline with a still flexible threshold that allowed selection of a relatively broad array of putative CQAs for MSC(AT) immunomodulatory or angiogenic basal fitness. The genes and soluble factors (measured under either licensed or unlicensed conditions), as well as morphological features were assigned minimization or maximization functions in the desirability analysis based on whether the correlation to functional outcomes was positive or negative ( Tables 2 and 3 ). For example, markers with negative correlations to inflammation-resolving MΦ or pro-angiogenic HUVEC PC1 composite scores (i.e., lower expression of the marker correlated to greater immunomodulatory/angiogenic function) were assigned minimization functions so that lower expression of the marker corresponded to a higher immunomodulatory or angiogenic desirability score. Furthermore, the R² values were used as indicators of the amount of variation in the data explained by the model for each gene/soluble factor/morphological feature in correlation with either immunomodulatory or angiogenic functional outcomes. Thus, these values were applied as individual weightings for each gene, soluble factor, or morphological feature. For example, expression of THBS1 by licensed MSC(AT) correlated most strongly with inflammation-resolving MΦ PC1 composite scores (R² = 0.5481); the R² value was used as a relative weighting such that THBS1 expression levels contributed more strongly to the overall desirability score relative to the other genes, soluble factors, and morphological features included in the analysis. Given that putative CQAs would require limits or ranges in order to be practically used, we provided a range of data outputs for each MSC(AT) readout (gene, soluble factor, and morphological feature) that correlated with immunomodulatory pro-resolving MΦ polarization or HUVEC tube formation ( Tables 2 and 3 ). These values inform the upper and lower limits of the putative CQA matrix readouts corresponding to MSC(AT) immunomodulation ( Table 2 ) or angiogenic fitness ( Table 3 ).

Using the panels of MSC(AT) genes, soluble factors, and morphological features that correlated with inflammation-resolving MΦ PC1 scores (Licensed panel: ACTA2, ANGPT1, CCN2, CXCL8, EDN1, PDCD1LG2, THBS1, TNFAIP6; Unlicensed panel: ACTA2, ANG, ANGPT1, CCN2, EDIL3, PDGFA, TGF-β, THBS1, TSG101, VEGFA; Morphological features: cell diameter, cell circularity), desirability analysis revealed that MSC(AT) cultured under 3D-N configurations had significantly higher overall immunomodulatory desirability scores relative to 2D-H and 2D-N configurations ( Figure 7A ), further corroborating that 3D-N MSC(AT) displayed augmented immunomodulatory properties. No significant differences in overall immunomodulatory desirability scores could be detected between donors.

Similar desirability analyses were performed using the angiogenic markers that correlated with angiogenic HUVEC PC1 scores (Licensed panel: SMAD7, HGF, ANG, IDO1, TSG101, CXCL8, VEGF protein; Unlicensed panel: SOX9, TNFAIP6, VEGF protein). The analysis revealed that donor differences dominated over the desirability rankings with Donors 1, 2, and 3 displaying significantly higher scores relative to Donor 4 ( Figure 7B ), further suggesting that the effect of donor predominated over variations in CPP conditions when considering MSC(AT) angiogenic functionality. Donor 5 was excluded from the desirability analysis due to an incomplete dataset. No significant differences were observed for overall angiogenic desirability scores across variations in CPP conditions.

Altogether, using the matrix of putative CQAs that included MSC(AT) genes and soluble factors, morphological features, and immunomodulatory/angiogenic functional readouts, desirability analysis allowed us to empirically rank MSC(AT) immunomodulatory and angiogenic fitness across varying CPP conditions that enhance immunomodulation and angiogenic potency, and across MSC(AT) donors. The results showed that MSC(AT) cultured under 3D-N conditions displayed the highest overall immunomodulatory ranking, while specific adipose tissue donors had highest angiogenic desirability rankings.