Designing Topographically Textured Microparticles for Induction and Modulation of Osteogenesis in Mesenchymal Stem Cell Engineering

Mesenchymal stem cells have been the focus of intense research in bone development and regeneration. We demonstrate the potential of microparticles as modulating moieties of osteogenic response by utilizing their architectural features. Topographically textured microparticles of varying microscale features were produced by exploiting phase-separation of a readily-soluble sacrificial component from polylactic acid. The influence of varying topographical features on primary human mesenchymal stem cell attachment, proliferation and markers of osteogenesis was investigated. In the absence of osteoinductive supplements, cells cultured on textured microparticles exhibited notably increased expression of osteogenic markers relative to conventional smooth microparticles. They also exhibited varying morphological, attachment and proliferation responses. Significantly altered gene expression and metabolic profiles were observed, with varying histological characteristics in vivo. This study highlights how tailoring topographical design offers cell-instructive 3D microenvironments which allow manipulation of stem cell fate by eliciting the desired downstream response without use of exogenous osteoinductive factors.


Introduction 37
Microparticles, also known as microcarriers, have gained significance as building blocks for tissue 38 engineering strategies, in bioinks for three-dimensional (3D) bioprinting and for large-scale expansion 39 of anchorage-dependent cells (1). Since they allow the formation of interconnected porous scaffolds and 40 spatiotemporal release of bioactive factors, microparticles are an attractive tool for engineering complex 41 tissues and biological interfaces. Understanding the correlation between microparticle-based cues and 42 associated cellular responses is critical to achieve predictable outputs for translational applications. 43 Tailoring surface properties of microparticles to direct differentiation in 3D provides the opportunity to 44 transform cell delivery systems from passive mechanical supports to functional components of cell 45 expansion processes and regenerative therapies. 46 Mesenchymal stem cells (MSCs), also known as multipotential stromal cells or mesenchymal stromal 47 cells, are widely used in tissue engineering due to their multi-lineage potential and immunomodulatory 48 effects (2). Since material properties influence MSC behaviour, tailoring the physical properties of 49 microparticles makes it possible to achieve the desired biological responses. Topographical properties 50 of the substrate play a key role in determining cellular responses (3), which have been shown to 51 sometimes be more dominant than biochemical cues and stiffness (4). Therefore, the aim of this study 52 was to explore the utilization of 3D architectural features of microparticles (i.e. topographical cues) for 53 osteoinduction as a potential alternative for biochemical factors. We demonstrate the potential of 54 capitalizing on cell-substrate interactions to develop cell-instructive microparticles, whereby tailored 55 microparticle design can allow control over stem cell fate. This approach will facilitate the application 56 of the 'understand-predict-control' engineering design principle within stem cell biology in a 3D 57 platform, leading to improved microscale culture protocols for drug development and therapeutic 58 applications. We have previously reported strategies to modify microparticle surfaces to improve cell 59 adhesion (5), but the use of microcarriers with controlled topographies to drive stem cell differentiation 60 has not been investigated. To test the capability of topographical designs on microparticles to induce 61 osteogenesis, a range of viability and osteo-specific gene, protein and mineralization assays were 62 performed using primary human mesenchymal stem cell (hMSCs) in vitro, and within non-healing 63 murine radial bone defects in vivo. 64 there were no significant differences in Young's moduli after incubation with cell culture media (Fig. cell attachment on microparticles at 4 hours post-seeding was significantly higher than on planar 2D 139 discs, with higher total cell attachment on dimpled microparticles relative to smooth ones (Fig 2E). Data 140 showed significantly lower normalized attachment densities to dimpled microparticles and 2D planar 141 controls relative to smooth microparticles (Fig 2F). Results were confirmed using the PrestoBlue assay, 142 a metabolic activity-based measure of cell attachment (Fig. 2G), where normalized cell attachment was 143 significantly lower on textured microparticles relative to smooth ones (Fig. 2H). 144 Cell morphology was influenced by different topographies, with cells spreading on smooth surfaces and 145 adopting more rounded or elongated morphologies on textured microparticles (Fig. 3A). Differences 146 observed in cell spreading was hypothesized to be due to preferential use of different integrins for 147 adhesion to varying topographical designs. Therefore, cell adhesion was investigated further to identify 148 which integrin(s) mediated cell-microparticle interactions. The ability of anti-integrin blocking 149 antibodies to interfere with hMSCs adherence to smooth and dimpled microparticles, along with flat 150 tissue culture-plastic (TCP) substrates, was studied. All tested anti-integrin antibodies blocked ≈75% of 151 cell adhesion to smooth microparticles and ≈60% of adhesion to TCP substrates (Fig. 3B). In contrast,152 results suggested that hMSCs attachment to dimpled microparticles was mediated primarily via α5 and 153 αvβ3 only among integrins tested, with a significant difference in hMSCs attachment to smooth and 154 dimpled microparticles with anti-α2 blocking antibodies (Fig. 3B). This demonstrates that binding to 155 topographically textured particulate surfaces significantly alters integrin binding, and therefore impacts 156 downstream signaling. 157 Cells proliferated on microparticles for 15 days, where PrestoBlue indicated consistently lower 158 metabolic activity on dimpled microparticles at day 12 relative to day 8 (Fig. 3C). This was hypothesized 159 to be due to either: (A) a decrease in cell number after day 8, which was not reflected in DNA-based 160 quantification on day 14 ( Fig. 3E) nor images of GFP-labelled hiMSCs on dimpled microparticles-161 containing discs (Fig. 3D); or (B) due to a decrease in cell metabolic activity. Cells were capable of 162 increased proliferation at later time-points (Fig. 3C), signifying a transient metabolic decrease. 163 Additionally, cell aggregation was observed on textured microparticles (Fig. 3D), while MSCs spread 164 evenly over smooth microparticle discs. DNA-based quantification of cell numbers 14 days post-seeding 165 showed similar cell numbers on all substrates, with marginally higher cell numbers on angular 166 microparticles compared to 2D discs (p=0.1; Fig. 3E). There were no significant differences in cell 167 numbers on microparticle-based substrates at day 21 ( Fig. 3F) and were generally lower on 168 microparticles than control, yet not statistically significant.
In order to delineate potential effects of any residual fusidic acid present after the fabrication process 207 from the influence of topographical cues on stem cell fate, dimpled microparticle-conditioned media 208 was employed to explore its effects on hMSCs differentiation. Two cell seeding densities were 209 investigated to account for variability in initial cell attachment on different microparticle surfaces. As 210 displayed in Supplementary Fig. 3C-D, there was no influence of the microparticles-conditioned media 211 from smooth and dimpled microparticles on OCN expression or mineralization at days 14 and 21, 212 respectively. 213

Culturing hMSCs On Topographically Textured Microparticles Significantly Alters Their Metabolic 214
Profile 215 Alterations in the genomic and proteomic landscapes are accompanied by changes in metabolic 216 pathways, making metabolomics an appropriate method to assess cellular phenotype. Extracellular 217 metabolite levels can be linked with intracellular metabolism (16). Samples of culture media were 218 analyzed by liquid chromatography (LC)-mass spectrometry (MS) at day 15 post-culture to investigate 219 adaptive metabolic changes occurring in hMSCs cultured on various microparticle designs and 2D. Data 220 was analyzed using principal component analysis (PCA) and orthogonal partial least-squares 221 discriminant analysis (OPLS-DA) to determine overall biological variation and identify differences in 222 metabolic response. To avoid limitations associated with normalization to protein content, normalization 223 was carried out to DNA quantification at the same time-point using IDEOM (17). PCA scores plot 224 displays reproducible clustering of biological replicates for each group, and highlighted distinct 225 differences in metabolic profiles of hMSCs cultured on different substrates (Fig. 5B). Unconditioned 226 fresh media samples showed distinct clustering from samples in the presence of hMSCs 227 ( Supplementary Fig. 4C). Models were evaluated using cross-validation and goodness of fit (R 2 ) and 228 predictive ability (Q 2 ) (0.98 and 0.97, respectively), indicating successful establishment of reliable 229 models. PCA showed that exposure to either smooth or textured microparticles resulted in cells 230 transitioning to a metabolic state different from that of TCP-cultured cells (Fig. 5B). Additionally, 231 dimpled microparticle samples were clearly separated from smooth microparticle samples on the OPLS-232 DA scores plot, indicating altered metabolic states of hMSCs cultured on dimpled microparticles relative 233 to smooth ones (Fig. 5C). Both microparticle types were also significantly separated from TCP-cultured 234 cells on OPLS-DA scores plots (Supplementary Fig. 4D). Mass ions with VIP scores > 1.0 were 235 considered key differential metabolites, which was combined with Students t-test with false discovery 236 rate correction to assess statistical significance. biomarkers of the influence of surface topographical design of microparticles. Metabolites including 239 stearic acid, oleic acid, L-alanine and creatinine (Supplementary Table 3) were significantly higher in 240 culture media of hMSCs on dimpled compared with smooth microparticle samples and controls. Two 241 metabolites were found at remarkably higher concentrations in the media from cells cultured on textured 242 microparticle samples relative to smooth microparticles. These were a trihydroxy epi-vitamin D3 243 derivative (194.8 and 124.1 fold higher in normalized and non-normalized data of dimpled samples 244 relative to smooth, respectively, p<0.0001) and PG (20:1(11Z)/0:0), a glycerophospholipid (LPG(20:1); 245 70.6 and 45.0 fold higher in normalized and non-normalized data of dimpled compared to smooth, 246 respectively; p<0.0001). Palmitic, stearic and oleic acids are major fatty acid components in human 247 plasma lysophosphatidic acid (LPA)(18), which were also revealed to be significantly increased in the 248 metabolic profile of dimpled media samples relative to smooth (Supplementary Table 3). ERK, 249 interleukin-1 (IL-1) and JNK pathways were found to be central within the major signaling networks 250 identified by Ingenuity Pathway Analysis (Supplementary Fig. 5). 251 As anticipated during active differentiation (19), key pathways such as aminoacyl tRNA biosynthesis, 252 amino acid metabolism and energy-based pathways were significantly upregulated overall in hMSCs 253 cultured on dimpled compared with smooth microparticles (Fig 5D,E). Ingenuity pathway analysis 254 (IPA) of canonical signalling suggested the osteo-inductive effect of dimpled microparticles may occur 255 via enhanced activity of iNOS (Fig. 5D). Metabolic pathways of the identified differentially abundant 256 metabolites between smooth and dimpled microparticles media samples were also examined using 257 MetaboAnalyst. Pathway significance was determined from pathway enrichment analysis, and impact 258 values were determined based on the influence of metabolite(s) on a pathway's function. A significantly 259 affected pathway was arginine and proline metabolism, including L-glutamate-5-semialdehyde and L-260 citrulline. Top upstream regulators, identified by IPA based on expected effects between transcriptional 261 regulators and known target genes in the Ingenuity ® database, included ZC3H10, consistent with its 262 activated state. 263

Textured Microparticle Designs Induce Distinct Bone Regeneration Patterns in Vivo 264
To evaluate the bone-forming potential of topographically-patterned microparticles in vivo, a murine 265 non-healing radial bone defect model was used (20). Polyimide sleeves, with pores along the sides, were 266 filled with either smooth, dimpled or angular microparticles in a collagen gel carrier (Fig. 6A). Low 267 elastic modulus collagen type I was used as a biocompatible carrier system. As the aim was to evaluate 268 the osteogenic reconstruction potential of microparticle design, eight weeks were deemed adequate to 269 interpret differences based on histology. eight weeks post-implantation. Quantification of bone volume by μCT ( Fig. 6B and Supplementary 274 Fig. 6A) revealed that only BMP-2-loaded collagen resulted in significant repair and bridging of the 275 defect. As expected, microparticles and collagen carrier alone did not promote the same level of bone 276 growth as the BMP-2 impregnated collagen gel, although dimpled microparticles displayed a slightly 277 higher mean bone volume relative to other designs (Fig. 6B). 278 Histological analysis at 8 weeks showed fibrous tissue with no significant bone formation within the 279 defect for all the conditions except BMP-2, where both bone formation and bone marrow establishment 280 in the center of the defect were noted, including hematopoietic elements (labelled BM; Fig. 6C) and 281 adipocytes (Fig. 6C). Higher-magnification images within the defect confirmed different tissue 282 characteristics among microparticles and the plain collagen carrier. With plain collagen, we observed 283 mostly fibrous connective tissue with patches of inflammatory granulation tissue (Fig. 6C) and almost 284 no collagen deposition in the ECM, as indicated by Picrosirius Red staining (Fig. 6D). Despite the 285 presence of tightly packed microparticles within the implant tube, cell infiltration was clearly noted 286 within the defects (yellow arrowheads; Fig. 6C). Some bone formation could be observed in defects 287 containing smooth microparticles, possibly attributable to the stiffness of PLA stimulating osteogenesis 288 to some degree (4). With dimpled and angular microparticle-containing implant tubes, cells were 289 organized into larger endosteum-like structures as in the bone marrow, displaying osteoblasts and 290 osteocytes in lacuna, and compact bone ( Fig. 6C and Supplementary Fig. 6C). Abundant collagen 291 deposition, indicated by Picrosirius Red staining, was noted in regions that exhibited new bone formation 292 ( Fig. 6D). However, there was an evident lack of hematopoietic bone marrow elements observed in the 293 defects treated with dimpled microparticles compared to smooth and angular microparticles, despite the 294 presence of supportive networks of newly-formed trabeculae for hematopoiesis to occur (Fig. 6C). 295 Alcian Blue staining (Supplementary Fig. 6B) indicated the presence of sulfated glycosaminoglycans, 296 which was especially notable within defects containing angular microparticles. Additionally, the 297 presence of a high number of well-organized blood vessels within defects containing both types of 298 textured microparticles was observed (black arrowheads in Fig. 6C and Supplementary Fig. 6C). 299 design criteria for microparticle-based environments to promote ex vivo control of cell fate for stem cell biomanufacturing and disease model development. In this work, we investigate the feasibility of 306 exploiting cell-substrate interactions to develop cell-instructive platforms for driving osteogenic 307 differentiation of hMSCs in 3D without the use of costly growth factors. We provide evidence herein of 308 the in vitro topographically-patterned 3D niche microenvironments driving MSCs to go down an 309 osteogenic lineage. Our findings indicate that MSC morphology, adhesion, proliferation, osteogenic 310 lineage specification, and metabolic profile can be regulated using microengineered substrate-based cues 311 in the form of microparticle topographical design. 312 When employing microparticles as cell carriers, cell adhesion is critical. Increased total cell attachment 313 to microparticles compared to 2D controls is expected due to their increased surface area-to volume 314 ratio. Significantly lower normalized attachment densities to dimpled microparticles relative to smooth 315 microparticles agrees with a previous study describing anti-adhesive properties of pits between 3-10 µm 316 (23), which is similar to dimple sizes described herein. Results were confirmed using a metabolic 317 activity-based measure of cell attachment, where normalized cell attachment was significantly lower on 318 textured microparticles relative to smooth ones. Metabolism has also been reported to be lower in cells 319 exposed to topographies relative to control substrates (24). 320 Morphological differences in the cultured MSCs can be explained by how the surfaces regulate 321 intracellular tension. On the smooth microparticles, the stiff, smooth surfaces will resist contractile 322 forces exerted by MSCs to cluster integrins, which facilitate actin polymerization and act as anchoring 323 points. However, direct physical interference of the micro-scale topographical patterns with the 324 establishment and maturation of focal adhesion blocks cell spreading (23). Differences observed in cell 325 morphology and spreading was also hypothesized to be due to the preferential use of different integrins 326 for adhesion to varying topographical designs. Integrins initiate numerous downstream signaling 327 pathways that regulate cellular processes, including differentiation (22). Results suggested that hMSCs 328 attachment to dimpled microparticles was mediated primarily via α5 and αvβ3 only among integrins 329 tested, with a significant difference in hMSCs attachment to smooth and dimpled microparticles with 330 anti-α2 blocking antibodies. This demonstrates that binding to topographically textured particulate 331 surfaces significantly alters integrin binding, and therefore impacts downstream signaling. Both α2 and 332 α5 integrins are mechanosensitive molecules which enhance osteogenic differentiation in response to 333 different spherical spatial boundary conditions (25). Moreover, signaling through αVβ3 integrin plays a 334 significant role in promoting osteogenesis (26). 335 The transient metabolic decrease of hMSCs cultured on the dimpled microparticles may reflect 336 osteoblastic differentiation and/or calcium internalization at that time point (27,28). The higher cell 337 numbers on angular microparticles may be due to reduced physical tension experienced by cells on the flat regions available, which may have allowed cells to grow faster compared to cells on convex surfaces (29). Cell aggregation observed during proliferation on textured microparticles may be due to varying 340 patterns of cell adhesion caused by somewhat variable dimple sizes. Some dimples may have been more 341 conducive for induction of differentiation than others (30). This may have led to creation of 342 heterogeneous microenvironments within the same sample as early as day 3. Since adhesion-based 343 signaling contributes to MSC differentiation in monolayer cultures (31), data herein suggests that 344 cellular mechanoreceptors can be exploited to induce osteogenesis in 3D systems. Osteogenic 345 differentiation is enhanced on compact 3D structures, where cells display smaller cell areas and higher 346 major axis length (3), similar to cell morphologies observed on dimpled microparticles. Additionally, 347 MSCs cultured on disordered topographies display elevated levels of osteocalcin (21). The results 348 reported herein demonstrate that hMSCs differentiated towards an osteogenic lineage on textured 349 microparticles, with varying levels of osteocalcin expression and mineralization. In the absence of 350 exogenous osteo-inductive supplements, dimpled microparticles successfully induced osteogenic 351 differentiation of hMSCs in basal media, as indicated by the significant increase in levels of OCN 352 expression after 2 and 3 weeks of culture. It is noteworthy that smooth microparticles showed slightly 353 higher expression of osteocalcin than flat 2D discs, which correlates with reports on the ability of curved 354 surfaces to promote osteogenesis (14). Mineralization was notably higher on textured microparticles. 355 Interestingly, markedly higher mineralization was observed on angular microparticles, despite lower 356 levels of OCN expression. This can be correlated to previous studies in vivo, where osteocalcin-null 357 mice were reported to exhibit increased mineral-to-matrix ratio (32), while mice overexpressing 358 osteocalcin showed normal mineralization (33). 359 Differential gene expression between hMSCs cultured on TCP, smooth and dimpled microparticles was 360 identified. Genes for COL10A1 were differentially upregulated in dimpled microparticle samples 361 relative to TCP. COL10 is associated with endochondral ossification and directing bone mineralization 362 (34). Cells cultured on smooth microparticles displayed an upregulation of BGN, which regulates 363 collagen fibrillogenesis (35). TGFBR2, encoding for transforming growth factor-beta (TGF-β) receptor 364 type 2, was upregulated in cells cultured on both smooth and dimpled microparticles, suggesting the 365 possibility of stiff, curved surfaces priming cells to be more responsive to TGF-β regulation of 366 osteogenesis. A change in expression of several osteogenesis-related genes were notably different on 367 dimpled versus smooth microparticles. Altered genes included COL4A3, which encodes for a type IV 368 collagen component making up epithelial and endothelial cell basement membranes (36). 369 Downregulated genes in MSCs cultured on dimpled microparticles included EGF, which promotes ex 370 vivo expansion of MSCs without triggering differentiation, maintaining its stemness (37), and GDF10, also downregulated in cells cultured on dimpled samples only, which may be related to its recently analyses suggest that hMSCs are both a producer and target of vitamin D3 metabolites when cultured on 376 dimpled particulate surfaces, which suggests a potential role of vitamin D metabolism in topographically 377 induced differentiation. While some mammalian cells can make their own 1,25-dihydroxyvitamin D 378 (1,25(OH)2D3) and respond to it (40) (47). Carnitine has also been linked to osteogenic differentiation 398 (48). Ketone body metabolism is a significant contributor to mammalian energy metabolism within 399 extrahepatic tissues when glucose is not readily available (49). Top upstream regulators identified by 400 IPA included ZC3H10, consistent with its activated state. ZC3H10 is a poorly characterized RNA-401 binding protein which regulates mitochondrial biogenesis (50), which correlates with the importance of 402 mitochondrial dynamics in osteogenesis (51). More studies are needed to reveal how it regulates investigate the contribution of topographically textured microparticles to bone regeneration without 408 incorporation of any exogenous variables, such as cells or growth factors. With textured microparticle-409 containing implant tubes, cells were organized into larger endosteum-like structures and abundant 410 collagen deposition. However, there was an evident lack of hematopoietic bone marrow elements 411 observed in the defects treated with dimpled microparticles compared to smooth and angular 412 microparticles, despite the presence of supportive networks of newly formed trabeculae to support 413 hematopoiesis. We hypothesize that incorporating low quantities of growth factors within the 414 microparticles will promote synergistic effects of topographical design and microparticle-mediated 415 growth factor presentation (20), which will be investigated in future studies. This confirms that the 416 appropriate microparticle design can maximize bone regeneration potential in clinical applications and 417 is vital to support optimal bone growth. This should be chosen based on its ability to support new bone 418 formation, alongside promotion of hematopoiesis and bone marrow restoration. 419 Conclusions 420 This study offers novel insights into the substantial impact of topographical patterning of microparticles 421 on hMSCs response. We demonstrate herein how hMSCs morphology, adhesion, proliferation, 422 osteogenic lineage specification, metabolic profile and in vivo histological characteristics can be 423 modulated using microparticle-based 3D topographical design. Such micro-scale control represents a 424 crucial advance in our ability to study and regulate stem cell-microenvironment interactions in 3D. 425 Designing cell-instructive microparticles with defined topographies provides a promising platform for 426 investigating how microenvironmental cues modulate the development of 3D-engineered bones and to 427 create in vitro high-throughput drug screening array systems. Combining such platforms into next-428 generation organoid and niche bioengineering protocols will enable improved system control and 429 increase the ability to probe the multivariable complexity of biological systems. Overlaying 430 microparticle-based bioengineering techniques to pattern micro-tissues and develop 'multiplexed' 431 organoids can improve fidelity of developmental models, acting as mechanical cue providers to emulate 432 the native cellular microenvironment for the development of in vitro models for basic and translational 433 stem cell research. Moreover, this study provides key insights into the design of instructive, injectable 434 scaffolds for bone regenerative repair. 435 containing 1g of PLA in 5 mL of dichloromethane ((20% w/v;DCM;≥99.8,Fisher Scientific)  Sigma-Aldrich) as stabilizer. The resulting emulsion was stirred continuously at 500 rpm at room 443 temperature for at least 4 hours to allow for solvent evaporation. To remove residual PVA, microparticles 444 were centrifuged at 4500 rpm for 5 min and subsequently washed with deionized (DI) water three times. 445 Cell strainers with pore sizes of 40 and 100 µm were used to separate microparticles in this size range 446 and the collected microparticles were freeze-dried for storage at -20°C. 447 To produce textured microparticles, fusidic acid (FA; 98%, Acros Organics or F0756-#SLBN8134V, 448 Sigma-Aldrich) was incorporated in the organic phase as previously described (52)  Poietics™ protocols. 520

Assessment of Cell Attachment and Metabolic Activity 521
Cells were seeded on sintered discs in cell-repellent CellStar® 96-well plates (Greiner Bio-One) at 3×10 4 522 cells/well and incubated for 4 hours. Discs were then washed gently with PBS to remove non-adherent 523 cells. DNA-based quantification of cell attachment was determined by CyQUANT NF Cell Proliferation 524 Assay (ThermoFisher Scientific, UK) using a Tecan Infinite M200 microplate reader (Tecan, UK), with 525 λexc/λem 485/535 nm. DNA content was correlated to the number of cells through the use of a reference 526 standard curve. hMSCs cultured on flat and topographically-textured microparticle discs were also 527 assessed for metabolic activity using PrestoBlue TM (Invitrogen, UK) at 4 hours post-seeding (for 528 assessment of initial cell attachment; seeded at 3×10 4 cells/well) and after 1, 5, 8, 12 and 15 days (to 529 assess proliferation; seeded at 1×10 4 cells/well), according to the manufacturer's protocol. Briefly, 530 culture medium was replaced by PrestoBlue™:culture medium (1:9) and incubated in the dark for 531 1.5 hours at 37 °C. Duplicate 100 µL aliquots of supernatant were assessed for fluorescence at at λexc/λem 532 560/590 nm. DNA quantification of the lysates collected at days 14 and 21 was carried out using the 533 Quant-iT™ PicoGreen ® dsDNA assay (Invitrogen), and concentrations were calculated using a standard 534 curve generated from DNA standard provided. Proliferation was also observed on the discs using 535 immortalized human mesenchymal stem cells (hiMSCs) and imaged using a Nikon SMZ1500 dissection 536 microscope.
(MAB1987, Merck), anti-α5 (MAB1956Z, Merck), anti-αvβ3 (14-0519, eBioScience) integrin antibodies 540 or anti-mouse isotype IgG antibody (M6898, Sigma-Aldrich) for 30 min at 37°C, prior to seeding in 96-541 well plates containing discs or control wells pre-coated with 3% (v/v) BSA in PBS. Cell suspensions of 542 4.5×10 5 cells/mL were used. After incubating for 1 hour at 37°C, non-adherent cells were removed by 543 gently rinsing with PBS several times. The 1 hour period for evaluating cell adhesion was selected to 544 avoid cells secreting a significant amount of matrix nor significantly change their integrin expression 545 profile while adherent(55). CyQUANT® NF Cell Proliferation assay was used for cell quantification, 546 expressed as the proportion of cells attached relative to cell numbers attached in the isotype IgG control 547 after subtraction of background values of blank discs. 548

Determination of Osteogenesis 549
For differentiation experiments, 3250 cells were seeded per disc, which corresponds to a range of 2750-550 10,000 cells/cm 2 . 551

Quantitative Determination of Alkaline Phosphatase Levels 552
The Alkaline Phosphatase Detection Kit Fluorescence (Sigma-Aldrich, UK) was used according to the 553 manufacturer's protocol. Cultured cells, 7 days post-seeding, were lysed using 1% Triton X-100 for 20 554 minutes, followed by three freeze-thaw lysis steps and centrifugation. Samples were incubated with the 555 non-fluorescent 4-methylumbelliferyl phosphate disodium salt as substrate, and resultant fluorescence 556 was measured using a plate reader. ALP activity was normalized to total DNA content, determined using 557 the Quant-iT PicoGreen dsDNA Assay Kit (Invitrogen, UK). 558

Assessment of Osteocalcin Expression 559
At day 14, hMSCs were rinsed with warm PBS and fixed with 3.7% (w/v) paraformaldehyde in 560 deionized water for 40 min. Cells were permeabilized using 0.1% (w/v) Triton-X 100 in PBS (Sigma-561 Aldrich) for 30 min. Non-specific binding sites were blocked by incubation in 10% (v/v) normal donkey 562 serum (D9663; Sigma-Aldrich) and 1% (v/v) bovine serum albumin (BSA) in PBS for 1 h. Cells were 563 then incubated with anti-Osteocalcin antibody (AB10911; 1:250; Merck Millipore) overnight at 4°C. 564 Donkey anti-rabbit IgG-FITC secondary antibody (1:500; Invitrogen) was added for 2 h. Samples were 565 counterstained with NucBlue® Fixed Cell ReadyProbes (Thermo-Fisher) and visualized using the 566 confocal unit LSM 780 of the Zeiss Elyra PS1 microscope. Quantification of osteocalcin in conditioned 567 media was carried out using the Human Osteocalcin Instant ELISA Kit (eBioscience) at the indicated times according to the manufacturer's instructions. Data analysis was performed using the OsteoImage assay (Lonza) was used after 3 weeks of culture. Mineralization was measured 572 quantitatively by spectrophotometer at 492 nm excitation and 520 nm emission wavelengths. Results 573 were normalized to DNA content, quantified using Quant-iT PicoGreen assay (Thermo Fisher 574 Scientific). Presence of extracellular calcium deposits was also verified using von Kossa staining (silver 575 nitrate solution with exposure to UV for 60 min then sodium thiosulfate solution; Sigma Aldrich) and 576

SEM. 577
Gene Expression Array 578 Total RNA was extracted from hMSCs cultured in basal culture medium for 2 weeks using 579 Samples were sequentially dehydrated with ethanol, mounted onto aluminum stubs and sputter-coated 591 with gold for 180 s prior to imaging on a JEOL 6060LV variable pressure scanning electron microscope 592 (Jeol UK Ltd.) at 10kV. 593

Microparticle-conditioned media 594
To prepare microparticle-conditioned media, culture media was incubated with either smooth or dimpled 595 microparticle discs for 14 or 21 days (same length of time used to study OCN expression and 596 mineralization, respectively). This media was periodically and immediately (without storage) transferred 597 to cells cultured on TCP after filtration through a cell strainer (pore size 40 µm), to observe the influence 598 of any residual fusidic acid on differentiation.
to pre-cooled tubes and stored at −80 °C. Fresh culture medium samples and methanol samples were 604 processed in parallel as no-cell controls. Samples were prepared as six biological replicates (two donors) 605 and the experiment run in two independent repeats. An equal mixture of all samples was prepared as a 606 pooled quality control for assessment of instrument performance(56). For metabolite foot-printing, LC-607 MS was performed on an Accela LC system coupled to an Exactive MS (ThermoFisher Scientific, UK), 608 as previously described.(57) Spectral data was acquired in full scan ion mode (m/z 70-1400, resolution 609 50,000) in positive and negative electrospray ionization modes. Probe temperature and capillary 610 temperature were kept at 150 and 275 °C, respectively. Chromatographic separation was carried out 611 using a ZIC-pHILIC (5μm column, 150 mm × 4.6 mm, Merck Sequant) maintained at 45°C and flow 612 rate of 300 μLmin −1 as previously described.(58) Mobile phase consisted of (A) 20 mM ammonium 613 carbonate in water and (B) 100% acetonitrile, eluted with a linear gradient from 80% B to 5% B over 15 614 min, followed by a 2 min linear gradient from 5% B to 80% B, and 7 min re-equilibration with 80% B 615 (injection volume= 10 μL; 4°C). LC-MS data were processed using XCMS for untargeted peak-616 picking(59), and peak matching was carried out using mzMatch.(60) IDEOM (v20) was used for putative 617 metabolite identification.(61) Level 1 metabolite identification was performed by matching accurate 618 masses and retention times of 268 authentic standards, which were analyzed using the same analytical 619 conditions according to the metabolomics standards initiative(62, 63). Level 2 putative identification 620 was considered when standards were unavailable and predicted retention times were used as an 621 orthogonal means to improve metabolite identification. Supervised orthogonal partial least squares 622 discriminant analysis (OPLS-DA) was initially performed using SIMCA-P version 13.0.2 (Umetrics AB, 623 Sweden) for general visualization of metabolite differences and to observe differences between control 624 and treated samples. Quality of the models was evaluated based on R 2 (goodness-of-fit) and Q 2 625 (predictive ability). For detection of key discriminatory metabolites, mass ions were selected by variable 626 importance in projection (VIP) values, where VIP >1 were considered potential biomarkers. Univariate 627 analysis was used as a final feature selection. Student t-test with false discovery rate (FDR) correction 628 was carried out using MetaboAnalyst(64) to evaluate levels of significant differences between controls 629 and microparticle-cultured media, as well as pathway analyses. Analyses were also carried out using 630 acid) with 3 mL ice-cold 0.1 M sodium hydroxide and 0.5 mL 10x DMEM. Microparticles were 637 sterilized by ultraviolet radiation. Microparticles were prepared in collagen solution at a final 638 concentration of 620 mg/mL, and 3 µL of the microparticle suspensions were placed into the implant 639 tubes. Collagen gel loaded with BMP-2 (75 µg/mL, R&D Systems) was used as a positive control, while 640 plain collagen was used as a negative control. All materials were prepared the day before surgeries. 641 Experiments were performed under a project license issued under the ASPA (Animals Scientific 642 Procedures Act 1986) by the Home Office, UK. C57B1/6J male mice (8 to 10 weeks old; Charles River 643 laboratories) were anesthetized under isoflurane, and the right forelimb was shaved and swabbed with 644 isopropyl alcohol and iodine solution. After induction of anesthesia, mice were provided with 645 buprenorphine and carprofen administrated subcutaneously. A skin incision was made along the 646 forearm, and muscle tissue over the radius was blunt-dissected. A 2.5-mm defect was created in the 647 center of the radius using a double-bladed bone cutter. Implant tubes were placed into the defect by 648 fitting it at the proximal and distal ends of the radial defect, and the incision was then closed with 649 degradable Vicryl ® suture. Mice were monitored after surgery for signs of distress, movement and 650 weight loss. At the end of the experiment (8 weeks), mice were euthanized, and radial bones were 651 explanted and fixed in 10% neutral-buffered formalin solution. 652 High-resolution micro-X-ray-computed tomography system (micro-CT, Skyscan 1174) was used to 653 determine bone ingrowth within the defects. Fixed samples were scanned at a voltage of 50 kV, current 654 of 800 μA, and a voxel resolution of 9.86 μm. A 0.5mm aluminium filer was also applied. Transmission 655 images were reconstructed using Skyscan supplied software (NRecon) with the resulting two-656 dimensional image representing a single 9.86 μm slice. Quantitative analysis of bone ingrowth was 657 obtained using direct morphometry calculations in the Skyscan CTAn software package. 658 Bone samples from all treatment and control groups (n=5 each, and two samples with no defect) were 659 decalcified using 5% (v/v) formic acid and dehydrated in progressively higher concentrations