Assessing characteristics of RNA amplification methods for single cell RNA sequencing

We conducted a large-scale control experiment to assess the transfer function of three scRNA-seq methods and factors modulating the function. Our approach was to dilute bulk total RNA (from a single source) to levels bracketing single-cell levels of total RNA (10 pg and 100 pg) in replicates and amplifying the RNA to levels sufficient for RNA sequencing. Overall design: We performed replicate transcriptome amplifications of Universal Human Reference RNA (UHR) and Human Brain Reference RNA (HBR) that were diluted to single-cell and ten-cell abundances (10 and 100 picograms (pg.) total RNA or ~200,000 and 2 million mRNA molecules, respectively) and were amplified using three single-cell RNA amplification methods. Methods included the antisense RNA IVT protocol (aRNA), a custom C1 SMARTer protocol (SmartSeq Plus) performed on a Fluidigm C1 96-well chip, and a modified NuGen Ovation RNA sequencing protocol (NuGen). Bulk ribo-depleted UHR and HBR RNA were sequenced and served as a reference. The general experimental scheme was consistent for all dilution replicates; however, there were differences across experimental groups in the specifics of experimental protocols, necessitated by particular methodologies. Because of these experimental differences, head-to-head comparison of methods is not appropriate and our goal is to provide quantitative analyses of factors affecting individual methods. Current results should be used in experimental planning, data analysis, and method optimization rather than as a performance test of any particular method. Detailed experimental design: Each collaborating center obtained reference RNA with the same lot number for Universal Human Reference (UHR) RNA (Agilent 740000, Lot 0006141415) and Human Brain Reference (HBR) (Ambion AM6050, Lot-105P055201A) and performed replicate amplification using a single amplification method, detailed below. SmartSeq Plus: Reference RNA was diluted to an intermediate stock solution by serial dilution. A final 1000-fold dilution occurred on the C1 chip, such that individual wells in a given batch contained 9.99 pg. sampled from a common intermediate dilution. ERCC spike-in RNA mix 1 (Ambion 4456740) was also added for a final mass of approximately 7 femtograms (fg.) per sample, a 4,000,000x dilution from stock. Samples for each source RNA were prepared in single batches. After amplification, cDNA from the entire C1 96-well plate was quantified using picogreen. C1 chips with an average yield of less than 3 nanograms were discarded. The top 15 reactor wells by cDNA concentration were selected as representative 10 pg. samples for sequencing library preparation. Another 50 wells were selected by the same criteria. These were pooled in sets of 10, generating 5 100 pg. samples for each HBR and UHR. All samples for a given source were prepared in a single sequencing library preparation batch using Nextera XT C1 protocol. NuGen: HBR samples were prepared in a single batch using amplification protocol 1, generating 4 10 pg. and 4 100 pg. amplified replicates. UHR samples were prepared in two batches, using either amplification protocol 1 or 2, generating 15 10 pg. and 11 100 pg. samples. A single sequencing library preparation was performed for each batch of samples using either Lucigen NxSeq or NuGen Ovation Rapid protocol. aRNA: Amplification was performed as previously described (Morris J, Singh JM, Eberwine JH. Transcriptome analysis of single cells. J. Vis. Exp. [Internet]. 2011; Available from: http://www.jove.com/video/2634/transcriptome-analysis-of-single-cells). HBR samples were prepared in 4 batches from separate dilutions of reference RNA, generating 19 10 pg. and 3 100 pg. amplified replicates. ERCC spike-ins were added to 5 of the 10 pg. replicates before amplification at a dilution of 4,000,000x from stock. UHR samples were diluted and amplified in 2 batches from separate dilutions of reference RNA, generating 12 10 pg. and 7 100 pg. amplified replicates. A single sequencing library preparation was performed using Illumina TruSeq Stranded mRNA protocol modified to begin with amplified aRNA. A small numbers of reads were assigned to ERCC transcripts in replicates from the batch where ERCCs had been added that did not have spike-ins added (average of 0.5% of the number of reads assigned in spiked samples). 18 additional HBR 10 pg. replicates were amplified using aRNA for protocol optimization experiments. These samples were treated separately and were excluded from primary analysis. Bulk UHR and HBR: For each reference RNA, three sequencing libraries were generated from bulk material at the same laboratory as the SmartSeq Plus replicates. Cytoplasmic and mitochondrial ribosomal RNA (rRNA) were depleted using Ribo-Zero Gold as part of Illumina TruSeq Stranded Total RNA protocol. Samples were sequenced on Illumina HiSeq 2000. Because of differences in experimental design, direct comparison across methods of precision and the effect of input RNA abundance is difficult. For example, input RNA amount as a factor have different meanings for the different amplification methods: for SmartSeq Plus, because 100pg samples were constructed by pooling 10 pg. samples after cDNA amplification, any resulting effects involve library construction, while for aRNA and NuGen resulting effects reflect both cDNA amplification steps and library steps.

138

Dueck HR, Ai R, Camarena A, Ding B, Dominguez R, Evgrafov OV, Fan JB, Fisher SA, Herstein JS, Kim TK, Kim JM, Lin MY, Liu R, Mack WJ, McGroty S, Nguyen JD, Salathia N, Shallcross J, Souaiaia T, Spaethling JM, Walker CP, Wang J, Wang K, Wang W, Wildberg A, Zheng L, Chow RH, Eberwine J, Knowles JA, Zhang K, Kim J