Scaffold-free human mesenchymal stem cell construct geometry regulates long bone regeneration

Biomimetic bone tissue engineering strategies partially recapitulate development. We recently showed functional restoration of femoral defects using scaffold-free human mesenchymal stem cell (hMSC) condensates featuring localized morphogen presentation with delayed in vivo mechanical loading. Possible effects of construct geometry on healing outcome remain unclear. Here, we hypothesized that localized presentation of transforming growth factor (TGF)-β1 and bone morphogenetic protein (BMP)-2 to engineered hMSC tubes mimicking femoral diaphyses induces endochondral ossification, and that TGF-β1 + BMP-2-presenting hMSC tubes enhance defect healing with delayed in vivo loading vs. loosely packed hMSC sheets. Localized morphogen presentation stimulated chondrogenic priming/endochondral differentiation in vitro. Subcutaneously, hMSC tubes formed cartilage templates that underwent bony remodeling. Orthotopically, hMSC tubes stimulated more robust endochondral defect healing vs. hMSC sheets. Tissue resembling normal growth plate was observed with negligible ectopic bone. This study demonstrates interactions between hMSC condensation geometry, morphogen bioavailability, and mechanical cues to recapitulate development for biomimetic bone tissue engineering.

L ong bones of the appendicular skeleton such as the femur provide load-bearing structure and carry out critical metabolic functions 1 . Mesenchymal cell condensation in the early limb bud initiates long bone morphogenesis to form the cartilage anlage that gives rise to endochondral bone formation 2,3 . This process is dependent on local morphogen gradients and mechanical forces in utero (reviewed in ref. 4 ). Postnatal bone retains the remarkable ability to heal via endochondral ossification without fibrous scarring [5][6][7] ; however, perturbations of the fracture site can interfere with the repair process when bone defects reach a critical size, which can contribute to non-union 8 .
Traditional scaffold-based bone tissue engineering approaches aim to create implantable constructs through a combination of biocompatible scaffolds, regeneration-competent cells, and/or bioactive cues 9 . Owing to the anatomy of the femoral diaphysis, studies have explored pre-formed tubular constructs in subcutaneous and orthotopic environments. Natural coral [10][11][12] and synthetic polymers 13,14 have been investigated as tubular scaffold materials for mesenchymal stem cells (MSCs). Furthermore, cellfree tubular composites of synthetic polymers and bone morphogenetic protein (BMP)-2 or BMP-7 [15][16][17][18] have been tested. Limitations of such strategies include (i) the use of scaffolds designed to match the properties of mature rather than developing tissues, (ii) the necessity to pre-culture MSCs for 2-3 weeks in an induction medium to stimulate osteogenic pre-differentiation, (iii) the delivery of a single morphogen, (iv) scaffold interference with critical cell-cell interactions, (v) scaffold degradation rate that is not in sync with new tissue formation, and (vi) potential immunogenicity or the cytotoxicity of the scaffold or its degradation byproducts.
Biomimetic bone tissue engineering approaches that recapitulate normal tissue development and remodeling have gained considerable interest as of late [19][20][21] . Scaffold-free cartilage templates derived from self-assembled human MSC (hMSC) condensations have been shown to progress through in vivo endochondral ossification, contingent on in vitro morphogen priming [22][23][24][25][26][27][28] . Limitations of this approach include (i) the necessity to pre-culture hMSC condensations for ≥3 weeks in an induction medium supplied with morphogens (e.g., transforming growth factor (TGF)-β1) to stimulate cartilage template formation, and (ii) and the inability to generate tubular tissue constructs. To that end, we recently reported a modular, scalable, scaffold-free system for engineering hMSC rings that can be assembled and fused into tubular structures [29][30][31] . The incorporation of TGF-β1-presenting gelatin microspheres for in situ chondrogenic priming 32 and BMP-2-presenting mineral-coated hydroxyapatite microparticles to induce bony remodeling of the cartilaginous template 33 circumvents the need for lengthy predifferentiation and enables early in vivo implantation. Using this approach, we recently demonstrated functional femoral bone defect healing via endochondral ossification with microparticlecontaining hMSC condensate sheets for a localized presentation of TGF-β1 34 or TGF-β1 + BMP-2 35 contingent on in vivo mechanical loading. Such high-density hMSC sheets are produced by seeding the cell-microparticle suspension atop standard Transwell membrane inserts. The cells coalesce through cell-cell interactions within 2 d of culture to yield a multi cell-layer condensate sheet with uniformly incorporated microparticles that can be lifted from the Transwell with ease and used for long bone defect healing applications. However, the effects of implant geometry-mimicking the geometry of long bone diaphyses with a channel for nutrient transport and endogenous cell invasion-on the healing outcome have not been assessed thus far.
In this study, we aimed to recapitulate the cellular, biochemical, and mechanical environment of the endochondral ossification process during early limb development for large long bone defect regeneration. We tested the hypotheses that (i) microparticlemediated in situ presentation of TGF-β1 + BMP-2 to engineered hMSC condensate tubes induces the endochondral cascade in vitro and in a subcutaneous environment in vivo to a greater degree than the presentation of either morphogen alone, and that (ii) the geometry of TGF-β1 + BMP-2-presenting hMSC tubes, stabilized with custom compliant fixation plates that allow delayed initiation of in vivo loading, facilitate enhanced endochondral healing of critical-sized femoral segmental defects compared to loosely packed hMSC sheets with identical cell number and morphogen dosing.

In vitro maturation of engineered tubular condensations.
Endochondral bone formation is initiated by condensation and chondrogenic lineage commitment of mesenchymal cells. Therefore, we seeded a suspension of hMSCs and microparticles to locally present (1) TGF-β1, (2) BMP-2, or (3) TGF-β1 + BMP-2 from within the maturing construct in custom agarose molds to facilitate tissue ring self-assembly by day 2, as described recently 30 . Tubular hMSC constructs were then created by placing several ring building blocks in direct contact with each other on horizontal glass tubes for an additional 6 d to facilitate tissue fusion (Supplementary Fig. S1 and Fig. 1a). We first investigated the effects of in situ morphogen presentation on hMSC tube lineage commitment at day 8 of culture. No gross morphological differences between groups were observed at this time point (Fig. 2a). Transcript analysis of key differentiation markers, normalized to controls without growth factors, revealed only minimally elevated mRNA expression of the chondrogenic genes sex-determining region Y-box 9 (SOX9), aggrecan (ACAN), and collagen type 2A1 (COL2A1), and the early osteogenic gene alkaline phosphatase (ALP) in BMP-2-loaded hMSC tubes compared to TGF-β1-presenting constructs (Fig. 2b). In contrast, TGF-β1 + BMP-2 dual presentation significantly potentiated SOX9, ACAN, COL2A1, and ALP mRNA levels compared to both TGF-β1-or BMP-2-only constructs (Fig. 2b). No differences in Runt-related transcription factor-2 (RUNX2) and collagen type 1A1 (COL1A1) expression were observed across groups (Fig. 2b). Immunoblot analysis showed SMAD3 phosphorylation in TGF-β1-loaded hMSC tubes; p-SMAD5 was not induced similar to controls without growth factor (Fig. 2c-e). In contrast, BMP-2 presentation induced significant SMAD3 phosphorylation and notable SMAD5 phosphorylation compared to TGF-β1-loaded hMSC tubes (Fig. 2c-e). While TGF-β1 + BMP-2 co-presentation did not further augment SMAD3 phosphorylation vs. BMP-2loaded hMSC tubes, SMAD5 phosphorylation was significantly potentiated compared to either TGF-β1-or BMP-2-only constructs (Fig. 2c-e and Supplementary Fig. S10a). Histologically, engineered hMSC tubes displayed comparable 3-dimensional cellular organization across groups with relatively evenly distributed gelatin microspheres and no substantial glycosaminoglycan (GAG) or mineral deposition. Of note, the relatively proportionally dispersed Alizarin Red-staining observed in all groups was localized to incorporated mineral-coated hydroxyapatite microparticles ( Fig. 2f and Supplementary  Fig. S2a-c). Together, these findings suggest that while hMSC tubes across all groups were phenotypically undifferentiated at day 8 of culture, in situ presentation of TGF-β1, BMP-2, or TGF-β1 + BMP-2 imparted robust chondrogenic lineage priming in a morphogen-dependent manner.
Next, we assessed the effects of microparticle-mediated TGF-β1, BMP-2, or TGF-β1 + BMP-2 presentation on stimulating hMSC tube differentiation. Constructs were cultured for 2 weeks in basal medium followed by 3 weeks in osteogenic medium (Supplementary Fig. S1 and Fig. 1a), conditions previously shown to facilitate robust chondrogenic and osteogenic fate specification in high-density hMSC constructs 30,36,37 . Gross morphological evaluation revealed comparable width and height of hMSC tubes across groups at 5 weeks (Fig. 3a-c). Histologically, engineered hMSC tubes displayed distinct patterns of tissue differentiation. While all constructs were relatively similar in size and cellularity across groups, presentation of TGF-β1 induced more robust and homogenous GAG deposition, exhibiting cells with chondrocytelike morphology in Safranin-O-stained regions, compared to BMP-2-or TGF-β1 + BMP-2-loaded constructs (Fig. 3d, e and Supplementary Fig. S3m, n). Limited mineral deposition was observed with TGF-β1 presentation alone, similar to day 8 constructs. Conversely, BMP-2 and TGF-β1 + BMP-2 presentation induced substantial new mineral deposition by week 5 compared to TGF-β1-only hMSC tubes, consistent with the biochemical assessment (Fig. 3d, e and Supplementary Fig. S3m, n). Immunohistochemistry revealed robust Col II and Col X staining across groups, indicating the presence of both mature and hypertrophic chondrocytes. In contrast, Col I staining was only observed in BMP-2 and TGF-β1 + BMP-2-loaded hMSC tubes ( Supplementary Fig. S4a-d). Tissue tubes using hMSCs derived from three different donors were prepared for biochemical analysis. The data showed significantly higher GAG/DNA and absolute GAG content with TGF-β1 presentation compared to BMP-2-or TGF-β1 + BMP-2-loaded constructs (Fig. 3f, j and Supplementary Fig. S3a, e, i). Conversely, BMP-2 and TGF-β1 + BMP-2 presentation promoted significantly higher Ca 2+ /DNA and BMP-2-only presentation promoted significantly greater total Ca 2+ content relative to TGF-β1-only hMSC tubes (Fig. 3g, k and Supplementary Fig. S3b, f, j). ALP/DNA and absolute ALP activity were elevated with BMP-2 presentation, reaching significance with dual morphogen delivery vs. TGF-β1 (Fig. 3h, l and Supplementary Fig. S3c, g, k). No differences in DNA content were noted across groups ( Fig. 3i and Supplementary  Fig. S3d, h, l). Together, these findings suggest that in situ presentation of TGF-β1, BMP-2, or TGF-β1 + BMP-2 stimulated chondrogenesis, chondrocyte hypertrophy, and osteogenesis indicative of endochondral ossification of hMSC tubes by week 5 in vitro in a morphogen-dependent manner.
with osteoid occupying the transitional region, suggestive of endochondral ossification ( Fig. 5a and Supplementary Fig. S5d). Nearly identical tissue formation patterns in a morphogendependent manner were observed at week 6 ( Fig. 5b and Supplementary Fig. S5e). Across groups and time points, new tissue was largely comprised of human cells, as evidenced by in situ hybridization for human Alu repeats, while a mix of human and mouse cells contributed to the fibrous tissue infiltrating portions of the luminal space as well as surrounding the hMSC tube constructs (Fig. 5c, d). Immunohistochemistry corroborated the histological findings at 3 and 6 weeks. Col II and Col X staining as a proxy for identifying mature and hypertrophic chondrocytes, respectively, was most intense in areas with prominent GAG matrix, whereas Col I staining was strongest in regions with extensive mineral deposition (Supplementary Fig. S6a-d). Together, these findings suggest that localized presentation of TGF-β1, BMP-2, or TGF-β1 + BMP-2 stimulated in vivo cartilage and bone formation in hMSC tubes in a morphogen-dependent manner that was guided by the implant architecture. Importantly, TGF-β1 + BMP-2-loaded constructs formed a robust cartilaginous template that was actively remodeled into mineralized trabecular bone tissue through endochondral ossification in a subcutaneous environment. The presence of cortical bone-like compartments further suggests engagement of intramembranous ossification. Therefore, for subsequent studies, we focused on dual morphogen presentation.
In vivo performance of engineered tubular condensationsorthotopic implantation. To evaluate the role of engineered condensation geometry on the induction of orthotopic bone regeneration, day-8 TGF-β1 + BMP-2-loaded hMSC tubes were implanted in critical-size femoral segmental defects in 12-weekold athymic nude rats (Fig. 1c). The hMSC tubes' regenerative potential was compared to that of day-2 TGF-β1 + BMP-2-loaded hMSC sheets comprising matching cell numbers and morphogen-loaded microparticles. We previously reported successful calvarial 37 and femoral segmental 35 defect healing with dual morphogen-loaded hMSC sheets. For calvarial bone defects, the soft hMSC sheets can be implanted directly in the wound site without the need for additional supportive materials to keep the constructs in place. For femoral bone defects, however, the hMSC sheets are placed into thin, perforated, electrospun polycaprolactone (PCL) nanofiber mesh tubes 17,38,39 (Supplementary Fig. S7a) to maintain construct shape and location and aid in handling/implantation, all while not hindering the constructs' regenerative capacities. Femora were stabilized with custom fixation plates that permit dynamic tuning of plate compliance in vivo 34,35,38,40,41 . Similar to our prior studies, plates were initially implanted in a locked configuration (i.e., 0-4 weeks) for stable fixation, and then they were unlocked at week 4 to initiate ambulatory load transfer (i.e., 4-12 weeks) ( Supplementary  Fig. S7b), recently shown to enhance the restoration of bone function using growth factor-loaded hMSC sheets 34,35 .
Recognizing the differences in pre-culture times of the TGF-β1 + BMP-2-loaded condensations prior to implantation, 8 days for hMSC tubes and 2 days for hMSC sheets, we evaluated one randomly selected construct per group prior to implantation. Histologically, hMSC tubes (~550 µm wall thickness) appeared denser with noticeable GAG deposition in the outermost tissue layer, but cells did not display mature chondrocyte morphology ( Supplementary Fig. S8a, b). In contrast, hMSC sheets (~250 µm thickness) were more loosely packed within the PCL nanofiber mesh tube and showed no evidence of GAG deposition. In both groups, ARS staining was comparable and limited to incorporated mineral-coated hydroxyapatite microparticles. In situ hybridization for human Alu repeats confirmed the cells' identity ( Supplementary Fig. S8a, b). Immunoblot analysis revealed robust SMAD3 and SMAD5 phosphorylation in both groups, with significantly higher levels in hMSC tubes vs. sheets (Supplementary Figs. S8c-e and S10b).
Longitudinal in vivo microCT analysis showed increased bone formation with both TGF-β1 + BMP-2-loaded hMSC tubes and sheets upon implantation in femoral segmental defects over time. Bone volume levels at week 4 were comparable between groups; however, hMSC tubes continued to stimulate bone formation in a near-linear fashion upon load initiation at this time point through week 12, reaching significance compared to week 4 ( Fig. 6a, d). In contrast, the rate of hMSC sheet-induced bone formation noticeably decreased after week 4 (Fig. 6a, d). Coinciding with plate unlocking, bone volume accumulation rate with hMSC tubes markedly increased between weeks 4 and 8; opposite trends were noted with hMSC sheets. By week 12, bone volume accumulation rates were comparable between groups (Fig. 6b). None of the defects achieved bridging, defined as mineralized tissue fully traversing the defect, at week 4. Fifty percent of defects implanted with TGF-β1 + BMP-2-loaded hMSC tubes were bridged by week 8, whereas hMSC sheets yielded only 33% bridging. By week 12, defect bridging reached 75% with hMSC tubes exhibiting an increasing trend over time, whereas levels remained unchanged with hMSC sheets (Fig. 6c).
High-resolution ex vivo microCT analysis at 12 weeks demonstrated relatively uniform proximal-to-distal bone distribution in defects treated with either hMSC tubes or sheets with new bone largely confined to the 5-mm defect diameter (Fig. 7a, b and Supplementary Movie S3). TGF-β1 + BMP-2-loaded hMSC tubes induced a greater average bone volume fraction compared to hMSC sheets (Fig. 7c), consistent with the longitudinal in vivo assessment. Similarly, while not reaching significance, hMSC tube-treated defects displayed increased average trabecular number and connectivity density vs. hMSC sheets ( Fig. 7d and Supplementary Fig. S9a). The expected inverse relationship was observed for trabecular separation, while trabecular thickness and degree of anisotropy were equivalent across groups (Fig. 7e, f and Supplementary Fig. S9b). MicroCT-based assessment of structural properties at 12 weeks showed enhanced mean and minimum polar moment of inertia (pMOI) with hMSC tubes compared to hMSC sheets ( Supplementary Fig. S9c, d). Of note, both hMSC tubes and sheets induced negligible ectopic bone formation (i.e., bone extending beyond the 5-mm defect diameter; Fig. 7g), with overall levels markedly lower compared to those recently obtained with BMP-2 delivered on collagen 35 , suggesting a positive safety profile of BMP-2 delivered from scaffold-free, self-assembled hMSC condensations.
To assess whether in situ presentation of TGF-β1 + BMP-2 to engineered hMSC constructs facilitated endochondral regeneration, we performed histological analyses of the defect tissue at week 12. Both hMSC tubes and sheets induced robust ossification; well-defined trabeculae containing lacunae-embedded osteocytes were found along with cartilaginous structures at different stages of remodeling ( Fig. 7g and Supplementary Fig. S9e). Engineered hMSC tubes promoted the formation of growth plate-like, transverse cartilage bands on both the proximal and distal sides connecting the newly formed tissue with the intact bone ends. These remarkable structures exhibited zonal organization of both mature and hypertrophic chondrocytes and were aligned along the principal ambulatory load axis. A prominent GAG matrix was found embedded in trabecular bone with osteoid occupying the transitional region ( Fig. 7g and Supplementary Fig. S9e), indicative of endochondral ossification. No such structures were evident in any of the hMSC sheet-implanted defects. Regenerate bone tissue in both groups at 12 weeks was composed of rat cells and, as demonstrated by in situ hybridization for human Alu repeats, remaining human cells (Fig. 7i). Immunohistochemistry indicated the most intense Col II and Col X staining for mature and hypertrophic chondrocytes, respectively, in areas exhibiting robust GAG matrix ( Supplementary Fig. S9f). Lastly, quantitative 2-D histomorphometry of new bone tissue within the femoral defects revealed significantly greater bone area fraction with hMSC tubes compared to hMSC sheets (Fig. 8).
Together, these findings suggest that localized presentation of TGF-β1 + BMP-2 to hMSC condensations stimulated in vivo cartilage and bone formation in an orthotopic environment, independent of implant geometry. However, defined tubular hMSC condensations, exhibiting a greater degree of chondrogenic priming at the time of implantation, facilitated more robust healing compared to loosely packed hMSC sheets with bony bridging achieved in the majority of critical-size defects. New bone was formed through endochondral ossification in both groups; however, only hMSC tubes induced regenerate tissue that was remarkably similar to that observed with hMSC condensations containing 50% more cells and morphogens 35 resembling the normal growth plate architecture.

Discussion
In this study, we report the capacity of scaffold-free hMSC tubes to generate bone tissue in vivo through endochondral ossification after only 8 days of pre-culture necessary to allow for tube formation by fusion of individual hMSC rings, and the ability to heal critical-sized rat femoral segmental defects. We employed a biomimetic bone tissue engineering strategy 19 to partially recreate the cellular, biochemical, and mechanical environment of the endochondral ossification process during long bone development. Specifically, we utilized (i) cellular self-assembly to form tubular mesenchymal condensations, (ii) microparticle-mediated TGF-β1 + BMP-2 presentation to activate specific morphogenetic pathways in situ, and (iii) delayed in vivo mechanical loading to augment defect healing. As such, our approach produced a selforganized tubular construct capable of autonomously progressing through formation, maturation, and differentiation into a structured tissue in a manner comparable to normal embryonic development, a concept referred to as developmental engineering 23,42 , while approximating the geometry of the adult femoral diaphysis with a channel for nutrient transport and endogenous cell invasion.
Following a similar biomimetic strategy, others have used scaffold-free, self-assembled hMSC condensations to form cartilage templates that can undergo hypertrophy and progress through endochondral ossification in vivo to form ossicles in a subcutaneous environment, but only after lengthy TGF-βmediated chondrogenic priming in vitro [22][23][24][25][26][27] . The first study that investigated orthotopic implantation of scaffold-free hMSC condensations and showed successful healing of rat femoral segmental bone defects stabilized with internal fixation plates 28 also required lengthy in vitro pre-differentiation. We recently demonstrated femoral bone defect healing via endochondral ossification using microparticle-containing hMSC condensate sheets for localized presentation of TGF-β1 34 or TGF-β1 + BMP-2 35 that was contingent on in vivo loading. No lengthy predifferentiation of the cellular implants was required. Despite these efforts, bona fide tubular tissue constructs have not been developed using such protocols. This is in contrast to multiple cell-based studies employing traditional bone tissue engineering approaches 9 to form tubular constructs. Natural coral exoskeleton from Porites sp. [10][11][12] , and synthetic polymers such as poly lactic-co-glycolic acid (PLGA) 13 or polycaprolactone (PCL) 14 have been investigated as tubular scaffolds for MSCs. Coral/MSC tubes produced mineralized tissue with shape and structure comparable to native long bones in vivo in a subcutaneous environment [11][12][13] , and facilitated near complete healing of sheep metatarsal segmental bone defects stabilized with anchor plates 10 . Similarly, implantation of PLGA tubes combined with delayed, minimally-invasive percutaneous injection of MSCs stimulated bone regeneration of sheep tibial segmental defects stabilized with dynamic compression plates 14,43 .
Although not without merit, these studies were hampered by (i) the use of scaffolds designed to match the properties of mature Fig. 6 In vivo microCT evaluation of femoral defect healing induced by engineered hMSC condensation tubes and sheets. Longitudinal morphometric analysis of a bone volume and b bone volume accumulation rate at weeks 4, 8, and 12 in defects implanted with hMSC tubes or sheets containing TGF-β1 + BMP-2-loaded microparticles. Analyzed by two-way ANOVA with Tukey's post hoc test (p < 0.05 or lower considered significant). c Longitudinal determination of defect bridging, defined as mineral fully traversing the defect (n = 4 (tubes) or 3 (sheets) biologically independent samples per group; blue lines/circles = hMSC tubes; cyan lines/squares = hMSC sheets). Significance of trend analyzed by chi-square test (**p < 0.01). d Representative 3-D microCT defect reconstructions selected based on mean bone volume. Scale bar, 5 mm. Individual data points are shown with mean ± SD.  Recently, we reported a modular, scalable, scaffold-free system for engineering hMSC rings that can be assembled and fused into tubular structures [29][30][31] . Here, we also employed gelatin microspheres 32 for early TGF-β1 presentation 36 , and mineral-coated hydroxyapatite microparticles 33 for sustained BMP-2 presentation 36 . This combination exerts potent in situ chondrogenic priming effects on early hMSC condensations, consistent with our recent study 30 . Importantly, this approach renders time-and cost-intensive pre-differentiation of the cellular constructs obsolete. Both TGF-β1 and BMP-2 have essential roles during embryonic skeletal development and postnatal bone homeostasis; TGF-β signaling is paramount during pre-cartilaginous mesenchymal cell condensation 46 , whereas BMP-2 acts upstream of the key transcription factors sex-determining region Y-box (SOX) 5, 6, and 9 during early chondrocyte differentiation 47,48 and is a potent osteogenic growth factor. Here, we showed that localized presentation of TGF-β1 + BMP-2 stimulated cartilage and bone formation in hMSC tubes in vivo in a subcutaneous environment guided by the implant geometry. A robust cartilage template exhibiting zonal architecture with both mature and hypertrophic chondrocytes that was actively remodeled into trabecular bone through endochondral ossification was observed, and the presence of cortical bone-like structures indicated engagement of intramembranous ossification with spatial fidelity. Upon implantation in femoral segmental defects, stabilized with custom internal compliant fixation plates that allow for the delayed commencement of in vivo mechanical loading after 4 weeks of stable fixation 34,35,38,41 , engineered hMSC tubes induced more robust bone healing compared to loosely packed hMSC sheets with defect bridging achieved in the majority of critical-size defects. Using minimization of the Akaike's information criterion 49 , we recently showed that defect bridging, minimum pMOI, and bone volume fraction can serve as accurate predictors of maximum torque and torsional stiffness 34,35 . While these biomechanical properties were not directly assessed here, our findings of hMSC tubes inducing~2.3-fold greater defect bridging,~3.0-fold greater minimum pMOI, and~1.2-fold greater bone volume fraction compared to hMSC sheets are consistent with these prior observations and indicate likely enhanced biomechanical competency of experimental limbs at 12 weeks following hMSC tube implantation. New bone was formed through endochondral ossification in both groups; however, only hMSC tubes induced regenerate tissue that partially resembled the normal growth plate architecture. Failure of hMSC sheets in the present study to facilitate the formation of these structures upon implantation that were observed in recent related studies using larger hMSC sheets 34,35 may be due to differences in cell number and morphogen dosing. Compared to these previous studies, here we implanted smaller constructs comprised of hMSC sheets that contained~50% fewer cells (3.2 × 10 6 vs. 6.0 × 10 6 ) and were loaded with~50% less morphogens (TGF-β1: 1.0 vs. 1.8 µg; BMP-2: 1.0 vs. 1.9 µg) to match the hMSC tube composition. Examining bone volume fraction as a key morphometric readout, hMSC sheets here exhibited~2-fold lower values vs. the larger sheets reported recently 35 , while bone volume fraction levels with hMSC tubes were only~1.6-fold lower. Together, this suggests that implant geometry can positively affect the healing outcome in this model; however, uncovering the exact contributions and interactions of cell number, morphogen dosing, and construct geometry will require further investigation.
A key consideration of implementing cell condensation technology whether in a tube or sheet geometry, is how the constructs are handled during both harvesting and implantation into defects. Much of the early work of cell sheet technology was done using poly(N-isopropylacrylamide) (PIPAAm), a thermoresponsive polymer that changes hydrophilicity in response to temperature changes allowing for the release of cell sheets cultured on it 50 . This thermoresponsive polymer allows for harvesting thin cell sheets (i.e., no more than a few cell layers thick) for utilization in tissue engineering applications. Other approaches have utilized cell sheets supported by biodegradable scaffolds for orthopedic and dental tissue engineering applications 51,52 . However, recent studies 30,34,35,37 including the one described herein have shown that cell sheets made from hMSCs grown on Transwell membranes can be generated of sufficient thickness to be harvested by simply peeling them off of the membranes as early as 2 d after initial cell seeding, and then they can be manipulated individually. Similarly, hMSC condensate rings can be manually assembled into tubes after 2 d, and resulting tubes can be implanted without a supporting biomaterial, if desired.
Limitations to our orthotopic study include (i) the difference in pre-culture time between hMSC tubes and sheets, and (ii) the small number of femoral defect samples. Regarding the former, for consistency with previous in vivo studies 34,35,37 , we elected to implant day 2 hMSC sheets in comparison to day 8 hMSC tubes. In doing so, we were able to perform correlative analyses of hMSC sheetinduced defect healing across studies, and to provide context for interpreting the proof-of-principle results obtained with hMSC tubes. The two types of hMSC condensations used in the current femoral defect study were matched in cell number, microparticle concentration, and morphogen dosing; however, the additional 6 days in culture facilitated greater chondrogenic priming of hMSC tubes vs. sheets. No detailed assessment of ECM present in the cellular constructs at the time of implantation was performed beyond histological staining for GAGs. Biochemical analysis of hMSC tubes during the fusion process revealed no detectable amounts of GAG until 3 d after self-assembly (i.e., 5 d in culture) 30 . Therefore, we reasoned that the contribution of potential ECM contained within the hMSC condensations at implantation is likely minimal. However, the presence of live cells was recently demonstrated to be important for defect healing; we showed that devitalization of hMSC sheets by freeze-thaw cycling (which will leave any available ECM physically present/functional) substantially reduced bone formation compared to live-cell controls 34 . It is because of these previous findings that the primary focus of the study was to compare scaffold-free, self-assembled hMSC condensations of different geometries following thorough characterization of the hMSC tubes. It would be interesting to investigate to what degree the extra pre-culture time contributed to the enhanced regenerative effects in comparison to the implant geometry, possibly with an in vivo comparison of decellularized hMSC condensations to assess the relative contribution of ECM vs. cells.
Regarding the latter limitation, ten rats required euthanasia 24 h post-surgery; animals presented with marked gastric distention due to ingested hardwood bedding material. This unexpected pica behavior was traced back to buprenorphine administration, as reported by others 53,54 . Consequently, the sample size may have hindered the statistical differences observed between groups in the femoral defect experiment. Using all available specimens, quantitative bone histomorphometry revealed significantly greater defect healing with hMSC tubes vs. hMSC sheets. It would be worthwhile to expand upon the proof-of-principle findings described here in future studies, in particular, to further investigate the degree of restoration of limb biomechanical propertiesconsidered the ultimate functional test. Furthermore, longer observation times beyond 12 weeks might help address whether tubular condensations are capable of facilitating complete cortical remodeling and re-establishment of a medullary canal.
In conclusion, we describe a biomimetic bone tissue engineering approach that recapitulates certain aspects of the normal endochondral cascade in the developing limb. Implantation of chondrogenically primed hMSC tubes, achieved through in situ morphogen presentation rather than lengthy pre-culture, in large long bone defects that will not heal if left untreated stimulated endochondral ossification for significant bone regeneration with delayed in vivo mechanical loading. The tubular hMSC system is of clinical relevance for long bone regeneration, and our findings advance the present understanding in the field of developmental engineering.
hMSC isolation and expansion. Human bone marrow-derived mesenchymal stromal/stem cells (hMSCs) were derived from the posterior iliac crest of three healthy donors (M/26, F/25, and M/49 years of age) under a protocol approved by the University Hospitals of Cleveland Institutional Review Board. Cells were isolated using a Percoll (Sigma-Aldrich, St. Louis, MO) density gradient and cultured in low-glucose Dulbecco's modified Eagle's medium (DMEM-LG; Sigma-Aldrich) containing 10% pre-screened fetal bovine serum (FBS; Sigma-Aldrich), and 1% penicillin/streptomycin (P/S; Fisher Scientific). The first media change washed away non-adherent cells, and adherent cells received fresh media supplemented with 10 ng/ml fibroblast growth factor-2 (FGF-2, R&D Systems, Minneapolis, MN) 55,56 .
Hydroxyapatite microparticle mineral coating and BMP-2 loading. Mineralcoated hydroxyapatite microparticles (MCM) were kindly provided by Dr. William L. Murphy (University of Wisconsin, Madison, WI). A preparation using low carbonate (4.2 mM NaHCO 3 ) coating buffer and detailed characterization has been reported previously 33,36 . Lyophilized MCM from the same batch as used in our previous studies 30,35,36 , were loaded with a 100 µg/ml solution of rhBMP-2 (Dr. Walter Sebald, Department of Developmental Biology, University of Würzburg, Germany; 1.6 or 6.4 µg/mg) in PBS for 4 h at 37°C. BMP-2-loaded MCM were then centrifuged at 800×g for 2 min and washed 2× with PBS. Unloaded MCM without growth factor were incubated with PBS only and treated similarly.
Preparation of microparticle-incorporated hMSC condensate tubes. Expanded hMSCs (4.0 × 10 5 cells/construct; passage 4) were thoroughly mixed with TGF-β1loaded GM (0.4 µg/mg; 0.3 mg/construct) and BMP-2-loaded MCM (1.6 or 6.4 µg/ mg; 0.08 mg/construct) in basal medium. Fifty microliters of the suspension were seeded in the agarose culture wells 29-31 and allowed to self-assemble for 2 days with feeding after 1 day (Supplementary Fig. S1). Subsequently, tissue rings were removed from the culture wells and transferred to 2 mm diameter borosilicate glass tubes (Adams & Chittenden Scientific Glass, Berkeley, CA) for fusion in 3-, 4-, or 8-ring hMSC condensate tubes. Tissue tubes were cultured horizontally on modified custom polycarbonate holders 29,30 in 60 mm Petri dishes (Corning) containing 10 ml basal medium for 2 weeks, before switching to osteogenic medium for 3 weeks. This time course was previously shown to facilitate robust endochondral ossification 30,36 . Osteogenic medium was identical to basal medium but contained 0.17 mM L-ascorbic acid-2-phosphate and 5 mM β-glycerophosphate 30,36,37 . Respective media were replaced every 2 d. hMSC tubes incorporated with TGF-β1loaded GM and unloaded MCM, or unloaded GM and BMP-2-loaded MCM were prepared and cultured in a similar fashion. Of note, hMSC condensate tubes incorporated with unloaded GM and unloaded MCM were prepared and served as controls for qRT-PCR and immunoblot analysis. For in vivo studies, tubular constructs (subcutaneous: 1.6 × 10 6 cells; orthotopic: 3.2 × 10 6 cells) were cultured for a total of 8 days in basal medium prior to implantation, previously shown to be sufficient for fusion into tubes 30 .
Gross morphological assessment. Gross macroscopic images of in vitro hMSC condensate tubes at 8 days (N = 3) and subcutaneous hMSC condensate tube explants at 3 and 6 weeks (N = 6-8) were taken with a smartphone camera (Apple Inc., Cupertino, CA) and calibrated using a metric ruler held within the frame of the image. ImageJ software (v1.53a, National Institutes of Health, Bethesda, MD) was used to assess the tissue area of the hMSC condensate tubes after initial calibration. Gross macroscopic images of in vitro hMSC tubes at 5 weeks (N = 3) were taken with a digital SLR camera (Canon, Melville, NY). Tube wall width and height measurements were obtained using calipers (Fowler High Precision, Newton, MA) at the 12, 4, and 8 o'clock positions 30 .
Quantitative reverse transcription-polymerase chain reaction (qRT-PCR) analysis. Day 8 hMSC condensate tubes were homogenized in TRI Reagent (Sigma-Aldrich) for subsequent total RNA extraction and cDNA synthesis (iScript™ kit; Bio-Rad, Hercules, CA). One hundred nanograms of cDNA were amplified in duplicates in each 40-cycle reaction using a Mastercycler (Eppendorf, Hauppauge, NY) with annealing temperature set at 60°C, SYBR® Premix Ex Taq™ II (Takara Bio Inc., Kusatsu, Shiga, Japan), and custom-designed qRT-PCR primers (Supplementary Table S1; Life Technologies, Grand Island, NY) 30,34,35,37 . Transcript levels were normalized to GAPDH and mRNA fold-change calculated relative to control hMSC condensate tubes without growth factor using the comparative C T method 59 .
Histological and immunohistochemical analysis. Week 5 in vitro hMSC condensate tubes (N = 3), weeks 3 and 6 subcutaneous explants (N = 3), and day 8 hMSC condensate tubes and sheets (N = 1) were fixed in 10% neutral buffered formalin (NBF) for 24 h at 4°C before switching to 70% ethanol. Femora at 12 weeks were fixed in 10% NBF for 72 h at 4°C and then transferred to 0.25 M ethylenediaminetetraacetic acid (EDTA) pH 7.4 for 14 d at 4°C under mild agitation on a rocker plate, with changes of the decalcification solution every 3-4 days. Following paraffin processing, 5 µm sections were cut using a microtome (Leica Microsystems Inc., Buffalo Grove, IL) and stained with Hematoxylin & Eosin (H&E), Safranin-O/Fast-green (Saf-O), Alizarin Red S (ARS), Masson's Trichrome (TC), or picrosirius red (PSR; Polysciences, Inc., Warrington, PA). For constructs used as orthotopic implants within nanofiber mesh tubes, hMSC tube wall width/ thickness at day 8 and hMSC sheet thickness at day 2 were determined on H&Estained photomicrographs of transaxial and sagittal sections (N = 1; an average of 6 measurements per sample) using ImageJ software. In situ hybridization to detect human Alu repeats was performed using an ISH iVIEW Blue Plus Detection Kit (Ventana Medical Systems Inc., Tucson, AZ) and a BenchMark ULTRA automated IHC/ISH slide staining system (Ventana) according to the manufacturer's instructions. Immunohistochemistry was performed to examine the presence of collagen (Col) types I, II, and X. All sections were deparaffinized and rehydrated with decreasing concentrations of ethanol, and endogenous peroxidase activity was quenched by submerging slides in 30% (v/v) hydrogen peroxide/methanol (1:9) for 10 min. For epitope retrieval, sections were digested with pronase (1 mg/ml; Sigma-Aldrich) in PBS at RT for 15 min. Rabbit anti-human Col I (ab138492) 1:2000, rabbit anti-human Col II (ab34712) 1:1000, and rabbit anti-human Col X ((ab58632) 1:200; all from Abcam, Cambridge, MA) were used as primary antibodies. Rabbit IgG (Vector Laboratories) served as a negative control. The Histostain-Plus Bulk kit (Invitrogen; ThermoFisher Scientific, Waltham, MA) with aminoethyl carbazole (Invitrogen) was utilized according to the manufacturer's instructions with Fast-green counterstain. Slides were mounted with glycerol vinyl alcohol (Invitrogen), and light microscopy images, employing linearly polarized light for PSR-stained sections, were captured using an Olympus BX61VS microscope (Olympus, Center Valley, PA) with a Pike F-505 camera (Allied Vision Technologies, Stadtroda, Germany).
Subcutaneous implantation-subcutaneous bone formation model. Nine-weekold male NCr-nude mice were obtained from the Athymic Animal and Xenograft Core Facility at Case Western Reserve University (CWRU). Animals were anesthetized using 2% isoflurane. Small skin incisions were made on the dorsal sidẽ 15 mm from the midline. Subcutaneous pockets were generated using blunt dissection and 4 hMSC condensate tubes (2 per side) were implanted per mouse approximately 30 mm from each other, cranial to caudal. Two constructs from the same group were on the same side in each animal, which were evenly matched with other groups among animals for homogenous pairing distribution between TGF-β1 -BMP-2, TGF-β1-BMP-2/TGF-β1, and BMP-2-BMP-2/TGF-β1. The incisions were closed and animals were given subcutaneous injections of 0.1 mg/kg buprenorphine at 0 and 12 h postoperative (post-op). All procedures were performed in strict accordance with the NIH Guide for the Care and Use of Laboratory Animals, and the policies of the CWRU Institutional Animal Care and Use Committee (IACUC) (Protocol No. 2014-0096). Animals were euthanized at 3 and 6 weeks by CO 2 asphyxiation followed by cervical dislocation. Skin flaps were opened and hMSC condensate tube explants retrieved.
Nanofiber mesh production. Nanofiber meshes were formed by dissolving 12% (w/v) poly-(ε-caprolactone) (PCL; Sigma-Aldrich) in 90/10 (v/v) hexafluoro-2propanol/dimethylformamide (Sigma-Aldrich). The solution was electrospun at a rate of 0.75 ml/h onto a static aluminum collector. 9 mm × 20 mm sheets were cut from the product, perforated with a 1 mm biopsy punch (VWR, Radnor, PA) ( Supplementary Fig. S7), and glued into tubes around a 4.5 mm mandrel with UV glue (Dymax, Torrington, CT). Perforated PCL nanofiber mesh tubes were sterilized by 100% ethanol evaporation under UV light over-night and washed 3× with sterile PBS before use. One day 8, microparticle-incorporated hMSC condensate tube or three day 2 hMSC condensate sheets combined, each comprised of a total of 3.2 × 10 6 cells, were added into a sterile perforated PCL mesh tube for orthotopic implantation.
Orthotopic implantation-femoral segmental defect model. Critical-sized (8 mm) bilateral segmental defects were created in the femora of 12-week-old male Rowett nude (RNU) rats (Taconic Biosciences Inc., Hudson, NY) using an oscillating saw under isoflurane anesthesia 34,35,38,40,41,63 . Animals received subcutaneous injections of 4 mg/kg lidocaine with 2 mg/kg bupivacaine for local block. Anterolateral incisions were made over the length of each limb, and the vastus intermedius and vastus lateralis muscles were blunt-dissected to expose the femur. Limbs were stabilized by custom internal fixation plates that allow controlled transfer of ambulatory loads in vivo (Supplementary Fig. S7) 34,35,38,40,41,63 and secured to the femur by four bi-cortical miniature screws (J.I. Morris Co, Southbridge, MA). Animals were given subcutaneous injections of 0.04 mg/kg buprenorphine every 8 h for the first 24 h post-op with 4 mg/kg carprofen every 24 h for 72 h. In addition, 5 ml of 0.9% NaCl were administered subcutaneously to aid in recovery. All procedures were performed in strict accordance with the NIH Guide for the Care and Use of Laboratory Animals, and the policies of the CWRU IACUC (Protocol No. 2015-0081). Animals were euthanized at 12 weeks by CO 2 asphyxiation followed by cervical dislocation. Hind limbs were excised and femora retrieved.
In vivo microCT. In vivo microcomputed tomography (microCT) scans were obtained in RNU rats at 4, 8, and 12 weeks to assess longitudinal femoral defect healing 34,35 . Data were acquired using an Inveon microPET/CT system (Siemens Medical Solutions, Malvern, PA) at 45 kVp, 0.2 mA, and 35 μm isotropic voxels, and reconstructed using system-default parameters for analyzing bone and accounting for the metal in the fixation plates. DICOM-exported files were processed for 3-D analysis (CTAn software, Skyscan; Bruker, Billerica, MA) using a gauss filter at 1.0-pixel radius and a global threshold range of 28-255 for all samples. This segmentation approach allowed viewing of the normal bone architecture in the binary images as seen in the original reconstructed images 64 . One hundred and eighty slices in the center of each defect were analyzed using a 10 mm diameter circle centered on the medullary canal to assess bone volume. A binary bridging score was assigned by two independent, blinded observers, and determined as mineralized tissue fully traversing the defect. Representative 3D images were created using Dornheim Segmenter software (Dornheim Medical Images, Magdeburg, Germany), selected based on average bone volume values.
Ex vivo microCT. Ex vivo microCT scans were obtained at 3 and 6 weeks (mouse; subcutaneous model) or at 4, 8, and 12 weeks (rat; orthotopic model). Data were acquired using a Skyscan 1172 microCT scanner (Skyscan; Bruker) with a 0.5 mm aluminum filter at 75 kVp and 0.1 mA. Subcutaneous explants wrapped in gauze were placed in a plastic sample holder and scanned in PBS at 10 µm isotropic voxels, 1170 ms integration time, rotation step of 0.5°, and frame averaging of 5. All subcutaneous explants were scanned within the same container using the same scanning parameters. Femora wrapped in gauze were placed in a plastic sample holder with the long axis oriented parallel to the image plane and scanned in PBS at 20 µm isotropic voxels, 330 ms integration time, rotation step of 0.5°, and frame averaging of 5. All femora were scanned within the same container using the same scanning parameters. All scans were then reconstructed using NRecon software (Skyscan; Bruker) with the same reconstruction parameters (ring artifact reduction of 5, beam hardening correction of 20%). For 3-D analysis (CTAn software, Skyscan; Bruker), a gauss filter at 1.0-pixel radius and a global threshold range of 65-255 were used for all samples. This segmentation approach allowed viewing of the normal bone architecture in the binary images as seen in the original reconstructed images 64 . For subcutaneous explants, varying numbers of slices were analyzed in an 8 mm diameter circle to account for the differences in size. For femora, 325 slices in the center of each defect were analyzed using a 10 mm (total) or 5 mm (defect) diameter circle centered on the medullary canal. Bone volume, bone volume fraction, polar moment of inertia, and the morphometric parameters trabecular number, trabecular thickness, trabecular separation, degree of anisotropy, and connectivity density were calculated 64 . Representative 3D images were created using CTAn software (Skyscan), selected based on average bone volume values.
Bone histomorphometry. The standard two-dimensional (2-D) histomorphometric parameter 65,66 bone area fraction, which by definition is numerically identical with the corresponding bone volume fraction, was assessed in the center of each femoral defect (H&E-stained sagittal sections, three equal-sized square regions of interest spanning the length of the defect; Fig. 8a) according to established protocols [67][68][69] . First, black-and-white thresholded image masks of the H&Estained color images were created using the wand tool (tolerance: 45) in Photoshop CC 2018 software v19.1.3 (Adobe Systems, San Jose, CA) to highlight areas of bone. Bone tissue was designated in black, while the remaining tissue was designated in white, creating the mask for analysis (Fig. 8a). Next, using ImageJ software, the entire image area was calculated as tissue area and, using the default wand tool, the combined black-colored areas were summarized as bone area to calculate bone area fraction (i.e., an average of three regions of interest per sample before an average of all samples per group).
Statistics and reproducibility. Fold-changes over control in mRNA expression and ratios of phosphorylated SMADs/total SMADs, and differences in gross morphological hMSC condensate tube wall width and height, and biochemical markers were analyzed by Student's unpaired t-test, one-way or two-way analysis of variance (ANOVA) with Tukey's multiple comparison post hoc test, as appropriate. Differences in gross morphological tissue area, and ex vivo microCT bone volume, and 3-D morphometry in hMSC condensate tube explants were analyzed by two-way ANOVA with Tukey's multiple comparison post hoc test. Bridging of femoral defects was determined by chi-square test for trend for each group. Differences in longitudinal in vivo microCT bone volume and bone volume accumulation rate at 4, 8, and 12 weeks were assessed by two-way ANOVA with Tukey's multiple comparison post hoc test. Ex vivo microCT bone volume fraction and 3-D morphometry, and 2-D bone histomorphometry in femoral defects at 12 weeks were determined by Student's unpaired t-test. Sample sizes were chosen based on previous studies from our group. No data were excluded from the analyses. Investigators were blinded to group allocation during data collection and analyses. All data are shown with mean ± SD, some with individual data points. The significance level was set at p < 0.05 or lower. Groups with shared letters have no significant differences. GraphPad Prism software v6.0 (GraphPad Software, La Jolla, CA) was used for all analyses.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.