Ultrahigh-Throughput Mammalian Single-Cell Reverse-Transcriptase

Jul 26, 2013 - To image the thermocycled droplets, 10 μL of emulsion was pipetted into Countess chambered coverglass slides (Invitrogen). The slides ...
2 downloads 0 Views 3MB Size
Download PDF
Article pubs.acs.org/ac

Ultrahigh-Throughput Mammalian Single-Cell Reverse-Transcriptase Polymerase Chain Reaction in Microfluidic Drops Dennis J. Eastburn,† Adam Sciambi,† and Adam R. Abate* Department of Bioengineering and Therapeutic Sciences, California Institute for Quantitative Biosciences, University of California, San Francisco, California 94158, United States S Supporting Information *

ABSTRACT: The behaviors of complex biological systems are often dictated by the properties of their heterogeneous and sometimes rare cellular constituents. Correspondingly, the analysis of individual cells from a heterogeneous population can reveal information not obtainable by ensemble measurements. Reverse-transcriptase polymerase chain reaction (RT-PCR) is a widely used method that enables transcriptional profiling and sequencing analysis on bulk populations of cells. Major barriers to successfully implementing this technique for mammalian single-cell studies are the labor, cost, and low-throughput associated with current approaches. In this report, we describe a novel droplet-based microfluidic system for performing ∼50000 single-cell RT-PCR reactions in a single experiment while consuming a minimal amount of reagent. Using cell type-specific staining and TaqMan RT-PCR probes, we demonstrate the identification of specific cells from a mixed human cell population. The throughput, robust detection rate and specificity of this method makes it well-suited for characterizing large, heterogeneous populations of cells at the transcriptional level.

C

injected with reagent, and sorted at kilohertz rates, holding potential for performing millions of single-cell reactions at unprecedented throughput. However, an obstacle to realizing the potential of this approach is that, at concentrations of a single mammalian cell in a microdroplet, cell lysate is a potent inhibitor of RT-PCR.9−11 To avoid mammalian cell lysate inhibition of RT-PCR, previous drop-based methods have utilized large droplets (2 nL), in which the lysate concentration is no longer inhibitory, or agarose droplets that can be solidified, rinsed, and stained with DNA dyes.12,13 Methods using large droplets have only been able to analyze ∼100 cells in total, while the agarose method is unable to use TaqMan probes or cell staining, precluding correlation of specific cell types with associated transcriptional targets. An alternative strategy for performing single-cell RT-PCR on mammalian cells is to isolate the cells in microwells fabricated into an elastomeric device. This approach allows robust and specific single-cell transcriptional profiling; however, because each microwell and its control valves must be

ellular heterogeneity and its impact on biological function and disease is becoming increasingly important for questions in human immunology, stem cell biology, and cancer research. By transcriptionally analyzing individual cells with reverse-transcriptase polymerase chain reaction (RT-PCR), it is possible to identify rare cells or transient cell states that are unobservable when studying the entire population in bulk.1−4 However, obtaining meaningful information on these cells necessitates tools capable of high-throughput analysis. Current methods for manipulating, isolating, and transcriptionally profiling single mammalian cells with RT-PCR are cumbersome and limited in throughput, enabling the examination of just hundreds of individual cells. The ultrahigh-throughput capability of droplet-based microfluidics is ideal for single-cell analysis applications.5−7 These microfluidic techniques rely on microdroplets, tiny spheres of aqueous liquid ranging from 1 to 100 μm in diameter, to encapsulate biological components in an oil-based emulsion.8 The drops serve, essentially, as very tiny “test tubes,” compartmentalizing millions of reactions. A major advantage of this approach is that a minimal amount of reagent is used, greatly reducing the cost for a given experiment. In addition, with microfluidic techniques, the drops can be formed, split, © 2013 American Chemical Society

Received: July 5, 2013 Accepted: July 26, 2013 Published: July 26, 2013 8016

dx.doi.org/10.1021/ac402057q | Anal. Chem. 2013, 85, 8016−8021

Analytical Chemistry

Article

PCR mix flow rate was selected such that the final drops contained a 1× RT-PCR reaction mix. TaqMan RT-PCR Reactions. The sequence of amplification primers used for the RT-PCR reactions were as follows: EpCAM Forward 5′-CCTATGCATCTCACCCATCTC-3′, EpCAM Reverse 5′-AGTTGTTGCTGGAATTGTTGTG-3′; PTPRC/CD45 Forward 5′-CCATATGTTTGCTTTCCTTCTCC-3′, PTPRC/CD45 Reverse 5′-TGGTGACTTTTGGCAGATGA-3′; GAPDH Forward 5′-TGAGTCCTTCCACGATACCA-3′, GAPDH Reverse 5′-ACCATGAGAAGTATGACAACAGC-3′. The sequences of TaqMan probes are as follows: EpCAM 5′-/HEX/ATCTCAGCC/ZEN/TTCTCATACTTTGCCATTCTC/IABkFQ/-3′; PTPRC/CD45 5′-/HEX/CCTGGTCTC/ZEN/CATGTTTCAGTTCTGTCA/IABkFQ/-3′; GAPDH 5′-/HEX/AGATCATCA/ZEN/ GCAATGCCTCCTGCA/IABkFQ/-3′. Premixed amplification primers and TaqMan probes were ordered as a PrimeTime Standard qPCR assay from Integrated DNA Technologies (IDT) and were used at the suggested 1× working concentration. Superscript III reverse transcriptase (Invitrogen) was used to enable first stand cDNA synthesis. Following picoinjection of RT-PCR reagents, drops were collected in PCR tubes and transferred to a T100 Thermal Cycler (BioRad). Reactions were incubated at 50 °C for 15 min followed by 93 °C for 2 min and 45 cycles at 92 °C for 15 s and 60 °C for 1 min. To prevent evaporation of PCR reagents from the microfluidic drops, we placed the heated lid on the thermocycler set to 105 °C. Fluorescence Imaging. To image the thermocycled droplets, 10 μL of emulsion was pipetted into Countess chambered coverglass slides (Invitrogen). The slides were imaged on a Nikon Eclipse Ti inverted microscope using conventional widefield epifluorescence with a 4× objective. Fluorescence filters were chosen to optimize the signal intensity and minimize background fluorescence due to spectral overlapping of the dyes used in the multiplexed reactions. The images were captured using NIS Elements imaging software from Nikon. Analysis of Fluorescence Images. Drops imaged by the fluorescence microscope were identified using MATLAB. For each field of view (Figure 2a containing ∼500 drops), we captured a brightfield image and fluorescence images at three wavelengths (HEX, Calcein green, and Calcein violet). When processing a field, we first subtracted a smooth background from each fluorescence channel to remove largescale nonuniformities in the excitation light source. The droplet boundaries were then located in the bright field image (Figure S2a of the Supporting Information) and, for each fluorescence channel, drop intensities were measured by fitting a paraboloid to the drop intensity profiles. We size-gated (Figure S2b of the Supporting Information) to exclude drops too small or large. Each replicate for a particular probe incorporated between 30 and 50 fields of view, for a total of 15000−25000 drops, and we performed cluster analysis on each of the three pairs of channels to determine probe efficiency for both cell types (Figure 2, panels b−d). This analysis used the empty drops as standards and, accordingly, we shifted the center of the low fluorescence, heavily populated cluster of each plot (lower left of Figure 2, panels b−d) to zero. We automatically set the fluorescence thresholds, above which a drop is counted as fluorescing, using the spread of the empty-drop clusters along each axis.

fabricated and individually controlled, throughput is also limited to just a few hundred cells in total.2,3,11 To enable expression analysis of large numbers of mammalian cells in a heterogeneous population, new methods are needed that combine the throughput of droplet-based microfluidic techniques with the specificity of microwell reactions. Here, we describe a novel platform that combines ultrahighthroughput droplet-based microfluidic techniques with highfidelity TaqMan RT-PCR. Our platform enables specific detection of different human cell types in a heterogeneous sample using single cell RT-PCR at a throughput of more than 2 orders of magnitude beyond existing methods.

■

MATERIALS AND METHODS Microfluidic Device Fabrication. The PDMS devices were fabricated by pouring uncured PDMS (10:1 polymer to cross-linker ratio) over photolithographically patterned layers of photoresist (SU-8 3025) on a silicon wafer. After curing in an oven at 80 °C for 1 h, we extracted the molds with a scalpel and punched access holes using a 0.75 mm biopsy core. The devices were then bonded to glass slides after a surface treatment of 1 mbar O2 plasma for 20 s in a 300 W plasma cleaner. To make the channels hydrophobic, as needed for droplet-based experiments, we treated them with Aquapel by flushing Aquapel through the channel, blowing the channels clear with air, and baking the device at 80 °C for 10 min. Cell Culture and Cell Staining. Human PC3 prostate cancer and Raji B-lymphocyte cell lines were cultured in appropriate growth medium supplemented with 10% FBS, penicillin, and streptomycin at 37 °C with 5% CO2. Prior to cell staining, Raji cells were pelleted and washed once in phosphate buffered saline (PBS). Confluent and adhered PC3 cells were first trypsinized prior to pelleting and washing. Cells were stained in 1 mL Hank’s balanced salt solution with 2 μM Calcein Violet AM or Calcein Green AM for 30 min at room temperature. Following staining, cells were washed with PBS and then resuspended in PBS that was density matched with OptiPrep solution prior to encapsulation in microfluidic drops. Operation of Microfluidic Devices. Cell suspension and lysis buffer (100 mM Tris pH 8.0, 2% Tween-20, proteinase K 1.5 μg/μL) were loaded into 1 mL syringes and injected at equal flow rates into a coflow dropmaker using syringe pumps (New Era) controlled with custom LabVIEW software. Flow focusing using fluorinated oil (FC40) with 5% PEG-PFPE amphiphilic block copolymer was used to generate the microdroplet emulsion. The dimensions of the device and flow rates of the reagents were adjusted to obtain the desired 40 μm drop size. Emulsions containing encapsulated cells were collected into a 1 mL syringe and heated in a 50−55 °C water bath for 15 min to enable proteinase K digestion of cell lysate. The temperature was then raised to 90 °C, and the syringe was incubated for an additional 10 min to heat-inactivate the proteinase K. Following the heat bath, drops were reinjected into the dilution and picoinjection device. Reinjected drops were paired with dilution drops containing water. To apply the electric field for drop merger and subsequent picoinjection, we filled the electrode and surrounding moat channels with a 3 M NaCl solution, having a conductivity of ∼0.1 S/cm. We energized the electrode with 20 kHz, 300 VAC signals generated by a fluorescent light inverter (JKL Components Corp) attached via an alligator clip to the syringe needle. The picoinjection 2× RT8017

dx.doi.org/10.1021/ac402057q | Anal. Chem. 2013, 85, 8016−8021

Analytical Chemistry

Article

Figure 1. Microfluidic workflow and devices. Arrows indicate direction of flow. (a) Workflow of our microfluidic platform. (b) Picture of the dilution and injection device connected to syringe pumps. Colored liquid was used to aid visualization of the moat (orange) and droplet channels (blue). (c) Scale representation of the two microfluidic devices used in this report. (d−i) Microscope images of sections of the fully functioning microfluidic device are shown (scale bars are 60 m). Drops are pseudocolored for clarity. (d) Dropmaker used to encapsulate cells, visible in image, with lysis buffer prior to offline thermal incubation. (e) Following proteinase K inactivation, drops are reinjected and spaced with oil before (f) pairing with dilution drops. (g) Paired drops are merged with an electric field. After dilution, a portion of the lysate is (h) split off and subsequently (i) picoinjected with 2× RT-PCR reagents. The drop splitting step prior to picoinjection reduces the total RT-PCR reagent needed and generates a smaller more stable drop size for thermocycling. Picoinjected drops are collected into a tube, thermocycled, and analyzed for fluorescence.

Ultrahigh-Throughput PMT Detection of Drops. A schematic of the ultrahigh-throughput detection setup is depicted in Figure S3a of the Supporting Information. We positioned the device using lamp illumination filtered through a 785 nm band-pass filter, a 40× objective in an inverted microscope (Motic), and a camera behind a 740 nm dichroic beamsplitter. Imaged drops appeared as seen in Figure S3b of the Supporting Information. Drop fluorescence was excited by two lasers (405 nm for Calcein Violet excitation and 532 nm for HEX excitation), whose beams were first positioned by two adjustable mirrors before being combined by a 427 nm dichroic. The beam pair was reflected off both a 405/532 nm dual notch filter and the 740 nm illumination dichroic and focused through the objective. The laser spots, visible with the camera, were aligned with the channel by adjusting the channel position and focus (Figure S3b of the Supporting Information). The longer-wavelength emission was returned through the objective, reflected by the 740 nm dichroic, and transmitted through the 405/532 nm notch filter. From there, the Calcein Violet and HEX signals were separated with a 552 nm dichroic, with the Violet passing through a 448 nm bandpass filter to one photomultiplier tube (PMT, Thorlabs) and the HEX through a 572 nm bandpass to a second PMT. Inline Detector Data Analysis. The PMT voltages were processed by a digital-to-analog converter card (National Instruments) and field-programmable gate array (National Instruments) and recorded at a 20 kHz acquisition rate as a time series using LabVIEW. After recording, MATLAB was used to extract the drop fluorescences from the time series data. All drops were detectable due to the background HEX signal of the uncleaved probe (Figure 3b), and we thresholded this minimum signal to locate drop boundaries within the series. We gated drops according to their duration (Figure S3c of the Supporting Information) and estimated their fluorescence by averaging the middle third of its sampled points. These points

were plotted on heat maps and gated according to the minima between the four quadrant clusters.

■

RESULTS AND DISCUSSION

To overcome cell lysate-mediated inhibition of RT-PCR in microdroplets, we have developed a microfluidic workflow, outlined in Figure 1a, which encapsulates and proteinase K digests individual cells in drops (left) before diluting the resultant lysate and injecting PCR reagents (right). These successive steps are performed in devices made from polydimethylsiloxane (PDMS) molds bonded to glass slides. A photograph of the dilution and injection device with inlet and outlet tubing attached is shown in Figure 1b, and a scaled schematic of the channel layout is given in Figure 1c. Accompanying the workflow in Figure 1a are microscope images of each step in progress (Figure 1, panels d−i). To begin, a cell suspension is coflowed into a microfluidic drop maker with lysis buffer containing proteinase K (Figure 1d). Due to the laminar flow in the channels, mixing of the cells and lysis buffer is prevented until the cells are encapsulated. The cells are encapsulated at limiting dilution such that most drops are empty but ∼1−5% contain single cells, in a process governed by Poisson statistics. Following lysis and proteinase K digestion off-chip, the lysate-containing drops are introduced into a second device (Figure 1e), paired with a large water drop (Figure 1f), and then merged with an electric field for an ∼20fold dilution (Figure 1g). After in-droplet mixing, a portion of the diluted lysate is split off (Figure 1h), whereupon RT-PCR reagents are added with picoinjection via a second electric field (Figure 1i). The processed drops are collected in PCR tubes, thermocycled, and analyzed for transcript amplification. An important advantage of this workflow is that the drops retain their compartmentalization at all times; this enables the use of cell-viability dyes and fluorescent antibodies that are incompat8018

dx.doi.org/10.1021/ac402057q | Anal. Chem. 2013, 85, 8016−8021

Analytical Chemistry

Article

ible with methods that break the drops to remove cell lysate or introduce DNA dyes.13 As a demonstration of the specificity of our approach, we used it to screen a mixed population of human cells. Due to their reliance on both amplification primers and gene-specific fluorescence probes, TaqMan assays are ideal for highspecificity transcript detection. Previously, we demonstrated the ability to characterize specific expression of PTPRC transcripts from Raji B-lymphocyte total RNA using digital droplet TaqMan assays.14 This study was enabled by the sensitivity of TaqMan assays in drops, which allowed detection of individual transcripts. In the present study, however, this extreme sensitivity can lead to the detection of free transcripts or dead cells encapsulated at the initial step of our workflow. To identify and exclude drops encapsulating only free transcripts or dead cell debris, we labeled Raji cells with calcein green viable cell dye prior to encapsulation in drops.15,16 We also included calcein violet dye-labeled PC3 prostate cancer cells that do not express PTPRC.17 With this dual-cell staining method, we are able to further assess the specificity of our method by correlating PTPRC TaqMan probe signal with a cell population expressing the PTPRC transcript (green Raji cells) and a cell population lacking PTPRC expression (violet PC3 cells). The two cell types were mixed, processed with our platform, thermocycled and the resultant emulsion was imaged with a fluorescence microscope (Figure 2a). Calcein green fluorescent drops containing Raji cell lysate frequently displayed the presence of PTPRC probe HEX fluorescence; however, calcein violet fluorescent drops indicating the presence of PC3 lysate did not correlate with the PTPRC probe signal. Similar control experiments where reverse transcriptase was left out of the PCR reaction mix further demonstrated the specificity of the PTPRC TaqMan probe in our workflow (Figure S1 of the Supporting Information). To quantify the specificity of our single-cell RT-PCR method, we acquired hundreds of images of the thermocycled drops and analyzed them using a custom MATLAB script to measure drop intensities (Figure S2 of the Supporting Information). This data is compiled and shown as scatterplots in Figure 2, panels b−d. To classify each cell based on expression of PTPRC and correlate this with the cell stain, we specified thresholds for the fluorescence values using a statistical algorithm. Using this method, we first confirmed that Raji and PC3 cells were rarely coencapsulated within the same drop (Figure 2b). We then determined PTPRC probe fluorescence in the presence of PC3 cell lysate and observed little correlation between the two (Figure 2d). Conversely, we observed strong correlation between the presence of Raji cell lysate and PTPRC expression (Figure 2c), demonstrating that this probe is a faithful marker for Raji cells. Using replicates of this experiment, we calculated a detection rate of 85% for Raji cells and 0.3% for PC3 cells using PTPRC (Figure 2e). To further ensure that our method is specific for cell detection and to eliminate the possibility that biochemical differences between the lysates of Raji and PC3 are the source of the measured correlation, we also tested a TaqMan probe for GAPDH. In contrast to the PTPRC expression, GAPDH expression is observed in both PC3 cells and Raji cells. This probe thus allowed us to target both cell types for detection. With the same procedure as before, we determined the correlation GAPDH expression in PC3 and Raji cells. By analyzing mixtures of PC3 and Raji cells in different replicates, we observed that GAPDH expression is detected in 88% of

Figure 2. Single-cell RT-PCR assays are highly specific. (a) Merged brightfield and fluorescence image showing PTPRC TaqMan probe (red) detection of Raji (green), but not PC3 (blue), lysate in drops. Individual fluorescent channels from the dashed region are shown below. The scale bar is 100 μm. (b−d) Scatterplots of drop fluorescence obtained from microscope images demonstrate PTPRC probe specificity for Raji cell lysate. (b) Raji and PC3 coencapsulation is rare. (c) PTPRC fluorescence is seen in 87.6% of drops with Raji cell lysate, but only (d) 0.3% of drops with PC3 cell lysate, excluding the two drops that show Raji and PC3 coencapsulation. (e) Detection rates of Raji or PC3 cells are shown for PTPRC and GAPDH TaqMan probes.

drops containing PC3 cells and in 99% of drops containing Raji cells (Figure 2e). These results indicate that we can detect transcripts from both cell types using this method; furthermore, this method enables targeted and highly specific detection of a distinct cell type based on a unique transcriptional marker. In addition to allowing the specific detection of cells, our platform is capable of operating at ultrahigh-throughput rates, a distinctive advantage of droplet-based microfluidic workflows. Due to the number of drops generated by our microfluidic system (∼300000/h), it is not possible to image all drops with fluorescence microscopy. Consequently, we built a highthroughput fluorescence-based droplet detector that excites fluorescent dyes in the drops with lasers and records the resultant emitted light with photomultiplier tubes (depicted in Figure 3a and in more detail in Figure S3 (panels a−c) of the Supporting Information). We configured the optics to observe 8019

dx.doi.org/10.1021/ac402057q | Anal. Chem. 2013, 85, 8016−8021

Analytical Chemistry

Article

Figure 3. Ultrahigh-throughput inline fluorescence detection. (a) Schematic of inline fluorescence drop detection after RT-PCR. (b) Representative sample of the data collected using the inline detector. Fluorescence drop signals are recorded as peaks in PMT voltage versus time. Raji positive drop fluorescence (blue peaks) correlates with PTPRC probe fluorescence (red peaks). (c) A longer time series, containing (b) in the gray region, indicating the data quality. Blue arrows show examples of false negatives demonstrating the relative rarity of these events.

Raji cells labeled with calcein violet stain and HEX fluorescence from the PTPRC TaqMan probe. Time series data from drops run through our inline detector show fluorescence peaks as a function of time, where the amplitudes of the peaks are proportional to the droplet intensities for a given channel, as shown in Figure 3 (panels b and c). In this experiment, Raji cells were encapsulated into 10.4% of all analyzed drops. Following normalization, gating, and correlation analysis of the peak values, we observed that 75.0% of drops with Raji cell lysate also had a PTPRC probe signal (Figure 4a). As a control,

approaches lack the throughput or specificity to analyze large numbers of cells. The challenge to date has been the potent inhibition of RT-PCR in picoliter volumes. Our system overcomes this inhibition by using microfluidic devices to proteinase treat cells, dilute cell lysate, split drops, and picoinject RT-PCR reagents. This allows us to perform reactions in small, thermostable drops with cell lysates that are optimized for RT-PCR. Additionally, with microfluidics, we are able to process cells at extremely high throughput, analyzing ∼50000 cells per run. With higher encapsulation rates using inertial ordering, microfluidic device improvements to accommodate higher flow rates, and the implementation of parallelization, we anticipate increasing the number of cells analyzed by 100-fold. This should enable >1 million single cell RT-PCR reactions in a few hours. Such ultrahigh-throughput single-cell RT-PCR will enable the detection and analysis of circulating tumor cells (CTCs) and other rare cell types in mixed populations using gene expression analysis. Our system can unambiguously identify unique cell types from heterogeneous populations of cells based on gene expression. This is enabled by the simultaneous use of live cell stains and TaqMan probes in our microfluidic workflow that maintains drop compartmentalization at all times. The ability to correlate the presence of a previously intact cell and transcript amplification in the same microdroplet is a critical control that has been lacking from prior attempts at single-cell RT-PCR in emulsion-based systems. In direct support of the importance of this control, we observed TaqMan signal in some drops that did not contain significant concentrations of lysate. Amplification in these drops is likely due to free transcripts from cell lysis prior to encapsulation and is efficient due to the absence of inhibitory lysate in these drops. Despite careful handling of the cells and the use of buffers to minimize cell stress, such lysis is inevitable due to the sheer number of cells analyzed by our system. Consequently, “lysate free” RT-PCR amplification represents a significant source of false positive contamination in droplet-based microfluidic workflows. Only by identifying drops that started with encapsulated intact cells is it possible to account for this potential source of error. Additionally, through the use of fluorescence-based dielectrophoretic droplet sorting, these false positive “lysate-free” drops can be eliminated, enabling subsequent downstream

Figure 4. Ultrahigh-throughput detection of Raji cells with single-cell RT-PCR reactions. Fluorescence data acquired with the ultrahighthroughput inline detector was plotted using a heat map. (a) 47078 Raji cell lysate drops (calcein violet fluorescence) also have a PTPRC probe signal (HEX fluorescence) following single cell RT-PCR in the presence of reverse transcriptase and inline detection. (b) No-RT control drops show minimal PTPRC probe fluorescence with inline detection.

we repeated this experiment with RT-PCR reagent lacking reverse transcriptase. As expected, these control drops showed minimal PTPRC probe signal and correlation with Raji lysate, demonstrating that the TaqMan signal we observed was specific (Figure 4b). In this single high-throughput experiment, we performed 47078 single-cell RT-PCR reactions in 2 h of device run time while using only 400 μL of total RT-PCR reagent.

■

CONCLUSIONS Robust, single-cell RT-PCR is essential for gene expression analysis of heterogeneous populations of cells, but available 8020

dx.doi.org/10.1021/ac402057q | Anal. Chem. 2013, 85, 8016−8021

Analytical Chemistry

Article

(12) Mary, P.; Dauphinot, L.; Bois, N.; Potier, M. C.; Studer, V.; Tabeling, P. Biomicrofluidics 2011, 5, 24109. (13) Zhang, H.; Jenkins, G.; Zou, Y.; Zhu, Z.; Yang, C. J. Anal. Chem. 2012, 84, 3599. (14) Eastburn, D. J.; Sciambi, A.; Abate, A. R. PLoS One 2013, 8, e62961. (15) Papadopoulos, N. G.; Dedoussis, G. V.; Spanakos, G.; Gritzapis, A. D.; Baxevanis, C. N.; Papamichail, M. J. Immunol. Methods 1994, 177, 101. (16) Pulvertaft, J. V. Lancet 1964, 1, 238. (17) Kaighn, M. E.; Narayan, K. S.; Ohnuki, Y.; Lechner, J. F.; Jones, L. W. Invest. Urol. 1979, 17, 16. (18) Agresti, J. J.; Antipov, E.; Abate, A. R.; Ahn, K.; Rowat, A. C.; Baret, J. C.; Marquez, M.; Klibanov, A. M.; Griffiths, A. D.; Weitz, D. A. Proc. Natl. Acad. Sci. U.S.A. 2010, 107, 4004.

analysis of only the drops containing successful single-cell RTPCR reactions.18 The microfluidic system described here should accelerate the investigation of cellular heterogeneity by enabling new tools and approaches for characterizing large cell populations at the level of individual cells. In addition to identifying cell types or states through the use of TaqMan assays, this method is ideal for mitigating intercellular transcript amplification bias prior to microarray analysis or sequencing studies. Furthermore, our microfluidic workflow could be modified to barcode and amplify all of the transcripts within a single drop, thereby permitting ultrahigh-throughput transcriptome sequencing of single cells.

■

ASSOCIATED CONTENT

■

AUTHOR INFORMATION

S Supporting Information *

Three figures that further detail the methods used in this study and an additional control experiment that strengthens the conclusions of the main paper. This material is available free of charge via the Internet at http://pubs.acs.org. Corresponding Author

*E-mail: [email protected]. Tel: (415) 476-9819. Author Contributions †

These authors contributed equally.

Notes

The technology described in this paper is the subject of a patent filing. The authors declare the following competing financial interest(s): the authors have filed a patent application on methods detailed in this manuscript.

■

ACKNOWLEDGMENTS We thank Phil Romero and John Haliburton for reading the manuscript. Data for this study were acquired at the Nikon Imaging Center at UCSF/QB3. This work was supported by an NSF Career award (DBI-1253293), a Research Award from the California Institute for Quantitative Biosciences (QB3), the Bridging the Gap Award from the Rogers Family Foundation, a New Frontiers Research Award from the UCSF/Sandler Foundation Program for Breakthrough Biomedical Research, and a grant from the University of California Proof of Concept Program.

■

REFERENCES

(1) Bendall, S. C.; Nolan, G. P. Nat. Biotechnol. 2012, 30, 639. (2) Kalisky, T.; Blainey, P.; Quake, S. R. Ann. Rev. Genet. 2011, 45, 431. (3) Kalisky, T.; Quake, S. R. Nat. Methods 2011, 8, 311. (4) Levsky, J. M.; Singer, R. H. Trends Cell Biol. 2003, 13, 4. (5) Guo, M. T.; Rotem, A.; Heyman, J. A.; Weitz, D. A. Lab on a Chip 2012, 12, 2146. (6) Novak, R.; Zeng, Y.; Shuga, J.; Venugopalan, G.; Fletcher, D. A.; Smith, M. T.; Mathies, R. A. Angew. Chem., Int. Ed. 2011, 50, 390. (7) Vyawahare, S.; Griffiths, A. D.; Merten, C. A. Chem. Biol. 2010, 17, 1052. (8) Teh, S. Y.; Lin, R.; Hung, L. H.; Lee, A. P. Lab on a Chip 2008, 8, 198. (9) Arezi, B.; McCarthy, M.; Hogrefe, H. Anal. Biochem. 2010, 400, 301. (10) Hedman, J.; Radstrom, P. Methods Mol. Biol. 2013, 943, 17. (11) White, A. K.; VanInsberghe, M.; Petriv, O. I.; Hamidi, M.; Sikorski, D.; Marra, M. A.; Piret, J.; Aparicio, S.; Hansen, C. L. Proc. Natl. Acad. Sci. U.S.A. 2011, 108, 13999. 8021

dx.doi.org/10.1021/ac402057q | Anal. Chem. 2013, 85, 8016−8021