Strategy for achieving standardized bone models

Abstract Reliably producing functional in vitro organ models, such as organ‐on‐chip systems, has the potential to considerably advance biology research, drug development time, and resource efficiency. However, despite the ongoing major progress in the field, three‐dimensional bone tissue models remain elusive. In this review, we specifically investigate the control of perfusion flow effects as the missing link between isolated culture systems and scientifically exploitable bone models and propose a roadmap toward this goal.


| KEY PARAMETERS IN PERFUSION BIOREACTORS
2.1 | Flow rate, circulation velocity, and shear stress Perfusion bioreactors supply cells with culture medium at a selected flow rate, which is the volume of medium perfused through the scaffold in a given amount of time. However, cells do not respond directly to flow rate values but to the resulting chemical and mechanical environments.
As stated, the culture medium circulation velocity defines the convective transfer of soluble substances. This velocity defines mass transport rates across the scaffold in association with diffusion phenomena, therefore playing a major role in defining the chemical environment of cells. Regarding mechanical stimuli, they are assumed to be mainly transmitted to bone cells through fluid flow and matrix deformation (Goggin, Zygalakis, Oreffo, & Schneider, 2016;Gusmao & Belangero, 2009;Owan et al., 1997;Paul, Malhotra, & Muller, 2018;You, Weinbaum, Cowin, & Schaffler, 2004). Simplified relationships between flow rate, circulation velocity, and shear stress are reported in the following sections.

| Circulation velocity
Applied to homogeneous and unidirectional perfusion, the simplified continuity equation states that: where Q is the flow rate and v is the average velocity of the fluid flowing through an open cross-section of surface S (Figure 1a).
Thus, the average fluid circulation velocity within the scaffold is directly tied to its dimensions and porosity (Figure 1c) through the relationship: (3) Figure 1d shows the average velocity generated by a 1 ml/min flow rate through circular scaffolds for different common values of scaffold diameter and porosity percentage. For the same flow rate, differences in scaffold size and porosities can easily generate a tenfold difference in average velocity.

| Shear stress
When the thickness of the extracellular matrix (ECM) and cell layers is significantly smaller than the pore diameter, we can use the shear stress applied to the scaffold walls to estimate the flow-induced shear stress applied to the cells. In cylindrical channels, wall shear stress τ can be extrapolated from the liquid dynamic viscosity μ, the diameter of the channel d (Figure 2a), v through the Poiseuille-Hagen law: By locally approximating scaffold pores to cylinder fragments, with d being the pore diameter, we obtain the following relationships: which directly connects local shear stresses to the selected flowrate, medium viscosity, scaffold dimensions, pore sizes, and porosity percentage. Figure 2b shows the average shear stresses generated by a 1 ml/min flowrate through 1 cm 2 cross-section scaffolds for different common values of pore diameters and porosity percentage. For this same flowrate, the variations in scaffold internal architecture produce average shear stresses ranging from 3.7 to 26.7 mPa.

| Scaffold architectural features defining flow effects
How a given flowrate will translate into mass transport rates and shear stress depends on a combination of the scaffold architectural features and bioreactor characteristics (Du, Ushida, & Furukawa, 2015). The architectural features orienting cell fate by defining flowinduced mechanical and chemical environments are summarized below.

| Shape and dimensions
Scaffolds can range from a few millimeters in size (Grayson et al., 2008;Jagodzinski et al., 2008) to several centimeters (Li, Tang, Lu, & Dai, 2009;Liu et al., 2012), greatly varying the area of the cross-section exposed to flow. Variations in scaffold shape and bioreactor chamber designs (especially if and how the scaffold is sealed) also define preferential flow pathways. For example, culture medium can sometimes bypass the scaffold porosity (Figure 3a) or be forced in specific flow configurations ( Figure 3b). Figure 3c shows a bioreactor where proper perfusion is ensured by press-fitting the scaffold into a custom silicone cassette.

| Porosity
The macroscopic structure produced by a network of pores is often described using porosity values, expressed as a percentage of the volume of voids over the total volume of the scaffold and often ranging from 50% to 90% (Gariboldi & Best, 2015). Porosity can be F I G U R E 2 Impact of scaffold properties on flow-induced shear stress. (a) Shear stress τ inside the channels is proportional to fluid velocity v and inversely proportional to the channel diameter d. Shear stress is presented in decreasing order from τ 1 (max) to τ 3 (min). (b) Approximation of shear stress in mPa in a scaffold of a cross-section A = 1 cm 2 perfused at a flowrate of 1 ml/min, and different values of porosity (p) and pore diameter (d in µm), calculated using Equation ( is directly conducive to tissue in growth. Porosity is only one of the numerous parameters that may be used to describe the porous architecture of a scaffold (e.g., interconnectivity, pore orientation, tortuosity, pore and interconnection shape). Used alone, porosity is a poor predictor of biological responses. In particular, mass transport and shear stress values, which are key factors affecting cell fate and tissue development, cannot be evaluated based on the porosity value alone; other architectural parameters must be provided (Ashworth, Best, & Cameron, 2014;Bohner, Loosli, Baroud, & Lacroix, 2011).

| Macropore size and geometry
As stated in Section 2.1, for a given velocity, shear stress is inversely proportional to the channel diameter (d). Therefore, pore size is a significant parameter to know and control (if possible) to determine the relationship between cell behavior and flow effects. Unfortunately, mainly due to the manufacturing techniques used to produce scaffolds, their porous architectures are always more complex than an arrangement of straight channels, as schematized in Figure 4b-d . Random macropore distribution, size, orientation, shape, and so forth are predominant in scaffolds used for bone tissue engineering (BTE), which mainly consist of interconnected pores (Figure 4e,g) or intertwined fibers (Figure 4f,h). Macroporosity is almost systematically approximated as spherical with a unique dimension, the "mean diameter," or in the best case, a diameter distribution (Bohner et al., 2011). This simplification is not representative of the actual macropore geometry and does not include the interconnection features, which are defining parameters for local circulation velocities and shear stresses.
F I G U R E 3 Impact of scaffold and culture chamber designs on flow pathways. Shape and dimensions of the scaffold with respect to the chamber design greatly modify the area and the volume of the scaffold truly perfused; culture medium can flow (a) preferentially all around the scaffold (a1) from the middle (Liu et al., 2012) or (a2) at the periphery (streamline simulation; Cruel et al., 2015), (b) at specific location (Grayson et al., 2010) or (c) through the entire construct  F I G U R E 4 Macroporous structure. Representation of the types of pore space depending on their connection to the surface of the material (open, closed, and blind-ended; Ashworth et al., 2014) for (a) an ideal structure composed of channels, (b) a tortuous porous network and a scaffold composed of (c) spherical interconnected pores or ( (Bancroft et al., 2002;Cartmell, Porter, Garcia, & Guldberg, 2003;Grayson et al., 2008;Li et al., 2009;Sonnaert et al., 2017), and second that cells can be surprisingly sensitive to moderate variations in flowrate (Cartmell et al., 2003;Su, Wang, & Chou, 2014). Bancroft et al. (2002) cultivated rat marrow stromal osteoblasts in perfused fiber mesh titanium scaffolds and found that an increase in flowrate from 0.3 ml/min to 1 ml/min generated an over six-fold increase in the calcium content of cultured scaffolds. Conversely, in Cartmell et al. (2003), increasing the flowrate from 0.1 to 0.2 ml/min caused a fourfold decrease in the total DNA. These studies illustrate that cell behavior, especially viability and differentiation rates, can be significantly altered by subtle changes in flowrates.
Cell behavior is not related to flowrate in a linear manner. In the same study by Bancroft et al. (2002), increasing the flowrate from 1 ml/min to 3 ml/min "only" doubled the calcium content. In Cartmell et al. (2003), no significant change in DNA content or OCN and Runx2 expression was observed between 0.01 ml and 0.1 ml/min, whereas a sharp decrease in DNA and an increase in OCN and Runx2 were observed at 0.2 ml/min (Figure 5a). These results suggest that cells are particularly responsive to ranges and thresholds of stimuli.
F I G U R E 5 Influence of the combination of shear stress and mass transport on cells fate. (a) Total DNA content (cell proliferation) and Runx2 gene expression measured after 7 days of culture of MC3T3-E1 immature osteoblast-like cells in human trabecular bone scaffold under static or dynamic medium flow conditions (from Cartmell et al., 2003). (b) The existence of thresholds for mass transport rates and shear stresses visualized here could explain the marked changes observed between perfusion flowrates of 0.1 and 0.2 ml/min. (c) Appropriate levels of shear stress can have significant osteogenic effects, but higher values cause cell damage and detachment (McCoy & O'Brien, 2010). In this hypothesis, achieving higher shear stresses at a given flowrate would allow benefiting simultaneously from the osteogenic effects of shear stress and mass transport HADIDA AND MARCHAT | 255

| Confusion surrounding flowrate
Given the significance of flowrate values, a priority in the field of in vitro BTE is determining the "optimal flowrate" for osteogenesis.
However, as described in Section 3.1, studies aiming to assess flowrate effects obtain contradictory results regarding the optimal perfusion rate and describe different effects for identical flowrates.
For instance, using a static culture as a control, Gomes, Holtorf, Reis, and Mikos (2006)  and internal architecture than those in their reference study (Baas et al., 2010;el Haj et al., 1990).
As explained in Section 2.1, for a given flowrate, differences in the scaffolds architecture result in different local fluid speeds and shear stresses. The combined influence of both should be systematically considered when interpreting perfusion study results. In Cartmell et al. (2003), the upregulated expression of OCN and Runx2, which are characteristic responses to shear stress exposure (Wittkowske et al., 2016), was observed at 0.2 ml/min ( Figure 5a). According to the authors, the sharp decrease in cell viability also observed for this flowrate may be linked to the increased shear stress. However, the osteogenic levels of shear stress are not usually correlated with a decrease in cell viability (Bancroft et al., 2002;Chen et al., 2016;Farack et al., 2011;Grayson et al., 2008;Holtorf, Sheffield, Ambrose, Jansen, & Mikos, 2005;Kleinhans et al., 2015;Li et al., 2009;Liu et al., 2012;Su et al., 2014), and we hypothesize that the shear stress level required to elicit an osteogenic cell response in the scaffolds used is reached only for flowrates inducing mass transport levels already detrimental to cells ( Figure 5b). Although relevant mass transport rates enhance cell viability and proliferation, excessively high rates have inhibitory effects that may be linked to cell signaling disruptions (Grayson et al., 2011;Li et al., 2009).
This idea would be consistent with the structure of the scaffolds used, which have a relatively high porosity (82%) and an unusually large mean pore size (645 µm), resulting in lower shear stresses at a given flow rate (see Section 2.1). Based on this observation, the protocol parameters could be adjusted by using scaffolds with a smaller pore size or a more viscous culture medium (c.f. Section 5.1.3) to obtain a better combination of shear stress and mass transport effects, as proposed in Figure 5c. Therefore, approaching flowrate in terms of resulting circulation velocities and shear stresses would offer additional insight into result interpretation and understanding why perfusion can reduce cell viability (Bartnikowski, Klein, Melchels, & Woodruff, 2014;Cartmell et al., 2003;Jaasma & O'Brien, 2008;McCoy, Jungreuthmayer, & O'Brien, 2012). Similarly, studies investigating optimal scaffold pore sizes and porosity Gomes et al., 2006;McCoy et al., 2012) in perfusion bioreactors should consider that the observed results may not actually be only related to the scaffold features but also to the corresponding shear stress and mass transport resulting from the perfusion of this scaffold at the selected flow rate (see also Section 4.4.1). Thus, a challenging target in BTE is the determination of these optimal shear stress and circulation velocity ranges for a given cell type.

| OBSTACLES IN DEFINING OPTIMAL FLOW EFFECTS
In the context of the ongoing pursuit of optimal operating conditions that will yield the desired levels of tissue performance or functioning, the determination of optimal shear stress and circulation velocity ranges for a given cell type constitutes both a challenging target and a major potential milestone in BTE.

| Using reference shear stress values
Fixing or identifying optimal shear stress and circulation velocity in complex 3D systems from approximated variables only (e.g., pore size, flow homogeneity in the scaffold) is arduous. Therefore, the BTE community primarily relies on different values from the literature.

| Biomimetics
In 1994, Weinbaum, Cowin, and Zeng (1994)  Therefore, there is a priori no scientific incentive to reproduce this range of shear stress in perfusion experiments.

| Two-dimensional systems
Parallel flow chambers were specifically introduced to study shear stress effects. In a rectangular section, wall shear stresses are reliably defined by the following equation: with Q being the flowrate, µ being the medium viscosity, w being the width, and h being the height of the flow chamber ( Figure 7a).
In these devices, the shear stresses exerted on cells are approximately equal to the chamber wall shear stresses, allowing fine tuning of these parameters. Two-dimensional (2D) plates are a F I G U R E 6 Weinbaum lacunar-canalicular porosity model. (a) Diagram showing a simplified trabecular cross-section with bone lining cells and osteocytes submitted to oscillatory axial and bending loads as studied by Weinbaum et al. (1994). The fluid displacement caused by the loading in the periosteocytic space (1) of the canaliculi (2)

| Heterogeneity issues
Depending on the properties of these heterogeneities (e.g., disparity with mean value and prevalence), scaffold "average" a pipe of length L and diameter d, the resistance R is given by the following equations: For a given pressure drop, the flowrate distribution between two possible pathways of resistance R 1 and R 2 (Figure 8a) is then given by the relationship: As channel diameter ratios are raised to power 4, even small heterogeneities in pore and pore interconnection sizes can cause significant flow redistributions in a scaffold and redefine the local cell environment.
In Maes et al. (2012), 4-mm thick scaffolds manufactured by gel casting have a declared pore size of 270 µm. However, whole scaffold µ-CT revealed the presence of macropores spanning upto 1.6 mm, that is, six times the declared 270 µm pore size and 43% of the scaffold total height. The simulated streamlines clearly show that preferential flow pathways are defined by these macropores, while the rest of the scaffold is comparatively undersupplied (Figure 8b). In those conditions, the mechanical environment is not defined by the specified scaffold properties but by the uncontrolled heterogeneities associated with the manufacturing process.

| Distribution issues
The appeal of some irregular structures is in the way they mimic the complexity of the physiological environment. However, random and complex structures are ineffective when trying to understand cell responses to specific stimuli. Most current monitoring techniques (e.g., alkaline phosphatase activity, ARN, and bone-specific protein expression) allow only the study of cell populations as a whole, thus averaging out the effects of the local flow ranges found across a scaffold (Voronov, VanGordon, Sikavitsas, & Papavassiliou, 2010). Therefore, scaffolds generating

| Presentation
To estimate fluid circulation velocity and shear stress levels inside the scaffold from the flowrate value and scaffold properties, Goldstein et al. (2001) introduced a simplified scaffold model. In this approximation, the complex scaffold geometry is simplified by reducing the interconnected pore network to a bundle of parallel, cylindrical channels whose diameters are equal to the scaffold average pore size (Figure 9; Grayson et al., 2008). This model allows the direct use of the simple relationships between the setup parameters and resulting flow effects presented in Section 2.1 (cf. Eq. (1) to (5)). Currently, shear stress values can be predicted through CFD through 3D models computed from scaffolds Cioffi, Boschetti, Raimondi, & Dubini, 2006;Jungreuthmayer et al., 2009;Maes et al., 2012;McCoy et al., 2012;VanGordon et al., 2011). The steadily increasing computational power and availability of high-end CFD software make these simulations the most reliable tool at our disposal to assess mechanical constraints in complex geometries. By comparing shear stresses predicted by CFD data with corresponding mean shear stress values predicted by the Goldstein approximation, we found that although suitable in some studies (Cioffi et al., 2006;Jungreuthmayer et al., 2009;Maes et al., 2012), the approximation could also generate heavily inaccurate results, as highlighted in Figure 10. F I G U R E 1 0 Goldstein approximation robustness. Histogram of the log2 ratio of the computational fluid dynamics (CFD)-generated shear stress values over the average shear stress in the scaffold predicted by the Goldstein approximation. Log2 values vary between −5.5 and 1.8, indicating that depending on the study, the Goldstein approximation gives shear stress values ranging from 45 times lower (Kleinhans et al., 2015) to three times higher (McCoy et al., 2012) than those predicted by CFD used in recent studies (Kleinhans et al., 2015;Vetsch, Betts, Muller, & Hofmann, 2017) to encompass scaffold hydraulic properties (e.g., porosity and permeability) while reducing the complexity of the simulation. Figure 11a shows the shear stress distribution at the median transverse cut plane predicted by this model for a flowrate of 12 ml/min (Vetsch et al., 2017). Figure 11b shows the shear stress distribution in a similar scaffold generated by finite element analysis (FEA) based on the actual scaffold geometry for a lower flowrate of 0.3 ml/min. With similar architectural features, shear stress should be proportionally higher for a 12 ml/min flowrate than for 0.3 ml/min (Equation (5)

| Other causes of interference in the interpretation of results
In addition to altered flow effects, sources of variability in cell behavior can stem from other scaffold architectural features as well as the intrinsic chemical properties of the scaffolds and culture media or mechanical stimulation methods. The significant impact of each of these features on in vitro bone growth has been well documented, but it seems to be generally overlooked by teams with unrelated specializations.
Apart from the permeability, mainly controlled by the macroscopic porous network, the size and geometry of macropores and interconnections were also proved to influence cell colonization, tissue growth and osteogenesis, for example, concave surfaces being greater for tissue growth, osteogenesis and microcapillary-like structure self-assembly than convex surfaces (Bianchi et al., 2014;Gariboldi, Butler, Best, & Cameron, 2019;Juignet et al., 2017;Rumpler, Woesz, Dunlop, van Dongen, & Fratzl, 2008).

| Influence of the chemical environment on cell responses to mechanical stimuli
Material surface physicochemical properties (e.g., solubility, wettability, hydrogen potential) and culture medium have a defining impact on in vitro cell behavior and shear stress responses.

Shear stress responses conditioned by surface chemistry
Cells approaching an implant material do not make direct contact with its surface but interact with a layer of proteins rapidly adsorbed from the culture medium (Anselme, 2000). Because cells depend on specific proteins for anchorage and extracellular cues, the composition and conformation of adsorbed proteins at the material surface are key mediators of cell behavior Fourel et al., 2016;Streuli, 2009;Wilson, Clegg, Leavesley, & Pearcy, 2005). Material intrinsic properties and particularly surface properties (e.g., surface chemical functional groups) affect the key early event of protein adsorption and subsequent cell adhesion, growth, and differentiation (Anselme, Ponche, & Bigerelle, 2010;Vitte, Benoliel, Pierres, & Bongrand, 2004), as well as their response to shear stress

Influence of medium composition
Culture medium ionic environment, pH value, and especially osteogenic supplementation such as dexamethasone, β-glycerophosphate, and ascorbic acid have been shown to have a significative impact on cell proliferation, differentiation, and overall bone development (Brunner et al., 2010;Monfoulet et al., 2014;Nishimura et al., 2015;Vetsch, Paulsen, Muller, & Hofmann, 2015), but also on cells response to a dynamic perfusion environment. In the presence of osteogenic supplementation human mesenchymal stem cells subjected to shear stress demonstrated significantly stronger increases in growth and alkaline phosphatase activity (Farack et al., 2011). Fetal bovine serum (FBS), another standard supplement of cell culture media, leads to significant differences in experimental outcomes and have in some instances been shown to cause spontaneous mineralization in silk-fibroin scaffolds, even without cells present (Vetsch, Paulsen, et al., 2015).
Furthermore, FBS-associated RNA is coisolated with cell culture-derived extracellular RNA and interferes with the downstream RNA analysis. FBS transcripts can also be taken up by cultured cells and affect the results of gene expression profiling technologies (Wei, Batagov, Carter, & Krichevsky, 2016). In secretome profiling, proteins contained in the FBS often mask the proteins secreted by cells, concealing their identification by mass spectrometry (Nonnis et al., 2016). Ill-defined medium supplementation and recurrent variability in serum batch composition (Brunner et al., 2010) introduce several unknown variables into the cell culture system and might be a major reason why different laboratories are unable to reproduce data published in the literature (Vetsch, Paulsen, et al., 2015).
Additionally, most biological in vitro assays are performed under an atmospheric oxygen concentration (pO 2 = 20%-21%). However, the native environment of bone stem cells contains much less oxygen (e.g., between 1.3% and 4.2%; Spencer et al., 2014). This "in situ normoxia" (Ivanovic, 2009)  Another popular stimulation methodology is to submit cells to the dual effect of perfusion and cyclic mechanical loading of the scaffold; submitting cells to both flow-induced shear stresses and substrate deformation is an attempt to better reproduce in vivo stimuli David et al., 2008;Dumas et al., 2009;Stops, Heraty, Browne, O'Brien, & McHugh, 2010;Zong ming et al., 2013).
Nonetheless, to our best knowledge, fluid movement due to substrate deformation has never been quantified, making the resulting cell responses much more arduous to interpret as the flow patterns remain unknown.

| Defining culture standards
As previously explained, the most common approach since the introduction of 3D perfusion bioreactors has simply been to test combinations of cells and scaffolds and link the observed biological effects to the corresponding flowrate value. Unfortunately, the conclusions of these studies are nongeneralizable as they cannot be related to fluid circulation speed and shear stress values, which are two key factors independently affecting bone cell behaviors and tissue development.
Therefore, the first challenging target in the 3D controlled culture system is the determination of the ranges of both shear stress and mass transport values responsible for cell survival and osteogenic stimulation, as outlined in Figure 5c. This challenge can be overcome by the purposeful design of shear stress distribution within the scaffolds at a given velocity. Specifically, narrowing the ranges of both parameters to a homogeneous mechanical environment would allow their direct association with the observed biological response. How a given flowrate will translate into this combination of fluid circulation speed (determining nutrient transport, waste management and paracrine communication mechanisms) and shear stress depends on the bioreactor design and scaffold properties.

| System design
In addition to a relevant stimuli regimen, a necessary condition for controlling the scaffold mechanical environment through the flowrate is to ensure that all the medium is actually flowing through the scaffold. The scaffolds could be press-fitted into custom designed sealing systems to ensure proper perfusion and prevent undesired flow pathways (Du et al., 2015). Leaks can be reduced by adjusting the fluidic circuit to decrease hydrostatic pressure build-ups and ensuring that components do not deteriorate over the culture duration (Allori et al., 2016). Homogeneous perfusion also requires a homogeneous flow pattern to expose the whole scaffold surface to equivalent flowing conditions. Thus, the bioreactor chamber must be designed in accordance with its expected operating flow rate range to ensure adapted velocity fields throughout the scaffolds. Moreover, depending on the flow conditions, ill-designed chambers can generate swirls. Swirls can lead to disruptive flow patterns and may lead to medium stagnation, compromising the culture (Freitas, Almeida, & Bartolo, 2014;Vetsch, Hofmann, & Müller, 2015).
Therefore, in the development phase, computational simulation of the velocity fields must be performed on the full bioreactor chamber volume to define its suitable design with respect to the expected operating flow parameters and scaffold properties.

| Scaffold properties
Scaffold architectural features and the ability to accurately control them play a central role in achieving an exploitable mechanical environment (Choi, Zhang, & Xia, 2010). We find essential to note that randomness often seems to be mistaken for homogeneity (Liu et al., 2018;Maes et al., 2012;Qian, Yuan, Zhimin, & Anchun, 2013).
Conventional scaffold manufacturing methods (e.g., particle leaching and fiber meshing) can create only structures with high variability in shape and size within their macroporous network, resulting in "uniformly dispersed" shear stresses and fluid velocities within the scaffolds at best (Liu et al., 2018).  (Allori et al., 2016;Sonnaert et al., 2017). Moreover, most of the time, flow simulations in AM scaffolds are run with the original 3D model (Allori et al., 2016;Guyot et al., 2014;Sonnaert et al., 2017), neglecting that 3D-printed scaffolds can display significant intersample variability and deviation from their original design (Marin & Lacroix, 2015).
For generating ranges narrow enough to discriminate the values of shear stress and fluid velocity to which a given cell type is most responsive, an accessible design would comprise cylindrical channels of equivalent diameters arranged alongside the flow direction, actually generating the ideal environment assumed in the Goldstein approximation. The large class of periodic minimal surfaces is also particularly interesting for these in vitro applications. Indeed, triply periodic minimal surfaces ensure a regular macroporous network and can be extensively manipulated, allowing for easier design of shear stress distributions.
Moreover, the permeability of these structures is more than tenfold greater than that within a scaffold with random-pore architecture of comparable porosity and pore size (Melchels et al., 2010).

| Disentangling flow effects
As shear stress is proportional to both fluid velocity and viscosity, we can use these parameters to separate the effects of mass transport and shear stress on tissue growth. Increasing the medium viscosity by adding dextran, which does not seem to have an impact on hBMSC cultures at low concentrations (Li, Dai, & Tang, 2008), exposes cultured cells to increased levels of fluid shear stress while maintaining essentially constant chemotransport conditions for nutrient delivery and waste removal. Flowrate and viscosity can also be modified simultaneously to expose cells to various speeds of circulation while maintaining constant shear stress. This strategy established that shear stress and mass transport levels are independent biological stimuli (Li et al., 2009;Sikavitsas, Bancroft, Holtorf, Jansen, & Mikos, 2003). The systematic application of this approach in scaffolds generating narrow and predictable ranges of shear stress and velocity would allow the determination of the optimal fluidic environment for bone growth and more generally, a better understanding of the effects of culture conditions on a given cell type in 3D perfused systems. Since not all cells react in the same way to chemical and mechanical stimuli, it is essential to validate the parameters of the system with the cell type(s)  most adapted to the scientific question or intended application. As a key component of the 3D culture system, the cells used have to be thoroughly sorted and selected, for instance with antibodies targeting cell-specific surface markers (Camilleri et al., 2016;Zhang et al., 2019).

| Perspective
The next challenge is the large-scale implementation of standardized 3D

| Automation
Reasonably, automation has been repeatedly proposed as a solution for standardization and cost reduction (Martin, Smith, & Wendt, 2009;Martin, Wendt, & Heberer, 2004;Nerem, 2014;Salter et al., 2012;Tandon et al., 2013;Yeatts & Fisher, 2011). However, until recently, process automation was associated with high upfront cost and required specific sets of skills to implement. Currently, the increased availability of programmable commercial modular solutions dedicated to fluidics (e.g., Elvesys®, Fluigent®, Cellix®) and accessible microcontrollers (e.g., Arduino®, BeagleBone®, Raspber-ryPi®) associated with generic fluidics components (e.g., electrovalves, switches, manifolds) paves the way in every laboratory for easier automation of tasks, such as medium sampling and renewal, or even critical steps such as scaffold cell seeding. Many cell seeding protocols do not grant a homogeneous cell distribution, and involve time consuming procedures and technical handling of the seeded scaffold which places unnecessary stress on cells and increases contamination risks. In contrast, automated bioreactor systems can deliver safe and standardized production of engineered tissue constructs, maximizing prospective scale-up and cost-effectiveness in the long term (Martin et al., 2009).

| Managing culture evolution
In continuous perfusion studies, the flowrate remains constant for the duration of the experiment. However, as bone growth progresses, the construct initial porosity decreases, and pores either decrease in size or become completely obstructed, modifying the mechanical environment to which cells are exposed over time (cf. Section 2). Considering a constant flowrate into a cylindrical channel of initial diameter d 0 constricted into a HADIDA AND MARCHAT | 265 channel of diameter d t after a period (t) of ECM deposition, the resulting circulation velocity v t and shear stress at the ECM surface τ t in this very simplified model are given by the following equations, obtained by developing Equations (1) and (5) In experiments involving significant tissue growth in regard to the scaffold's available space, fixing the flowrate systematically initiates an accelerating increase in fluid velocity and shear stress, potentially offering a partial explanation for why many bone cell populations seem to die down after a few days or weeks in vitro. Although working with a constant perfusion flow rate is the most popular methodology, pumps with integrated pressure control can provide a constant pressure drop. In the simplified model described above, the relations between the evolving fluid velocity, shear stress, and the initial conditions in this configuration are then given by the following equations, obtained by developing Equations (5) and (8)with a constant ΔP: Moreover, these models approximate the ECM surface to a moving solid boundary, whereas interstitial flow within the ECM is a key element intervening in the osteocytic differentiation of embedded cells. Compared to the flow-induced shear stress at the surface of the ECM, interstitial flow within the ECM generates significantly higher levels of shear stress for embedded cells (Guyot, 2015), which are also more likely to be in a bridged configuration ( Figure 7b). In addition, to shear stress, bridged cells are also submitted to drag forces (You et al., 2004), which entail greater levels of deformation. Thus, surface shear stress decreasing proportionally to ECM growth (Equation 11) offers an interesting configuration that should be investigated.
Combined with live cell monitoring and automation, numerical bone growth models (Guyot, 2015), integrating the deposition of ECM and its impact on the macroporous network and resulting flow environment, will help to adjust and maintain optimal flow parameters throughout the culture. in conventional 2D (138) or static 3D (e.g., gels; Vazquez et al., 2014), have provided valuable data on osteoblast-osteoclast interactions and emphasized the role of osteocytes as sensors and orchestrators of the function of both osteoblasts and osteoclasts Florencio-Silva et al., 2015;Owen & Reilly, 2018;Zhu et al., 2018).
These combinations of tissue models could also be used as operative platforms for the evaluation of the efficacy, safety, and toxicity of drug candidates (Edington et al., 2018;Ishida, 2018;Kimura, Sakai, & Fujii, 2018;Tetsuka, Ohbuchi, & Tabata, 2017) and medical devices (Guan et al., 2017), potentially boosting research time and cost efficiency while coming into the scope of the 3Rs principles (replace, reduce, and refine). Although the opportunities offered by these models are game-changing, published proof-of-concept studies and prototypes have yet to switch from technology research to actual biological, clinical, and biomedical applications (Junaid et al., 2017;Kimura et al., 2018).

| CONCLUSION
The osteogenic effects of perfusion flow have been discussed for two decades, with the aim of delivering proper guidelines regarding the adequate parameters for bone tissue growth in 3D perfused systems.
In this review, we identified multiple factors contributing to this limitation.
A given flowrate breaks down into different shear stress and circulation velocity levels depending on the scaffold features, independently defining the mechanical and chemical cellular environments. Thus, determining generalizable osteogenic culture conditions will require a shift in focus from determining the optimal flowrate in a given setup to defining the optimal combination of shear stress and circulation velocity for a given cell type. A tighter control over the cell environment can be achieved through replacing commonly used approximations by more rigorous culture and scaffold design including flow simulations ahead the system implementation. When possible, implementing automation will ensure higher degrees of repeatability, lighten the culture workload, and provide exciting perspectives when combined with live monitoring of cell activity.
Achieving control over the cell environment and resulting translatability will provide a solid basis for deepening our understanding of the relations between various culture parameters and biological responses in BTE. Eventually, the understanding of these relations will steer the development of new approaches to bone diseases, replacement, and interactions with other organs.