Tian Yu, Chong Tang, Ying Zhang, Ruirui Zhang, Wei Yan, Microfluidics-based digital quantitative PCR for single-cell small RNA quantification, Biology of Reproduction, Volume 97, Issue 3, September 2017, Pages 490–496, https://doi.org/10.1093/biolre/iox102
Download citation file:
© 2019 Oxford University PressClose
Quantitative analyses of small RNAs at the single-cell level have been challenging because of limited sensitivity and specificity of conventional real-time quantitative PCR methods. A digital quantitative PCR (dqPCR) method for miRNA quantification has been developed, but it requires the use of proprietary stem-loop primers and only applies to miRNA quantification. Here, we report a microfluidics-based dqPCR (mdqPCR) method, which takes advantage of the Fluidigm BioMark HD system for both template partition and the subsequent high-throughput dqPCR. Our mdqPCR method demonstrated excellent sensitivity and reproducibility suitable for quantitative analyses of not only miRNAs but also all other small RNA species at the single-cell level. Using this method, we discovered that each sperm has a unique miRNA profile.
Real-time quantitative polymerase chain reaction (PCR) (qPCR) has been widely used for quantitative analyses of DNA and RNA. Despite the increasing usage of next-generation sequencing technologies, qPCR remains a convenient and effective means for quantitative analyses . The conventional qPCR methods require internal controls for relative quantification, and for this purpose, housekeeping genes are usually chosen; however, many of them also display variable levels among different or even the same cell types under various physiological conditions  and thus, tend to cause errors in quantitative analyses [3,4]. Digital qPCR (dqPCR) analyses, as a new technology for nucleic acid detection and quantification, are characterized by absolute quantification without the need for internal controls. Digital qPCR is based on the concept that each template molecule is partitioned into a single microreaction, in which interference from other nucleic acids is minimal, leading to a much greater sensitivity and accuracy compared to the conventional qPCR methods . For small noncoding RNA (sncRNA) quantification, mainly two qPCR methods exist: the first relies on ligation of a universal adaptor to the 3΄ end of a sncRNA by terminal transferase; after reverse transcription, the cDNAs for the sncRNA are amplified using a downstream universal primer from the adaptor and an upstream primer from the sncRNA . The specificity of this method is relatively low because of the use of the universal primers. The second employs a stem-loop primer that anneals to the 3΄ end of an microRNA (miRNA), and this method has demonstrated a much greater specificity . However, both methods require microgram levels of starting total RNAs or sncRNAs, and the reproducibility is poor when sncRNA input is low, which may explain why a reliable single-cell sncRNA quantitation method has not yet been available. Recently, a droplets-based dqPCR method (called ddqPCR herein), which combines the droplets-based template partition with the use of stem-loop primers, has greatly improved the reproducibility of microRNA detection and quantitation, especially when only a limited amount of starting materials are available . Unfortunately, the design of the stem-loop primers is proprietary, and the primers are only available for known miRNAs. For analyzing other small RNA species, e.g. tRNA-derived small RNA (tsRNAs) , piwi-interacting RNA (piRNAs) , mitochondrial DNA-encoded small RNA (mitosRNAs) , small nucleolar RNA (snoRNAs) , etc., there is no dqPCR method available for now. Moreover, the “stem-loop” method cannot differentiate closely related miRNAs with only one or two nucleotide differences. To overcome these shortcomings, we developed an alternative dqPCR method, in which single-template molecules are partitioned into micro chambers instead of droplets followed by high-throughput dqPCR. This microfluidics-based dqPCR (called mdqPCR herein) method has demonstrated excellent reproducibility and sufficient sensitivity for single-cell sncRNA detection and quantitation. Moreover, it does not require special primers and can distinguish sncRNAs with only one or two nucleotide differences.
Materials and methods
Mice were housed in a temperature- and humidity-controlled animal facility at the University of Nevada, Reno, with free access to water and food. Adult male mice of 8–12 weeks of age were used for collecting epididymal spermatozoa. Mouse use (Protocol#: 00494) was approved by Institutional Animal Care and Use Committee of the University of Nevada, Reno and are in accordance with the “Guide for the Care and Use of Experimental Animals” established by National Institutes of Health (1996, revised 2011).
Proof-of-concept accuracy test
Four synthetic miRNAs (miR-34b, miR-34c, miR-191, and miR-16) were purchased from the Integrated DNA Technologies, Inc. (for sequences see Supplementary Table S1). A three-fold serial dilution (100 pM, 33.3 pM, 10 pM, 3.33 pM, 1 pM, 0.33 pM, 0.1 pM, and 0.033 pM) was performed for each of the four synthetic miRNAs in triplicates. Each dilution was polyadenylated in a reaction containing 2.5 μl miRNAs, 1 μl first-strand buffer (Clontech), 0.5 μl 2.5 mM LD_CDS primer (Supplementary Table S2), 0.25 μl RNaseOUT (Invitrogen), 0.5 μl polyA polymerase (NEB), and 0.5 μl ATP, at 37°C for 30 min followed by incubation at 65°C for 5 min. The polyadenylated RNAs were then reverse-transcribed in a reaction containing 1 μl first-strand buffer (Clontech), 1 μl DTT, 0.25 μl SMARTScribe Reverse Transcriptase (Clontech), 1 μl dNTP, 0.5 μl RNaseOUT (Invitrogen), and 1.25 μl H2O, at 42°C for 30 min followed by inactivation at 85°C for 5 min. The sncRNA cDNA templates were then subjected to mdqPCR using the Fluidigm BioMark HD system, as described below.
Microfluidics-based digital quantitative PCR using the Fluidigm BioMark HD system
To achieve one single cDNA molecule per micro chamber in a Fluidigm 37K IFC chip (48.770 digital array IFC), the cDNA template dilution factors were determined using the dilution factor calculation sheet (Supplementary File S1), which is created based on the protocol provided by the Fluidigm. Sample preparation, priming, partition, amplification, and data collection were performed following the manufacturer's instructions for the 37K digital PCR workflow (Fluidigm). TaqMan Gene Expression qdPCR was performed using a common “Trueseq anti” primer (Supplementary Table S2) and sncRNA-specific primers (Supplementary Tables S3–S5), as well as a TaqMan probe (Supplementary Table S2) on the BioMark HD system following the manufacturer's protocol (Fluidigm). Data mining was conducted according to the user guide for digital PCR analyses (Fluidigm).
Single-sperm capture using the Fluidigm C1 system
Male adult C57Bl/6J mice were euthanized and the epididymis was dissected and transferred into 2.5 ml of calcium-free phosphate-buffered saline (PBS, pH7.4). The epididymis was further dissected into smaller pieces followed by incubation in a humidity incubator at 37°C for ∼10 min. The supernatants (∼2 ml) were collected and washed with PBS for five times by repetitive centrifugation (×100g for 5 min) and resuspension. The washed epididymal spermatozoa were counted by an Automated Cell Counter (TC20, Bio-Rad), and then diluted to a concentration of ∼100 sperm/μl. The single-cell capture was performed following the protocol provided by the Fluidigm.
Single-sperm microfluidics-based digital quantitative PCR
Single-sperm lysis, polyadenylation, reverse transcription, and preamplification were all performed using the Fluidigm C1 Single Cell Auto Prep System (Fluidigm). The open source option of the C1 system allows for customized programing such that the captured single cells can be processed in four micro chambers in succession. The program was designed for single-sperm sncRNA mdqPCR using the C1 Script Builder (Supplementary File S2), and the reactions and conditions in each of the chambers used are illustrated in Supplementary Figure S1. In general, a single sperm is first captured in chamber C containing 4.5 nl of C1 cell wash buffer; the single-sperm cell is then transferred to chamber E1 containing 4.5 nl of 2× lysis buffer (0.5% Salkosyl, 0.05 M Tris–HCl pH8, 0.05 M KCl, and 0.2 M DTT), 1 nl C1 loading buffer, and 3.5 nl water at 70°C for 10 min; this condition had been validated to lead to a complete lysis of the sperm head, which is essential for a complete release of sperm-borne RNAs. Polyadenylation reaction occurs in a 9 nl volume in chamber E2, which contains 1 nl 40% Tween 20, 0.5 nl C1 loading buffer, 3 nl first-strand buffer (Clontech), 1 nl ATP, 1 nl Escherichia coli poly(A) polymerase (NEB), 0.5 nl Clontech RNase Inhibitor, and 2 nl HPLC-purified LD_CDS primer (Supplementary Table S2), at 37°C for 30 min followed by inactivation at 65°C for 5 min. Finally, reverse transcription is conducted at a 9 nl reaction volume in chamber E3, containing 0.5 nl C1 loading buffer, 3 nl Clontech first-strand buffer, 1 nl 0.1 M DTT, 2 nl dNTP, 0.5 nl Clontech RNase Inhibitor, and 2 nl SMARTScribe Reverse Transcriptase (Clontech), at 42°C for 60 min followed by inactivation at 85°C for 5 min. Preamplification of the cDNAs was performed for 25 cycles in chambers E4 and E5 using the miRNA-specific/sense primer (1 mM) and the “Truseq anti short” primer (24 mM) following the protocol for LongAmp Taq polymerase (NEB). The preamplified cDNA templates were then subjected to mdqPCR using the BioMark HD system, as described above.
Microfluidics-based digital quantitative PCR-based small noncoding RNA quantitation using pooled sperm cells
We also performed mdqPCR using 1/1000 of preamplified small RNA cDNAs from 1000 sperm instead of single sperm. For the pooled 1000 sperm, all major steps (lysis, polyadenylation, reverse transcription, and preamplification) were conducted in regular 0.5 ml tubes in exactly the same manner as those for single sperm by scaling up the reaction volumes by 1000 times. An aliquot containing 1/1000 of the preamplified small RNA cDNAs was used to conduct mdqPCR in six replicates using the Fluidigm BioMark HD system, as described above.
The workflow and proof-of-concept accuracy testing
The workflow of our mdqPCR analyses for sncRNAs consists of three steps: cDNA synthesis, template partition, and high throughput dqPCR (Figure 1A). Small noncoding RNAs are first polyadenylated using poly-A polymerase, followed by reverse transcription using oligo-dT flanked by an adaptor sequence (Supplementary Table S2). The sncRNA cDNAs of each sample are then partitioned into 770 micro chambers using a Fluidigm dqPCR 37K chip (48.770 digital array IFC), with each chamber at a volume of 0.85 nl and a total volume of 0.662 μl for each sample. Since hundreds of chambers are used for partition, the template molecules follow a Poisson distribution and thus, the number of chambers with a possibility of having a single template molecule can be calculated . To achieve one template molecule per chamber, we designed an easy-to-use, Microsoft Excel calculator fort determining the dilution factors (Supplementary File S1) based on the protocol provided by the Fluidigm. Once single molecule participation is completed, the chip is subjected to TaqMan-based high-throughput qPCR using the Fluidigm BioMark HD system (Figure 1A).
To evaluate the performance of mdqPCR, we chose to analyze four miRNAs known abundantly expressed in mouse sperm (miR-34b, miR-34c, miR-191, and miR-16) . We synthesized these four miRNAs (Supplementary Table S1), and a serial three-fold dilution was prepared for each of the four synthetic miRNAs in triplicates. Microfluidics-based digital quantitative PCR was conducted using the “Truseq anti” primer (Supplementary Table S2) and synthetic miRNA-specific primers (Supplementary Table S3) on the Fluidigm BioMark HD system. Excellent correlations were observed between tested and quoted concentrations for all four synthetic miRNAs, as evidenced by an average R2 value of ∼0.97 (Figure 1B–E). The mean coefficients of variation ranged between 3% and 22% in this method, which is much lower than those reported for the ddqPCR method (22%–50%) .
Specificity and sensitivity of the microfluidics-based digital quantitative PCR method
The biggest advantage of our mdqPCR method lies in the real-time monitoring of the amplification curves for each template molecule, which allows for determination of amplification properties for each of the sncRNAs templates, including those with one or two nucleotide differences due to point mutation, insertion, and deletion events. For example, if a mismatch is present, the amplification curve tends to shift to the lower right or is relatively flat (Figure 2A). To evaluate the specificity of our method, we synthesized miRNAs containing two point mutations (Supplementary Table S1). When primers matching the original (mutation-free) miRNAs were used to amplify the mutant miRNAs, the amplification curves indeed either shifted to the lower right or displayed minimal amplification, suggesting nonspecific detection (Figure 2B). Thus, data points showing heterogeneous amplification curves can be easily identified and eliminated. To determine the cut-off factor that can balance between sensitivity and specificity, we generated a receiver operating characteristic (ROC) curve by plotting the true-positive rate (reflecting sensitivity) against the false-positive rate (representing specificity) at various threshold settings. As shown in the ROC curve, for each pair of miRNAs (authentic miRNA vs. its mutant), the linear regression of Ct and endpoint values revealed a sensitivity of 0.96 and a specificity of 0.99 (Figure 2C). Single point mutations at different positions of the miRNAs were also tested, and a specificity of ∼0.8 and a sensitivity of ∼0.9 were achieved after data training (Supplementary Figure S2).
Preamplification tends to cause bias and our mdqPCR method was designed to overcome this potential problem by using “nested primers” for the preamplification and the subsequent digital detection steps (Supplementary Table S2). To demonstrate this, we performed mdqPCR using either 10 pg of mouse sperm miRNAs for 10, 15, and 20 cycles of preamplification, or 10 ng mouse sperm miRNAs without preamplification. The R2 values between the two assays were 0.93 and 0.96, suggesting that the preamplification step indeed does not cause significant bias (Figure 2D). To explore whether two sncRNAs with similar sequences could interfere with each other's amplification during mdqPCR, we mixed miR-34b or miR-34c with its mutant form (Supplementary Table S1) at a molar ratio of 1:1 followed by mdqPCR. Neither R2 nor standard deviation values were affected by the mutant miRNAs bearing one or two point mutations (Figure 2E and F), implicating that the mutant miRNAs have little or no effects on the quantitation of the authentic miRNAs.
Microfluidics-based digital quantitative PCR-based single-sperm miRNA profiling
Given the high sensitivity and specificity, we further explored whether mdqPCR could be used for single-cell sncRNA quantitation. Single mouse sperm were captured using the Fluidigm C1 Single Cell Auto Preparation System. Sperm captured in the Fluidigm C1 open-source, programmable chips were further processed for lysis, RNA release, polyadenylation, reverse transcription, and preamplification using a customized program (Supplementary File S2). A set of miRNA-specific, shorter primers (Supplementary Table S4) and the “Trueseq anti short” (Supplementary Table S2) primer were used for preamplification for 25 cycles, whereas the primers used for subsequent mdqPCRs were 4 nucleotides longer at the 3΄ ends (Supplementary Table S5). Use of the “nested primers” (Figure 3A) serves as a stringent filter to prevent the background noises (i.e. templates containing mismatches due to potential amplification errors and/or mis-annealing of the preamplification primers) introduced during preamplification from getting amplified in the final digital detection step. The preamplified cDNAs were then partitioned in the 37K dqPCR chips followed by dqPCR. We examined levels of eight miRNAs in either six replicates containing 1/1000 of small RNA cDNAs from ∼1000 pooled mouse sperm cells each or 12 single-sperm cells. Interestingly, similar profiles were detected for all six replicates (Figure 3B), whereas each single sperm displayed a unique profile (Figure 3C).
Single-cell capture followed by high-throughput qPCR have been increasingly popular for the past several years, and many research institutions and laboratories do own the Fluidigm C1 Single Cell Auto Prep system and/or BioMark HD PCR system. Based on these microfluidics-based instruments, we developed this novel digital PCR protocol, which is comparable to, if not better than, the droplets-based dqPCR method in sensitivity and reproducibility . Therefore, the mdqPCR method can be a powerful alternative to the droplets-based dqPCR system for small RNA quantification. This method adds a new application to the Fluidigm BioMark HD system, and it can be coupled with either the C1 single-cell capture system or any other single-cell capture approach (e.g. FACS) for single-cell small RNA quantification analyses.
This method has several advantages over the droplets-based digital qPCR method  (Supplementary Table S6). First, stem-loop primers are not needed, and one can design their own primers for cDNA preparation for not only miRNAs but other small RNAs as well. The design of nested primers (primers for dqPCR are 4 nt longer than those for preamplification at the 3΄ ends) can minimize preamplification bias and enhance specificity, which is particularly critical for single-cell sncRNA quantitation. Second, the droplets-based digital qPCR collects only one data point at the end of amplification for each molecule, which generates ∼10 MB data. In contrast, our mdqPCR method gathers ∼100 data points/features with ∼1 GB data for each molecule. The large amount of data can be used for machine learning and “big data” analyses to enhance the assay accuracy in the future. Moreover, real time monitoring of the Ct and endpoint values for each molecule allows the investigators to determine and eliminate false-positive data points at the data-calling step, thus enhancing the reliability of the quantitative data. In both sampling and real experiments, we observed consistently <1% of the assays showing heterogeneous amplification curves indicative of aberrant amplification and detection. The low frequency of false-positive data points largely benefits from the nested primer design, which significantly enhances the specificity of the final digital amplification. Third, the sensitivity and reproducibility are high enough for small RNA profiling using single cells. Regular qPCR for low input sncRNA quantification has been problematic due to poor reproducibility. Digital qPCR has made accurate small RNA quantification possible. The automated single-cell capture followed by cDNA preparation procedures all occur within the concealed micro channels and chambers without manual handling. The preamplified sncRNA cDNAs in single-cell capture chips can then be directly loaded onto the dqPCR 37K chip (48.770 digital array IFC) for partition and dqPCR. These procedures are largely automated, thus minimizing technical errors often introduced during manual handling. Lastly, by modifying the protocol presented here, one can easily develop novel methods for single-cell sncRNA-Seq, which remains unavailable for now.
The detection sensitivity of a Fluidigm 37K dqPCR chip is 1.5 molecule/μl per reaction when the standard protocol is used, whereas it is 0.25 molecule/μl per reaction in the Bio-Rad droplet dqPCR method . The difference results from the fact that the Bio-Rad droplet qdPCR system usually generates ∼20,000 droplets per template, whereas 770 reaction chambers are used for single molecule partition in our method. However, by loading a sample to more chambers, the same sensitivity of 0.25 molecule/μl can easily be reached (Supplementary Table S6). It is noteworthy that the sensitivity of our standard protocol is sufficient for majority of sncRNA quantitation analyses.
The open source option of the Fluidigm C1 Single Cell Auto Prep system made the design of customized protocol possible. In the protocol presented here, a complete lysis of sperm chromatin was achieved, which is difficult because of the high degree nuclear condensation in sperm nuclei. By lysing the sperm chromatin/head, sperm-borne RNA contents can be released completely . Interestingly, each sperm appears to contain variable amounts of the eight miRNAs examined, whereas the profiles of the eight miRNAs are similar among six replicates of the pooled sperm samples. This finding demonstrated the power of single-sperm miRNA profiling because cellular heterogeneity could be masked when many cells are pooled for small RNA profiling, whereas unique miRNA signatures could only be revealed through single cell profiling. It is intriguing that each sperm has a unique miRNA profile because it suggests that sperm-borne small RNAs may contribute to the genetic and epigenetic diversity of the offspring given that these miRNAs have been shown to play a role during fertilization and early embryonic development, as well as intergenerational epigenetic inheritance [16–18].
miRNAs are important regulators of cell functions in reproductive system [19,20]. Sperm-borne miRNAs and endo-small interfering RNA (siRNAs) have been found important for fertilization and embryonic development . Other small RNA species, e.g. piRNAs, snoRNAs, and transfer RNA (tRNAs), have also been found in sperm , some of which appear to be of functional significance [22–24]. While other methods using the proprietary stem-loop primers, e.g. droplet-based dqPCR, could only detect a limited number of well-studied miRNAs, this microfluidics-based dqPCR method can be used for detecting all small RNA species at the single-cell level, thus opening doors to a wide array of single-cell studies in the future.
Supplementary data are available at BIOLRE online.
Supplementary Figure S1. Arrangement and usage of micro chambers in mdqPCR using the Fluidigm C1 open source programmable chips. Chamber C is for capturing a single cell. Chamber E1 is used for cell lysis and release of RNA contents. Chamber E2 is for polyadenylation reactions through which the polyA tail is added to small RNAs. The reverse transcription is performed in chamber E3. Both chambers E4 and E5 are used for PCR amplification.
Supplementary Figure S2. The ROC curve summarizing the linear model fitness for the training data for single point mutations at different positions. The specificity of ∼0.8 and sensitivity of ∼0.9 were achieved after data training.
Supplementary Table S1. Sequences of synthetic sncRNAs.
Supplementary Table S2. Sequences of the RT primer, the antisense/universal primers, and the TaqMan probe used in mdqPCR.
Supplementary Table S3. Sequences of the synthetic miRNA-specific primers used in mdqPCR assays.
Supplementary Table S4. Sequences of the miRNA-specific/sense primers used for preamplification.
Supplementary Table S5. Sequences of miRNA-specific/sense primers used in mdqPCR assays.
Supplementary Table S6. Comparison between mdqPCR-based and other existing low input small RNA detection and quantification methods.
Author Contributions: CT and WY conceived the research. TY, YZ, and RZ performed the experiments. All participated in data analyses. CT and WY wrote the manuscript. All reviewed the manuscript.