Predictions for optimal mitigation of paracrine inhibitory signalling in haemopoietic stem cell cultures
 Joseph D Berry^{1},
 Pankaj Godara^{2},
 Petar Liovic^{1} and
 David N Haylock^{2}Email author
DOI: 10.1186/s1328701500487
© Berry et al.; licensee BioMed Central. 2015
Received: 21 October 2014
Accepted: 10 March 2015
Published: 16 April 2015
Abstract
Introduction
Recent studies in the literature have highlighted the critical role played by cell signalling in determining haemopoietic stem cell (HSC) fate within ex vivo culture systems. Stimulatory signals can enhance proliferation and promote differentiation, whilst inhibitory signals can significantly limit culture output.
Methods
Numerical models of various mitigation strategies are presented and applied to determine effectiveness of these strategies toward mitigation of paracrine inhibitory signalling inherent in these culture systems. The strategies assessed include mixing, mediaexchange, fedbatch and perfusion.
Results
The models predict that significant spatial concentration gradients exist in typical cell cultures, with important consequences for subsequent cell expansion. Media exchange is shown to be the most effective mitigation strategy, but remains labour intensive and difficult to scaleup for large culture systems. The fedbatch strategy is only effective at very small Peclet number, and its effect is diminished as the cell culture volume grows. Conversely, mixing is effective at high Peclet number, and ineffective at low Peclet number. The models predict that cell expansion in fedbatch cultures becomes independent of increasing dilution rate, consistent with experimental results previously reported in the literature. In contrast, the models predict that increasing the flow rate in perfused cultures will lead to increased cell expansion, indicating the suitability of perfusion for use as an automated, tunable strategy. The effect of initial cell seeding density is also investigated, with the model showing that perfusion outperforms dilution for all densities considered.
Conclusions
The models predict that the impact of inhibitory signalling in HSC cultures can be mitigated against using media manipulation strategies, with the optimal strategy dependent upon the protein diffusion timescale relative to the media manipulation timescale. The key messages from this study can be applied to any complex cell culture scenario where cellcell interactions and paracrine signalling networks impact upon cell fate and cell expansion.
Introduction
Haemopoietic stem cells (HSCs) ultimately give rise to all blood cells, and as a consequence hold great promise for ex vivo production of mature blood cells for blood transfusion. However, the quantity of HSCs able to be harvested from patients is insufficient to generate the enormous numbers of cells required for clinical use. Hence, there is a critical need to increase the number of HSCs for mature cell biomanufacture [1]. A common approach used to expand HSCs is to culture them under static conditions in fully defined media without serum but with supplementation of early acting synergistic factors that promote HSC survival and proliferation [26]. Protocols for the expansion of HSCs are usually formulated to ensure that sufficient growth factors are provided in the initial cell culture medium for the duration of the culture period. More sophisticated ex vivo culture systems utilise various feeding strategies to provide sustained levels of the key haemopoietic growth factors required for maximal cell production at the end of the culture period.
The fate, proliferation and differentiation of HSCs within ex vivo culture systems are ultimately determined by the interplay between the intrinsic properties of the HSCs and a multitude of extrinsic signals that collectively influence growth. Simplistically, extrinsic cues can be considered as either stimulatory or inhibitory and the relative magnitudes of these competing influences will determine HSC response. Until recently, HSC expansion strategies have focussed mainly on what combination and concentration of stimulatory regulators need to be provided to ensure optimal cell proliferation and selfrenewal decisions. However, recent studies from Zandstra and colleagues have highlighted the influence of combinations of cellsynthesised inhibitory proteins present at subthreshold levels that significantly limit expansion of HSC and their immediate progeny [710]. These negative feedback regulatory loops are important in HSC cultures, especially those where cells are seeded at high density, resulting in minimal distance between precursor cells and/or their nascent progeny.
Bioreactor systems for HSC growth and expansion should be designed to provide adequate amounts of stimulatory factors and cytokines, glucose and other essential metabolites to promote the survival and division of cells, but also remove or mitigate the effect of cellsynthesised inhibitors. Approaches for mitigating the effect of these inhibitors include but are not limited to: mixing and redistribution of cells and media; removal of inhibitory factors by media exchange; removal of inhibitory factors by continuous media perfusion; dilution of inhibitory factors by continuous media addition (fedbatch culture); sequestration of inhibitory factors; and targeted molecular negation of specific inhibitory factors.
Models that predict growth of HSC cultures under different mitigation strategies enable rapid and efficient identification of optimal operating conditions for ex vivo culture systems, eliminating the need for expensive and timeconsuming device prototyping. Kirouac et al. formulated a HSC growth model to describe regulation of HSC populations through intercellular communication caused by cellsecreted factors [9]. This was initially applied to static cultures (with and without media exchange), and was extended by Csaszar and colleagues to include perfused and fedbatch cultures [8]. The model is formulated under the assumption that these secreted factors are instantaneously redistributed throughout the cell culture. This assumption implies that the proteins secreted by cells are effectively massless, and concentration gradients are not present in typical cell cultures.
Herein we describe a mathematical approach for considering the impact of inhibitory proteins on HSC growth, based on consideration of their molecular mass and size. We use analytical and numerical models to predict optimal strategies for mitigating the effects of inhibitory signalling on the expansion of HSC populations.
Materials and methods
The equations and models used in this study are presented in Additional file 1. To determine the evolution of the secreted factor ϕ in static, mixed, fedbatch and mediaexchange cell cultures under the assumption of fixed and uniform cell density, the governing equations were discretised and solved using the FiPy partial differential equation solver [11]. The domain was split into 100 equal elements, and a time step of 300 seconds was used. When the assumption of fixed cell number was relaxed, the Gear method was used to solve the governing equations for these cultures as an initial value problem, available as a standard numerical library in Python (http://www.python.org). The method of lines was used to discretise the transient diffusion equation spatially, allowing it to be solved using the Gear method. The domain was split into 200 elements to accommodate the growth of the domain in the Eulerian frame.
The governing equations for the perfusion culture were solved using ANSYSCFX (ANSYS, Canonsburg, PA, USA), with a constant time step of 400 seconds. The symmetry of the geometry was exploited to reduce computational expense. The mesh size used was approximately 27,000 nodes, and the mesh was graded to ensure that large spatial gradients of ϕ near the bottom of the cell culture were sufficiently resolved. Tests were performed to ensure that the mesh resolution was of adequate size to provide accurate results. Tests were also performed to ensure that, with Q = 0, the growth model for both the perfused and fedbatch cultures gave the same result at the static culture limit.
Results
In contrast, fedbatch and perfusion cultures as considered in this study are both continuous strategies. For fedbatch cultures, medium is assumed to be continuously added at a constant rate of Q ml/day. No medium is removed from the culture, and thus the inhibitory protein concentration is diluted as the culture volume increases. In practice, it is difficult for dilution to be a continuous process: Csaszar and colleagues performed experiments using a semicontinuous dilution regime with 30 minute dilution cycles [8]. In our models we assume that dilution acts continuously, which gives an upper limit for the performance of fedbatch systems. Further, the models here assume that the action of dilution imparts no mixing on the media already present, which is a reasonable assumption if medium is added at very low flow rates (see Additional files 2 and 3). In a perfused culture, fresh medium is continually delivered to the culture system, and spent medium is continually removed, at the constant rate of Q ml/day, such that the cell culture volume remains constant over time.
We are assuming in the numerical models that the cell redistribution due to the media manipulations has a negligible effect on the subsequent growth of the cell population. This is a reasonable assumption for both the perfusion and dilution strategies, because the shear stresses exerted on the cells for the flow rates considered are extremely low. However, cell redistribution of the cells is a natural consequence of the mixing and media exchange strategies. The effect of this redistribution is unknown, and very difficult to quantify because it is an inherent part of practical media manipulation. The action of cell redistribution may have a positive effect on cell expansion because it exposes cells to new neighbours, but it may also be deleterious because the close cell contact necessary for effective juxtacrine and paracrine signalling will be periodically disrupted. We also assume here that laboratory incubator vibrations have negligible effect on mixing in the cell cultures, because they are illdefined and can be easily negated with the use of vibration damping systems.
Two types of fedbatch devices are considered in this study: a circular well with the same footprint as a single well on a standard tissue culture 24well plate, and a 12 ml culture bag. Cells are considered to reside on the bottom surface of each device due to the action of gravity. The surface area of the well is 2 cm^{2}, and the surface area of the culture bag is 30 cm^{2}. For the perfused culture scenarios considered, the inlet and outlet pipes are located halfway up the well, and the top surface of the culture is considered to be a solid wall. If Q = 0 ml/day for either the fedbatch or the perfused culture, the system reduces to a static culture. The well culture device is assumed to be manufactured from thin polydimethylsiloxane, which allows oxygen to permeate through the walls of the device [1214], therefore placing no practical limitation owing to oxygen diffusion considerations on the height of the media within the device.
Physical properties of known haemopoietic stem cell inhibitory factors
Inhibitory factor  Molecular weight (kDa)  Diffusivity (m ^{ 2 } /second) 

TGFβ  26  6.43 × 10^{−11} 
TNFSF9  23.8  6.76 × 10^{−11} 
MIP1α  8  1.23 × 10^{−10} 
MIP1β  8  1.23 × 10^{−10} 
IP10  9  1.16 × 10^{−10} 
NAP2  8  1.23 × 10^{−10} 
SPARC  40  5.07 × 10^{−11} 
PDGFCC  13.4  9.28 × 10^{−11} 
Parameter  Symbol  Value  Units 

Maximum proliferation rate of Lin^{−} cells  u _{ max }  6.26  1/day 
Maximum proliferation rate of Lin^{+} cells  u _{+}  0.204  1/day 
Maximal proliferation compartment  n _{ max }  5.32  – 
Proliferative decay term  D _{ gr }  3.38  – 
Selfrenewal probability  f _{ max }  0.634  – 
Selfrenewal decay term  D _{ sr }  1.96  – 
First SF1 Hill coefficient  k _{1}  0.614  – 
Second SF1 Hill coefficient  k _{s}  1.08  – 
SF2 Hill coefficient  k _{2}  0.555  – 
SF3 Hill coefficient  k _{3}  0.625  – 
SF4 Hill coefficient  k _{4}  0.533  – 
SF1 concentration inducing 50% maximal SF2 secretion  L _{ s }  0.915  pg/ml 
Cell cycle Hill coefficient  k _{ t }  5  – 
Time for 50% of cells to enter cycle  τ _{ D }  2  Days 
SF1 secretion rate  r _{1}  1.94 × 10^{−6}  pg/cell/day 
SF2 secretion rate  r _{2}  2.40 × 10^{−6}  pg/cell/day 
SF3 secretion rate  r _{3}  4.88 × 10^{−7}  pg/cell/day 
SF4 secretion rate  r _{4}  4.34 × 10^{−7}  pg/cell/day 
Fedbatch strategy is preferable for mitigating inhibitory signalling if diffusion is not considered
Considering the effects of diffusion gives rise to significant spatial concentration gradients
To test the validity of the instantaneous redistribution assumption, we can evaluate the importance of this by considering the inhibitory protein TGFβ, with a molecular weight of 26 kDa corresponding to a diffusivity of 6.5 × 10^{−11} m^{2}/second. The height of 1 ml liquid in a 24wellplate well is approximately 5 mm, giving a characteristic time for diffusion of τ _{ D } = L ^{2}/D ~ 4.5 days. This time is significant in comparison with the cell culture time scale of τ _{culture} = 8 to 16 days. Therefore, significant concentration gradients will be present in typical cell cultures, and diffusive transport needs to be accounted for in growth models in order to accurately assess the effectiveness of mitigation strategies.
By taking into consideration the protein diffusivity we are able to predict the concentration of protein at any distance from the cellular source. To illustrate this, Figure 2b shows the concentration profile of TGFβ after 1 day in a 24wellplate well geometry for a culture of 10^{5} cells secreting TGFβ at a fixed rate of r _{ ϕ } = 1.94 × 10^{−6} pg/cell/day. The concentration of TGFβ at the bottom of the well is approximately 0.45 pg/ml, which is double the value predicted by the instantaneous redistribution model. Including the effects of diffusion thus gives rise to significant spatial concentration gradients, which need to be considered when comparing strategies to mitigate the effects of different inhibitory proteins.
Assuming instantaneous redistribution of inhibitory signals overpredicts the effectiveness of fedbatch cultures
The concentration of inhibitory protein at the bottom of a cell culture will thus increase with culture time as ϕ _{cell} ∝ t ^{1/2}.
The uniform dilution assumption can be considered to give a bestcase prediction for practical dilution experiments where medium is added to the top or to the side of the culture. Notably, the effect of diffusion causes the inhibitory signal concentration to increase with time, whereas the instantaneous redistribution assumption predicts that the cell exposure concentration becomes constant, and the value it predicts is far below the value predicted when diffusive effects are considered (Figure 3). Hence, assuming instantaneous redistribution of inhibitory signals throughout the cell culture leads to an underprediction of the concentration of inhibitory proteins exposed to the cells and therefore an overprediction of the effectiveness of a fedbatch culture.
When spatial gradients of concentration are taken into account, perfusion is a more effective means of reducing exposure of cells to paracrine inhibitory signalling than dilution
The optimal inhibition strategy depends upon the diffusive time scale, measured by the Peclet number
Here τ _{ d } is the diffusive time scale, τ _{ s } is the media manipulation time scale and L is the characteristic length scale, which here is the media height h. The media manipulation time scale is dependent on the mitigation strategy being considered. For fedbatch and perfused cultures, the media manipulation time scale is τ _{ s } = V _{0}/Q. For mixed cultures, the media manipulation time scale is τ _{ s } = 1/f _{mix}. For the mediaexchange strategy, where αV _{0} ml is exchanged every 1/f _{ME} days, the mediamanipulation time scale is τ _{ s } = 1/αf _{mix} .
The limit of Peclet number Pe = 0 corresponds to the instantaneous redistribution assumption. To allow direct comparison of the relative effectiveness of each mitigation strategy, the media manipulation time scale is chosen as τ _{ s } = 1 day. Thus, the flow rate of the fedbatch and perfused cultures is Q = V _{0} ml/day, and the mixing frequency is f _{ME} = 1/day. Two different mediaexchange strategies with τ _{ s } = 1 day are considered: full media exchange daily (α = 1, f _{ME} = 1/day); and half media exchange every 12 hours (α = 0.5, f _{ME} = 2/day). Four different cell cultures are considered: a 12 ml culture bag of initial volume 1 ml, and a well with three different initial volumes of 0.2, 1, and 5 ml. These correspond to initial media heights of 0.33, 1, and 5 mm respectively.
The 0.2 ml culture in the 24wellplate well gives Pe = 0.18, and thus the effect of mixing is still negligible. The fedbatch strategy is more effective at mitigating inhibitory signalling than media exchange for the first 4 days of this culture, but is less effective for longer culture times. For the first day of cell culture, the fedbatch strategy outperforms perfusion, but is much less effective thereafter. Increasing the culture volume to 1 ml in the well also increases the Peclet number to 4.45, indicating that the media manipulation time scale is now faster than the diffusive time scale. As a consequence, mixing becomes a feasible mitigation strategy, because the inhibitory protein only has time to partially diffuse throughout the cell culture before mixing occurs again. For this culture scenario, perfusion and media exchange are the most effective strategies, both clearly outperforming dilution. The dilution strategy is only marginally effective at mitigating inhibitory signalling for this media height because the secreted protein is not able to diffuse away from the cells over the life of the culture.
The act of dilution increases the Peclet number over time, negating the influence of dilution as a mitigation strategy
The diffusive time scale becomes slower in comparison with the dilution time scale over the cell culture period, indicating that the effect of dilution is diminished. This is shown clearly in Figure 5, where the positions of the free surface and the diffusive front are marked in the cell culture. Here the diffusive front is defined as the height in the cell culture where the concentration is 5% of the value at the bottom. After 1 day of culture, there is medium present in the well with very low concentrations of TGFβ relative to the concentration at the bottom of the well. The diffusive time scale at this point in time is approximately nine times greater than the dilution time scale (Pe = 8.9), and there has not been sufficient time for the protein to diffuse uniformly throughout the entire culture. Thus, there are already regions within the cell culture medium with very low concentrations of inhibitory protein. This represents a sink for the inhibitory proteins to diffuse into, and as such there is no benefit in addition of more media by dilution. This is clearly evident in Figure 5: as the cell culture period increases so does the Peclet number, and the position of the diffusive front moves further below the position of the free surface, indicating that there is medium already available within the cell culture for the inhibitory signal to diffuse into. At day 2 the Peclet number is 13.4, and at day 8 the Peclet number of the culture is approximately 40.
Cell growth within a perfused culture increases with increasing flow rate, and perfusion is able to sustain expansion at higher cell seeding densities
Mixing and media exchange strategies become more effective as the Peclet number increases
Discussion
This study has presented numerical models of mixed, mediaexchange, fedbatch and perfused cultures that take into account the finite mass of secreted paracrine signals, allowing spatial concentration gradients to form within cell cultures. A growth model is also developed that takes into account the finite mass of secreted factors. The models presented here show that the optimal mitigation strategy is dependent on the Peclet number, which measures the size of the diffusive time scale relative to the mediamanipulation time scale. When the Peclet number is very small, the inhibitory protein redistributes quickly throughout the cell culture, and the fedbatch approach is a useful strategy. However, we have shown that there is limited scope for tunable feedback control in fedbatch cultures unless the Peclet number remains very small over the life of the cell culture. Even for low initial Peclet numbers (Pe_{0} ≿0.2), the effectiveness of dilution diminishes over the cell culture period. This is because the action of dilution causes the cell culture volume to continually increase, and consequently the Peclet number also increases with time. This increase in Peclet number over time diminishes the effectiveness of dilution as a means for reducing the concentration of inhibitory proteins where cells are actually growing.
As the Peclet number increases, the size of the diffusive time scale increases relative to the dilution time scale, and there are regions of media with very low inhibitory concentration already present in the cell culture. Hence, the addition of more fresh media to the culture has negligible effect. This prediction is consistent with previous experimental data on fedbatch cultures. Further, the tunability of fedbatch cultures is also restricted by the dilution Peclet number increasing with flow rate as Pe ∼ Q ^{2}, whereas for other mitigation strategies with constant cell culture volume the Peclet number is only linearly proportional to the flow rate. For moderate to high Peclet numbers, the numerical models predict that media exchange is the best strategy, but as a strategy it is labour intensive, especially if the frequency of exchange is shorter than 1 day. Further, media exchange is too difficult to implement in largescale culture systems. At moderate to high Peclet numbers, mixing represents the most economical method, with significant mitigation predicted without need for extra media to be added to the cell culture.
It has been shown here that perfusionbased systems are more effective at mitigating the effects of inhibitory signalling in HSC cultures than dilutionbased approaches. Other strategies where cell culture volume is kept constant are more amenable to tunability than dilution, especially perfusion. Increasing the flow rate increases the effectiveness of the perfusion strategy, and thus has significant scope as an automated, tunable strategy for optimising cell expansion. Furthermore, the perfusion culture scenarios considered here are not optimised. Significant improvement in efficiency may be achieved through enhancing the design of the perfusion device and its operating conditions. There are extensive reports on the fluid flow and nutrient distribution within perfusion bioreactors [2328]. Using the growth model presented here to predict the distribution of inhibitory factors and the subsequent effect on cell expansion in these devices will allow efficient assessment of design and operating conditions to optimise expansion whilst eliminating the need for expensive and timeconsuming prototype iterations.
The growth model as it stands will underestimate cell expansion in perfused cultures because perfusion, shown here to effectively lower the concentration of endogenous inhibitory factors, will also act to lower the concentration of endogenous stimulatory factors and hence also lower the predicted cell expansion. However, the growth model does not take into account exogenous stimulatory factors that are already present at very high concentrations in the supplied media. The effect of perfusion on the stimulatory factor concentration will thus be minimised, as fresh stimulatory factors will be brought in to replace those swept away. Kirouac presented a variation of the original model that did not include the positive proliferation feedback factor ϕ _{3}, showing that it was a slightly better fit to the experimental data [29]. However, both stimulatory factors were retained in the presented model to fully capture the effect of positive feedback loops observed in vivo [30]. There is thus a need to reformulate the growth model to remove the influence of positive feedback loops, and to account for exogenous stimulatory factors already present, in order to fully capture the overall effect of these types of mitigation strategies. There is also a need to reformulate the growth model in terms of celltosurfacearea ratio, rather than cell seeding concentration. For example, cells in a culture seeded at 10^{5} cells/ml are packed much closer together in a standard tissue culture 24wellplate well than on the surface of a 12 ml culture bag. Assuming that the cells are of diameter ~15 μm, this corresponds to an initial average distance between cells of ~40 μm in the well, and ~170 μm in the culture bag. Reformulating the growth model to account for the celltosurface area ratio will allow more accurate predictions of cell growth, and also capture the effect of heterogeneous inhibitory signal distribution within perfused and diluted cultures.
The growth model used here suggests that there is an upper limit to growth for loosely adherent cells within a specified surface area, independent of the mitigation strategy used. Quantification of this upper limit would require new estimates of the growth model parameters. However, it can be conjectured that once the surface area of a cell culture has been exhausted (that is, the cells become fully confluent), cells will begin to stack one upon another. As a result, strategies such as perfusion and dilution will be unable to mitigate against paracrine signalling for cells not on the uppermost layer. Thus, supplementary strategies such as cell redistribution and mixing of the entire cell culture may become critical at high cell densities to push cell expansion beyond the limit possible with perfusion/dilution alone.
The optimal approach to mitigate inhibitory signalling may consist of changing from one strategy to another during the life of the cell culture depending upon levels of paracrine signalling present, and the Peclet number of the cell culture. For example, dilution could be used initially as it appears to be the most effective strategy in the early stages of cell culture, and then perhaps perfusion, media exchange or mixing could be utilised to further enhance cell expansion. The models described herein provide the basis for a detailed cost analysis, specifically for the expensive growth factors that will be consumed at different amounts depending on the culture manipulation and feeding strategy being deployed. However, more knowledge of specific inhibitory proteins and how they act in combination is required to determine optimal mitigation strategies. This is particularly relevant in cultures where a heterogeneous mix of stem cells and differentiated cells are present. In this case – as often occurs with the asynchronous division and differentiation of HSCs – maturing, differentiated cells can generate a mixture of inhibitory proteins and these may act cooperatively to inhibit nascent primitive haemopoietic progenitors present in culture.
Future work will focus on developing a more generalised growth model, which also takes into account the difference in diffusivities of different inhibitory factors. In particular, a significant improvement would be the inclusion of stochastic modelling of individual factor trajectories, such as the approach developed by Moledina and colleagues [31], giving rise to cell population heterogeneity in the culture.
Conclusions
Numerical models of HSC cultures that incorporate inhibitory signalling and the effects of finite protein size are presented. The models are applied to determine the effectiveness of various strategies used to mitigate the effects of inhibitory signalling inherent in these cultures. The strategies assessed include mixing, mediaexchange, fedbatch and perfusion approaches. The results show that the optimal approach to mitigate inhibitory signalling in HSC cultures is dependent upon the relative sizes of the protein diffusion time scale relative to the time scale of the media manipulation.
Endnote
^{a}Here and throughout the paper the secretion rate is expressed in terms of actual cell number X, rather than the normalised cell number \( \tilde{X} \) used by Kirouac and colleagues [9].
Abbreviations
 HSC:

haemopoietic stem cell
 ME:

media exchange
 Pe:

Peclet number
 SF:

secretion factor
 TGFβ:

transforming growth factor beta
Declarations
Acknowledgements
JDB acknowledges Capability Development Fund support in the area of computational fluid dynamics now based in the Commonwealth Scientific Industrial Research Organisation (CSIRO) Mineral Resources Flagship. PG was supported by an Office of the Chief Executive CSIRO Science Leaders postdoctoral fellowship. DNH is a CSIRO Office of the Chief Executive Science Leader and an Australian Research Council Future Fellow, and is supported by a Science and Industry Endowment Fund project grant.
Authors’ Affiliations
References
 Lee EJ, Godara P, Haylock D. Biomanufacture of human platelets for transfusion: rationale and approaches. Exp Hematol. 2014;42:332–46.View ArticlePubMedGoogle Scholar
 Hofmeister CC, Zhang J, Knight KL, Le PJ, Stiff P. Ex vivo expansion of umbilical cord blood stem cells for transplantation: growing knowledge from the hematopoietic niche. Bone Marrow Transplant. 2007;39:11–23.View ArticlePubMedGoogle Scholar
 Simmons PJ, Haylock DN. Use of hematopoietic growth factors for in vitro expansion of precursor cell populations. Curr Opin Hematol. 1995;2:189–95.View ArticlePubMedGoogle Scholar
 Haylock DN, Nilsson SK. Expansion of umbilical cord blood for clinical transplantation. Curr Stem Cell Res Ther. 2007;2:324–35.View ArticlePubMedGoogle Scholar
 Chou S, Chu P, Hwang W, Lodish H. Expansion of human cord blood hematopoietic stem cells for transplantation. Cell Stem Cell. 2010;7:427–8.View ArticlePubMed CentralPubMedGoogle Scholar
 Audet J, Miller CL, Eaves CJ, Piret JM. Common and distinct features of cytokine effects on hematopoietic stem and progenitor cells revealed by dose–response surface analysis. Biotechnol Bioeng. 2002;80:393–404.View ArticlePubMedGoogle Scholar
 Csaszar E, Chen K, Caldwell J, Chan W, Zandstra PW. Realtime monitoring and control of soluble signaling factors enables enhanced progenitor cell outputs from human cord blood stem cell cultures. Biotechnol Bioeng. 2014;111:1258–64.View ArticlePubMed CentralPubMedGoogle Scholar
 Csaszar E, Kirouac DC, Yu M, Wang WJ, Qiao WL, Cooke MP, et al. Rapid expansion of human hematopoietic stem cells by automated control of inhibitory feedback signaling. Cell Stem Cell. 2012;10:218–29.View ArticlePubMedGoogle Scholar
 Kirouac DC, Madlambayan GJ, Yu M, Sykes EA, Ito C, Zandstra PW. Cell–cell interaction networks regulate blood stem and progenitor cell fate. Mol Syst Biol. 2009;5:293.View ArticlePubMed CentralPubMedGoogle Scholar
 Madlambayan GJ, Rogers I, Kirouac DC, Yamanaka N, Mazurier F, Doedens M, et al. Dynamic changes in cellular and microenvironmental composition can be controlled to elicit in vitro human hematopoietic stem cell expansion. Exp Hematol. 2005;33:1229–39.View ArticlePubMedGoogle Scholar
 Guyer JE, Wheeler D, Warren JA. FiPy: partial differential equations with python. Comput Sci Eng. 2009;11:6–15.View ArticleGoogle Scholar
 Leclerc E, Sakai Y, Fujii T. Cell culture in 3dimensional microfluidic structure of PDMS (polydimethylsiloxane). Biomed Microdevices. 2003;5:109–14.View ArticleGoogle Scholar
 Kim L, Toh YC, Voldman J, Yu H. A practical guide to microfluidic perfusion culture of adherent mammalian cells. Lab Chip. 2007;7:681–94.View ArticlePubMedGoogle Scholar
 Mehta G, Mehta K, Sud D, Song JW, BersanoBegey T, Futai N, et al. Quantitative measurement and control of oxygen levels in microfluidic poly (dimethylsiloxane) bioreactors during cell culture. Biomed Microdevices. 2007;9:123–34.View ArticlePubMedGoogle Scholar
 Hatzfeld J, Li ML, Brown EL, Sookdeo H, Levesque JP, O’Toole T, et al. Release of early human hematopoietic progenitors from quiescence by antisense transforming growth factor beta 1 or Rb oligonucleotides. J Exp Med. 1991;174:925–9.View ArticlePubMedGoogle Scholar
 Kirouac DC, Ito C, Csaszar E, Roch A, Yu M, Sykes EA, et al. Dynamic interaction networks in a hierarchically organized tissue. Mol Syst Biol. 2010;6:417.View ArticlePubMed CentralPubMedGoogle Scholar
 Francis K, Palsson BO. Effective intercellular communication distances are determined by the relative time constants for cyto/chemokine secretion and diffusion. Proc Natl Acad Sci U S A. 1997;94:12258–62.View ArticlePubMed CentralPubMedGoogle Scholar
 Incropera FP, DeWitt DP. Fundamentals of heat and mass transfer. Hoboken, NJ: John Wiley & Sons; 2002.
 Crampin EJ, Gaffney EA, Maini PK. Reaction and diffusion on growing domains: scenarios for robust pattern formation. Bull Math Biol. 1999;61:1093–120.View ArticlePubMedGoogle Scholar
 Madzvamuse A, Maini PK. Velocityinduced numerical solutions of reaction–diffusion systems on continuously growing domains. J Comput Phys. 2007;225:100–19.View ArticleGoogle Scholar
 Taylor GI. Dispersion of soluble matter in solvent flowing slowly through a tube. Proc R Soc A. 1953;219:186–203.View ArticleGoogle Scholar
 Aris R. On the dispersion of a solute in pulsating flow through a tube. Proc R Soc A. 1960;259:370–6.View ArticleGoogle Scholar
 Horner M, Miller WM, Ottino J, Papoutsakis ET. Transport in a grooved perfusion flatbed bioreactor for cell therapy applications. Biotechnol Prog. 1998;14:689–98.View ArticlePubMedGoogle Scholar
 Korin N, Bransky A, Khoury M, Dinnar U, Levenberg S. Design of well and groove microchannel bioreactors for cell culture. Biotechnol Bioeng. 2009;102:1222–30.View ArticlePubMedGoogle Scholar
 Sandstrom CE, Bender JG, Miller WM, Papoutsakis ET. Development of novel perfusion chamber to retain nonadherent cells and its use for comparison of human ‘mobilized’ peripheral blood mononuclear cell cultures with and without irradiated bone marrow stroma. Biotechnol Bioeng. 1996;50:493–504.View ArticlePubMedGoogle Scholar
 Walker GM, Zeringue HC, Beebe DJ. Microenvironment design considerations for cellular scale studies. Lab Chip. 2004;4:91–7.View ArticlePubMedGoogle Scholar
 Titmarsh D, Hidalgo A, Turner J, Wolvetang E, CooperWhite J. Optimization of flowrate for expansion of human embryonic stem cells in perfusion microbioreactors. Biotechnol Bioeng. 2011;108:2894–904.View ArticlePubMedGoogle Scholar
 Hutmacher DW, Singh H. Computational fluid dynamics for improved bioreactor design and 3D culture. Trends Biotechnol. 2008;26:166–72.View ArticlePubMedGoogle Scholar
 Kirouac D. Intercellular feedback in hematopoiesis. Toronto: University of Toronto; 2009.Google Scholar
 Eaves AC, Eaves CJ. Maintenance and proliferation control of primitive hematopoietic progenitors in longterm cultures of human marrowcells. Blood Cells. 1988;14:355–68.PubMedGoogle Scholar
 Moledina F, Clarke G, Oskooei A, Onishi K, Günther A, Zandstra PW. Predictive microfluidic control of regulatory ligand trajectories in individual pluripotent cells. Proc Natl Acad Sci U S A. 2012;109:3264–9.View ArticlePubMed CentralPubMedGoogle Scholar
 Doran MR, Aird IA, Marturana F, Timmins N, Atkinson K, Nielsen LK. Bioreactor for blood product production. Cell Transplant. 2012;21:1235–44.View ArticlePubMedGoogle Scholar
 Preston BN, Comper WD, Hughes AE, Snook I, van Megen W. Diffusion of dextran at intermediate concentrations. J Chem Soc Faraday I. 1982;78:1209.View ArticleGoogle Scholar
Copyright
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.