Improving single-cell transcriptome sequencing efficiency with a microfluidic phase-switch device

Abstract

In this paper, we present a novel method to improve the efficiency of single-cell transcriptome sequencing for analyzing valuable cell samples. The microfluidic device we designed integrates multiple single-cell isolation chambers with hydrodynamic traps and achieves a nearly 100% single-cell capture rate and minimal cell loss, making it particularly suitable for samples with limited numbers of cells. Single cells were encapsulated using a novel phase-switch method into picoliter-sized hydrogel droplets and easily recovered for subsequent reactions. Minimizing the reaction volume resulted in a high reverse transcription (RT) efficiency for RNA sequencing (RNA-Seq). With this novel microfluidic platform, we captured dozens of hESC cells (H9) simultaneously and obtained live cells in individual picoliter volumes, thus allowing for the convenient construction of a high-quality library for deep single-cell RNA-Seq. Our single-cell RNA-Seq results confirmed that a spectrum of pluripotency existed within an H9 colony. This integrated microfluidic platform can be applied to various cell types for the investigation of rare cellular events, and the phase-switch single-cell processing strategy will improve the efficiency and accessibility of single-cell transcriptome sequencing analysis.

Introduction

Single-cell transcriptome analysis has attracted extensive interest as it is a powerful tool for quantifying transcriptional heterogeneity in a population of cells^{1, 2}, which is critical for research on cancer, developmental biology, and stem cells^{3, 4}. The characterization of the whole transcriptome of individual cells by single-cell sequencing is essential for studying rare and valuable cells^5–7. Extensive investigations into single-cell gene expression have the potential to reveal rare cell populations, but require a low-cost and high-efficiency method for single-cell isolation and sample preparation⁸. Emerging microfluidic technologies provide a powerful platform for high-throughput single-cell isolation, as they allow multiplexing, precise volume control, and reduced sample consumption⁹. Commercial single-cell technologies implementing microfluidics, such as BD FACSAria, Fluidigm C1, Silicon Biosystems DEPArray, and AVISO CellCelector, have relatively high equipment and reagent costs^10–13. In order to lower the cost per sample, the isolated single-cells can be barcoded by mixing a cell and a barcoding bead in a microwell or droplet, thus enabling hundreds to thousands of cells to be pooled for one RNA-Seq reaction^14–16. However, a large number of input cells are often required due to cell lost, and limited their applications in analysis of rare cells. Various multiplexed valve-based microfluidic systems have been developed for single-cell genetic analysis, such as RT-PCR and digital PCR^17–19, as well as for sample preparation for single-cell sequencing^{20, 21}. These methods enable precise single-cell manipulation and integrate many pre-sequencing steps in the device, such as cell lysis, mRNA extraction, reverse transcription, cDNA fragmentation, and library preparation. Among these pre-sequencing reaction steps, controlling the efficiency of reverse transcription (RT) is essential for high-quality library construction in single-cell sequencing²². Existing valve-based microfluidic systems usually generate a large volume for each single-cell reaction; however, large reaction volumes result in lower mRNA concentrations, which are undesirable for RT reaction²³. Here, we report a novel phase-switch microfluidic processor that can perform nanoliter RT reactions for high-quality cDNA generation. Since RT highly depends on mRNA concentration, minimizing the reaction volume could improve the yield of cDNA and thus reduce amplification bias^{24, 25}. This approach allows for the analysis of rare cell samples with reduced cost and improved efficiency.

In this study, we demonstrate that our microfluidic phase-switch platform, which integrates multiple single-cell isolation chambers and on-site droplet generators, is capable of trapping multiple single-cells in a highly efficient manner and encapsulating them in picoliter volume hydrogels. The microfluidic phase-switch platform provides a highly controllable way to perform single-cell transcriptome sequencing analysis. The hydrogel encapsulation strategy presents a unique approach for single-cell sample preparation²⁶. The cells remain viable in individual hydrogels and are easy to recover from the microwells in the platform, enabling convenient subsequent single-cell manipulation²⁷. The phase-switch mechanism minimizes the carry-over volume of each cell and enables RT in nanoliter volumes with a very high mRNA concentration, which is essential for obtaining high-quality cDNA and constructing a library for deep RNA-Seq. In principle, this microfluidic platform has the potential to be expanded to the investigation of rare cellular events in many different cell types with greatly improved cell-capture efficiency and accessibility of single-cell transcriptome sequencing.

Experimental

Phase-switch microfluidic device design

The phase-switch single-cell processor was composed of a multi-layer microfluidic device for single-cell capture and thermoplastic microwells for single-cell collection. As shown in the schematic diagram in Figure 1a, the microfluidic device consisted of a three-layer elastomeric structure and a thermoplastic substrate. In the polydimethylsiloxane (PDMS) elastomeric device, the top layer contained the fluidic channels, the middle layer contained the pneumatic control channels with push-up designs, and the bottom layer was a blank layer with through-hole structures. The device was designed to perform single-cell isolation and picoliter droplet generation by pneumatic control, as illustrated in Figure 1b. The PDMS chip was reversibly bonded to a cyclic olefin copolymer (COC) substrate with microwell structures to collect the isolated single-cells. We designed the device with a hybrid PDMS-COC configuration; PDMS was chosen for its intrinsic advantage for valve actuation, and COC is a suitable material for the multi-well layer due to its excellent biocompatibility, low water absorption, and high transparency.

Figure 1.

Schematic diagram of the microfluidic single-cell phase-switch device. (a) Design of the multiplexed single-cell processor. Inset shows single-cells captured in the hydrodynamic cell traps. (b) Workflow of single-cell processing using phase-switch units. The oil was first loaded into the channel, followed by droplet generation using valve-controlled microfluidics.

Microfluidic device fabrication

Standard multilayer soft lithography was used for the microfabrication of the phase-switch device. The silicone molds for casting PDMS elastomeric devices were made by photolithography with an EVG mask aligner (EVG 610, EV Group, Austria). The pneumatic layer (20 μm) was fabricated by SU8–2035 photoresist (Microchem). For the fluidic layer, the mold was prepared by two-step lithography. Briefly, the 7 μm high cell trap feature was first patterned with SU8–2010 photoresist (MicroChem), then the 15 μm thick channel feature was made in AZ-50XT positive photoresist (AZ Electronic Materials). All silicon molds were treated with trimethylchlorosilane by gas phase silanization prior to making the PDMS chip.

To prepare the fluidic layer, PDMS (RTV615, ratio 5:1) was poured on the mold, degassed, and cured for 50 min at 80°C. The pneumatic layer was made by spin-coating PDMS (RTV615, ratio 20:1) and baked for 40 min at 80°C. After baking, the PDMS fluidic layer was peeled off and aligned to the pneumatic layer, and then baked again at 80°C for 60 min. Meanwhile, another Si wafer with pillar structures was prepared and spin-coated with a thin layer of PDMS (RTV615, ratio 20:1) to generate a blank layer with through-hole structures. The two-layer PDMS chip was separated from the pneumatic mold, mounted onto the blank layer with through-hole structures, and baked for 4 hours at 80°C. Finally, the three-layer bonded structure was treated with oxygen plasma using a reactive ion etcher (RIE-10NR, Samco, Japan), reversibly bonded to a COC substrate with microwell structures, and baked overnight at 80°C. The microwell structures (32 wells with 2 mm diameter) on the COC substrate were made using a computer controlled micro-milling system.

Microfluidic device operation

The semi-automated device was operated by a homemade pneumatic control system with a touchscreen, a single chip microprocessor, and solenoid valves. Five pneumatic valves operated the microfluidic device, and a control program was written in the microprocessor for chip automation. The solenoids were connected to the microfluidic control line ports by Tygon tubing. The control lines were filled with FC-40 electronic liquid (Fluorinert^™, 3M, USA) and actuated at a pressure of 25 psi.

Single-cell isolation

Human embryonic stem cell (hESC) H9 lines (WiCell Research Institute, Madison, WI) were cultured using the recommended feeder-free protocol^{28, 29}. Cells were trypsinized and resuspended in 1× phosphate buffered saline (PBS, pH 7.4) for single-cell capture experiments. The microfluidic device was first rinsed with 1× PBS and degassed by driving trapped air through the walls of the gas-permeable device. Then a gelation buffer was produced with Na-alginate (Sigma Aldrich) dissolved in deionized water (2% w/w) and mixed with 2M CaCO₃ nanoparticles, and cells were diluted in the gelation buffer to form the cell suspension. The cell suspension was injected into the inlet and captured by passive cell trapping. After most cell traps were occupied by single cells, oil was loaded into the flow channels at a pressure of approximately 2–3 psi, replacing the cell suspension. Then the cell chamber valves were closed to isolate each individual cell in a chamber, and oil was pushed at a pressure of 7 psi from the left side into the cell chambers. Finally, in each cell capture unit, a picoliter droplet was generated to encapsulate only one cell, and subsequently collected in each individual microwell on the thermoplastic plate. The volume of the cell suspension (200–300 pL) was precisely controlled by the dimension of the cell trapping structure in the chamber. Meanwhile, since 0.1% glacial acetic acid (Sigma Aldrich) was added to the pushing oil, gelation of the droplet occurred when loading into the microwells, because the suspended nanoparticles dissolved due to a pH drop initiated by the acetic acid, thus forming the hydrogel droplets containing single-cells.

Library construction and sequencing

Single-cell hydrogel droplets were retrieved from the microwells on the COC plate, then a chelating buffer (55mM sodium citrate, 30mM EDTA, 150mM NaCl) was introduced to depolymerize the hydrogels containing single-cells. RT reactions were performed as described previously to generate cDNA for library construction^{23, 30}. For each single-cell sample, cDNA was obtained with REPLI-g WTA Single Cell Kit (Qiagen) according to the manufacturer’s protocol. Fragmentation of amplified cDNA was performed with NEBNext dsDNA Fragmentase (NEB). For each sample, 100 ng of small fragments (50−500bp) were used for RNA-Seq library preparation with the NEBNext Library Prep kit (NEB). All libraries were quantified using BioAnalyzer 2100 (Agilent) and KAPA Library Quantification Standards Kits for Illumina platforms (KAPA) and were sequenced on the Illumina HiSeq 2000 platform (Illumina, USA).

Sequencing data analysis

Before analyzing the sequencing data, the raw reads were filtered according to sequencing quality, adaptor contamination, and duplicated reads, so only high-quality reads remained. The RNA-Seq data were analyzed with Partek Flow version 4 (Partek Inc., USA). Reads with Phred scores less than 20 were trimmed from both ends of the raw sequencing reads, and trimmed reads shorter than 25 nt were excluded from further analyses. Both pre- and post-alignment QA/QC was carried out with default settings. Trimmed reads were mapped onto human genome hg38 using Tophat 2.0.8 as implemented in Flow with default settings, using Gencode 20 annotation (www.gencodegenes.org) as guidance. Gencode 20 annotation was used to quantify aligned reads to genes/transcripts using the Partek E/M method³¹. Read counts per gene were normalized for all samples using Upper Quartile normalization³² and analyzed for differential expression using Partek’s Gene Specific Analysis method. Genes with less than ten reads in any sample were excluded.

Results and discussion

Single-cell encapsulation strategy

In the microfluidic platform (Figure 2a), cells were first isolated and immobilized in individual traps in the cell capture chamber. In order to capture a live mammalian cell in a defined volume reaction system for RNA-Seq, we developed a novel phase-switch method to convert single-cells from a water-based medium into picoliter-sized hydrogel droplets. Unlike PCR which is performed in cycles, the reverse transcription (RT) reaction is a one-step process that depends on RNA concentration. Therefore, controlling the volume of individual single-cell reactions is the key factor for successful RNA-Seq preparation^{24, 25}. Preferably, the volume of the cell solution should be limited to 200–300 pL for a sufficiently high mRNA concentration in the RT reaction.

Figure 2.

Illustration of the single-cell droplet encapsulation process. (a) Photograph of the microfluidic single-cell device. Inset shows a single cell that has been captured by the cell trap. (b) Pneumatic controller with a microfluidic chip. (c) The single-cell capture efficiencies for various numbers of cells loaded into the microfluidic device. (d) Multiple series of optical micrographs showing the sequential single-cell droplet encapsulation process. The scale bar is 200 μm. (e) The single-cell encapsulated hydrogels in multiple collection microwells. The scale bar is 20 μm.

Multilayer microfluidic devices with sieve structures were previously used to carry out single-cell capture^{17, 18}. However, the previous method pushed out the captured single-cell with reaction buffer, the volume of the single-cell suspension was hard to control. To precisely control the volume of the cell suspensions, we used oil to encapsulate the captured single-cells, and subsequently generated hydrogel droplets containing single-cells with defined volumes. The hydrogel droplets kept the cells viable for a few days, and they were easy to recover from the oil phase for sequencing analysis. Our microfluidic processor executed a critical step of the single-cell transcriptome analysis by capturing each cell in a small volume, resulting in great improvement of the RT reaction.

Each of the single-cell extraction experiments was monitored step by step in real time under a microscope, while a homemade pneumatic control system controlled the operation of the device (Figure 2b). First, single cells were captured individually in the sieve structure in each capture unit. The sieve structure was designed to accommodate the size of human embryonic stem cells (hESCs). The cell loading process lasted for less than two seconds, with almost every hydrodynamic cell trap ultimately being occupied by a single cell, as shown in the inset of Figure 2a. We loaded various numbers of single hESCs into the microfluidic device and found that the cell loss was minimal. Figure 2c shows the capture efficiencies with standard deviations when different numbers of cells were loaded into the microfluidic device. For each cell loading number, 3 runs of experiments were performed. When the cell loading numbers ranged from 1 to 29 cells, the capture rate was 100%, when the cell loading number was 30 cells, 31 cells and 32 cells, the capture rate varied from 96.7% to 98.9%.

After single-cell isolation, the next step was single-cell encapsulation. To encapsulate an hESC in a volume-defined droplet, we used 3M fluid (Novec 7500) containing 1 % pegylated surfactant as the oil phase. The whole single-cell droplet encapsulation process is depicted in Figure 2d. First, oil filled all the flow channels except the center channel next to the sieve structure (Figure 2 panel d1-d3). After filling with oil, each cell loading channel was partitioned by valves, and oil was pushed from the left side to create a droplet containing a single cell at the sieve structure (Figure 2 panel d4). The whole encapsulation process is demonstrated in supplementary Movie S1^§. The volume of the cell encapsulating droplet was precisely defined by the size of the center channel next to the sieve structure. In this device, the droplet volume was calculated to be 270 pL with a CV < 5%. We measured the droplet volume variation for 150 individual cells and the standard deviation of each droplet volume was less than 5%. The single-cell droplets were initially kept in liquid form, then the pushing oil containing acetic acid surrounded the droplets in the microwells and created a pH drop, initiating the gelation of the droplets. We used a live cell staining reagent, Vybrant™ DyeCycle™ Green Stain (ThermoFisher, USA) to test the viability of cells. The single-cell encapsulated hydrogels were observed in multiple collection microwells (Figure 2e), the cell staining results show that all cells remained viable in the hydrogels. The size of the hydrogel was about 80 μm in diameter, showing that cells were encapsulated in picoliter volumes. Homogenous droplets were generated simultaneously in this microfluidic device for sequencing analysis, thus facilitating single-cell manipulation and improving experimental efficiency.

Single-cell transcriptome sequencing

With this microfluidic device, multiple live single-cells were encapsulated in 270 pL droplets for next-generation sequencing on an Illumina Hiseq 2000 Instrument (BGI, USA). After retrieving the single-cell hydrogel droplets from the COC micro-well plate, a chelating buffer was introduced to depolymerize the hydrogels and release the single cells. A series of reactions was carried out, including cell lysis, cDNA synthesis, amplification, purification, and library construction. The resulting library was subsequently subjected to transcriptome sequencing. The sequencing data were first analysed using Circos plots, which represented the coverage areas of whole transcriptome sequencing, and then mapped to the reference genome using the software Circos-0.64. The coverage of each cell was calculated and plotted with the rings from inside out representing single cells 1 to 6, 7 to 12, 13 to 18, and 19 to 24, respectively. The outermost ring represents the chromosomes of the human reference genome (Figure 3).

Figure 3.

Circos plots representing the coverage of the transcriptome sequencing. The rings from inside out represent single-cell numbers 1 to 6, 7 to 12, 13 to 18, and 19 to 24, respectively. The outermost ring represents the chromosomes of the human reference genome.

The coverage of the transcriptome of all single-cells was well distributed in the 24 single-cells, which demonstrated that high quality and consistent cDNAs from all 24 single-cells were obtained. The mapping rate to the human genome was up to 90%, which demonstrated that the defined, picoliter-scale carry-over volumes increased the RNA concentration and led to highly effective RT reactions and identified more genes than previous studies^33–35.

Gene expression heterogeneity in hESCs

Information on the gene expression of hESCs is critical for understanding the normal and pathological development of human cells and tissues. Due to the prior limitations of analytical methods, the H9 cell line was originally assumed to be a pure cell population^36–38 until a single-cell molecular profiling study was performed in 2008³⁹. With our microfluidic platform, we were able to take a close look at the heterogeneity of gene expression among the cells within an hESC colony.

Noticeable gene expression heterogeneity among the 24 single-cells was revealed by the clustering diagram in Figure 4. We analyzed the sequencing data using the heatmap package in R version 3.3.1. The cut-off threshold was set at >10 reads. We identified variation among the 24 individual hESCs in the expression levels of 95 known pluripotency genes (Figure 4a). For example, the pluripotency gene POU5F1 (also known as OCT4) was typically expressed at a high level but was differentially expressed among the 24 cells. This result agrees with previous immunocytochemistry results⁴⁰. Besides POU5F1, there were also other pluripotency genes, such as TDGF1 and DMNT3b, that were expressed at high but varying levels among the 24 cells. In addition, we observed that some of the pluripotency genes were expressed at low levels, such as NANOG and GDF3 (Figure 4a). In our microfluidic phase-switch device, since the minimized carry-over volume is essential for obtaining high-quality cDNA, even genes with low expression levels could be efficiently detected, without being drowned out by the signals of highly expressed genes as in previous studies^{41, 42}. The expression profiles of 57 pluripotency genes with low expression levels were further analyzed with unsupervised hierarchical clustering (Figure 4b). These data suggested that individual H9 cells express pluripotency genes at various levels, and may represent a spectrum of pluripotency. This observation of various degrees of pluripotency within an H9 colony may explain why such a colony can undergo spontaneous differentiation. It is important to identify the heterogeneity of H9 cell lines. Many biological assays currently considered the H9 cells as a homogeneous population and have difficulty to explain the variations in results from different laboratories. The heterogeneity of H9 cells underscores the importance of selecting control H9 cells for matching these in experimental conditions. Therefore, our deep single-cell RNA-Seq method provided a new insight into the heterogeneity within an H9 population.

Figure 4.

Differential gene expression of pluripotency genes within 24 individual hESCs. Unsupervised hierarchical clustering of (a) all 95 pluripotency genes, and (b) 57 pluripotency genes with low expression levels.

Conclusions

In summary, we have developed a microfluidic phase-switch device to isolate single cells in very small volumes and perform single-cell whole transcriptome analysis. This platform has close to a 100% single-cell capture rate, and it is particularly suitable for analysis of rare cell samples. With a novel phase-switch mechanism and a hybrid design, single-cells are captured and converted to picoliter-sized hydrogel droplets for further analysis. The subsequent nanoliter RT reaction results in high-quality cDNA production from single-cells and allows deep sequencing, as demonstrated by our analysis of multiple H9 hESCs. Up to 32 single-cell assays can be performed in parallel in a single device, and the device is scalable to accommodate more single-cells if needed, which improves experimental efficiency. The sequencing results in the study reported here confirmed that individual hESCs differentially express pluripotency genes, which indicates that there is a spectrum of pluripotency among cells within an H9 colony^43–45.

Overall, the microfluidic device we developed provides a cost-effective and robust platform for deep single-cell RNA-Seq, which is difficult to realize with other microfluidic devices. Because of the small dead-volume of the device and individual loaded cells can be monitored to prevent cell loss, this device is most suitable for single-cell RNA-seq of rare cells such as circulating tumor cells isolated from blood. The hybrid PDMS-COC device with phase-switch droplet generation design achieved single-cell manipulation with easy operation, and provided high-quality cDNA ready for deep RNA-Seq. This multiplexed microfluidic platform can be applied to the study of various types of rare cells, and the phase-switch single-cell processing strategy for transcriptome sequencing will have a broad impact on molecular biology.