ty control of transcriptomic data RNA-seq analysis showed that the transcriptomes of lumbar 5 DRG from five adult male mice were homogenous. Immunostaining of CGRP and NF200, combined with IB4 fluorescent labeling showed that ~35% of neurons were order JW 55 positive for IB4, ~41.2% were positive for CGRP and ~46.4% were positive for NF200. Approximately 20% of IB4-positive neurons expressed CGRP, whereas ~5.6% expressed NF200. Approximately 28% of CGRP-positive neurons expressed NF200. A small number of neurons were triple labeled. A DRG-specific approach was designed to increase the efficiency of single-cell RNA-seq used for neuron typing. First, IB4-positive neurons were identified among neurons freshly dissociated from lumbar DRG by IB4 fluorescence labeling. Under a microscope, neurons without identifiable satellite cells attached to the cell surface were selected and aspirated with glass pipettes. IB4-positive small neurons, IB4-negative small neurons and large neurons contributed ~1/3 of the selected neurons. The total number of neurons used was determined by the results of transcriptomic analysis. We profiled neurons until the number of neuron clusters extracted from the data plateaued and the correlation among neurons stabilized. Sequencing libraries were constructed from the cDNA of 203 neurons collected from 19 mice. To ensure quality of the samples, FPKM values for house-keeping genes actin B or glyceraldehyde-3-phosphate dehydrogenase in each neuron were required to be around 4 675 3 456 or 367 268, respectively. Datasets for the corresponding 197 neurons, including 64 IB4-positive neurons, 69 IB4-negative small neurons and 64 large 
neurons, were processed for transcriptome analysis. To obtain single-neuron transcriptomes, we examined the correlation between sequencing depth and the number of genes. We found that 1, 5, 10, 20, 30 and 40 million mapped reads per neuron detected 8 906 79, 9 781 127, 10 158 133, 11 142 223, 11 450 256 and 11 Chang-Lin Li et al. npg 85 npg Types of primary sensory neurons 86 694 288 genes, respectively. The number of mapped reads for each detected gene correlate linearly with the total number of mapped reads. Thus, the probability of detecting low-abundance genes depends on the depth of RNA-seq. Therefore, 30 million reads was considered the minimum number of mapped reads to achieve maximal mapping while maintaining efficiency. The average number of mapped reads was 58.2 million for a single neuron. The number of detectable genes for each neuron ranged from 7 972 to 13 960 per neuron and 20 794 in total. To evaluate the effect of sequencing depth on gene number, six neurons were resequenced to a depth of 181 million reads three times more than the average sequencing depth for all neurons. The libraries obtained from deeper sequencing runs shared 99.8% similarity with the libraries obtained by lower depth sequencing of the same neuron. Thus, a sequencing depth of 30 million FPKM was sufficient. Finally, to examine the quality of RNA-seq, cDNA from one neuron was divided into two equal parts and processed for RNA-seq. The transcriptomic datasets of two libraries derived from the same neuron shared 99.4% similarity, suggesting a high quality of RNAseq. Gene modules identified by weighted gene co-expression network analysis We performed PubMed ID:http://www.ncbi.nlm.nih.gov/pubmed/19822652 principal component analysis on all 197 single-neuron transcriptomes as previously reported. Genes with the highest loading in the first three principal components were analyze