Analytical Biochemistry 427 (2012) 178–186

Contents lists available at SciVerse ScienceDirect

Analytical Biochemistry

journal homepage: www.elsevier.com/locate/yabio

Comparison of reverse transcription–quantitative polymerase chain reaction methods and platforms for single cell gene expression analysis ⇑ Bridget C. Fox, Alison S. Devonshire , Marc-Olivier Baradez, Damian Marshall, Carole A. Foy

LGC Limited, Teddington, Middlesex TW11 0LY, UK article info abstract

Article history: Single cell gene expression analysis can provide insights into development and disease progression by Received 6 February 2012 profiling individual cellular responses as opposed to reporting the global average of a population. Reverse Received in revised form 10 May 2012 transcription–quantitative polymerase chain reaction (RT–qPCR) is the ‘‘gold standard’’ for the quantifi- Accepted 13 May 2012 cation of gene expression levels; however, the technical performance of kits and platforms aimed at sin- Available online 19 May 2012 gle cell analysis has not been fully defined in terms of sensitivity and assay comparability. We compared three kits using purification columns (PicoPure) or direct lysis (CellsDirect and Cells-to-CT) combined Keywords: with a one- or two-step RT–qPCR approach using dilutions of cells and RNA standards to the single cell Single cell level. Single cell-level messenger RNA (mRNA) analysis was possible using all three methods, although RT–qPCR mRNA the precision, linearity, and effect of lysis buffer and cell background differed depending on the approach RNA standard used. The impact of using a microfluidic qPCR platform versus a standard instrument was investigated for potential variability introduced by preamplification of template or scaling down of the qPCR to nanoliter volumes using laser-dissected single cell samples. The two approaches were found to be comparable. These studies show that accurate gene expression analysis is achievable at the single cell level and high- light the importance of well-validated experimental procedures for low-level mRNA analysis. Ó 2012 Elsevier Inc. All rights reserved.

For the majority of gene expression studies, a population mean most commonly used approach in research laboratories [10]. The is taken as a measure of specific gene expression levels. Although MIQE (minimum information for publication of quantitative real- this method is suitable for many studies, it does not yield informa- time PCR experiments) guidelines for reporting of RT–qPCR experi- tion on the differences in expression between cells within a heter- ments [11] call for thorough characterization of RT–qPCR assays ogeneous population or account for transcriptional responses and upstream sample processing steps such as cell isolation and lysis occurring in ‘‘transcriptional bursts’’ [1,2]. Hence, single cell gene that may affect RT–qPCR performance. Validation of the quantitative expression analysis may yield greater insights into cell biology, accuracy of single cell RT–qPCR is important due to the low levels of and this approach has many applications in the fields of diagnos- messenger RNA (mRNA) levels in each sample [12], whereas techni- tics [3], embryology [4], stem cell biology, and tissue engineering cal validation is less straightforward due to the heterogeneous nat- [5]. ure of the samples and limited scope for technical replicates [13]. Global analysis of single cell transcriptomes is possible using There are currently a variety of techniques available to isolate DNA microarray [6] and, more recently, RNA–seq technologies single cells, including micromanipulation, laser dissection micros- [7,8]. However, reverse transcription–quantitative polymerase copy, and FACS (fluorescence-activated cell sorting) [14,15]. chain reaction (RT–qPCR)1 is the ‘‘gold standard’’ for quantitative Whichever single cell isolation method is employed, it is vital to measurement of transcript abundance in single cells [9] and the employ lysis conditions that ensure complete cell lysis [16]. There are several lysis methods in use—some that enable the use of the

⇑ Corresponding author. Fax: +44 0208 9432767. lysate directly in RT–qPCR [10,13] and others that involve cell lysis E-mail address: [email protected] (A.S. Devonshire). and subsequent purification of the total RNA prior to RT–qPCR [17]. 1 Abbreviations used: RT–qPCR, reverse transcription–quantitative polymerase For all of these procedures, there is a need to understand any bias chain reaction; mRNA, messenger RNA; cDNA, complementary DNA; NIST, National that could be introduced due to incomplete lysis, differing elution Institute of Standards; CE, cell equivalent; HFF-1, human foreskin fibroblast-1; yields from purification steps, and potential inhibition of the RT– DMEM, Dulbecco’s modified Eagle’s medium; PBS, phosphate-buffered saline; Cq, quantification cycle; IVT, in vitro transcribed; COL6A1, collagen-type VI alpha 1; qPCR by the presence of lysis components. M6PRBP1, mannose-6-phosphate receptor binding protein 1; TNFR12A, tumor In addition to stringent cell isolation and lysis conditions at low necrosis factor receptor superfamily member 12A; CSF3, colony-stimulating factor mRNA levels, special attention must be given to factors that can 3 (granulocyte); ANOVA, analysis of variance; SD, standard deviation; MMLV, lead to unreliable RT–qPCR data. Several studies have shown that Moloney murine leukemia virus; tRNA, transfer RNA.

0003-2697/$ - see front matter Ó 2012 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.ab.2012.05.010 RT–qPCR methods for single cell analysis / B.C. Fox et al. / Anal. Biochem. 427 (2012) 178–186 179 at low complementary DNA (cDNA) levels, inhibition of real-time qPCR assays PCR by can be significant [18,19]. In addition, obtaining reliable data from copy numbers below 102 per qPCR has Predesigned TaqMan qPCR assays were supplied by Applied Bio- been shown to be problematic and the technical error associated systems (Foster City, CA, USA) as a 20 premix containing both with both the RT and the PCR steps can become more significant primers and FAM–nonfluorescent quencher (NFQ) probe (see Sup- as the template level decreases [13]. plementary Table 1 in supplementary material). The efficiencies of Alongside reagent kits applicable to single cell gene expression these assays were characterized with a dilution series of HFF-1 analysis, microfluidic PCR platforms such the BioMark (Fluidigm) cDNA (see Supplementary Methods and Supplementary Table 2 [20] and OpenArray (Life Technologies) [21] facilitate thousands in supplementary material). Primers and hydrolysis probe (FAM– of assays to be performed in parallel. This technology has been TAMRA) to ERCC-84 (GenBank accession: DQ883682) were used applied to single cell analysis using both digital PCR (dPCR) [22] as described previously [26] and supplied by Sigma–Aldrich (Poole, and qPCR approaches. The latter enables the analysis of tens UK). The final concentrations of each primer and probe for all as- to hundreds of gene targets in single cells to be performed says were 900 and 250 nM, respectively. RT–qPCR and qPCR were [5,23,24]. Preamplification of template is required for these performed in MicroAmp Optical 96-Well Reaction Plates (Applied high-throughput approaches at the single cell level, although Biosystems). PCR-based preamplification has been suggested to introduce amplification bias [25]. Although the BioMark platform has been Comparison of Cellsdirect, Cells-to-CT, and PicoPure RT–qPCR kits characterized for larger inputs of template RNA [26], the effect using a cell dilution series of preamplification and the scaling down of reaction volumes to the nanoliter scale has not been investigated for the low-level For cell dilutions, the number of HFF-1 cells was determined by mRNA quantities found in a single cell. Recent advances in micro- averaging five counts using a hemocytometer, and the cell suspen- fluidic approaches have also applied miniaturization not only to sion was then diluted in phosphate-buffered saline (PBS) by serial qPCR but also to the cell capture, lysis, and RT steps of single cell dilution to 103,102, 10, and 1 cell(s)/ll(n = 3). Lysates were pre- analysis. Microfluidic chip- and droplet-based technologies enable pared and RT–qPCR was performed using the CellsDirect (Invitro- effective concentration of the mRNA transcripts found in a single gen, Paisley, UK), Cells-to-CT (Applied Biosystems), and PicoPure cell [27–29]. (Arcturus/Life Technologies)/Message Sensor (Ambion) kits accord- In this study, we compared three commercially available RT– ing to the manufacturers’ instructions as described below. For all qPCR kits designed for low-level mRNA analysis (CellsDirect, three kits, 20% of the original lysate was used per RT–qPCR (Cells- Cells-to-CT, and a combination of the PicoPure purification col- Direct and Message Sensor) or per qPCR (Cells-to-CT) (n = 1 RT– umns and Message Sensor RT–qPCR kit) for their suitability for sin- qPCR or qPCR per gene assay). Real-time PCR was performed on gle cell gene expression analysis. The CellsDirect and Cells-to-CT the ABI PRISM 7900HT Sequence Detection System (Applied Bio- kits are designed for direct input of the cell lysate in the RT–qPCR, systems). A DNA standard curve composed of a 10-fold serial dilu- whereas the PicoPure/Message Sensor kit combination involves tion of GAPDH or B2M DNA standards (see ‘‘Preparation of DNA purification of total RNA prior to the RT–qPCR. Dilutions of fibro- standards’’ section below) from 106 to 10 copies (n = 3) was per- cells to the single cell level were used to test all technical fac- formed on each qPCR plate to quantify the number of cDNA copies tors contributing to the qPCR measurement, namely efficiency of produced by each RT reaction. In this and other experiments, raw lysis/extraction, RT, and qPCR steps. In addition, an RNA standard data from the 7900HT system were processed using SDS (Sequence (ERCC-84) from the set of controls developed by the External Detection System) software (version 2.3) with automatic baseline

RNA Controls Consortium (ERCC), an ad hoc group of 70 members setting and manual threshold setting. Cq (quantification cycle) val- from private, public, and academic organizations led by the Na- ues were analyzed further in Microsoft Excel 2003. tional Institute of Standards (NIST) [30], was employed to measure the sensitivity of RT–qPCR and the effect of components such as ly- CellsDirect sis buffer on the reactions. CellsDirect lysis buffer was prepared by mixing 10 parts resus- One of the kits (CellsDirect) was chosen to compare differ- pension buffer with 1 part lysis enhancer. Then 1 ll of cell suspen- ences between performing analysis using a standard qPCR sion was added to 23 ll of lysis buffer and incubated on ice for platform and the BioMark instrument based on its sensitivity 10 min, followed by cell lysis by incubation at 75 °C for 10 min. Ly- for low-level mRNA quantification and suitability for high- sates were frozen at –80 °C. Then 1 ll of 50 mM magnesium sulfate throughput approaches. Adaptation of the method to include a solution (provided with the kit) was added to the 24 ll of cell ly- preamplification step prior to qPCR, which is required for single sate prior to RT–qPCR. One-step RT–qPCR was performed with as- cell analysis on the BioMark platform, was compared with one- says specific to B2M or GAPDH using 5 ll of cell lysate in 20-ll step RT–qPCR using a dilution series of ERCC-84 RNA standard reaction volumes. Thermal cycling conditions were as follows: RT in a background corresponding to a single cell equivalent (CE). (42 °C for 15 min), followed by reverse transcriptase inactivation Laser-dissected single cells were collected to assess the impact (95 °C for 2 min), followed by 40 cycles of PCR (95 °C for 15 s, of microliter versus nanoliter qPCR volume on the analysis of 60 °C for 60 s). an endogenous transcript. Cells-to-CT Cell lysates were prepared by the addition of 1 ll of cell suspen- Materials and methods sion to 15.4 ll of lysis solution and mixed by pipetting, followed by a 5-min incubation step at room temperature. Stop solution Cell culture (1.6 ll) was added to stop the lysis reaction, followed by a 2-min incubation at room temperature prior to freezing at –80 °C. Two- Human foreskin fibroblast-1 (HFF-1) cells were maintained in step RT–qPCR was performed with separate RT (n = 1) and qPCR Dulbecco’s modified Eagle’s medium (DMEM, ATCC cell line (n = 1) steps. RT reactions were performed with 9 ll of cell lysate SCRC-1041, LGC Standards, UK) containing 10% fetal calf serum in 20 ll reaction volumes incubated at 37 °C for 60 minutes fol- (FCS, PAA Laboratories, Austria) incubated at 37 °C in a humidified lowed by heat inactivation for 5 minutes at 95 °C. qPCR was per- atmosphere of 5% CO2. formed with 8 ll of RT reaction and TaqMan assays to B2M and 180 RT–qPCR methods for single cell analysis / B.C. Fox et al. / Anal. Biochem. 427 (2012) 178–186

GAPDH in 20-ll reaction volumes with thermal cycling conditions were performed: The first PCR round was performed using 1 llof of 50 °C for 2 min, 95 °C for 10 min, followed by 40 cycles of 95 °C cDNA, and the second round was performed using 50 ng of purified for 15 s and 60 °C for 60 s. (Note: The proportions of lysate in the RT PCR product from the first round of PCR. For ERCC-84, one round of reaction and cDNA in the qPCR are within the manufacturer’s rec- PCR was performed with 1 ll of cDNA. The PCR conditions were as ommendations of a maximum of 45% for both stages.) follows: 95 °C for 5 min, followed by 35 cycles of 95 °C for 30 s, 60 °C for 30 s, and 72 °C for 2 min, followed by 72 °C for 10 min. PicoPure After each PCR, products were purified using the QIAquick PCR Cell suspension (1 ll) mixed in 100 ll of extraction buffer was purification kit (Qiagen). Quantity was determined using the Nano- incubated for 30 min at 42 °C, followed by centrifugation (3000g, Drop 1000 spectrophotometer, and expected product length was 2 min) to separate cell debris, and the supernatant was then frozen confirmed using the 2100 Bioanalyzer. at –80 °C. RNA purification was performed by mixing the sample 1:1 with 70% ethanol, followed by wash steps using wash buffer Investigating RT–qPCR precision and potential inhibitors 1 (100 ll) and wash buffer 2 (2 100 ll) and elution of the RNA with 30 ll of elution buffer. To assess RT–qPCR precision for different levels of transcript abundance, serial dilutions of ERCC-84 RNA standard to 104,103, Message Sensor and 102 copies were prepared in lysis buffer specific to the CellsDi- One-step RT–qPCR was performed using 6 ll of RNA eluted rect or Cells-to-CT kit, ensuring that the same proportion of lysis from the PicoPure columns in a 20-ll reaction volume containing buffer was present in each RT–qPCR and the same as that used pre- assays to B2M or GAPDH, Platinum Taq DNA polymerase, and viously for the cell dilutions. In parallel with the aim of investigat- ROX (Invitrogen) in addition to the components of the Message ing the effect of lysis buffer on RT–qPCR, RT–qPCRs were also Sensor RT kit (10 RT buffer, dNTP mix, RNasin, and RT). Thermal performed with the ERCC-84 serial dilution in the absence of lysis cycling conditions were as follows: RT (50 °C for 15 min), followed buffer. For testing of qPCR inhibition by reverse transcriptase, by reverse transcriptase inactivation (95 °C for 5 min), followed by ERCC-84 DNA standards were diluted to 104,103, and 102 copies 40 cycles of PCR (95 °C for 15 s, 60 °C for 60 s). in lysis buffer and RT–qPCR was performed in the presence and ab- sence of the reverse transcriptase enzyme (this was substituted Preparation of IVT RNA standards with nuclease-free water in the latter case). To investigate the im- pact of cell background, 102 copies of ERCC-84 RNA standard were In vitro transcribed (IVT) ERCC-84 RNA was produced from spiked into lysis buffer containing variable concentrations of cell ERCC-84 plasmid DNA (courtesy of Marc Salit, NIST, USA). ERCC- lysate equivalent to 200, 20, 2, 0.2, and 0 cells per qPCR. The cell 84 plasmid DNA was cleaved into a single linear strand using lysate was prepared by dilution of the 104 cell sample lysate, with BamHI restriction endonuclease enzyme (, 1 ll of the diluted cell lysate being used in place of 1 ll of lysis buf- Hitchin, UK). RNA was in vitro transcribed from the linearized fer. Dilutions prepared in the CellsDirect kit lysis buffer were plasmid using a MEGAscript T7 kit (Ambion) followed by DNase heated at 75 °C for 10 min before use in RT–qPCR, in line with pre- treatment and cleanup using RNeasy columns (Qiagen, Hilden, paring cell lysates with this kit. For the one-step CellsDirect kit, Germany). RNA concentration and length were measured using three replicate RT–qPCR assays were performed for each RNA sam- the Nanodrop 1000 spectrophotometer (ThermoScientific) and ple, whereas for the two-step Cells-to-CT kit, two 20-ll RT reac- 2100 Bioanalyzer (Agilent) systems, respectively. tions were performed and pooled, followed by three replicate qPCR assays. Preparation of DNA standards Comparison of one-step RT–qPCR with preamplification of template GAPDH, B2M, and ERCC-84 DNA standards were prepared to quantify cDNA copies and estimate RT efficiencies of the RT–qPCR Dilutions of ERCC-84 RNA standards to 104,103,102, and 10 kits tested. DNA standards were produced by two-step RT–PCR copies were prepared in CellsDirect lysis buffer containing pooled from 200 ng of HFF-1 total RNA (GAPDH and B2M) or 200 ng of HFF-1 RNA equivalent to that of a single cell per 8 ll. Six replicate IVT ERCC-84 RNA. Primer pairs were designed using NCBI pri- one-step RT–qPCR or RT–preamplification reactions (one-step RT– mer–BLAST to produce of similar length to the B2M PCR with 18 cycles of PCR) were performed for each dilution. One- and GAPDH RNA transcripts and were obtained from Sigma–Al- step RT–qPCR was performed according to the CellsDirect protocol. drich: GAPDH forward, 50-AGCCTCCCGCTTCGCTCTCT-30, and re- RT–preamplification was also performed using the CellsDirect kit verse, 50-CAATGCCAGCCCCAGCGTCA-30 (PCR product size: according to a protocol adapted from the BioMark Advanced Devel- 993 bp); B2M forward, 50-TAAGTGGAGGCGTCGCGCTG-30, and re- opment Protocol 5, with 20-ll sample RT–preamplification master verse, 50-AACCAGATTAACCACAACCATGCCTT-30 (PCR product size: mixes being prepared using 8 ll of lysate, 2 RT–PCR buffer, 0.2 ll 930 bp). The ERCC-84 primers used were as follows: forward, 50- of SuperScript III RT/Taq polymerase enzyme mix, and 20 ERCC- GCG AAT TGT CTG GGG CCT CGT T-30, and reverse, 50-CCC CAC 84 gene-specific assay (final concentration 0.05). The volume CCT GCA ACT GAG AAG-30 (PCR product size: 970 bp). was made to 20 ll using nuclease-free water. RT–preamplification was performed on the GeneAmp PCR system 9700 (Applied Biosys- cDNA synthesis for preparation of DNA standards tems) under the following conditions: 50 °C for 15 min, followed First-strand synthesis was performed using the TaqMan Reverse by 95 °C for 2 min, followed by 18 cycles of 95 °C for 15 s and Transcription Reagents kit (Applied Biosystems) using the reverse 60 °C for 4 min. Reactions were diluted 1:5 in Tris–EDTA (ethylene- primer (200 nM final concentration) for each gene in 20-ll reac- diaminetetraacetic acid) (pH 8.0, Fluka, 100 ll final volume) and tions. RT incubation conditions were as follows: 48 °C for 30 min, stored at –20 °C. Preamplified cDNA samples were analyzed by followed by 95 °C for 5 min. qPCR using the 7900HT real-time PCR cycler with incubations of 50 °C for 2 min and 95 °C for 10 min, followed by 45 cycles of PCR for preparation of DNA standards 95 °C for 15 s and 60 °C for 60 s, using TaqMan Universal PCR Mas- PCR was performed using the HotStar Taq Plus kit (Qiagen) in ter Mix (Applied Biosystems) and 2.25 ll of preamplified sample in 50-ll reaction volumes containing 150 nM of the respective primer a 20-ll final volume. Triplicate qPCRs were performed for each pairs. For the GAPDH and B2M DNA standards, two rounds of PCR preamplified sample. RT–qPCR methods for single cell analysis / B.C. Fox et al. / Anal. Biochem. 427 (2012) 178–186 181

Laser capture microdissection of HFF-1 cells A CellsDirect 6 HFF-1 cells were revived from cryopreservation in liquid nitro- y = 1.23x + 1.69 gen by rapid thawing at 37 °C for 2 min and seeded in one T-175 R² = 0.972 flask (Corning). After 24 h, the cells were washed once in PBS and 5 enzymatically detached with trypsin (Sigma), counted, and seeded onto laser microdissection membrane slides (Molecular Machines 4 and Industries, Glattbrugg, Switzerland) at a density of 30,000 cells/mm2. Cells were allowed to adhere overnight in cul- 3 ture medium (DMEM + 10% fetal bovine serum [FBS]) at 37 °Cin a saturated humidified atmosphere with 5% CO2. After 24 h, the y = 1.25x + 1.42 (cDNA copies) 2 cells were incubated with 2 mM calcein–AM (Invitrogen) for 10 R² = 0.977 30 min, washed once in PBS, briefly fixed in absolute ethanol at log room temperature for a few seconds, and air-dried. Dry slides were 1 stored at 4 °C prior to laser microdissection. Then 100 HFF-1 cells were captured in 20 ll of CellsDirect lysis buffer in the lid of a 0 0.2-ml nuclease-free PCR tube using a Zeiss Axio Observer laser 0123 capture microscope using calcein fluorescence (488 nM excitation log (number of cells) wavelength) to identify single cells. The cells were lysed by heating 10 at 75 °C for 15 min. B Cells-to-CT Comparison of BioMark and 7900HT qPCR systems 6 y = 1.15x + 1.66 A serial dilution of a pool of 100 laser-captured HFF-1 single 5 R² = 0.994 cells was used to evaluate the BioMark real-time PCR system with the 7900HT. The lysate was diluted 1:1 with CellsDirect lysis buffer 4 before performing a 5-fold serial dilution to 22.5, 4.5, and 0.9 CEs per RT–preamplification reaction. Duplicate RT–preamplification 3 reactions were performed in a total volume of 22 ll containing a mix of 11 TaqMan assays (0.05) (see ERCC-84 and assays listed (cDNA copies) 2 y = 1.22x + 0.78 in Supplementary Table 1) and 0.2 ll of SuperScript III/Platinum 10 R² = 0.999 Taq enzyme mix with cycling conditions as in the ‘‘Comparison log of one-step RT–qPCR with preamplification of template’’ section 1 above. Preamplified samples were analyzed using the 7900HT and BioMark qPCR platforms. qPCR measurement of GAPDH was 0 performed on the 7900HT as described in that section with dupli- 0123 cate qPCRs per preamplified sample. log (number of Cells) The same samples were analyzed for the expression of GAPDH, 10 B2M, collagen-type VI alpha 1 (COL6A1), mannose-6-phosphate receptor binding protein 1 (M6PRBP1), tumor necrosis factor C Message Sensor receptor superfamily member 12A (TNFR12A), and colony-stimu- 6 lating factor 3 (granulocyte) (CSF3) on the BioMark platform. As- y = 1.02x + 2.22 says were prepared for loading onto BioMark 48.48 Dynamic 5

Arrays by diluting each 20 assay 1:1 with Assay Loading Reagent ) R² = 0.990 (Fluidigm). For sample loading, a premix consisting of 10 parts 4 TaqMan Universal Mastermix (Applied Biosystems):1 part Sample Loading Reagent (Fluidigm) was prepared, and 2.75 ll of this pre- mix was mixed vigorously with 2.25 ll of preamplified cDNA by 3 y = 1.14x + 2.08 vortexing and centrifuging to pool contents. Then 5 ll of each as-

(cDNA copies R² = 0.987 say or each sample mix was loaded per assay or sample inlet well, 2 10 respectively. Nine replicate qPCRs were performed for each pream- log plified sample consisting of three replicate assay inlets and three 1 replicate sample inlets. Thermal cycling conditions were identical to those described for the 7900HT. Cq data were generated using BioMark Real-Time PCR Analysis software (version 2.1.1) with glo- 0 0123 bal automatic threshold and linear baseline correction settings and a default quality threshold of 0.65. Data were exported for process- log10(number of Cells) ing using Microsoft Excel 2003. Fig.1. Comparison of RT–qPCR kits using an HFF-1 dilution series. RT–qPCR was performed on a 10-fold dilution series of HFF-1 cells from 103 cells to 1 cell using Statistical analysis three RT–qPCR kits and assays for GAPDH (filled diamonds) and B2M (open triangles): (A) CellsDirect; (B) Cells-to-CT; (C) PicoPure/Message Sensor. Mean cDNA copy numbers from three replicate cell samples are shown, with the Linear regression analysis of log (mean cDNA copy number) 10 exception of B2M single cell data point (n = 2) (B). Slopes and R2 values from linear versus log (cell number) (see Fig. 1 in Results) and of C versus 10 q regression of log10(cDNA copies) against log10(number of cells) are displayed for log2(RNA/DNA copy number) (see Figs. 2 and 4 in Results) were GAPDH (dashed line) and B2M (solid line). 182 RT–qPCR methods for single cell analysis / B.C. Fox et al. / Anal. Biochem. 427 (2012) 178–186

A 40 B 40

38 38 y = -1.0434x + 44.023 36 y = -0.831x + 41.92 36 R² = 0.946 R² = 0.951 34 34 q q 32 32 C C

30 30 y = -1.279x + 44.55 R² = 0.989 28 y = -0.9042x + 39.456 28 R² = 0.9972 26 26

24 24 6 8 10 12 14 6 8 10 12 14

log2(copy number) log2(copy number)

CD40 40

38 38 y = -1.0509x + 44.394 R² = 0.9791 36 36

34 34 q q 32 y = -1.1057x + 39.971 32 C C R² = 0.9802 30 30 y = -1.0044x + 41 28 28 R² = 0.9858

26 26

24 24 6 8 10 12 14 6 8 10 12 14

log2(copy number) log2(copy number)

E 40 F 40

38 38

36 36 q q C C 34 34 *

32 32

30 30 0 0.2 2 20 200 00.22 20200 CE CE

Fig.2. Investigating the effects of lysis buffer, reverse transcriptase, and cell background. RT–qPCRs were performed with the CellsDirect (A, C, E) or Cells-to-CT (B, D, F) kit. (A, B) Measurement of 104,103, and 102 copies of ERCC-84 RNA standard was compared in the presence (filled symbols, line) or absence (open symbols, dashed line) of lysis buffer. (C, D) qPCR efficiency of both kits was investigated by measurement of 104,103, and 102 copies of ERCC-84 DNA standard (filled symbols, solid line) (both kits) and in 2 the absence of reverse transcriptase (open symbols, dashed line) (Cells-to-CT only) (D). Data points show individual Cq values. Slopes and R values from linear regression of 2 log2(template copy number) versus Cq values are displayed. (E, F) ERCC-84 RNA standard (10 copies) was measured with varying concentrations of CEs. Average Cq values ± standard deviations (n = 3) and results of one-way ANOVA (⁄P < 0.005) are displayed. performed using Microsoft Excel 2003, with statistical analysis of Cq versus log2(DNA copy number) using the equation PCR effi- performed using GraphPad Prism (version 5a). Efficiency of PCR– ciency (%) = 100 (2(–1/S) 1) (Fig. 2C and D). The theoretical dou- amplification of DNA template was calculated from the slope (S) bling of DNA template with each PCR cycle corresponds to a PCR RT–qPCR methods for single cell analysis / B.C. Fox et al. / Anal. Biochem. 427 (2012) 178–186 183

40 A 24 y = -1.0244x + 41.386 R² = 0.9811 35 22 y = -1.0604x + 22.574 30 R² = 0.9931

q 20 C q 25 C

20 18

y = -1.031x + 28.523 15 R² = 0.9781 16 -1 0 1 2 3 4 5

10 log2(CE) 02468101214 log (RNA copy number) 2 B 24

Fig.3. Comparison of one-step RT–qPCR with preamplification using the CellsDirect kit. A 10-fold dilution series of ERCC-84 RNA standard from 104 to 10 copies was prepared in cell background equivalent to a single HFF-1 cell and six replicate RT– qPCR (open diamonds) or RT–preamplification reactions (open triangles) were 22 performed. Three replicate qPCRs per preamplified sample were performed. Data points (n = 6) represent individual RT–qPCR or mean qPCR values (n = 3) from each 2 RT–preamplification reaction. Slopes and R values are displayed from correlation of y = -1.0443x + 21.139

C values against log (RNA copy number). Arrows indicate delta C (DC ) ± standard q q 2 q q 20 R² = 0.9902 error between non- and preamplified sample series based on regression analysis. C efficiency of 100%. Significance values were defined by P values ob- tained by one-way and two-way analyses of variance (ANOVAs) performed using Microsoft Excel 2003 (one-way ANOVA: Fig. 2E 18 and F) and GraphPad Prism (version 5a) (two-way ANOVA: Fig. 2A–D). Comparison of slopes and intercepts obtained from nonamplified and preamplified templates (see Fig. 3 in Results) 16 was performed by regression analysis using R software. Weighted -1 0 1 2 3 4 5 regression was performed because it was noted that the variance increased with low copy numbers. Log2(CE)

Fig.4. Comparison of nanoliter- with microliter-scale qPCRs using a dilution series of laser-captured single cells. Lysate from a pool of 50 laser-captured HFF-1 single Results cells was diluted to the equivalent of 22.5, 4.5, and 0.9 cells and preamplified using the CellsDirect kit (n = 2). Preamplified sample was analyzed for GADPH expression Comparison of RT–qPCR kits using a cell dilution series in microliter or nanoliter qPCRs using the 7900HT (A) or BioMark (B) platform. Data

points represent mean Cq values for each preamplified sample (qPCR replicates: 2 To assess the impact of scaling down cell lysis, RT, and qPCRs to n = 9, BioMark); n = 2, 7900HT). Slopes and R values are displayed from linear regression of C values against log CE. the single cell range, we assessed the limits of detection of two ref- q 2 erence genes (GAPDH and B2M) with three commercially available RT–qPCR kits (CellsDirect, Cells-to-CT, and PicoPure/Message Sen- undetermined reading at the single cell level for B2M. With the sor), designed for the analysis of small numbers of cells, using a CellsDirect kit, the slopes of target copy number plotted against 10-fold cell dilution of fibroblasts from 103 cells to 1 cell (see Mate- cell number (1.23 for GAPDH and 1.25 for B2M vs. the theoretical rials and Methods). Intact cells at each of the respective cell num- slope of 1) indicate the least proportional response of the three ap- bers were used to ensure that the data obtained reflected the entire proaches (Fig. 1A). R2 values also suggest nonlinearity in B2M and process from cell lysis to Cq value. The results obtained for the cell GAPDH measurements across the range of cell numbers studied, dilution series are shown in Fig. 1. Linear regression was performed likely due to the drop in cDNA copies detected at 103 cells by logarithmic plots of cDNA copy numbers of GAPDH or B2M (Fig. 1A). The Cells-to-CT kit showed a more linear response than against the number of cells to gauge the linearity and proportion- the CellsDirect kit (R2 > 0.99) (Fig. 1B), although differences in the ality of the response across the measured range, as indicated by R2 measurements of the two target genes were apparent with this and slope (ideal slope = 1.0), respectively. Copy numbers of each kit compared with the other two approaches, indicating similar cDNA were measured against a DNA standard curve of GAPDH or copy numbers of B2M and GAPDH per cell. The PicoPure/Message B2M PCR product (see Materials and Methods) to estimate the Sensor method demonstrated the best overall efficiency, with the gross efficiency of lysis and RT steps, which is indicated by the highest number of cDNA copies measured for both genes tested intercept of the linear regression equation (Fig. 1). (intercept > 2.0, indicating > 102 B2M and GAPDH copies per single GAPDH and B2M transcripts were detected over the entire cell cell reaction) (Fig. 1C). The slopes of the linear regression plots dilution series for the CellsDirect and PicoPure/Message Sensor kits (B2M slope of 1.02, GAPDH slope of 1.14) (Fig. 1C) were also for all three single cell samples, whereas the Cells-to-CT kit gave an closer to the expected value of 1.0 compared with the other two 184 RT–qPCR methods for single cell analysis / B.C. Fox et al. / Anal. Biochem. 427 (2012) 178–186

approaches, suggesting that this methodology gave more propor- Cq values (P < 0.0001) and improved linearity in the absence of re- tional and accurate results. verse transcriptase (R2 = 0.986 vs. R2 = 0.979) (Fig. 2D). The effect of cell background on RT–qPCR was analyzed by mea- suring a fixed number of ERCC RNA copies in the presence of no Precision of RT–qPCR kits at single cell level background or 0.2, 2, 20, or 200 CEs (Fig. 2E and F). For the CellsDi-

rect kit, this analysis showed that no significant differences in Cq The performance of all three kits was assessed in terms of the values were observed in the presence of varying cell background number of successful RT–qPCR repeats and the variation between levels (Fig. 2E). In the results obtained by the Cells-to-CT kit, the replicate measurements (SD Cq) observed across a dilution series addition of cell background equivalent to 200 cells led to an 4 2 of ERCC-84 RNA standard from 10 to 10 copies (Table 1). For improvement in performance, with a reduction in mean Cq of 4 cy- the CellsDirect and PicoPure/Message Sensor kits, all of the RT– cles compared with smaller quantities of cell background qPCR repeats across the range gave a positive Cq value, whereas (P < 0.005) (Fig. 2F). for the Cells-to-CT kit, the 100% success rate was achieved upward of 103 copies. Overall, the precision of the CellsDirect kit was high- er than that of the other two approaches, with SD Cq below 0.25 for Evaluation of BioMark qPCR platform all three levels of RNA concentration (Table 1). Having characterized three different RT–qPCR methodologies suitable for analysis of several genes in single cells, the suitability Investigating potential inhibitors of RT–qPCR of a recently developed high-throughput qPCR platform for multi- target single cell analysis was also evaluated. Single cell analysis The results of the cell dilution series with the CellsDirect and using BioMark microfluidic real-time PCR arrays (‘‘dynamic ar- Cells-to-CT kits suggested possible inhibition of these two direct rays’’) requires preamplification of the sample to generate suffi- lysis kits by reaction components such as lysis buffer, reverse cient material for the analysis of multiple genes. Therefore, transcriptase enzyme, and cell background. To address this, RT– initially the use of preamplification prior to qPCR was investigated qPCR was performed using RNA or DNA standards (see Materials using ERCC-84 RNA standards and compared with a one-step RT– and Methods), varying these three parameters. A 10-fold serial qPCR approach, using the 7900HT system in both cases (Fig. 3). dilution of exogenous ERCC-84 RNA (‘‘RNA standard’’) was used For these studies, the CellsDirect kit was used for both RT–qPCR to investigate potential inhibition of the RT–qPCR process by the and RT–preamplification because this kit showed good linearity presence of lysis buffer (Fig 2A and B). Results were analyzed by and high precision at low copy numbers (Fig. 2 and Table 1). In linear regression of Cq values versus log2(RNA copy number), addition, this is a one-step RT–PCR kit that was amenable to rather than the usual convention for qPCR data of plotting Cq ver- high-throughput procedures. sus log10(DNA copy number/concentration) so that the slope re- Initially, the propensity for preamplification to introduce ampli- flects the linearity of both the RT and qPCRs (vs. the ideal slope fication bias of individual genes within a complex background of –1.0), as opposed to the efficiency of the qPCR assay alone. Inves- based on their abundance was investigated using a 10-fold serial tigation of the effect of lysis buffer on the CellsDirect kit revealed a dilution of exogenous ERCC-84 RNA standard from 104 to 10 copies significant decrease in Cq values of RT–qPCR in the presence of lysis in a single cell background equivalent. Preamplification of tem- buffer compared with the RNA template in water (–1.6 cycles, plate using the CellsDirect kit with 18 cycles of PCR followed by P < 0.0001), showing a beneficial effect of the lysis buffer in the quantification by qPCR was compared with one-step RT–qPCR

RT–qPCR (Fig. 2A). For the Cells-to-CT kit, mean Cq values of (Fig. 3). Both methods displayed good precision and linearity, with ERCC-84 RNA template in lysis buffer were significantly higher R2 values greater than 0.99 and slopes of –1.02 and –1.03 (vs. an than those in a water background (two-way ANOVA, P < 0.0001), ideal slope of 1.0) for the one-step RT–qPCR and preamplification particularly for 104 copies of RNA standard (P < 0.0001) (Fig. 2B). approaches, respectively. Regression analysis confirmed that there ERCC-84 RT–qPCR results using the Cells-to-CT kit in the presence was no significant difference between the slopes of the two data of lysis buffer were also less linear than in the absence of this com- series and indicated a consistent decrease in Cq values across the 2 ponent (R = 0.946 vs. 0.989) (Fig. 2B). copy number range (mean DCq of 12.93, standard error of 0.29) A dilution series of ERCC-84 DNA standard was performed to for the preamplified samples compared with the nonamplified measure the qPCR efficiency of the kits and possible inhibition of samples (Fig. 3). the qPCR step by the reverse transcriptase enzyme (Fig. 2C and Following the analysis of preamplification methodology on the D). Because the CellsDirect kit is a one-step kit with the reverse 7900HT platform, qPCR performance of the nanofluidic BioMark transcriptase and Taq polymerase combined, it was not possible Dynamic Arrays was compared with that of the standard microliter to investigate the effect of the reverse transcriptase in isolation; volume qPCR platform based on GAPDH measurements in a 5-fold however, the slope of the DNA standard dilution series (in lysis dilution series from a pool of 100 laser capture microdissection buffer) (Fig. 2C) suggested an acceptable qPCR efficiency (a slope (LCM)-collected single HFF-1 cells diluted to 22.5, 4.5, and 0.9 of –1.1 equates to efficiency of PCR amplification of 87%; see Mate- CEs (n =2)(Fig. 4). GAPDH expression was detectable in 0.9 CEs rials and Methods). For the Cells-to-CT kit, investigation of the ef- using both microliter- and nanoliter-volume qPCR platforms fect of RT enzyme on qPCR efficiency revealed significantly lower (Fig. 4A and B, respectively). Linear regression analysis of the

Table 1 Precision of CellsDirect, Cells-to-CT, and Message Sensor kits

RNA copies Cells Direct Cells-to-CT Message Sensor 100% positive SD 100% positive SD 100% positive SD p p 102 0.24 N/A 0.73 p p p 103 0.04 0.34 0.23 p p p 104 0.12 0.39 0.04

Note. SD, standard deviation. Triplicate RT–qPCRs were performed with different copy numbers of ERCC-84 RNA standard, and the numbers of positive reactions and SDs between replicate Cq values were measured for each kit. RT–qPCR methods for single cell analysis / B.C. Fox et al. / Anal. Biochem. 427 (2012) 178–186 185 dilution series indicated similar linearity between the two plat- verse transcriptase supplied in this kit, which has previously been forms, with R2 values of 0.990 and 0.993 and slopes of –1.06 and shown to increase the sensitivity of RT–PCR [33]. –1.04 for the 7900HT and BioMark, respectively (Fig. 4). BioMark Other possible causes of inhibition of RT–qPCR resulting in the analysis of an additional five gene targets expressed in HFF-1 less proportional slopes (further from the ideal slope of 1.0) seen cells revealed that two of these targets (B2M and COL6A1) were for the CellsDirect and Cells-to-CT kits (Fig. 1A and B) were further also detectable at the single cell level, although slope and linearity investigated (Fig. 2). The Cells-to-CT kit was significantly inhibited of regression analysis varied depending on gene target (see by the presence of the reverse transcriptase in the qPCR, a well- Supplementary Table 3 in supplementary material). Two genes of characterized phenomenon observed in previous studies interest (TNFR12A and M6PRBP1) were detected at the 4.5- and [18,19,34]. Conditions to alleviate the inhibition of reverse trans- 22.5-CE level, whereas CSF3, which was assumed to be expressed criptase on RT–qPCR include the use of carrier transfer RNA (tRNA). at low levels in HFF-1 cells (Supplementary Table 2), was not The use of carrier tRNA to reduce inhibition has been suggested to detected in any of the samples (Supplementary Table 3). be beneficial to reactions using avian myeloblastosis virus (AMV), RNase H– RT, but not MMLV H– RT [32,35,36], and an improved performance with the Cells-to-CT kit was observed by the addition Discussion of a 200-cell background equivalent (Fig. 2F); however, due to commercial restrictions, it was not known which RT enzyme was In this study, we have compared three RT–qPCR kits (CellsDi- used in the Cells-to-CT kit. The CellsDirect kit uses SuperScript rect, Cells-to-CT, and a combination of the PicoPure extraction kit III, which is an engineered MMLV RT, and increasing the cell back- with the Message Sensor RT–qPCR kit) for analysis of gene expres- ground was not observed to be beneficial with this kit (Fig. 2E); in- sion at the single cell level. deed, a reduction in efficiency at the 103 cell level was indicated for The sensitivity, linearity, and reaction efficiencies of the kits were the CellsDirect kit (Fig. 1A). This may be due to a reduction in lysis investigated using a cell dilution series and RNA dilution series efficiency with increasing cell number given that RT–qPCR using (Figs. 1 and 2). This approach ensured that the entire process from RNA standard was not affected (Fig. 2E). Finally, the CellsDirect cell lysis to Cq value was assessed, namely lysis/extraction, RT, and kit performed better in the presence of lysis buffer compared with qPCR steps; hence, any biases or inhibition imposed from the various water, whereas an increase in Cq values and reduced linearity were stages of process would affect the final Cq value. It should be noted observed for the results of the RNA dilution series in lysis buffer that whereas the efficiency of qPCR is normally modeled by linear compared with water for the Cells-to-CT kit (Fig. 2A and B, respec- regression of Cq values versus log10(DNA concentration), with a slope tively). This finding may be due to differential effects of lysis buffer of –3.3 indicating the ideal doubling of template with each PCR cycle components such as chaotropic salts and detergents; in theory, (efficiency = 2.0) [31], the efficiencies of the nonexponential lysis these agents may inhibit RT or PCR enzymes by causing partial pro- and RT reactions are described as the yield of cDNA molecules (% effi- tein unfolding, but optimized levels of guanidine thiocyanate have ciency for known RNA input) based on quantification of cDNA using a been shown to enhance RT efficiency compared with control reac- DNA standard curve [13,32]. RT efficiency may be influenced by mul- tions [13]. In addition, the relatively high proportions of cell lysate tiple factors, including transcript copy number as well as total input (45%) and RT reaction (40%) in the RT and qPCRs, respectively, for per reaction [32]. Therefore, the slope of linear regression relating the Cells-to-CT kit compared with the CellsDirect kit (where lysate reaction input in terms of cell numbers to measured cDNA copies formed 25% of the RT–qPCR) may have contributed to the inhibi-

(Fig. 1) or RNA copies to Cq values (Figs. 2–4) reflects any biases in tory effects observed for the former method. This is an important the multiple reaction stages, with significant deviation from the factor to consider when developing a workflow for single cell anal- ideal slope of 1.0 likely to affect the accuracy of gene expression ysis because larger volumes of lysis buffer for cell capture will quantification, whereas the intercept of the regression is more infor- necessitate a higher proportion of lysis reactions in the RT–qPCR. mative regarding the efficiency of cell lysis and RT. In summary, the above results based on cell and RNA dilutions sug- The cell dilution series (Fig. 1) demonstrated that all three kits gest that the Cells-to-CT kit may perform optimally with a higher cell assessed were capable of mRNA detection at the single cell level. number/total RNA level than that found in a single cell. In comparison, The precision of the kits in terms of transcript copy number was the PicoPure/Message Sensor kit would be suitable for single cell gene investigated using dilutions of an IVT RNA standard (ERCC-84) (Ta- expression studies with a relatively small sample size. However, due ble 1); the CellsDirect and PicoPure/Message Sensor kits again to the requirement for the purification step, this kit combination showed comparable sensitivity, detecting 102 copies, with the might not be ideal for high-throughput procedures. The CellsDirect CellsDirect kit demonstrating higher precision at this level of tran- kit also performed well at low cell and RNA copy numbers and did script abundance, whereas consistent detection (100% positive) not demonstrate inhibition due to lysis buffer or RT components was observed for the Cells-to-CT approach at 103 copies and above. (Fig. 2), suggesting its suitability for single cell analysis. The linearity of cDNA copy number values measured by RT–qPCR It has been suggested that to minimize technical noise associ- analysis of the cell dilution series using the three methods was ated with single cell RT–qPCR, the amount of RT product in the evaluated to assess the suitability of the kits for quantitative mea- qPCR should be maximized and technical replicates should be surements at the single cell level and to uncover any bias within avoided [13]. However, the number of genes that could be assessed the dynamic range caused by inhibition (Fig. 1). Linear regression would be low because the lysate could be split into only a few RT– analysis of GAPDH or B2M copy numbers versus cell number indi- qPCRs before quantification would become unreliable. Methods for cated that the PicoPure/Message Sensor kit demonstrated the most sample preamplification using specific target amplification over- proportional response of the three kits, whereas the slopes for the come this limitation. The emergence of microfluidic PCR platforms other two approaches were further from the ideal slope of 1.0. This and arrays, such as the BioMark system, also facilitates multigene may be due to more complete cell lysis using a separate extraction analysis alongside high sample throughput. The BioMark platform step (PicoPure) and a beneficial effect of using purified RNA for the has previously been shown to produce accurate RT–qPCR results RT and qPCR steps [18]. The overall reaction efficiency judged by across a range of target copy numbers, but not with total quantities the number of cDNA copies quantified using a DNA standard curve of template corresponding to the single cell level [26,37]. Single also indicated superior performance of the PicoPure/Message Sen- cell analysis using this platform increases the number of gene tar- sor kit combination in this respect (Fig. 1). This may be due to the gets; however, it also necessitates preamplification of template RNase H-plus form of Moloney murine leukemia virus (MMLV) re- RNA or cDNA prior to qPCR, which could introduce bias due to 186 RT–qPCR methods for single cell analysis / B.C. Fox et al. / Anal. Biochem. 427 (2012) 178–186 different levels of transcript abundance [25]; hence, this aspect [9] M. Bengtsson, A. Stahlberg, P. Rorsman, M. Kubista, Gene expression profiling was evaluated using dilutions of exogenous RNA standard in a in single cells from the pancreatic islets of Langerhans reveals lognormal distribution of mRNA levels, Genome Res. 15 (2005) 1388–1392. constant cell background equivalent to a single cell. These investi- [10] Y. Li, H. Thompson, C. Hemphill, F. Hong, J. Forrester, R.H. Johnson, W. Zhang, gations showed no detectable copy number bias for ERCC-84 D.R. Meldrum, An improved one-tube RT–PCR protocol for analyzing single- within the range from 10,000 to 10 copies, as indicated by good cell gene expression in individual mammalian cells, Anal. Bioanal. Chem. 397 (2010) 1853–1859. proportionality observed in the slopes of both dilution series and [11] S.A. Bustin, V. Benes, J. Garson, J. Hellemans, J. Huggett, M. Kubista, R. Mueller, consistent DCq values across the copy number range with and T. Nolan, M.W. Pfaffl, G.L. Shipley, J. Vandesompele, C.T. Wittwer, The MIQE without preamplification (Fig. 3). Interestingly, the variation guidelines: minimum information for publication of quantitative real-time PCR experiments, Clin. Chem. 55 (2009) 611–622. between replicates for low copy number is similar for both [12] M. Reiter, B. Kirchner, H. Müller, C. Holzhauer, W. Mann, M.W. Pfaffl, Quantification approaches (Fig. 3), indicating that this could stem from the RT noise in single cell experiments, Nucleic Acids Res. 39 (2011) e124. step, which is known to add significant variability for low copy [13] M. Bengtsson, M. Hemberg, P. Rorsman, A. Stahlberg, Quantification of mRNA in single cells and modelling of RT–qPCR induced noise, BMC Mol. Biol. 9 numbers of RNA and using single cell templates [12,38]. (2008) 63. The impact of scaling down qPCRs from standard microliter vol- [14] I.A. Kerman, B.J. Buck, S.J. Evans, H. Akil, S.J. Watson, Combining laser capture umes to the nanoliter range was investigated for low cell number microdissection with quantitative real-time PCR: effects of tissue applications by comparing the performance of the BioMark arrays manipulation on RNA quality and gene expression, J. Neurosci. Methods 153 (2006) 71–85. with that of the 7900HT in the quantification of GAPDH present in [15] M.J. Smith, A.C. Culhane, M. Donovan, J.C. Coffey, B.D. Barry, M.A. Kelly, D.G. a dilution series generated from laser-captured single cells (Fig. 4). Higgins, J.H. Wang, W.O. Kirwan, T.G. Cotter, H.P. Redmond, Analysis of GAPDH was detected at the single cell level on both platforms, and differential gene expression in colorectal cancer and stroma using fluorescence-activated cell sorting purification, Br. J. Cancer 100 (2009) the linearity of the RNA dilution series was also comparable, indi- 1452–1464. cating that there is no reduction in sensitivity or accuracy of single [16] K.E. Pierce, J.E. Rice, J.A. Sanchez, L.J. Wangh, QuantiLyse: reliable DNA cell RT–qPCR analysis by miniaturization of the qPCR. amplification from single cells, Biotechniques 32 (2002) 1106–1111. [17] Applied Biosystems, Arcturus PicoPure RNA isolation kit [product bulletin], publication CO14630 0810 (2010). Conclusions [18] O. Suslov, D.A. Steindler, PCR inhibition by reverse transcriptase leads to an overestimation of amplification efficiency, Nucleic Acids Res. 33 (2005) e181. [19] B. Liss, Improved quantitative real-time RT–PCR for expression profiling of Our study highlights that differences in the precision and linearity individual cells, Nucleic Acids Res. 30 (2002) e89. of RT–qPCR methodologies can have considerable impact on gene [20] R.M. Frederickson, Fluidigm: biochips get indoor plumbing, Chem. Biol. 9 (2002) 1161–1162. expression measurements of a single cell. Inhibitory effects due to [21] J.M. Dixon, M. Lubomirski, D. Amaratunga, T.B. Morrison, C.J. Brenan, S.E. Ilyin, reaction components such as lysis buffer and RT reagents can reduce Nanoliter high-throughput RT–qPCR: a statistical analysis and assessment, the sensitivity of detection by RT–qPCR. Assessment of technical as- Biotechniques 46 (2009) ii–viii. [22] L. Warren, D. Bryder, I.L. Weissman, S.R. Quake, Transcription factor profiling pects associated with the recently developed BioMark qPCR platform in individual hematopoietic progenitors by digital RT–PCR, Proc. Natl. Acad. indicates the suitability of this high-throughput approach for single Sci. USA 103 (2006) 17807–17812. cellanalysisandhighlightsthe importance ofusing externalRNAstan- [23] G. Guo, M. Huss, G.Q. Tong, C. Wang, L. Li Sun, N.D. Clarke, P. Robson, Resolution of cell fate decisions revealed by single-cell gene expression dards and reference materials for validation of methodologies. analysis from zygote to blastocyst, Dev. Cell 18 (2010) 675–685. [24] P. Dalerba, T. Kalisky, D. Sahoo, P.S. Rajendran, M.E. Rothenberg, A.A. Leyrat, S. Acknowledgments Sim, J. Okamoto, D.M. Johnston, D. Qian, M. Zabala, J. Bueno, N.F. Neff, J. Wang, A.A. Shelton, B. Visser, S. Hisamori, Y. Shimono, M. van de Wetering, H. Clevers, M.F. Clarke, S.R. Quake, Single-cell dissection of transcriptional heterogeneity The work described in this article was funded by the U.K. Na- in human colon tumors, Nat. Biotechnol. 29 (2011) 1120–1127. tional Measurement System. The authors acknowledge Jesus Min- [25] S.D. Ginsberg, RNA amplification strategies for small sample populations, Methods 37 (2005) 229–237. guez for valuable discussions on statistical analysis of qPCR data. [26] A.S. Devonshire, R. Elaswarapu, C.A. Foy, Applicability of RNA standards for We thank Alexandra Whale and Jim Huggett for detailed assess- evaluating RT–qPCR assays and platforms, BMC Genomics 12 (2011) 118. ment of the manuscript. [27] A.K. White, M. VanInsberghe, O.I. Petriv, M. Hamidi, D. Sikorski, M.A. Marra, J. Piret, S. Aparicio, C.L. Hansen, High-throughput microfluidic single-cell RT– qPCR, Proc. Natl. Acad. Sci. USA 108 (2011) 13999–14004. Appendix A. Supplementary data [28] P. Mary, L. Dauphinot, N. Bois, M.C. Potier, V. Studer, P. Tabeling, Analysis of gene expression at the single-cell level using microdroplet-based microfluidic technology, Biomicrofluidics 5 (2011) 24109. Supplementary data associated with this article can be found, in [29] H. Zhang, G. Jenkins, Y. Zou, Z. Zhu, C.J. Yang, Massively parallel single- the online version, at http://dx.doi.org/10.1016/j.ab.2012.05.010. molecule and single-cell emulsion reverse transcription polymerase chain reaction using agarose droplet microfluidics, Anal. Chem. 17 (2012) 3599– 3606. References [30] External RNA Controls Consortium, Proposed methods for testing and selecting the ERCC external RNA controls, BMC Genomics 6 (2005) 150. [31] M.W. Pfaffl, A new mathematical model for relative quantification in real-time [1] A.K. Dixon, P.J. Richardson, R.D. Pinnock, K. Lee, Gene-expression analysis at RT–PCR, Nucleic Acids Res. 29 (2001) e45. the single-cell level, Trends Pharmacol. Sci. 21 (2000) 65–70. [32] J.P. Levesque-Sergerie, M. Duquette, C. Thibault, L. Delbecchi, N. Bissonnette, [2] D. Longo, J. Hasty, Dynamics of single-cell gene expression, Mol. Syst. Biol. 2 Detection limits of several commercial reverse transcriptase enzymes: impact (2006) 64. on the low- and high-abundance transcript levels assessed by quantitative RT– [3] R. Todd, D.H. Margolin, Challenges of single-cell diagnostics: analysis of gene PCR, BMC Mol. Biol. 8 (2007) 93. expression, Trends Mol. Med. 8 (2002) 254–257. [33] S.K. Polumuri, A. Ruknudin, D.H. Schulze, RNase H and its effects on PCR, [4] C. Hartshorn, J.J. Eckert, O. Hartung, L.J. Wangh, Single-cell duplex RT–LATE– Biotechniques 32 (2002) 1224–1225. PCR reveals Oct4 and Xist RNA gradients in 8-cell embryos, BMC Biotechnol. 7 [34] L.N. Sellner, R.J. Coelen, J.S. Mackenzie, Reverse transcriptase inhibits Taq (2007) 87. polymerase activity, Nucleic Acids Res. 20 (1992) 1487–1490. [5] K.H. Narsinh, N. Sun, V. Sanchez-Freire, A.S. Lee, P. Almeida, S. Hu, T. Jan, K.D. [35] A. Stahlberg, M. Kubista, M. Pfaffl, Comparison of reverse transcriptases in Wilson, D. Leong, J. Rosenberg, M. Yao, R.C. Robbins, J.C. Wu, Single cell gene expression analysis, Clin. Chem. 50 (2004) 1678–1680. transcriptional profiling reveals heterogeneity of human induced pluripotent [36] D.P. Chandler, C.A. Wagnon, H. Bolton Jr., Reverse transcriptase (RT) inhibition stem cells, J. Clin. Invest. 121 (2011) 1217–1221. of PCR at low concentrations of template and its implications for quantitative [6] I. Tietjen, J.M. Rihel, Y. Cao, G. Koentges, L. Zakhary, C. Dulac, Single-cell RT–PCR, Appl. Environ. Microbiol. 64 (1998) 669–677. transcriptional analysis of neuronal progenitors, Neuron 8 (2003) 161–175. [37] S.L. Spurgeon, R.C. Jones, R. Ramakrishnan, High throughput gene expression [7] F. Tang, C. Barbacioru, Y. Wang, E. Nordman, C. Lee, N. Xu, X. Wang, J. Bodeau, measurement with real time PCR in a microfluidic dynamic array, PLoS One 3 B.B. Tuch, A. Siddiqui, K. Lao, M.A. Surani, mRNA-seq whole-transcriptome (2008) e1662. analysis of a single cell, Nat. Methods 6 (2009) 377–382. [38] J. Curry, C. McHale, M.T. Smith, Low efficiency of the Moloney murine leukemia [8] S. Islam, U. Kjällquist, A. Moliner, P. Zajac, J.B. Fan, P. Lönnerberg, S. Linnarsson, virus reverse transcriptase during reverse transcription of rare t(8;21) fusion Characterization of the single-cell transcriptional landscape by highly gene transcripts, Biotechniques 32 (2002) 768–770. multiplex RNA-seq, Genome Res. 21 (2011) 1160–1167.