Enhanced Signals and Fast Nucleic Acid Hybridization By Microfluidic Chaotic Mixing.

Communications
DNA Microarrays
DOI: 10.1002/anie.200503830
Enhanced Signals and Fast Nucleic Acid
Hybridization By Microfluidic Chaotic Mixing**
Jian Liu, Brian A. Williams, Richele M. Gwirtz,
Barbara J. Wold, and Stephen Quake*
Nucleic acid hybridization techniques are widely used in both
fundamental and clinical research to identify genes and
mutants, to map their correlations, and to analyze their
expression. DNA microarrays immobilize thousands of oligonucleotides, cDNA (c = complementary) clones, or polymerase-chain-reaction (PCR) products on the solid substrate,
thus providing a powerful tool for the large-scale detection of
target genes.[1, 2] However, hybridization in conventional
microarray experiments is performed in a diffusion-limited
manner, which is quite inefficient. The hybridization process
may take 8–24 h, during which period the characteristic
distance (1–3 mm) that a target DNA molecule can diffuse is
still one order of magnitude less than the typical size of most
microarrays (> 10 mm).[3, 4] Herein we describe an effective
answer to that problem using microfluidic chaotic mixing. Our
polydimethylsilicane (PDMS) devices use integrated peristaltic pumps to circulate the solution between two large
chambers, while chaotically mixing the components of the
solution in bridge channels at the same time. We demonstrate
that this approach dramatically enhances hybridization signals and improves sensitivity by nearly one order of magnitude relative to the conventional static-hybridization method
over the same length of time. Alternatively, for a desired
sensitivity, dynamic mixing can be used to accelerate the
hybridization process by a factor of three or more. This
approach offers many benefits, including high sensitivity,
rapid results, better reproducibility, low cost, compatibility
with commercial microarray slides, and ease of large-scale
integration.
[*] Prof. S. Quake
Department of Bioengineering and Howard Hughes Medical
Institute
James Clark Center E300, Stanford University
Stanford, CA 94305 (USA)
Fax: (+ 1) 650-736-1961
E-mail: [email protected]
J. Liu, Dr. B. A. Williams, R. M. Gwirtz, Prof. B. J. Wold
California Institute of Technology
Pasadena, CA 94305 (USA)
[**] This research was supported by the NIH Director’s Pioneer Award
(7DP1 0D000251-02) and NIH 1R01 HG002644-01A1. We thank
Yanxia Hao and Dr. Christopher S. Baker of the Genomics Core
Laboratory, Gladstone Institutes for providing mouse-printed
oligoarrays (17 K) and Frederick Balagadde of Caltech for technical
assistance.
Supporting information for this article is available on the WWW
under http://www.angewandte.org or from the author.
3618
2006 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Angew. Chem. Int. Ed. 2006, 45, 3618 –3623
Angewandte
Chemie
The fundamental problem faced by DNA-microarray
practitioners is how to continuously mix a sample solution at
low Reynolds number and transport the DNA molecules to
the proximity of the probes more rapidly than diffusion alone,
thereby increasing valid molar-hybridization events. Many
methods of active mixing have been developed, including
surface acoustic-wave microagitation,[5] bubble-induced
acoustic microstreaming,[3] alternative convection induced
through several ports,[4] “drain-and-fill” or air-driven bladders,[6, 7] magnetic stirring bars,[8] and shuttling sample plugs in
a serpentine microtrench.[9] Some researchers developed
electrokinetic methods to accelerate the transportation of
DNA molecules.[10–12] Although these methods all allow
shorter hybridization times and improved signals, they
appear to suffer from one or several shortcomings: inhomogeneous mixing,[3–5] small arrays,[10–12] relatively low sensitivity,[3, 4, 6, 8–12] or incompatibility with the widely used cyanine
(Cy)-dye microarray format.[10–12] On the other hand, chaotic
mixing has gained the interest of many researchers as a
promising method of mixing solutions in microchannels.[13–15]
We constructed a two-layer PDMS microfluidic device[16]
and sealed it to a spotted microarray slide to perform dynamic
hybridization. As shown in Figure 1 a, the fluidic layer of the
device contains two symmetric hybridization chambers
(6 mm < 6.5 mm < 65 mm, 5 mL). They are connected to each
other by bridge channels, the ceilings of which incorporate
protrusions with a herring-bone pattern to produce chaotic
mixing.[14] The bridge channels (400 mm wide, 40 mm high) are
connected to the chambers by bifurcating channels that
equalize the solution distributed into the chambers after
mixing. Four input/output through-holes with corresponding
micromechanical valves are used for loading samples or
disposing waste solutions. These valves are actuated to form
closed chambers during hybridization. Two sets of peristaltic
pumps[17] are integrated to circulate the fluid between the
hybridization chambers. The design allows different components in the solution to mix in a chaotic manner when they
pass through the bridge channels, and then to be delivered
through the hybridization chambers (Figure 1 b, c and Supporting Information). The peristaltic pumps create a fluid flux
of 5.2 nL s 1. It takes about 16 min to complete one
circulation of the two chambers. The flux can be further
raised by increasing the cross-sectional area of the individual
pumps.
We evaluated the efficiency of mixing by loading each
chamber half with the blank solution and half with the
solution containing fluorescent beads, then actuating the
pumps, and performing fluorescence measurements. A fluorescence inverted microscope with a charge-coupled device
(CCD) camera was set up to take images of the device.
Chaotic mixing was confirmed by independent observations
of the zigzag motion of the beads and their crossing each
other through the bridge channels (images not shown). The
fluorescence intensity was monitored in real time through 10
windows along the equator of one chamber. As shown in the
Supporting Information, within minutes chaotic mixing
dramatically decreased the fluorescence gradient along the
equator. The control experiment followed all the above
conditions except that a device without the herring-bone
Angew. Chem. Int. Ed. 2006, 45, 3618 –3623
Figure 1. Optical micrographs of the PDMS device. a) The inset shows
the 3D structure of the bridge channels with herringbone protrusions.
b) The chambers were loaded half with the red and half with the blue
solution. c) The peristaltic pumps (1–3 and 4–6) circulated the
solutions clockwise in the device. The herringbone protrusions in the
bridge channel produced chaotic mixing of the colors.
protrusions on the bridge channels was used. In this case, the
fluorescence difference along the equator was still substantial
even after the solutions have been circulated for 2 h, owing to
the absence of effective lateral mixing (see Supporting
Information). Therefore, the chaotic mixing introduced by
2006 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.angewandte.org
3619
Communications
the herring-bone protrusions is critical in homogenizing the
solutions in the chambers. Ripples in fluorescence intensity
were observed because the extra beadless solution initially
loaded in the bridge channels also took part in the fluid
circulation. Interestingly, they became a good indicator of the
periodicity of the circulation ( 8 min/ripple).
We performed a series of dynamic and static (control)
hybridizations for comparison. Two PDMS devices were
sealed onto a single home-spotted microarray slide (VEPO25C, CEL Associates, Inc.), thus covering areas of identical
probe patterns (70-mer DNA oligonucleotides). Each area
consisted of four identical blocks. As shown in Figures 2 a
and b, each block included 18 features (six spotting solutions
repeated three times). Six of them were negative-control
features. We prepared Cy3-labeled cDNA from the C2C12
mouse skeletal-muscle cell line, adding Arabidopsis thaliana
CAB (chlorophyll a/b-binding protein) spikes (cat. #2552201,
Stratagene) as a positive control. Details of total RNA
isolation, mRNA (m = messenger) extraction, and the
reverse-transcription protocols are described elsewhere.[18]
The cDNA sample was diluted into a series of solutions and
then separated into aliquots. They were spin-dried under
vacuum and kept at 4 8C before use. Two identical aliquots
were used to prepare the hybridization solutions with
ArrayHyb buffer (A-7718, Sigma-Aldrich Co.). They were
loaded into the PDMS devices sealed on the shared slide,
which was mounted on the flat bed of a thermocycler (PTC200, MJ Research) and prehybridized. We performed
dynamic hybridization by actuating the peristaltic pumps in
one of the devices, while static hybridization was performed in
the other device as a control. After hybridization at 52 8C for
2 h (or other specified time), the PDMS devices were peeled
away from the slide. The slide was immediately placed in a
plastic tube for programmed posthybridization washing
(AdvaWash 400, Advalytix). The slide was then spin-dried
in a centrifuge (5804R, Eppendorf) and scanned (ArrayWorx,
Applied Precision LLC) to obtain fluorescence images.
Hybridization with microfluidic chaotic mixing produced
stronger signals and better sensitivity than the static control
(Figure 2). When the amount of input molecules of CAB
spikes was decreased to 5.5 amol, the signal-to-noise (S/N)
ratio of the static method was slightly larger than 1, which
indicates that the signals were nearly indistinguishable from
the background at that point. However, the S/N ratio of
dynamic hybridization did not collapse to 1 until the amount
of input CAB molecules further decreased to 0.55 amol,
which is an enhancement in sensitivity of one order of
magnitude. To our knowledge, this sensitivity level is better
than any other reported method designed for active mixing in
hybridization. The above enhancement of signals was reproducible, as confirmed by independent hybridization experiments. Active mixing also reduced the spot-to-spot fluctua-
Figure 2. Signal improvement by microfluidic chaotic mixing. a) Dynamic- and b) static-hybridization blocks (each one of four identical) of the
home-spotted microarray. The diameter of the features is about 200 mm. c) Titration curves and d) hybridization kinetics, with data analyzed from
the features of myosin light-chain sense. Dynamic data are in red, static data are in black. The S/N ratio is calculated as the mean of the
fluorescence signals subtracted by the mean of the background signal, then divided by the standard deviation of the background signal.
3620
www.angewandte.org
2006 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Angew. Chem. Int. Ed. 2006, 45, 3618 –3623
Angewandte
Chemie
tion of the signals. The coefficients of variation (CV) of the
dynamic hybridization decreased to nearly half the values of
the conventional static method (Table 1 and Figure 2 c). The
results of hybridization kinetics (Figure 2 d) showed that
Table 1: The spot-to-spot coefficients of variation (n = 12) of dynamic
versus static hybridization.[a]
myogenin sense
MCK sense
myosin light-chain sense
CAB (positive control)
Dynamic
Static
0.11
0.08
0.12
0.10
0.18
0.27
0.24
0.23
[a] The negative-control solutions included a blank spotting buffer and a
solution containing oligonucleotides for the A. thaliana RCA gene (not
spiked in the sample). CAB = chlorophyll a/b-binding protein, MCK =
muscle creatine kinase.
dynamic mixing consistently produced signals with higher S/N
ratios than the static method. The signal from dynamic
hybridization for 2 h was nearly twice that from the static
control hybridized for 6 h. We noticed that the signals from
both methods decreased after > 6 h hybridization, which
might be attributed to partial dehydration of the arrays, as
PDMS is permeable to water vapor.
To evaluate the separate contributions of the circulatory
motion of the fluid and chaotic mixing to the signal enhancement, we performed side-by-side comparison experiments
that included an additional hybridization control with fluid
circulation but without chaotic mixing. Identical aliquots of
DNA target solutions were hybridized at 52 8C for 2 h under
three distinct conditions: static (control 1), simple fluid
circulation using the devices without the herring-bone
protrusions (control 2), and fluid circulation with chaotic
mixing. The experimental results (Table 2) clearly show that
chaotic mixing played a significant part in the overall signal
enhancement. Simple circulation of the fluid increased the
signal intensity to 1.6–2.3-fold those obtained from the static
control, whereas circulation with chaotic mixing improved the
signals 3.4–6.9-fold. Therefore, microfluidic chaotic mixing
has a major effect on the mass transfer of DNA targets to a
solid reactive boundary by effectively homogenizing the
solution.
We tested the compatibility of this approach with larger
arrays, and found that microfluidic chaotic mixing can
improve hybridization in high-density microarrays. We constructed a PDMS device with similar geometry but larger
hybridization chambers (7.5 mm < 36 mm < 65 mm, 35 mL). It
was sealed onto a commercial microarray (mouse-printed
Table 2: Increase (x-times) in the background-subtracted fluorescence of
dynamic compared with static hybridization.
myogenin sense
MCK sense
myosin light-chain sense
CAB (positive control)
Without
chaotic mixing
With
chaotic mixing
2.1
1.6
2.2
2.3
4.2
6.9
3.4
4.5
Angew. Chem. Int. Ed. 2006, 45, 3618 –3623
oligonucleotides array, 17 K, 70 mer, J. David Gladstone
Institutes). Approximately 9500 spots on the slide were
accessible within the hybridization chambers. All the experimental procedures followed that previously described,
except that two identical microarray slides were used for
comparison. The signals from hybridization with microfluidic
chaotic mixing were dramatically enhanced (Figure 3). For
individual features, we calculated the increase in the background-subtracted fluorescence by dynamic mixing over that
of the static method and summarized the data in a histogram
(Figure 4). The peak of the red columns (input-sample
concentration 1.6 ng mL 1) is located at a 3–4-fold increase.
When the input Cy3-labeled cDNA was diluted to 0.8 ng mL 1
(blue columns), the peak shifted to a 7–8-fold increase. This
result is consistent with our previous observation (Figure 2 c)
that the increase in the signals from dynamic over those from
static hybridization became larger as the concentration of the
input sample decreased. The signal enhancement of our
approach was comparable to or slightly better than other
reported values (mostly 2–5-fold) in the literature.[3, 4, 6–8]
The specificity of the signals of the dynamic versus those
of the static method was further examined (Figure 5) using
receiver-operating-characteristics
(ROC)
analysis.[19]
Dynamic hybridization reduced the overlap between the
intensity distributions of the negative-control and the positive-probe features. The increase in the area under the ROC
curves (Figure 5 b and d)[19] also suggests that the dynamic
approach allowed better signal specificity than the static
method.
We attempted to determine how many cDNA molecules
were actually hybridized onto the slide from the solution. The
nucleotide concentration of the CAB-spiked cDNA solution
was determined by UV/Vis spectroscopy. We manually
deposited volume-defined droplets of Cy3-labeled cDNA
molecules onto the slide by using microcapillary tubes
(0.2 mL, Drummond) and obtained a standard curve of the
fluorescence intensity versus the amount of deposited nucleotides. By estimating the CAB cDNA length, we calculated the
number of hybridized CAB molecules on a given spot. The
analysis was based on 2-h hybridization experiments. The data
(Table 3) reveal that only a small percentage of target DNA
molecules were actually hybridized with the static method,
with a large portion of them remaining in the solution.
Dynamic mixing can increase the percentage of molarTable 3: Percentages of hybridized CAB molecules out of the total
number of initial spikes.[a],[b]
Input CAB
[amol]
Dynamic
hybridization [%]
Static
hybridization [%]
55
28
5.5
2.8
1.1
9.4
4.1
1.2
1.6
1.9
1.3
0.66
0.20
0.20
0.54
[a] The yield of the labeling reaction was about 54 % according to
absorption measurements. [b] The full length of the CAB spikes
(500 base pairs) was used in the above estimation. This represents a
conservative lower limit as the hexamer-priming reaction may yield a
distribution of cDNA lengths.
2006 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.angewandte.org
3621
Communications
Figure 3. Fluorescence images of two side-by-side hybridization experiments (17-K microarray). a) Dynamic; b) static. c) and d) show magnified
views of block #33 from a) and b), respectively. A PDMS device with larger chambers was used in each experiment. Notice the gap between the
two chambers.
Figure 4. Histogram showing the increase in the background-subtracted fluorescence (F550 Mean B550 Mean) of dynamic over static
hybridization. The red and blue columns represent two sets of
independent hybridization experiments that used different concentrations of the Cy3-labeled cDNA (red 1.6 ng mL 1, blue 0.8 ng mL 1). The
wider spread in the distribution of the blue columns may result from a
greater fluctuation of signal intensities when the static method was
used with the more-dilute labeled target.
3622
www.angewandte.org
hybridization events several-fold. A new challenge may be to
make better use of the unhybridized target DNA molecules in
the solution.
We have demonstrated that microfluidic chaotic mixing
enhances hybridization signals 3–8-fold through the introduction of lateral mixing, the facilitation of the delivery of target
DNA molecules, and the increase in the number of molarhybridization events. Our approach has improved the sensitivity of DNA-microarray experiments by nearly one order of
magnitude and allowed better signal specificity than the static
method. The time taken for the hybridization step of the
conventional method has been shortened to 2 h with the new
approach. Our PDMS device is disposable and compatible
with high-density microarray slides. The device with larger
chambers has the potential to hybridize about 135 000 array
features in a single experiment, if new-generation arrayprinting tips[20] are used to spot ultrahigh-density microarrays
(25 000 spots cm 2).
Received: October 29, 2005
Revised: February 16, 2006
Published online: April 26, 2006
2006 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Angew. Chem. Int. Ed. 2006, 45, 3618 –3623
Angewandte
Chemie
[11] D. Erickson, X. Liu, U. Krull, D. Li, Anal. Chem. 2004, 76, 7269.
[12] C. Gurtner, E. Tu, N. Jamshidi, R. W. Haigis, T. J. Onofrey, C. F.
Edman, R. Sosnowski, B. Wallace, M. J. Heller, Electrophoresis
2002, 23, 1543.
[13] R. H. Liu, M. A. Stremler, K. V. Sharp, M. G. Olsen, J. G.
Santiago, R. J. Adrian, H. Aref, D. J. Beebe, J. Microelectromech.
Syst. 2000, 9, 190.
[14] A. D. Stroock, S. K. W. Dertinger, A. Ajdari, I. Mezic, H. A.
Stone, G. M. Whitesides, Science 2002, 295, 647.
[15] C. Simonnet, A. Groisman, Phys. Rev. Lett. 2005, 94.
[16] M. A. Unger, H. P. Chou, T. Thorsen, A. Scherer, S. R. Quake,
Science 2000, 288, 113.
[17] H. P. Chou, M. A. Unger, A. Scherer, S. R. Quake, Biomed.
Microdevices 2001, 3, 323.
[18] B. A. Williams, R. M. Gwirtz, B. J. Wold, Nucleic Acids Res.
2004, 32, e81.
[19] N. N. Khodarev, J. Park, Y. Kataoka, E. Nodzenski, S. Hellman,
B. Roizman, R. R. Weichselbaum, C. A. Pelizzari, Genomics
2003, 81, 202.
[20] M. O. Reese, R. M. van Dam, A. Scherer, S. R. Quake, Genome
Res. 2003, 13, 2348.
Figure 5. Distribution of positive/negative signals and ROC curves.
a) Distribution of the positive-probe signals * and the negative-control
signals ~ for the dynamic experiment. b) ROC curve for the dynamic
experiment, with an area under the curve of 0.92. c) Distribution of
positive/negative signals for the static experiment. d) ROC curve for
the static experiment, with an area under the curve of 0.73. Negativecontrol features (Operon) are sequences that were determined to be
absent from the mouse genome. Positive features were identified from
previous independent experiments (data not shown) as having signal
intensities greater than three standard deviations above the median
negative-control intensity in 72-h diffusion–hybridization experiments
with labeled 36-h C2C12 cDNA. The true-positive and true-negative
fractions are defined as the fraction of all positive-probe and negativecontrol features encountered at any given rank on the intensity scale,
respectively.
.
Keywords: analytical methods · DNA microarrays ·
gene expression · hybridization · mixing devices
[1] M. Schena, D. Shalon, R. W. Davis, P. O. Brown, Science 1995,
270, 467.
[2] D. J. Duggan, M. Bittner, Y. D. Chen, P. Meltzer, J. M. Trent, Nat.
Genet. 1999, 21, 10.
[3] R. H. Liu, R. Lenigk, R. L. Druyor-Sanchez, J. Yang, P.
Grodzinski, Anal. Chem. 2003, 75, 1911.
[4] M. K. McQuain, K. Seale, J. Peek, T. S. Fisher, S. Levy, M. A.
Stremler, F. R. Haselton, Anal. Biochem. 2004, 325, 215.
[5] Z. Guttenberg, A. Rathgeber, S. Keller, J. O. Radler, A.
Wixforth, M. Kostur, M. Schindler, P. Talkner, Phys. Rev. E
2004, 70.
[6] N. B. Adey, M. Lei, M. T. Howard, J. D. Jensen, D. A. Mayo,
D. L. Butel, S. C. Coffin, T. C. Moyer, D. E. Slade, M. K. Spute,
A. M. Hancock, G. T. Eisenhoffer, B. K. Dalley, M. R. McNeely,
Anal. Chem. 2002, 74, 6413.
[7] C. J. Schaupp, G. J. Jiang, T. G. Myers, M. A. Wilson, Biotechniques 2005, 38, 117.
[8] P. K. Yuen, G. S. Li, Y. J. Bao, U. R. Muller, Lab-on-a-Chip 2003,
3, 46.
[9] C. W. Wei, J. Y. Cheng, C. T. Huang, M. H. Yen, T. H. Young,
Nucleic Acids Res. 2005, 33, e78.
[10] C. F. Edman, D. E. Raymond, D. J. Wu, E. Tu, R. G. Sosnowski,
W. F. Butler, M. Nerenberg, M. J. Heller, Nucleic Acids Res.
1997, 25, 4907.
Angew. Chem. Int. Ed. 2006, 45, 3618 –3623
2006 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.angewandte.org
3623