We here describe the design of a 3D printed oscillating perfusion bioreactor, offering bidirectional fluid circulation in a cyclic manner (Supplementary Video S1). Our bioreactor chamber is composed of printed parts (Figures 1A, B and Supplementary Figure S1A) designed in Fusion360™ and printed in 1 h with a standard biodegradable bioplastic PLA (Castro-Aguirre et al., 2016) using a cost-effective Prusa™ MK3 printer. The printed parts consist of a scaffold holder and the upper and lower components of the perfusion chamber. The chamber components interlock through a thread to encapsulate the scaffolding material held by the scaffold holder. This junction is reinforced with a silicon O-ring. At both ends of the chamber, 3-way stopcocks create syringe-accessible links with the tubing. The later carries the media which oscillate through the chamber via a programmable pump. To prevent contamination, the system has two filters at both tubing ends, and disinfectant caps on the stopcock valves.

The design was fashion to accommodates the printing process (no hard curves or overhangs that necessitate support material) while supporting fluidic and airtightness (random seams placement, additional perimeters, grove and threads for tubing and silicon O-ring). Printing parameters were also optimized to favor printed layer-to-layer adhesion (listed in Supplementary Table S1). We also incorporated in the design a thread fitting 50 mL tubes at the bottom of the chamber, towards facilitating sample collection at the end of a culture process (Supplementary Figure S1B). Airtightness of printed chambers was first tested by gradual air pressure application for up to 3 bar (Figure 1C). No detectable leakage was observed during repeated test with a pool of 15 printed bioreactors. While having a better layer adhesion compared to autoclavable FDM filaments such as polypropylene and nylon, PLA cannot be autoclaved as its glass transition temperature is reached at 50°C. Hence, alcohol bath and UV exposure were investigated as other means of disinfection. Alcohol disinfection was performed with a 30 min immersion in 70% isopropanol (Figure 1D), previously reported to not degrade 3D printed PLA (Erokhin et al., 2019). UV light disinfection consisted in 30 min of exposure using a standard benchtop UV cabinet. Following treatment, disinfection was assessed by placing respective chambers for 24 h in flasks containing lysogeny broth (LB) at 37°C overnight under agitation. The growth of bacteria was assessed by optical density at 600 nm. Only the disinfection by isopropanol prevented bacteria growth, as opposed to UV light which did not reach the same level of disinfection. In addition, no mycoplasma contamination could be detected during culture (three experimental repeats, data not shown).

We thus here validate the design, printing, and disinfection of perfusion bioreactor chambers. In addition, we further developed a 3D printed bioreactor holder for placement in incubators (Figure 1E), as well as a bioreactor stand (Supplementary Figure S1C) for bench manipulation of the system. This offers easy handling and optimization of space for multiple bioreactor experiments.

We next aimed at validating our 3D printed system for the culture of human cells, towards the engineering of 3D microenvironments. To this end, we exploited a pre-established human mesenchymal stromal cell line (MSOD, Mesenchymal Sword Of Damocles) (Bourgine et al., 2014) (Supplementary Figure S2A). The MSOD line offers unlimited cell supply and standardization while maintaining properties of primary bone-marrow hMSCs. MSOD cells were dynamically seeded overnight on a collagen type I scaffold (Col1) and cultured for up to 2 weeks (Figure 2A). Collagen scaffolds have proven to be versatile materials thus exploitable in multiple organ/tissue modelling strategies (Meinel et al., 2004; Liu et al., 2009; Zhou et al., 2011; Ding et al., 2014; Chan et al., 2016). To assess cell colonization potential in long term culture, we perform a comparison between a Col1 and an alternative hexamethylene diisocyanate crosslinked collagen scaffold (CrL-Col1) within our system (Figure 2B, Supplementary Figure S2B–D,E). At 1 week of culture, both Col1 and CrL-Col1 scaffolds contained a 2-fold increase in cell number from input (5 million MSOD). However, at 2 weeks, an 8-fold increase was observed for Col1 while the CrL-Col1 scaffold presented a 12-fold increase. This disparity can be explained by the degradation of the Col1 scaffold, evident after 2 weeks of culture compared to CrL-Col1 (Supplementary Figure S2B).

Based on these results, the CrL-Col1 was thus selected for the rest of the study. We further characterized the porosity and provided previsously determined mechanical properties of our selected material (Supplementary Figure S2D, E, (Chaudhury et al., 2012)), revealing an interconnected pore network mainly comprised in the 50–100 µm range (>60%). Next, we demonstrated the versatility of the 3D printing approach towards the engineering of stromal environment of different dimensions. Bioreactor chambers capable of hosting CrL-Col1 scaffold of 6, 8, 10, and 12 mm were designed, and MSOD cells were cultured in the respective systems for a week. In contrast to the superficial scaffold colonization observed in 3D static culture (Figure 2C), MTT assay revealed the homogenous MSOD cell distribution within the scaffolding materials under dynamic perfusion culture (Figure 2C). This uniform distribution was further confirmed by confocal microscopy, with presence of MSOD cells detected across the CrL-Col1 material (Figure 2D) independently of the tissue size (Supplementary Figure S2C). This uniformity is crucial to ensure the functionalization of the scaffold with the supportive ability of the stromal cells, analogous to their role in the bone marrow tissue (Klimczak and Kozlowska, 2016).

We further assessed the capacity of removing proteins from 3D printed parts post-culture, towards demonstrating its reusability. To this end, we quantified protein content on the scaffold holder, as the bioreactor part primarily in contact with biological material (Supplementary Figure S2F). Prior to cleaning, scaffold holders carried 303.49 ± 70.79 µg of protein, while content on cleaned and freshly printed scaffold holders could not be detected (<20 µg). We next performed 4 successive rounds of 3D perfusion culture using the same 3 distinct bioreactors (Figure 2E), cleaned prior to each new experiment. We report a similar MSOD cell growth independently of the experimental round (CyQuant^TM quantification, Figure 2E), thus demonstrating the reusability of our 3D system.

Mesenchymal cells are an important component of the bone marrow niche, supporting the function of hematopoietic stem cells. Modeling the human BM niche ex vivo remains challenging, since HSCs can hardly be maintained in standard in vitro conditions (Kumar and Geiger, 2017). Here, we aimed at assessing the suitability of our 3D printed bioreactor for the engineering of BM microenvironment sustaining HSCs survival. To this end, MSOD cells were first cultured in our perfusion bioreactor for a week to functionalize the CrL-Col1 scaffold (6 mm). Subsequently, primary human CD34⁺ hematopoietic cells from cord-blood origin were added to the bioreactor, and cocultured in presence of low concentration of hematopoietic cytokines (Figure 3A). Within the engineered tissue (Supplementary Figure S3A), physical interactions between the mesenchymal fraction and hematopoietic cells (CD45⁺) could be evidenced using staining for the cytoskeletal protein tubulin (Figure 3B). Via confocal microscopy, a 3D stack of the established bone marrow microenvironment was reconstructed which not only further identified hMSCs-HSPCs interactions, but also highlighted the complex networks of intertwined hMSCs cytoplasmic protrusions (Supplementary Video S2, white arrows highlight interactions). After a week of coculture in a 3D printed bioreactor, the hematopoietic cells were harvested for quantitative phenotypic analysis by flow cytometry.

Tissues were recovered from the chamber and digested for cell retrieval, prior to immune staining for a panel of phenotypic markers indicating the lineage commitment, or absence thereof, within the hematopoietic CD34⁺ subpopulations (Figure 3C). Briefly, we first excluded MSOD, and dead cells based on positivity to GFP and Sytox Green. The negative fraction containing CD34⁺ cells was then further characterized for stem or commitment phenotypes (see Material and Methods section). As a 3D control, we similarly assessed the survival and development of blood cells cultured on the material only (“Scaffold”), thus deprived of MSOD cells.

As compared to the “Scaffold” condition, the number of CD34⁺ HSPCs retrieved from the “Niche” engineered tissue remained stable over the 7 days of coculture (18.06 ± 13.39 fold decrease in Scaffold against 1.22 ± .782 fold increase in Niche condition; Figure 3D). Yet, we observed a general decrease in the number of more committed myeloid-primed (CD34⁺CD45ra+CD10⁻CD33⁺) and lymphoid-primed (CD34⁺CD45ra+CD10⁺CD33⁻) progenitors in both the Scaffold and Niche settings (Figure 3D). The more stem CD34⁺CD45ra-population also decreased overtime in both conditions, though to a lower extend in the presence of MSOD (1.89 ± 1.09 fold decrease against 19.17 ± 6.29 in Scaffold). To further identify which stem populations were affected by this decline, we used the CD90 and EPCR markers to separate phenotypic HSCs (CD34⁺CD45ra-CD90+EPCR+) from the multipotent progenitors (MPPs) CD34⁺CD45ra-CD90+EPCR- and CD34⁺CD45ra-CD90-EPCR+ (Figure 3C). Importantly, MPPs EPCR+ were recently demonstrated to possess a higher MMPs re-population capacity (Fares et al., 2017; Anjos-Afonso et al., 2022).

We observed a stark difference in the stem subpopulation outputs during the 3D coculture between Niche and Scaffold conditions (Figure 3E). In both conditions, MPPs CD90+EPCR-drastically decreased as compared to the starting population, with a 18.65 (±13.05) fold decrease for Scaffold and 12.10 (±4.77) for Niche. However, while MPPs CD90-EPCR+ were maintained in Scaffold (1.61 ± .80 fold increase), we observe a 97.87 (±31.92) fold increase in presence of MSOD cells. Importantly, in contrast to Scaffold (1.873 ± .571 fold decrease), we measured a 13.30 (±2.25) fold increase in HSCs population when cocultured with MSOD cells. Remarkably, the observed hematopoietic support of MSOD in 3D coculture within our 3D printed perfusion bioreactor was not reflected in 2D culture (Supplementary Figure S3B), where MSOD does not exhibit striking hematopoietic support as compared to a no-stroma condition (Supplementary Figure SC-D).

Altogether, our data validate the generation of human bone marrow niche using a 3D perfusion bioreactor. In contrast to 2D culture and non-stromal conditions, we report superior maintenance of phenotypic HSCs and MPPs with high-repopulation potential on engineered microenvironments.

Experimental work carried out with primary human samples was approved by the regional and ethical committee for Lund/Malmö (Regionala Etikprövningsnämnden I Lund/Malmö), approval no. 2010-695. Informed consent was obtained from mothers of the umbilical cord blood (UBC) donors, and all samples were de-identified before use in the present study.

UBC was collected at Skåne University Hospitals and Helsingborg Hospital. Briefly, mononuclear cells were collected by Ficoll separation and CD34⁺ cells were isolated using the CD34 MicroBead kit (Miltenyi Biotec #130-702) according to the manufacturer’s instructions. UCB-CD34⁺ cells samples used in this study were originating from a pool of a minimum of 3 units, processed within 24 h after collection and with a minimum CD34⁺ purity of 94% and viability of at least 95%.

2D coculture was carried out in a Nucleon™ Delta Surface 12-well plate (ThermoFisher #140675). Briefly, 10.000 MSOD cells were cultured for 7 days in complete media (CM) composed of 500 mL of MEM-α (Gibco # 22571038) supplemented with 50 mL of tetracycline-free fetal bovine serum (FBS; ThermoFisher); 5 mL of sodium pyruvate (100 mM; Gibco #11360070); 5 mL of HEPES (1 M; Gibco # 15630080); 5 mL of Penicillin-Streptomycin-Glutamine (100x at 50 mg/mL; Gibco #10378016). Ascorbic acid (AA; 100 μM; Sigma #A8960-5G) was added at each media change which was performed every third day. On day 7, 4 Gy irradiation was performed using the CellRad X-ray source by Flaxitron, cells were then le ft to recover for another 24 h in fresh CM. 35.000 UCB-CD34⁺ cells were added in each well in 1 mL of Coculture media (CoM), composed of DMEM (High Glucose, no glutamine, no calcium; ThermoFisher #21068028); 20% BIT9500 Serum substitute (StemCellTechnologies #09500), 1% Penicillin-Streptomycin-Glutamine and 1% HEPES. Media change was performed every 2 days with the addition of .02% ß-mercaptoethanol (500X at 50 mM; ThermoFisher #31350010), human stem cell factor (SCF), thrombopoietin (TPO) and Fms-related tyrosine kinase 3 ligand (FLT3LG) at 10 ng/mL (all from Miltenyi Biotec, respectively #130-096-692, #130-095-745 and # 130-096-474) reconstituted in IMDM +10% Bovine Serum Albumin (StemCellTechnology #09300).

The design of the bioreactor chambers (See Figure 1A; Supplementary Figure S1A) was modelized using Fusion 360 from Autodesk and sliced using PrusaSlicer 2.4 (Prusa) for a .4 mm nozzle and .15 mm layer height (see full details and critical parameters in Supplementary Table S1). Once sliced, the model was loaded to a Prusa i3 MK3S+ and printed with a polylactic acid (PLA) 1.75 mm ± .03 mm natural/transparent filament (Verbatim #55317) on a steel bed coated with UHU Twist and Glue without solvent. Along with this publication, 3D printed files can also be found at the National Institute of Health (NIH) 3D print exchange repository (https://3dprint.nih.gov/users/cto-laboratory) as well as at our laboratory website (http://www.bourginelab.com/).

For disinfection, the PLA-printed parts of the bioreactor (See Figure 1A; Supplementary Figure S1A) were immersed for a minimum of 30 min inside a solution of 70% Isopropanol prior to use. Before assembly, the parts were dried inside a sterile ventilated hood for 5 min. All other components of the bioreactors (c.f. Tubing and silicone O-ring) were sterilized by autoclaving. The assembly of the bioreactor and media change was carried out as previously described (Dupard and Bourgine, 2021) under a sterile hood; for natural collagen I (Col1), a 6 mm diameter scaffold (BD, Avitene Ultrafoam) was used; for crosslinked collagen (CrL-Col1), a 6–12 mm diameter ZimmerPatch Collagen sponge (ZimmerBiomet #0101Z) was used as scaffolding material. The infuse/withdraw PHD ULTRA™ syringe pump from Harvard Apparatus was plugged to the bioreactor for the dynamic perfusion of the nutritive media. F or the commercially available components of the bioreactor displayed in Supplementary Figure S1A, the manufacturers, in relation to the numbering are the following: 5: Silicon O-ring (#11.2007.0728; NORMATEC^® O-ring; 13 × 2 mm); 6: Lower stage of the bioreactor chamber; 7: Filtropur S plus .2 µm (#83.1826.102, Sarsted); 8: Female luer thread (#FTLL055-6,005, Nordson Medical); 9 and 9’: Silicon tubing of respectively 16 and 30 cm length (#8060-0060 Nalgene^® 50, ThermoFisher; inner diameter of 1/4 in); 10: Male luer lock (#MTLL055-6005, Nordson Medical); 11: Discofix^® C safeflow closed system stopcocks with valves (#16494CSF B. Braun) with BD PureHub™ disinfecting caps (#306596, Becton Dickinson); 12: Female Luer Lug (#FTL210-6,005, Nordson Medical); 13: Silicon tubing of 10 cm length (#8060-0020 Nalgene^® 50, ThermoFisher; inner diameter of 1/16 in); 14: Male luer lock (#MTLL220-6005, Nordson Medical).

For contamination with mycoplasma, three chambers were left in culture with MSOD cells for 1 week. Ultimately, a 5 mL media sample was used for mycoplasma contamination assessment by the MycoplasmaCheck service from Eurofins Genomics. The performance of disinfection was evaluated after overnight incubation of 3D printed components in Lysogeny Broth (LB) medium at 37°C under 130 rpm shaking. Disinfection included 30 min exposure to UV light cabinet (#15572496, Fisher Scientific), or 30 min immersed in 70% isopropanol, controls consisted of immersion for 30 min in PBS. Bacterial growth was quantified through OD600 using the NanoDrop 2,000c from ThermoFisher.

Assembled chambers were occluded on one end and a syringe was plugged at the other end. To assess for airtightness the chamber was immersed in water and a gradual pressure of 1, 2 and 3 bar was applied through the syringe. If no air bubbles leaking from the chamber were observed, the chamber was considered airtight and fit for use in cell culture.

5 million MSOD cells were suspended in 7 mL of CM for 3D perfusion and in 35 µL of CM for 3D static culture. Media change in perfusion settings was performed as previously described (Dupard and Bourgine, 2021) every 3 day. For static 3D culture, a 2 mL media change was carried out similarly. For static 3D culture, seeding was perfomed by capilarity on scaffolds placed in 12 well-plates coated with 1% agarose. For 6 mm scaffolds in perfusion culture, a first overnight infuse/withdraw perfusion cycle speed of 2.8 mL/min with displacement goal at 2 mL allowed dynamic cell seeding on the scaffold; for the rest of the 3D culture the infuse/withdraw perfusion cycle speed was lower at .28 mL/min. Moreover, speed was adjusted for different scaffold diameter sizes to ensure equivalent shear stress and fluid dynamics across conditions: 4.95 mL/min and .495 mL/min for 8 mm; 7.75 mL/min and .775 mL/min for 10 mm; 11.2 mL/min and 1.12 mL/min for 12 mm scaffolds.

Scaffolds retrieved from the culture chamber were washed twice in pre-warmed PBS and incubated for 2 h at 37°C and 5% CO2 in DMEM (no phenol; Gibco #A1443001) with 50 μg/mL Thiazolyl Blue Tetrazolium Bromide (Sigma #M5655). Scaffolds were rinsed twice with pre-warmed PBS and cut in half in the median plane prior to imaging.

For cell number determination, scaffolds were washed twice in pre-warmed PBS before being submerged in a digestion solution containing 1 mg/mL of Proteinase K (Sigma #P2308), 10 μg/mL of Pepstatin A (Sigma #P5318), 1 mM EDTA (Sigma #03690) and 1 mM of Iodoacetamide (Sigma #I6125) in a Tris buffer (Sigma #T5912) pH 7.6. The digestion was carried out overnight before being used for CyQuant™ Cell Proliferation Assay (Invitrogen #C7026) according to the manufacturer’s protocol. For comparison, 5 million MSOD cells were subjected to the same conditions and used for the determination of the ratio between DNA content measured by CyQuant to cell number.

To ascertain the reusability of the bioreactor, 3 sets of printed components were reused up to 4 times for 3 days of perfusion culture with 05 million MSOD cells seeded in 6 mm Col1 scaffolds. Between each experimental round, printed components were washed overnight in a cold water bath with a sodium lauryl sulfate (SLS)-based soap (Tork #420701), and then rinsed abundantly under running cold water. At the end of culture, the cell number was determined by CyQuant^TM assay as described above. To quantify the level of protein adsorbed on the bioreactor printed components after culture, and after washing, we used 8 mm scaffold holders used in perfusion culture for 5 days with 2 million MSOD cells seeded on 8 mm Col1 scaffolds. Samples consisted of scaffold holders dipped in 3 × 2 mL of PBS (“Before cleaning”); washed overnight with SLS-based soap, dried and left 30 min in 70% isopropanol (“After cleaning”), and new scaffold holders (“Freshly printed”). Proteins adsorbed were detached and solubilized by sonication for 10 min in 2 mL of RIPA buffer (Sigma #R0278). Solubilized proteins were then quantified using the bicinchoninic acid assay (BCA) according to the manufacturer’s protocol (Merck Millipore #71285-3). Absorbance at 562 nm was measured with the SPECTROstar Nano from BMG LabTech.

3D coculture was performed as previously described (Dupard and Bourgine, 2021). Briefly, 35.000 UCB-CD34⁺ in CoM were injected in each perfusion bioreactor. 5 mL of media per bioreactor was changed every 2 days for 1 week, with re-injection in the system of the cellular populations retrieved from the withdrawn medium.

For 2D coculture, single-cell suspension was prepared by washing with PBS, followed by a 10 min digestion at 37°C in Trypsin-EDTA (.05%; Life Technology #25300054) supplemented with DNase I (.25 mg/mL; Thermofisher #10700595). Digestion was stopped by the addition of an equal volume of CoM. Cells were then resuspended in FACS buffer composed of PBS with 2% FBS and 1 mM EDTA and passed through a 40 µm nylon mesh. For 3D culture, PBS washes and trypsin digestion were carried out similarly under perfusion. All liquids were collected from the scaffold in a 50 mL tube and resuspended in FACS buffer. All samples were analyzed with an LSR Fortessa flow cytometer (BD Biosciences). Sytox Green (ThermoFisher #R37168) was used according to the manufacturer’s protocol and allowed the exclusion of both MSOD and dead cells from the analysis. Monoclonal antibodies against human CD34 (APC; 20:100; Clone: 581, BD #55824) and CD45ra (AF700; 10:100; BD #560673) allow the discrimination between committed (CD34⁺CD45ra+) and stem (CD34⁺CD45ra-) hematopoietic subpopulations. The committed population was further analysis for myeloid-lineage commitment with the expression of CD33 (BV650; 3:100; Biolegend #351904) surface marker (Myeloid-Primed: CD34⁺CD45ra-CD33⁺CD10⁻); while CD10 (BV421; 5:100; Clone: HI10a; Biolegend #312218) identified lymphoid lineage commitment (Lymphoid-Primed: CD34⁺CD45ra-CD33⁻CD10⁺). The stem cells compartment was further characterized with antibodies directed against EPCR (PE; 5:100; Biolegend #351904) and CD90 (PE-Cy7; 10:100; Clone: 5E10; Biolegend #561558) as previously described for in vitro hematopoietic culture76,77. Phenotypic hematopoietic stem cells (HSCs) were identified with the gating CD34⁺CD45ra-CD90+EPCR+; Multipotent progenitors population were defined as CD34⁺CD45ra- and further differentiated based on CD90 and EPCR expression (CD90+EPCR- and CD90-EPCR+). Samples were analyzed with FlowJo (version 10.7, BD Biosciences).

Scaffolds used for imaging were fixed overnight in fresh 4% Paraformaldehyde at 4°C. Fixed samples were then washed in PBS and embedded in a 4% Low gelling Temperature Agarose (#A9414, Sigma). Longitudinal 100 µm thick sections in the median plane of the scaffold were then cut using a 7,000 smz vibratome with stainless steel at 50 Hz frequency, 1.5 mm amplitude, and .05 mm/s speed. For MSOD distribution in the scaffold, as the green fluorescence protein is constitutively expressed, no antibody staining was necessary. Tubulin was detected using the primary Rat anti-Tubulin (1:1,000; Abcam #GR3208838-5) and the secondary antibody Donkey anti-Rat (Cy3; 1:200; Sigma #SAB4600131). For CD45, detection was performed with the primary Mouse anti-CD45 (1:100; Ebiosciences #14-0459-82) and the secondary antibody Donkey anti-Mouse (CF633; 1:200; Jackson ImmunoResearch #712-165-150). All sections were mounted with Prolong™ Glass Antifade (ThermoFisher #P36982). Image acquisition was done with a Stellaris 5 confocal microscope from Leica.

CrL-Col1 scaffold was mounted on a 12.5 mm aluminum stub. The scaffold was then sputtered with 10 nm Au/Pd (80/20) in a Quorum Q150 T ES turbo pumped sputter coater and examined in a Jeol JSM-7800 F FEG-SEM. Acquired micrographs of the top layer of the scaffold were used to determine pore diameters frequency via ImageJ (version 1.53f51). 127 pores were included in the analysis.

Statistical analysis was performed with R (version 4.1.3). Unless otherwise specified in figure legends, unpaired t-tests were used throughout the manuscript, with prior validation of test assumptions.