Quartz-Seq: a highly reproducible and sensitive single-cell RNA sequencing method, reveals non-genetic gene-expression heterogeneity

Abstract

Development of a highly reproducible and sensitive single-cell RNA sequencing (RNA-seq) method would facilitate the understanding of the biological roles and underlying mechanisms of non-genetic cellular heterogeneity. In this study, we report a novel single-cell RNA-seq method called Quartz-Seq that has a simpler protocol and higher reproducibility and sensitivity than existing methods. We show that single-cell Quartz-Seq can quantitatively detect various kinds of non-genetic cellular heterogeneity, and can detect different cell types and different cell-cycle phases of a single cell type. Moreover, this method can comprehensively reveal gene-expression heterogeneity between single cells of the same cell type in the same cell-cycle phase.

Figure 1

Schematic of the single-cell Quartz-Seq and Quartz-Chip methods. All of the steps of the whole-transcript amplification were executed in a single PCR tube. The first-strand cDNA was synthesized using the reverse transcription (RT) primer, which contains oligo-dT24, the T7 promoter (T7) and the PCR target region (M) sequences. After the first-strand synthesis, the majority of the RT primer was digested by exonuclease I, although it was not possible to eliminate the RT primer completely using this procedure. A poly-A tail was then added to the 3' ends of the first-strand cDNA and to any surviving RT primer. After the second-strand synthesis with the tagging primer, the resulting cDNA and the byproducts from the surviving primers contained the whole-transcript amplification (WTA) adaptor sequences, which include the RT primer sequence and the tagging primer sequence. These DNAs were used for the suppression PCR, which used the suppression PCR primer. Enrichment of the short DNA fragments, such as the byproducts, was suppressed. After the enrichment, the high-quality cDNA, which did not contain any byproducts, was obtained. The amplified cDNAs then had the T7 promoter sequence at the 3' ends of the DNA. These cDNAs were used for the Illumina sequencing and microarray experiments.

Figure 2

Reproducibility and sensitivity of single-cell Quartz-Seq. (a) Representative scatter plot of the gene-expression data from two replicate single-cell Quartz-Seq analyses of 10 pg of total embryonic stem (ES)-cell RNA (left panel). The blue line indicates a two-fold change, and the red line is a linear regression. Scatter plot of single-cell Quartz-Seq (whole-transcript amplification; WTA) and conventional RNA-seq (non-WTA) data using 1 µg of total ES-cell RNA (right panel). (b) Ratio of detected genes for three replicate Quartz-Seq analyses. Using two different independent Quartz-Seq experiments, 82.1% of the genes were detected by (left panel). The right panel shows the ratio of the genes detected by Quartz-Seq and conventional RNA-seq; more than 68.2% of the genes were detected by single-cell Quartz-Seq.

Figure 3

Comparison of the performances of Quartz-Seq, Quartz-Chip and other methods. (a) Box plot of Pearson correlation coefficients (PCCs) for the technical replication of Quartz-Seq and Smart-Seq with 10 pg diluted total RNA. We reanalyzed the following four original Smart-Seq datasets: mouse brain: MB; human brain (Nextera library preparation kit: HB (Nx); Universal human reference RNA: UHRR and UHRR with the Nextera library preparation kit: UHRR (Nx). The asterisk indicates the downsampling sequence reads from single-cell Quartz-Seq (paired-end (PE), 60 million reads, n = 3). (b) Box plot of PCCs between conventional RNA-seq and single-cell RNA-seq methods. (c) Comparison of our current and previous methods using 10 pg of total ES-cell RNA with GeneChip. Performance of (left) Quartz-Chip, and (right) the Kurimoto et al. method. The Kurimoto et al. data were reanalyzed using the original sets. The bar plots in the right panels show the numbers of genes that were detected with each method: the gray bars show the total number of genes, and the blue bars indicate the number of detected genes.

Figure 4

Single-cell Quartz-Seq detects different cell types. Results of single-cell Quartz-Seq with 12 embryonic stem (ES) cells and 12 primitive endoderm (PrE) cells. (a) Clustering of all samples. (b) Heat map of the marker genes for ES and PrE cells and the housekeeping genes. The bar plot in the right panel shows the mutual information (MI); a high degree of MI indicates high differential expression between two cell states. (c) Verification of the expression pattern between single cells using amplification-free single-cell quantitative (q)PCR. The gene-expression data for single cells correspond to the ES (n = 96 single ES cells in the G1 phase of the cell cycle), PrE (n = 96 single PrE cells in the G1 phase), ES200 (n = 40 single-cell-sized samples from pooled lysis of ES cells in the G1 phase), and PrE200 (n = 40 single-cell-sized samples from pooled lysis of PrE cells in the G1 phase) groups. The ES and PrE box plots represent the gene-expression variability, which includes the biological variability and the experimental error, while the ES200 and PrE200 box plots represent the gene-expression variability due to experimental error.

Figure 5

Single-cell Quartz-Seq reveals fluctuations in global gene expression in a single cell type in the same cell-cycle phase. (a) The × and Y axes represent the standard deviation (SD) of gene expression from the different datasets from single-cell Quartz-Seq with single embryonic stem (ES) cells that were all in the G1 phase of the cell cycle (n = 12, n = 8). (b) We detected expression of nine genes (Fn1, Zfp42, Sgk1, Tfrc, Utf1, Lefty1, Nanog, Sox2, and Spp1) by amplification-free single-cell quantitative (q)PCR with single ES cells in the G1 phase.The nine genes were selected from the ES cell differentiated genes (ES: fragment per kilobase of transcript per million fragments sequenced (FPKM) > 2; PrE: FPKM < 1). The plot shows the gene expression for the single-cell analysis of ES (n = 96 single ES cells in the G1 phase of the cell cycle) and ESA samples (n = 48 'averaged' single-cell samples from 300 pooled ES cells in the G1 phase). The ES single-cell box plots represent the gene-expression variability that contains the biological variability and the experimental error, while the ESA sample box plots represent the gene-expression variability due to experimental error. (c) The × axis represents the coefficient of variation (CV) of the gene expression from single-cell Quartz-Seq data with single ES cells in the G1 phase of the cell cycle (n = 12). The Y axis represents the CV of the gene expression from amplification-free single-cell qPCR with ES cells in the G1 phase of the cell cycle (n = 96). The CVs of the gene expression are plotted for the nine genes. The red lines represent regressions.