PAPER
www.rsc.org/loc | Lab on a Chip
Understanding microchannel culture: parameters involved in soluble factor
signaling{{
Hongmei Yu,a Caroline M. Alexanderbc and David J. Beebeac
Received 2nd January 2007, Accepted 27th March 2007
First published as an Advance Article on the web 19th April 2007
DOI: 10.1039/b618793e
While the importance of autocrine–paracrine signaling in vivo is clear, the ability to study the
effects of secreted endogenous factors in vitro is hampered by canonical culture platforms. In
multi-well plates, the large air–liquid interface gives rise to convective flows that continually mix
the fluid disrupting the local diffusion-based accumulation. Simple microchannels provide a more
controlled microenvironment that can be used to study secreted factor effects. Here, we utilize
microchannel culture to examine basic culture parameters and their interactions using normal
mammary gland epithelial cells (NMuMG). The following parameters were studied: (1) cell
density (80 vs. 240 cells mm22), (2) exogenous growth factors (epidermal growth factor [EGF] vs.
fetal bovine serum), (3) medium change frequency (1 h, 4 h, 12 h), and (4) culture platform
(microchannels vs. 96-well plates). The cells exhibited increased growth rates in microchannels as
compared to 96-well plates. Cell proliferation increased as the frequency of media change
decreased. For the microchannel geometries used, important threshold concentrations were
reached in a few hours. In aggregate, the results indicate that the function of the four factors and
their interactions on NMuMG growth are spatially and temporally related by molecular diffusion
in the controlled microchannel space. The convective-free microchannel environment may prove
useful for studying soluble factor signaling in vitro, and to test models and predictions of
autocrine–paracrine signaling.
Introduction
Local soluble factors play important roles in controlling cell
behavior in tissue microenvironments. These factors are
synthesized, transmitted, detected, and processed locally,
such that cells (homogenous or heterogeneous) without
direct contact can communicate within specified microenvironments. Locally derived soluble signals mediate many
biologic processes in vivo, such as pattern formation (e.g. sonic
hedgehog,1 bone morphogenetic proteins2), wound healing,
and angiogenesis (e.g. basic fibroblast growth factor,3 vascular
endothelial growth factor4). However, the specific and
complete pathways and mechanisms are typically not fully
understood, in part because the lack of spatial and temporal
control of parameters in canonical culture vessels makes
the study of autocrine–paracrine signaling challenging. An
improved culture system for studying soluble factor effects
would: (1) allow predictable soluble factor transport, (2)
provide spatial and temporal control of the microenvironment,
a
Department of Biomedical Engineering, University of Wisconsin–
Madison, Madison, WI, 53706, USA
b
McArdle Laboratory for Cancer Research, University of Wisconsin–
Madison, Madison, WI, 53706, USA
c
University of Wisconsin Comprehensive Cancer Center, University of
Wisconsin–Madison, Madison, WI, 53706, USA
{ This paper is part of a special issue ‘Cell and Tissue Engineering in
Microsystems’ with guest editors Sangeeta Bhatia (MIT) and
Christopher Chen (University of Pennsylvania).
{ Electronic supplementary information (ESI) available: Microchannel
array devices used in this study, Fig. S1–S4. See DOI: 10.1039/
b618793e
726 | Lab Chip, 2007, 7, 726–730
and (3) remove potential masking effects from undefined
exogenous factors (e.g. serum). While microchannels
inherently provide some of these attributes,5 few studies have
explored the culture parameters from the perspective of soluble
factor signaling.6,7
In this work, we performed multi-factor experiments to
evaluate the relative importance of various experimental
factors likely to be involved in soluble factor growth regulation
through their interactions in the diffusion dominant
microchannel culture environment. Four factors (and their
interactions) were examined in a full factorial randomized
experiment: (1) cell density (80 cells mm22 and 240 cells mm22),
(2) exogenous growth factors (EGF and fetal bovine serum
[FBS]), (3) medium change frequency (12 h, 4 h and 1 h), and
(4) culture platforms (microchannels and 96-well plates).
Materials and methods
Factorial experimental design allows for a more efficient
examination of multiple factors, providing information
about both the relative importance of individual factors and
interactions between factors.8 A randomized experimental
design was performed. The high and low levels of the factors
(medium change frequency, culture platforms, cell density and
exogenous growth factors) are defined in Table 1. All medium
contain DMEM and 10 mg ml21 insulin (survival factor for
NMuMG cells). There were three replicates for each condition.
PDMS microchannel plates (channel dimensions, 0.5 6
60 6 0.25 mm, W 6 L 6 H, see ESI Fig. S1{) were fabricated
as described previously.9 Normal mouse mammary gland
This journal is ß The Royal Society of Chemistry 2007
Supplementary Material (ESI) for Lab on a Chip
This journal is © The Royal Society of Chemistry 2007
Supplementary Fig. S1 The 96-well format microchannel arrays used in this study are shown. (A) The
layout shows that each of the 48 channels expand into a 2-well region of a standard 96-well plate, such
that the total surface area of every microchannel is the same as one well (30 mm2) of the 96-well plate.
(B) The fabricated PDMS microchannel arrays are spontaneously attached to an Omnitray and filled
with food color dye to compare with the 96-well plate (shown underneath). Due to the same layout of
the channel arrays and the wells and the plate dimension, both can be scanned with common plate
readers for high through readout (additional details have been previously reported 9).
Supplementary Fig. S2 Typical fluorescence images of NMuMG cells after the experiments. After
experiments, NMuMG cells are stained with Hoechst 33342 and imaged from the (A) 96-well plates
and (B) microchannels.
Supplementary Material (ESI) for Lab on a Chip
This journal is © The Royal Society of Chemistry 2007
Supplementary Fig. S3 The interaction between seeding density, culture medium, culture vessel and
medium change frequency (every 12 h and 1 h). The values and the ranks of the significant interactions
are shown in Table 2 of the manuscript. Three interactions out of the six 2-factor interactions are
significant: the interaction between the medium and culture vessels, the interaction between the
medium and medium change frequency, the interaction between medium and seeding density. The
matrix plot shows: 1) the interaction between the medium and culture vessels is the strongest
interaction, and also ranked as the 2nd most important factor among all the factors. The differences
between the growth of NMuMG in SFM and SM were smaller in microchannels than in the wells, and
SFM did not support the growth of NMuMG cells in the 96-well plates; 2) the interaction between the
medium and medium change frequency is the 2nd strongest interaction and also ranked as the 4th most
important factor in all the factors involved. The differences between SM and SFM became less when
the medium was changed every 1h compared to every 12h medium change; 3) the interaction between
seeding density and medium change frequency was the smallest significant interaction. The difference
between high and low density cultures were reduced after the medium changed frequency was
increased (12 h vs. 1 h).
Supplementary Material (ESI) for Lab on a Chip
This journal is © The Royal Society of Chemistry 2007
Supplementary Fig. S4 The interaction between seeding density, culture medium, culture vessel and
medium change frequency (every 12 h and 4 h). The values and the ranks of the significant interactions
were shown in Table 3 of the manuscript. Two interactions out of the six 2-factor interactions are found
to be significant: the interactions between the seeding density and culture vessel, and the interaction
between the medium and seeding density. The matrix plot shows: 1) the differences between channel
and wells were smaller for the high density cultures than for the low density cultures (120 cells mm-2
vs. 80 cells mm-2) and the growth of the low density cultures was worse in wells than in channels. This
effect was ranked as the 3rd most important effect; 2) high density cultures had increased overall growth
as compared to the low density cultures, but this difference increases in the SM vs. in the SFM. This
interaction is the least important factors of the five significant factors.
Table 1 Experimental design
Factors
Code
High level
Low level
Seeding density
Medium
Vessel
Frequency
D
M
V
F
240 cells mm22
SM (5% FBS)
Well
per 1 h (or per 4 h)
80 cells mm22
SFM (EGF)
Microchannel
per 12 h
epithelial cells (NMuMG, ATCC) were maintained in DMEM
with 10% FBS (Hyclone, SH30070.02) and 10 mg ml21 insulin
(Sigma, I-6634) in T25 flasks in 37 uC incubators with 5% CO2.
During experiments, cells were dissociated with 0.05% trypsin
for 2 min at 37 uC, washed and re-suspended in the culture
medium, diluted according to the experimental design and
seeded into the wells and channels (12 ml each). Medium was
added (36 ml) to each well in the 96-well plates to provide
sufficient coverage. After the cells attached (18–20 h), the cells
in one 96-well plate and one microchannel plate were fixed
with 95% methanol and stored at 4 uC (D1 samples), the rest of
the samples were treated as described in the experimental
design and fixed after an additional 24 h (D2 samples)
Readout methods used were similar to those previously
described9 and are described briefly here. Samples were stained
with nuclei dye (Hoechst 33342) at room temperature for
30 min, and detected at 350/460 nm with a monochromatic
microtiter plate reader (Tecan, XFLUOR4 Safire2 V4.62n)
with an 800 ms integration period. Fluorescence images were
acquired with a CCD camera (Diagnostics, RT3000) on a
scanning epi-fluoresence microscope (4 6 objective, Olympus,
IX70) driven by MetaMorph 6.0 (Universal Imaging Corp.).
Multi-parameter analyses on the ratios of D2 over D1
(24 runs) were performed with R 2.0.1.10 A series of power
transformations were performed to find the proper model.
The main effects and the two-factor interactions were then
determined and the significant factors (P , 0.05) were ranked
by their coefficients and plotted. The magnitudes of the effects
(coefficients) of the significant two-factor interactions were
also plotted.
Results
In the first experiment, NMuMG cells were cultured in the
wells (36 ml medium) and the channels (12 ml medium) with
equal surface area (30 mm2), four factors were evaluated with
medium changes every 12 h and every 1 h. The main factor
effects and interaction effects are plotted in Fig. 1A and B and
summarized in Table 2. The Hoechst 33342 stained images of
representative samples are shown in ESI Fig. S2{ and typical
detailed interaction data is shown in ESI Fig. S3{. First,
frequency was the most important factor. Increasing the
medium change frequency resulted in slower population
expansion within 24 h (every 1 h vs. every 12 h). Second, the
microchannels and 96-well plates had different effects on cell
growth. The growth of cells was higher in the microchannels
than in the wells during the 24–48 h culture period. Third,
the effects of cell density (80 vs. 240 cells mm22) and the
exogenous growth factors (5% FBS vs. EGF) were not
significant. Fourth, there was a strong interaction between
This journal is ß The Royal Society of Chemistry 2007
Fig. 1 The effects of factors and interactions on NMuMG cell
growth. All the plots show the cell growth during a 24 h period (the
ratio of population at 48 h over 24 h). (A) and (B) show the main
factor effects and the factor interactions in the NMuMG cultures for
12 h and 1 h medium change frequency. (A) The effects from the
medium change frequency (F, 1st) and the culture vessel (V, 3rd) were
significant. Cell growth significantly increased with 12 h medium
changes (vs. 1 h) and in microchannels (vs. 96-well plates). (B) Among
all two-factor interactions, the interactions between the medium and
culture vessels (M*V, 2nd), between the medium and medium change
frequency (M*F, 4th), and between the seeding density and medium
change frequency (D*F, 5th) are significant. The medium effects (SFM
vs. SM) were smaller in microchannels than in the wells (M*V) and
smaller with medium change every 1 h vs. every 12 h (M*F). The
density effects were smaller with medium changes every 1 h vs. every
12 h (D*F). The ranks, coefficients and p-values are summarized in
Table 2. (C) and (D) show the profiles of main factor effects and the
factor interactions in the NMuMG cultures for 12 h and 4 h medium
change frequency. (C) The growth of NMuMG cells are significantly
affected by all four factors: medium (1st), seeding density (2nd), culture
vessel (4th) and medium change frequency (5th). Cell growth
significantly increased in SM containing 5% FBS (vs. EGF only), at
240 cells mm22 seeding density (vs. 80 cells mm22), in microchannels
(vs. 96-well plates) and with 12 h medium change (vs. 4 h). (D) Among
all two-factor interactions, the interactions between the seeding density
and culture vessel (D*V, 3rd), between the seeding density and medium
(D*M, 5th) are significant. The density effects (240 vs. 80 cells mm22)
were smaller in microchannels than in the wells (D*V) and smaller in
SFM than in SM (D*M). The results are summarized in Table 3.
the medium and culture platforms, with a larger difference
between 5% FBS and EGF in the 96-well plates than in the
microchannels. Fifth, there was a weak interaction between the
Lab Chip, 2007, 7, 726–730 | 727
Table 2
Summary of the results for the 12 h–1 h medium change
Term
Rank
Mean D2/D1
D
M
F
V
D*M
D*F
D*V
M*F
M*V
F*V
Table 3
1
3
5
4
2
Coef
P-value
1.234
0.124
0.075
20.366
20.176
0.065
20.140
0.085
20.154
0.191
20.068
0.000
0.070
0.265
0.000
0.013
0.335
0.043
0.208
0.027
0.008
0.310
Summary of the results for the 12 h–4 h medium change
Term
Mean D2/D1
D
M
F
V
D*M
D*F
D*V
M*F
M*V
F*V
Rank
2
1
5
4
5
3
Coef
P-value
1.460
0.226
0.264
20.140
20.180
0.144
20.038
0.189
0.035
0.071
20.072
0.000
0.001
0.000
0.028
0.006
0.025
0.533
0.004
0.569
0.248
0.247
medium change frequency and medium. The differences
between serum free medium and serum containing medium
decreased when the medium was changed more frequently (1 h
vs. 12 h). Sixth, there was a weak interaction between seeding
density and medium change frequency. The effect of cell
density decreased as the medium change frequency increased.
In the second experiment, an intermediate medium change
frequency (4 h) was tested while keeping the total medium
volume (48 ml medium) and surface area (30 mm2) equal in the
microchannels and wells. In the microchannel cultures, a drop
of medium (36 ml) was placed on one port, increasing the
microchannel culture volume to 48 ml (including the 12 ml in
the channels). The entire medium in both platforms was
changed every 4 h and 12 h. The main factor effects and
interaction effects are plotted in Fig. 1C and D and summarized in Table 3. The detailed interaction data is shown in
ESI Fig. S4.{ First, the medium and cell seeding density were
the most important factors. Cells cultured in medium containing 5% FBS and at higher seeding density (240 cells mm22)
exhibited the highest proliferation. Second, the culture platform and medium change frequency were less important, but
cells cultured in the microchannels and with less frequent
medium change had better growth than in other conditions.
Third, there were significant interactions between the seeding
density and medium, and between the seeding density and
culture platforms. Decreasing seeding density decreased the
growth of the 96-well plate cultures more than in the
microchannel cultures. Similar to the results of the first
experiment, FBS promoted the growth of dense cultures more
than the sparse cultures (240 vs. 80 cells mm22). Finally, the
overall proliferation of cells with 4 h medium changes was
higher than those with 1 h medium changes.
728 | Lab Chip, 2007, 7, 726–730
Discussion
Many relevant soluble signals can diffuse from the cell
monolayer to the top of channels (250 mm) in less than 1 h
(e.g. insulin, 7.7 6 1025 mm2 s21 in free solution at 37 uC).11
Suppose that for a given factor, there is a threshold
concentration required to initiate a cellular response. The time
required to reach a specified concentration will be different
under different experimental conditions—namely, microchannels vs. open wells. In microchannels under no flow
conditions, the absence of large liquid–air interfaces provides
an environment largely free of convection in which the
transport of particles is governed primarily by diffusion.12,13
In such an environment, secreted factors will follow simple
diffusion until the presence of the channel top begins to cause
accumulation after y1 h. Importantly, larger particles will
move away from the cells more slowly than smaller particles.
In contrast, the forces at liquid–air interfaces create continual
convective flows in open wells.14,15 The convective flows
rapidly move both large and small particles away from the cell
surface and distribute them widely throughout the culture
volume and result in delayed signal thresholds in open wells.
Thus, one would expect the response to endogenous factors to
be delayed in open wells as compared to microchannels. Our
results are consistent with this scenario.
In the first experiment (1 h vs. 12 h media change frequency),
the medium change frequency and culture platforms were the
dominant factors. Furthermore, increasing the medium change
frequency (every 1 h) decreased the growth advantage
provided by FBS (compared with EGF only) and large seeding
density (240 vs. 80 cells mm22), and increased the differences
between 96-well plate and microchannel cultures. Additionally,
cell proliferation was higher in microchannels in medium with
EGF only, while the differences between two culture platforms
were small in FBS.
In the second experiment (4 h vs. 12 h media change
frequency), we wanted to explore the reasons for the smaller
growth rates of cells with 1 h medium change. One possibility
is that the frequent media changes imposed other environmental stresses on those cultures (e.g. temperature swings
during frequent removal from the incubator). However, these
systematic effects were blocked in the multi-parameter analysis
eliminating potential confounding effects. Another possibility
is the importance of the timing of the media change relative to
the production–consumption rates of secreted soluble factors.
That is, either the media change is too frequent (case 1) or not
frequent enough (case 2). In the first case, signal production
rates would be the limiting step in the reconstruction of the
local cellular environment (i.e. the medium change intervals
are too short for the signals to reach critical threshold
concentrations). In the second case, the secreted endogenous
factors would have already accumulated in the microchannels
in 1 h suggesting the signal production rate is large (or
consumption rate is low). The first experiment shows that cells
grew faster with less media change, suggesting that the soluble
growth signals were low in all cultures with 1 h medium change
and supporting first case scenario. The results of the second
experiment further support the first case (i.e. that signal
production is the limiting step). Under conditions where the
This journal is ß The Royal Society of Chemistry 2007
Fig. 2 Schematic representation of the experimental setups in the 12 h
vs. 4 h medium change frequency experiment. The addition of 36 ml
medium drop on one port expands the volume of microchannels (12 ml,
30 mm2) to the same as in the wells (48 ml), such that the cells cultured
in the two platforms are exposed to the same culture volume (48 ml)
and area (30 mm2). The entire 48 ml medium is refreshed every 4 h
or 12 h.
medium volume and surface area (i.e. same number of cells)
were equal, it is found that the growth of cells with 4 h medium
changes increased in all conditions and the effects between
different platforms still existed (Fig. 2C and D, Table 3).
Taken together, the results from both experiments illustrate
the importance in considering the physical differences between
microchannels and open well culture systems when designing
microchannel cell culture experiments. When the differences
between medium change frequency (4 h–12 h vs. 1 h–12 h) and
the total volume were small, the effects from the exogenous
factors and seeding density on cell growth were stronger than
those from the medium change frequency and culture platforms. Cell growth was improved in both platforms when
the medium was changed every 4 h rather than every 1 h,
suggesting that for this particular cell type and channel
dimensions, 1 h was not sufficient for the secreted factors
to reach essential concentrations. Low-density cultures
(80 cells mm22) grew better in the channels than in the wells
(in FBS) and high-density cultures (240 cells mm22) grew
similarly in both platforms in media containing FBS or EGF,
suggesting that the medium composition was similar in the
high-density cultures but different in the low-density cultures
in the two platforms. Thus, with the same exogenous factor
supply (per cell), it is likely endogenous factors are required for
exogenous growth factor function because the endogenous
factors accumulated in the high-density cultures but not in the
low-density cultures. The conclusions from the first experiment
agree with this point that rapid medium change (every 1 h)
prevented the accumulation of endogenous factors and led to
decreased effects of cell density (endogenous factors) and
exogenous growth factors. The different growth of the lowdensity cultures in the two platforms suggest the diffusion
dominant microchannel environment limits the dissipation of
endogenous factors from the cell local microenvironment and
provides growth advantages for low-density cultures. Thus, the
culture microenvironment is an important consideration that
can more strongly influence cell behaviour than other common
parameters (e.g. cell density and exogenous growth factors).
The constrained diffusion in microchannels enabled endogenous factors to accumulate in the local microenvironment,
This journal is ß The Royal Society of Chemistry 2007
such that cells were under the influences of local microenvironment but not the bulk medium and continue grow
even with limited exogenous growth factor supply, low cell
density or frequent medium change. In contrast, the local
microenvironment is disturbed in a 96-well plate/flask, such
that cells respond to what is in the bulk medium and change
the growth according bulk environments (e.g. exogenous
growth factors). It is important to note that the specific
time–geometry interactions are likely cell type specific,
providing an additional characterization tool for studying
cell interactions. Finally, one caveat is the role that the
selective absorption of molecules into PDMS may play in
modifying the culture microenvironment.16 The development
of non-PDMS microchannel devices will allow such comparisons to be made.
Conclusions
The multi-factor experiments provide information on the
process of endogenous soluble factors and their interactions
with other microenvironment factors in microchannel cultures.
The microchannel environments appear to favour the growth
of NMuMG cells by retaining endogenous growth factor
concentration in the local microenvironment. When there
are sufficient endogenous growth factors (e.g. high cell
density, fast production), the microchannels allow these
factors to accumulate within entire channels more rapidly
than in the wells due to the scale. When endogenous factors
are scarce (e.g. low seeding density, slow production), the
microchannels allow the retention of signals locally. The
optimal medium change frequency corresponds to the time
required to re-establish effective endogenous factor gradients
in the local microenvironment. Therefore, the optimal
frequency is an important measurement of the spatial (channel
vs. wells, and channel dimension) and temporal (diffusion
and accumulation) relationship between various microenvironment factors.
Acknowledgements
This work was funded by an Army Breast Cancer Research
Program grant #W81XWH-04-1-0572 and NIH grant #K25CA104162-02. We thank Dr Wei-Yin Loh (Department of
Statistics) for advice on the statistics, Dr Michael Hoffmann
(Department of Oncology), UW–Comprehensive Cancer
Center Small Molecule Screening Facility (SMSF), and Ben
Moga (Department of Biomedical Engineering) for useful
discussions on the experiments.
References
1 K. Saha and D. V. Schaffer, Signal dynamics in Sonic hedgehog
tissue patterning, Development, 2006, 133, 889–900.
2 A. Eldar, R. Dorfman, D. Weiss, H. Ashe, B. Z. Shilo and
N. Barkai, Robustness of the BMP morphogen gradient in
Drosophila embryonic patterning, Nature, 2002, 19(419), 304–8.
3 N. Ferrara and T. Davis-Smyth, The biology of vascular
endothelial growth factor, Endocr. Rev., 1997, 18, 4–25.
4 S. E. Hughes and P. A. Hall, Overview of the fibroblast growth
factor and receptor families: complexity, functional diversity and
implications for future cardiovascular research, Cardiovasc. Res.,
1993, 27, 1199–1203.
Lab Chip, 2007, 7, 726–730 | 729
5 G. M. Walker, H. C. Zeringue and D. J. Beebe, Microenvironment
design considerations for cellular scale studies, Lab Chip, 2004, 4,
91–97.
6 G. M. Walker, M. S. Ozers and D. J. Beebe, Insect cell culture in
microfluidic channels, Biomed. Microdev., 2002, 4, 161–166.
7 H. Yu, I. Meyvantsson, I. A. Shkel and D. J. Beebe, Diffusion
dependent cell behavior in microenvironments, Lab Chip, 2005, 5,
1089–1095.
8 G. P. Box, W. G. Hunter and J. S. Hunter, Statistics for
Experimenters, New York, Wiley, 1978.
9 H. Yu, C. M. Alexander and D. J. Beebe, A plate readercompatible microchannel array for cell biology assays, Lab Chip,
2007, 7, 388–391.
10 Note: R is a language and environment for statistical computing
and graphics, http://www.r-project.org/.
730 | Lab Chip, 2007, 7, 726–730
11 R. A. Freitas, Molecular Transport and Sortation in
Nanomedicine, Landes Bioscience, Austin, TX, USA, 1999,
ch. 3.
12 J. Atencia and D. J. Beebe, Controlled microfluidic interfaces,
Nature, 2005, 437, 648–655.
13 B. Rieger, H. R. Dietrich, L. R. Van Den Doel and L. J. Van Vliet,
Diffusion of microspheres in sealed and open microarrays,
Microsc. Res. Techniq., 2004, 65, 218–25.
14 K. P. Chen, Interfacial energy balance equation for surfacetension-driven Bénard convection, Phys. Rev. Lett., 1997, 78,
4395–4397.
15 E. Berthier, H. Yu and D. J. Beebe, in preparation.
16 M. W. Toepke and D. J. Beebe, PDMS absorption of small
molecules and consequences in microfluidic applications, Lab Chip,
2006, 6, 1484–1486.
This journal is ß The Royal Society of Chemistry 2007