Performance comparison of reverse transcriptases for single-cell studies

Performance comparison of reverse

transcriptases for single-cell studies

Zucha Daniel

1,2

, Androvic Peter

1,3

, Kubista Mikael

1,4

, Valihrach Lukas

Laboratory of Gene Expression, Institute of Biotechnology CAS, Biocev, 252 50 Vestec, Czech Republic;

Faculty

of Science, Charles University, 128 00 Prague, Czech Republic;

Laboratory of Growth Regulators, Faculty of

Science, Palacky University, 783 71 Olomouc, Czech Republic and

TATAA Biocenter AB, Gothenburg 411 03,

Sweden

ABSTRACT

Background: Recent technical advances allowing quantification of RNA from single cells are

revolutionizing biology and medicine. Currently, almost all single-cell transcriptomic protocols

rely on conversion of RNA to cDNA by reverse transcription (RT). However, RT is recognized

as highly limiting step due to its inherent variability and suboptimal sensitivity, especially at

minute amounts of RNA. Primary factor influencing RT outcome is reverse transcriptase

(RTase). Recently, several new RTases with potential to decrease the loss of information

during RT have been developed, but the thorough assessment of their performance is missing.

Methods: We have compared the performance of 11 RTases in RT-qPCR on single-cell and

100-cell bulk templates using two priming strategies: conventional mixture of random

hexamers with oligo(dT)s and reduced concentration of oligo(dT)s mimicking common single-

cell RNA-Seq library preparation protocols. Based on the performance, two RTases were

further tested in high-throughput single-cell experiment.

Results: All RTases tested reverse transcribed low-concentration templates with high

accuracy (R

> 0.9445) but variable reproducibility (median CV

= 40.1 %). The most

pronounced differences were found in the ability to capture rare transcripts (0 - 90% reaction

positivity rate) as well as in the rate of RNA conversion to cDNA (7.3 - 124.5 % absolute yield).

Finally, RTase performance and reproducibility across all tested parameters were compared

using Z-scores and validity of obtained results was confirmed in a single-cell model

experiment. The better performing RTase provided higher positive reaction rate and

expression levels and improved resolution in clustering analysis.

Conclusions: We performed a comprehensive comparison of 11 RTases in low RNA

concentration range and identified two best-performing enzymes (Maxima H-; SuperScript IV).

We found that using better-performing enzyme (Maxima H-) over commonly-used below-

average performer (SuperScript II) increases the sensitivity of single-cell experiment. Our

results provide a reference for the improvement of current single-cell quantification protocols.

Key words

Reverse transcription, reverse transcriptase, quantitative PCR, single cell, ERCC Spike-in,

RNA-Seq

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

INTRODUCTION

Single cell transcriptomics has emerged as a revolutionary technology transforming the

biomedical research. From the technical perspective, minute amounts of RNA found in single

cells have to be converted into cDNA, amplified and transformed into sequencing libraries.

Reverse transcription (RT) is a critical step in this process, as any RNA molecules that fail to

be initially captured are missing in the final data. Many factors can influence RT outcome, but

reverse transcriptase (RTase) is arguably the most prominent. Wild-type RTases are not

efficient and reliable enough for diagnostic and research applications. For in vitro purposes,

many properties, such as thermostability, fidelity, processivity, substrate binding affinity,

template switching activity and other properties are added or enhanced by engineering

1–4

. The

alteration of these intrinsic properties is closely linked to reaction outcome. Currently,

derivatives of Moloney Murine Leukemia Virus (MMLV) and Avian Myeloblastosis Virus (AMV)

RTases are the most common, although other promising sources of RTases have been also

identified

In addition to the selection of RTase also other factors must be considered when

planning an experiment, such as priming strategy or template concentration

6,7

. Oligo(dT)

primers target polyadenylated RNAs, while random sequence primers target all RNAs

including the abundant rRNA fraction. For the ability to enrich for polyadenylated RNAs,

oligo(dT)s have found the application in numerous RNA-Seq protocols

8–10

, whereas the

mixture of random hexamers with oligo(dT)s are predominantly used in RT-qPCR experiments

where it maximizes the reaction yield

. Gene-specific primers may be used as well to improve

the specificity of the reaction. They are the most popular for targeted RT-qPCR applications

mainly in diagnostics

6,7

or for specific applications such as miRNA quantification

12–14

Experimental parameters influencing the conversion of RNA into cDNA were studied

to a varying degree in past decades. The largest focus has been pointed on the choice of

reverse transcriptase. Generally, MMLV-derived RTases have been found to deliver superior

results

11,15–17

. Specifically, SuperScript II and SuperScript III (Thermo Fisher Scientific, USA)

have been several times highlighted for their reproducibility and sensitivity

15,16,18,19

. However,

not all RTases reported such constant quality. For example, performance of OmniScript and

SensiScript (both Qiagen, Germany) or AMV derived RTases have been heavily influenced

by experimental conditions, including the template input or by the laboratory conducting the

experiment

15,17,18

. Notably, RTases with template-switching properties required for certain

RNA-Seq protocols

4,20,21

were recently compared by Bagnoli et al. scoring Maxima H- (Thermo

Fisher Scientific, USA) and SmartScribe (Clontech, USA) as top performing candidates

Several studies also focused on the role of other reaction components on the reaction

outcome. The effectiveness of priming strategy was shown to be substantially gene-

dependent

6,7,22

but the discrepancies could be minimized by optimizing the primer

concentration

. Gene-related efficiency of RT had been notified on multiple

occasions

6,15,17,18,23

. Several reports suggested the addition of carrier molecules - tRNA

polyethylene glycol

or total extracted RNA

to increase reaction yield.

Although efforts have been made to characterize the influences of aforementioned

factors, little attention has been paid to characterize the performance of RT in low-RNA input

applications, such as single-cell RNA-Seq. The exception is a study performed by Levesque-

Sergerie et al. that partially mimicked single-cell conditions (detecting a single target in

concentration of thousands copies per reaction)

. Unfortunately, the experimental extent as

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

well as the number of tested RTases (five) were limited, therefore the study did not address

many aspects of the RT for limiting template concentrations.

Here, we decided to fill the gap and benchmark a broad spectrum of currently available

RTases in low-template applications. We systematically compared 11 RTases using

equivalents of single-cell as well as 100-cell samples with two priming strategies popular for

RT-qPCR and RNA-Seq applications. We characterized the RTases in terms of sensitivity,

accuracy, yield and reproducibility. In the single-cell model study we demonstrated RTase

influence on the experimental result quantified by differential positivity rates, higher expression

levels and data clustering. Our results uncovered the substantial differences between

individual RTases currently available and provide the data for an informed selection of the

best suited RTase for the particular application in, but also outside the field of single-cell

transcriptomics.

METHODS

Experimental design

The RTases were compared in two experiments: i) RTase benchmarking and ii) High-

throughput validation (Figure 1). The goal of RTase benchmarking was to test the performance

of 11 commercially available RTases (Table 1) in conditions mimicking single-cell and 100-

cell samples. The follow up validation experiment aimed to illustrate the impact of RTase in

gene expression profiling experiments using real single-cell samples. In both parts, two RT

priming strategies were utilized: 1) equimolar mixture of 50 µM random hexamers with 50 µM

oligo(dT)

(recommended concentration in RT-qPCR protocols (Supplementary protocols -

Supp. data 1)) and 2) 10µM oligo(dT)

(recommended concentration in single-cell RNA-Seq

protocols

RTase benchmarking

Template

External RNA Controls Consortium (ERCC) Spike-in (set 1) (Thermo Fisher Scientific, USA)

was used as primary template

. ERCC Spike-in set consists of unlabeled polyadenylated

transcripts of various lengths (250 to 2000 nucleotides) mimicking eukaryotic mRNAs. Due to

its known copy numbers and availability of DNA standards for absolute quantification of cDNA

molecules, it was possible to calculate RNA conversion rate (see Yield calculation, Methods

and ERCC Spike-in validation, Supp. data 2 for details). Stock ERCC Spike-in was 200,000×

diluted in TE-buffer supplemented with linear polyacrylamide (TE-LPA) for single-cell

conditions or 2,000× for bulk conditions and stored in aliquots (Table 2). For each experiment

a fresh aliquot was mixed with either 30 pg mouse cerebral RNA (used as a background)

mimicking a single-cell or 3 ng mimicking 100-cell bulk sample

(details in RNA material ,

Supp. data 1,2). In total, the ERCC Spike-in accounted for ~13 % of mRNA and ~0.5 % of

total RNA, respectively.

Reverse transcription

For each RTase, reaction conditions and thermal profile followed the recommended protocol

issued by manufacturer. General reaction components including nuclease-free water (NFW),

RT primers, dNTPs, dithiothreitol (DTT), and ribonuclease inhibitor (all Thermo Fisher

Scientific, USA) were supplied from a single stock. Total RT reaction volume was 5 µl. Each

experimental condition was run in 10 replicates unless stated otherwise.

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

An exemplary RT reaction consisted of: 2 µl of background RNA (3 ng for bulk and 30

pg for single-cell equivalents), 0.5 µl of ERCC Spike-in (2,000× or 200,000× diluted) and 2.5

µl of RT mastermix containing RT primers (50µM RT primers mixture or 10 µM oligo(dT)

10mM dNTPs, RTase specific buffer, 0.1M DTT - if requested, 10 U of RNaseOUT - if

requested, NFW and RTase). All RTs were performed in Bio-Rad C1000 Thermal Cycler (Bio-

Rad, USA). Prepared cDNA was 5× diluted in NFW and directly used in qPCR to avoid freeze-

thawing cycles. RTases included in this study are listed in Table 1. Detailed protocols are

listed in Supplementary protocols (Supp. data 1).

Table 1 List of benchmarked RTases.

Quantitative PCR

qPCR measurements followed a thoroughly validated protocol. A reaction volume of 10 µl

contained 2 µl of 5× diluted cDNA and 8 µl of qPCR mastermix, which consisted of 2.6 µl NFW,

5 µl TATAA SYBR Green mix (TATAA, Sweden) and 0.4 µl of 10 µM primers (Thermo Fisher

Scientific, USA). Cycling protocol consisted of initial enzyme activation at 95°C (t = 3 min),

followed by 45 cycles of denaturation at 95°C (t = 15 s), annealing at 60°C (t = 20 s) and

elongation at 72°C (t = 20 s). Melt curve analysis was performed in the temperature interval

of 65 to 95°C, using gradient of 0.5°C. Bio-Rad CFX 384 (Bio-Rad, USA) thermal cycler was

used for all measurements. CFX Manager Software (Bio-Rad, USA) and Project R were used

for data processing.

The initial comparison of 11 RTases was performed using five thoroughly validated

qPCR assays (Table 2). Detailed information on assay optimization and validation is listed in

ERCC Spike-in assay validation (Supp. data 1,2). For each assay, three performance metrics

were determined: efficiency (E), limit of detection (LOD) and limit of quantification (LOQ) (LOD

– LOQ tab, Supp. data 2). The assays were selected to target ERCC molecules present at

different abundancies (22 - 2822 copies per RT reaction in single-cell set-up) mimicking genes

expressed at different levels.

Reverse Transcriptase Manufacturer Origin RNase H activity

Reaction

temperature (°C)

Cost per 20 µl

reaction ($)

SuperScript II Thermo Fisher Scientific (USA) MMLV reduced 42 7.5

SuperScript III Thermo Fisher Scientific (USA) MMLV reduced 50 8.2

SuperScript IV Thermo Fisher Scientific (USA) MMLV reduced 50 8.2

Sensiscript Qiagen (Germany) unspecified present 37 6.0

PrimeScript Takara Bio Inc. (Japan) MMLV none 42 6.5

Maxima H- Thermo Fisher Scientific (USA) MMLV none 50 4.0

AccuScript Hi-Fi Agilent (USA) MMLV none 42 12.0

iScript Bio-Rad Laboratories (USA) MMLV present 42 6.5

MMLV Promega (USA) MMLV present 42 2.0

eAMV Merck KGaA (Germany) AMV unspecified 42 8.0

qScript Quanta Biosciences (USA) MMLV present 42 5.2

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

Table 2 Assay specifications. Limit of detection is the lowest concentration that gives rise to positive signal in 95

% of cases. Limit of quantification is defined as the lowest concentration producing a standard deviation (SD) < 0.5

among replicates. Amount of nucleic acid in bulk samples was 100 times higher.

Yield calculation

The quantification of absolute yield lies in determining relation between RNA input and cDNA

output reported by qPCR. Input RNA copy numbers were calculated from the ERCC

concentration provided by the manufacturer. The cDNA concentration was determined from

the measured Cq and a DNA standard curve covering the range from 2 × 10

to 2 × 10

-1

copies

(n = 4 replicates) (ERCC Spike-in assay validation, Supp. data 2). DNA standards were

prepared from PCR-enriched target sequences. The RT yield is estimated using Equation 1:

(Equation 1)

where a and b refer to slope and intercept of the particular ERCC Spike-in assay DNA standard

curve, respectively and Cq

RT=100%

is the Cq expected for 100% RT efficiency. Adjusting Cq by

-1 accounts for difference between single-stranded cDNA and double-stranded DNA standard

used in standard curve construction.

High-throughput validation on biological samples

Animals

All procedures involving the use of laboratory animals were performed in accordance with the

European Community Council Directive of 24 November 1986 (86/609/EEC) and animal care

guidelines approved by the Institute of Experimental Medicine, Academy of Sciences of the

Czech Republic (Animal Care Committee decision on 17 April 2009; approval number

85/2009). Double transgenic mice bearing SOD1(G93A) and GFAP/EGFP alterations were

used as a model organism. Mice with transgenic expression of mutant human superoxide

dismutase SOD1(G93A) exhibit phenotype similar to amyotrophic lateral sclerosis (ALS) in

humans, whereas GFAP/EGFP allows for visualization of astrocytes due to expression of

EGFP protein under human glial fibrillary acidic protein (GFAP) promoter. EGFP-positive

single-cells from dissected mouse tissue brain (3-months old) were collected using

fluorescence-activated cell sorting GRISORBEON system 1 (FACS;0.9.21 BDInflux) into 5 µl

lysis buffer (NFW + 1 mg/ml BSA). 96 single-cells were collected from both ALS and control

mice. Details regarding the single-cell suspension preparation and cell sorting may be found

elsewhere

Reverse transcription

Single-cells were reverse transcribed in a volume of 10 µl, consisting of: 5 µl of single-cell in

lysis buffer, 0.5 µl of 200,000× diluted ERCC Spike-in and 4.5 µl of RT mastermix. RT

Name

GenBank

accession

Length

[bases]

Single-cells:

transcript count

per RT

Single-cells: cDNA

count per qPCR

Limit of Detection

[molecules per qPCR]

Limit of Quantification

[molecules per qPCR]

Efficiency

ERCC-00084-01 DQ883682 970 22.0 1.8 5.7 6.4 0.95

ERCC-00095-01 DQ516759 495 88.2 7.1 5.7 100 0.863

ERCC-00092-01 DQ459425 1110 176.4 14.1 5.7 40 0.973

ERCC-00108-01 DQ668365 997 705.5 56.4 5.7 16 0.966

ERCC-00171-01 DQ854994 481 2821.9 225.8 5.4 6.4 0.978

Yield = 



 





















.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

mastermix preparation and thermal program followed RTase-specific protocol for Maxima H-

and SuperScript II (Supplementary protocols, Supp. data 1). RT mastermix contained: RT

primers (50µM RT primers mixture or 10µM oligo(dT)

), 10mM dNTPs, RTase specific buffer,

0.1M DTT (SuperScript II only), 20 U of RNaseOUT, NFW and RTase. Sample distribution

followed orthogonal experimental design.

Samples – preamplification and quality control

All cDNA samples were preamplified in 40 µl total reaction volume comprising of 4 µl undiluted

cDNA and 36 µl preAMP mastermix, which consisted of NFW, IQ Supermix buffer (Bio-Rad,

USA) and 250 nM primer mix of 78 assays (Thermo Fisher Scientific, USA) (both endogenous

and ERCC Spike-in assays). Detailed gene list is found in Primer sequences (Supp. data 1).

Thermal protocol comprised of heating at 95°C (t = 3 min), followed by 18 cycles of template

denaturation at 95°C (t = 20 s), annealing at 57°C (t = 4 min) and elongation at 72°C (t = 20

s). Template was immediately cooled on ice, 4× diluted in NFW and stored at -80°C.

4× diluted preAMP cDNA was used in qPCR quality control reactions. The expression

of 4 genes were measured – Gja1 (astrocyte marker), Cspg4 (NG2 cells), Vim and Slc1a3

(both commonly expressed in astrocytes). Only cells positive for Gja1 and concurrently

negative for Cspg4 were used for qPCR analysis; remaining two genes were used as cell

quality identifiers. preAMP reproducibility was also verified (preAMP validation, Supp. data 1,

2).

High-throughput qPCR

High-throughput measurements were conducted on a 96.96 Fluidigm BioMark platform

(Fluidigm, USA). Protocol was as described in Rusnakova et al.

. The cycling program

consisted of activation at 95°C (t = 3 min), followed by 40 cycles of denaturation at 95°C (t =

5 s), annealing at 60°C (t = 15 s) and elongation at 72°C (t = 20 s). After qPCR, melting curves

were measured between 60°C and 95°C with 0.5°C increments. Results were post-processed

based on melting curve inspection.

RESULTS

RTase benchmarking

In order to obtain a comprehensive view on the current state of commercially available RTases

for single-cell applications, the enzymes were benchmarked on templates regularly used in

single-cell RNA-Seq experiments (Figure 1 A). Single-cell or 100-cell bulk templates were

primed with two priming strategies: RT primers mixture (common for RT-qPCR) or oligo(dT)

priming (common for RNA-Seq). Performance of 11 RTases was assessed on 5 assays

targeting ERCC Spike-in molecules of different abundancy, allowing to investigate the range

from rare to highly-abundant transcripts (Table 2). Following the comparison on standardized

samples, the performance of two RTases was further validated on a single-cell model

experiment (High-throughput validation section) (Figure 1 B).

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

Figure 1 Experimental design.

Sensitivity

To assess reaction sensitivity, the rate of positive reactions for low and medium abundant

assays (ERCC 84 and 95) was determined. Reaction output was counted binary – positive or

negative reaction, independent of signal strength. Two main parameters influenced reaction

sensitivity – RT priming strategy and RTase itself.

Medium abundant assay ERCC 95 (88 molecules per RT) reported little differences

between priming strategies (Figure 2 A). For 9 out of 11 RTases recorded at least 80 % of the

reactions positive with both priming strategies. Out of the two remaining enzymes -

SuperScript II and eAMV, only the latter obtained less than 50 % of the reactions positive.

Four times less abundant template ERCC 84 (22 molecules per RT) proved to be more

challenging (Figure 2 B). The median positivity rate across all RTases was 30 %. Maxima H-

and SuperScript IV stood out of the comparison as their sensitivity was strongly enhanced with

RT primers mixture, reaching positivity rates of 90 %. With oligo(dT) priming their sensitivity

dropped to ~45 %. Similarly, the rate of positive reactions with SuperScript III dropped by more

than half when primed with oligo(dT). Notably, with SuperScript II, recommended in some

RNA-Seq protocols

20,27

, only ~20 % of the reactions were positive. The only AMV-derived

RTase in our comparison, eAMV, failed to show any positive reactions.

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

Figure 2 Sensitivity of the reaction is heavily influenced by the choice of RTase and priming strategy.

RTases denoted with “∆” doubled their rate of positive reactions with RT primers mixture. A Reaction positivity for

medium abundant template (n = 10 reactions per RTase). B Reaction positivity for low abundant template (n = 20

reactions per RTase).

Accuracy

Reliable and unbiased RT outcome is critically important for every study. Inconsistent RT

efficiency at any template concentration could potentially lead to false conclusions. To assess

the accuracy of the RTases over a wide concentration range, we merged single-cell and 100-

cell measurements and plotted RNA-molecule concentration (22 to 282,000 specific ERCC

Spike-in copies per RT reaction) versus cDNA concentrations estimated by qPCR. Absolute

cDNA concentrations were estimated using Equation 1 (see Methods), which accounts for

assay-specific differences in efficiency. Coefficients of Determination (R

) served as metric of

accuracy.

R6 + oligo(dT)

Oligo(dT)

ERCC 95 - 88 RNA transcripts per RT

Reverse Transcriptase

Reaction positivity [%]

SuperScript II

SuperScript III

SuperScript IV

SensiScript

PrimeScript

Maxima H-

eAMV

AccuScript

iScript

MMLV

qScript

100

ERCC 84 - 22 RNA transcripts per RT

Reverse Transcriptase

Reaction positivity [%]

SuperScript II



SuperScript III



SuperScript IV

SensiScript

PrimeScript



Maxima H-

eAMV

AccuScript

iScript

MMLV

qScript

100

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

All studied RTases reported accurate performance with both priming strategies. R

for

the RT primers mixture ranged from 0.9463 to 0.9896, and with oligo(dT)

priming the values

ranged from 0.9445 to 0.9825. Reproducibility varied considerably between single-cell and

bulk templates, however the expected performance remained linear across tested RNA input

range (Two representative RTases shown in Figure 3; remaining results are listed in Accuracy

– linearity tab, Supp. data 1). When looking at the reproducibility of RT replicates for each

RTase separately, the major effect of template abundancy was identified. Among tested

RTases, the number of cDNA copies per single-cell reaction varied in median by 40.06 %

(coefficient of variation - CV

), while for bulk samples median variation was substantially

lower at CV

= 10.27 %. The most reproducible RTases in single-conditions were Maxima H-

and SuperScript IV with median CV

of 29.15 % and 30.25 %, respectively. The least

reproducible RT reactions were observed for SuperScript II (CV

= 53.65 %) (Reproducibility

tab, Supp. data 1). In conclusion, although reproducibility of RT decreases with template

concentration, RT is overall substantially accurate even at single-cell levels regardless of used

enzyme or priming strategy.

Figure 3 RT performs with substantial accuracy even at single-cell conditions. Linear regression plots

reporting log

(captured ERCC copies) versus log

(input ERCC copies) for SuperScript II (top) and Maxima H-

(bottom) using RT primers mixture (left) and oligo(dT) primers (right). Graphs display the best linear fit and its 95

% confidence intervals. R

reflects the quality of fitted regression line. Reproducibility decreases at lower template

concentrations.

Yield

Reaction yield is one of the most critical performance metrics. Ideally the reaction would

approach 100 % value, however, previous studies have shown that the actual rate varies

SuperScript II - RT primers mixture

ERCC copies in RT (log

)

Reported ERCC copies (log

)

0.0 1.0 2.0 3.0 4.0 5.0 6.0

0.0

1.0

2.0

3.0

4.0

5.0

6.0

= 0.9583

SuperScript II - Oligo(dT) RT primers

ERCC copies in RT (log

)

Reported ERCC copies (log

)

0.0 1.0 2.0 3.0 4.0 5.0 6.0

0.0

1.0

2.0

3.0

4.0

5.0

6.0

= 0.9579

Maxima H- - RT primers mixture

ERCC copies in RT (log

)

Reported ERCC copies (log

)

0.0 1.0 2.0 3.0 4.0 5.0 6.0

0.0

1.0

2.0

3.0

4.0

5.0

6.0

= 0.9896

Maxima H- -Oligo(dT) RT primers

ERCC copies in RT (log

)

Reported ERCC copies (log

)

0.0 1.0 2.0 3.0 4.0 5.0 6.0

0.0

1.0

2.0

3.0

4.0

5.0

6.0

= 0.9813

A B

C D

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

substantially

15,18,19

. The use of ERCC Spike-ins and DNA standards enabled us to quantify

yields based on known target copy numbers per reaction (Equation 1, see Methods).

For single-cell template, significant differences in yield were observed between

RTases and priming strategies (two-way ANOVA, p

RTase

< 0.001, p

priming

= 0.005 and p

interaction

< 0.001). Reactions primed with RT primers mixture (Figure 4 A) were best processed by

SuperScript IV, Maxima H- and SuperScript III, as they reported average yields of 125 %, 102

% and 88 %, respectively. Theoretically, RTases with the strand displacement capability could

produce multiple cDNA copies from a single RNA transcript, thus reaching yields over 100 %.

The lowest yields with RT primers mixture were 24 % and 7 % for SuperScript II and eAMV,

respectively. With polyA priming, there was no RTase significantly outperforming all others

(Figure 4 B); SuperScript IV and Maxima H- showed the highest (71 % and 66 %) and eAMV

the lowest yield (14 %). To remove the influence of priming strategy and identify the

differences between RTases, one-way ANOVA was used separately for each priming strategy.

The effect of RTase was more pronounced for reactions with RT primers mixture (one-way

ANOVA, explained variation – R

= 0.8212) compared to polyA-targeted priming (one-way

ANOVA, R

= 0.6657). Significant mean differences based on the Tukey post-hoc multiple

comparison test (Tukey HSD) are indicated with letters in Figure 4.

For bulk samples, SuperScript IV and Maxima H- showed the highest yields of 138 %

and 118 % with RT primers mixture, respectively (Figure 4 C). On the other side, the lowest

yields with RT primers mixture were reported with MMLV and SensiScript enzymes (44 % and

43 %), respectively. With oligo(dT) primers, Maxima H-, SuperScript IV, SuperScript III and

qScript were the best scoring RTases yielding 71 %, 67 %, 67 % and 66 %, respectively

(Figure 4 D). SensiScript recorded the lowest yield of 40 %. Differences between RTases were

more prominent with RT primers mixture (one-way ANOVA, p < 0.001, R

= 0.9319) than with

oligo(dT)s (one-way ANOVA, p < 0.001, R

= 0.674).

Overall, the highest cDNA synthesis yields were obtained with SuperScript IV and

Maxima H-. In the opposite spectrum, eAMV was found to be the least yielding RTase. The

application of RT primers mixture resulted in the total yield over 100 % for some RTases, while

with oligo(dT) the maximal value was capped around 75 %.

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

Figure 4 Reaction yield significantly depends on the choice of RTase and priming strategy. Bar plots show

average yield per n = 10 RT replicates (1 RT replicate = average yield across 5 assays) with 95 % CI. Letters

indicate significant mean differences between RTases (p < 0.05, Tukey HSD post hoc comparison), with the highest

yield in bold. “†” marks unique significantly different mean value.

Performance reproducibility

To compare the overall performance of RTases relative to each other and independently of

transcript abundancy, we applied Z-score scaling

. Cq-derived Z-scores in the form of

boxplots (Figure 5) indicate the general RTase performance across all tested parameters

(higher Z-score indicates better performance), as well as the reproducibility of such

performance (spread of values). To highlight the difference in the reproducibility, Z-scores

across priming strategies were combined.

Maxima H- and SuperScript IV were the best performers among the studied RTases.

A median Z-score of 0.99 (interquartile range: 0.78 – 1.25) and 0.84 (0.59 – 1.09) was obtained

for single-cell measurements for SuperScript IV and Maxima H-, respectively. For bulk

measurements, SuperScript IV and Maxima H- reported to have median Z-score of 1.27 (0.94

– 1.62) and 1.18 (0.95 – 1.43), respectively. Among the other RTases, eAMV and SuperScript

II were the least suitable for single-cell conditions, scoring -1.58 (-2.21 - -1.02) and -0.74 (-

1.59 - -0.23) respectively.

The performance reproducibility varied significantly among tested RTases as well

(classical Levene’s test, p < 0.001 for both templates). The least variable performance,

measured as Z-score interquartile range, was recorded for SuperScript IV, Maxima H- and

Single-cell template - RT primers mixture

Reverse Transcriptase

Yield [%]

SuperScript II

SuperScript III

SuperScript IV

SensiScript

PrimeScript

Maxima H-

eAMV

AccuScript

iScript

MMLV

qScript

100

125

150

175

200

e†

Single-cell template - Oligo(dT) RT primers

Reverse Transcriptase

Yield [%]

SuperScript II

SuperScript III

SuperScript IV

SensiScript

PrimeScript

Maxima H-

eAMV

AccuScript

iScript

MMLV

qScript

100

125

150

175

200

Bulk template - RT primers mixture

Reverse Transcriptase

Yield [%]

SuperScript II

SuperScript III

SuperScript IV

SensiScript

PrimeScript

Maxima H-

eAMV

AccuScript

iScript

MMLV

qScript

100

125

150

175

200

A†

B†

Bulk template - Oligo(dT) RT primers

Reverse Transcriptase

Yield [%]

SuperScript II

SuperScript III

SuperScript IV

SensiScript

PrimeScript

Maxima H-

eAMV

AccuScript

iScript

MMLV

qScript

100

125

150

175

200

C D

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

PrimeScript in single-cells (Figure 5 A). In bulk samples, SensiScript, Maxima H- and

AccuScript were the most robust performers (Figure 5 B). Template abundancy had a

significant impact on the performance reproducibility of five RTases (indicated with “*” in Figure

5; classical Levene’s test, p < 0.05, adjusted with Bonferroni’s correction).

In summary, RTases significantly varied in their performance and reproducibility.

Among tested enzymes, Maxima H- and SuperScript IV were found to be superior to their

counterparts in both tested parameters and may be recommended for low RNA input

applications in a wide range of reaction conditions.

Figure 5 Best performing RTases retain consistently better performance even in single-cell conditions. Z-

scores inform about the reproducibility and overall relative performance of RTases in single-cell (A) and 100-cell

reactions (B). Unit of 1 Z-score equals to difference of 1 SD from the assay’s average Cq for given template and

given priming strategy. For enzymes denoted with “*”, performance with single-cell template was significantly less

reproducible than with bulk samples. Performance of the best RTases (scoring the highest Z-scores) was found to

be consistent, seen as small interquartile range.

High-throughput validation

The initial comparison of the 11 RTases allowed us to classify them based on their

performance. However, the comparison was based on artificial templates (ERCC Spike-in)

and the limited number of assays. Therefore, we evaluated two RTases, Maxima H- and

SuperScript II, in a routine high-throughput single-cell RT-qPCR profiling experiment based

on 78 assays (Figure 1 B). The two RTases were selected based on their performance (best-

performer vs below-average performer) and also based on their routine application in single-

cell RNA-Seq protocols

4,9,10,20,27

. In consistency with the first part of this study, two priming

strategies (RT primers mixture and oligo(dT)) were compared.

FASC-sorted astrocytes from healthy and ALS mouse brains were used as single-cell

samples. 15 wild-type and 5 ALS cells were analyzed with each RTase and priming strategy

(Figure 1). To minimize the biological variability, the quality of cells was pretested and only

cells passing the criteria were used for follow-up analysis (see Materials and methods).

Additional quality filter was performed after data acquisition (negative assays were discarded,

followed by melting curve analysis of assay specificity and control for astrocytic markers -

Gja1, Slc1a3 and Aqp4). After data pre-processing, we evaluated the RTases and priming

strategies on positive call rate, expression levels and cluster separation.

Single-cells

Reverse Transcriptase

Z-score

SuperScript II

SuperScript III *

SuperScript IV

SensiScript *

PrimeScript

Maxima H-

eAMV

AccuScript *

iScript

MMLV *

qScript *

-5

-4

-3

-2

-1

Bulks

Reverse Transcriptase

Z-score

SuperScript II

SuperScript III *

SuperScript IV

SensiScript *

PrimeScript

Maxima H-

eAMV

AccuScript *

iScript

MMLV *

qScript *

-5

-4

-3

-2

-1

A B

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

Positive call rate and expression levels

The number of positive reactions with each RTase and priming strategy was evaluated.

Maxima H- showed overall more positive reactions than SuperScript II (Figure 6 A, C). When

using RT primers mixture, Maxima H- yielded more positive calls for 55 % of the assays

(positivity rate increased on average by 12 % per assay); equal positive rate was found for 25

% of the assays and SuperScript II showed more positive calls for 20 % of the assays (Figure

6 A). Using oligo(dT)s the difference in RTase performance was magnified. Maxima H- gave

more positive calls for 74 % of the assays (on average 15 % more positive reactions), 17 % of

the assays had equal positive call rate and SuperScript II had more positive reactions only for

9 % of the assays (Figure 6 C).

Figure 6 Maxima H- delivers more positive reactions and stronger detection signal. The RTase influences

the amount of information obtained in high-throughput single-cell measurements (A,C – RT-qPCR reaction positive

call rate; B,D – Expression levels; Tukey boxplots).

RTase and priming strategy have impact not only on the reaction positivity but also on

the measured quantity as reflected by the Cq values. Consequently, RTase delivering stronger

signal is more favorable in both qualitative and quantitative applications. The expression data

was processed as relative quantities (RQs) calculating the level for each assay relative to the

least abundant reaction within the assay, for each RT priming strategy. Significantly higher

expression levels were obtained with Maxima H- with both priming strategies (Wilcoxon signed

rank test, p < 0.001) (Figure 6 B,D).

Assays

Difference in

assay positivity [%]

SuperScript II

n = 13

SuperScript II = Maxima H-

n = 16

Maxima H-

n = 35

Reverse Transcriptase

Relative quantity [log2]

SuperScript II Maxima H-

n = 64 genes

p-value < 0.001

Assays

Difference in

assay positivity [%]

SuperScript II

n = 5

Maxima H-

n = 43

SuperScript II = Maxima H-

n = 10

Reverse Transcriptase

Relative quantity [log2]

SuperScript II Maxima H-

n = 58 genes

p-value < 0.001

RT primers mixture

Oligo(dT)

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

Clustering

High-throughput gene expression experiments are typically analyzed using multivariate

statistics

29,30

. To test if the RTase also impacts the cluster separation, we applied the common

multivariate tool Principal Component Analysis (PCA). PCA was performed on Cq values that

were firstly transformed to RQs relative to the least expressed sample and then converted to

log

scale. Redundancy analysis (RDA) identified significant importance of the RTase for the

clustering for both priming methods, although values of explained variability (R

) were low

(RDA – covariate: enzyme; p = 0.036, R

= 0.0386 for RT primers mixture and p < 0.001, R

= 0.0825 for oligo(dT)s). As expected, larger proportion of the total variance in the data was

due to cell condition (control vs ALS). RT primers mixture recorded R

= 0.241 and oligo(dT)s

= 0.28 (RDA - covariate: cell treatment). The impact of the RTase may be also quantified

by the Euclidean distances between the two clusters’ centroids. When using the RT primers

mixture, the cluster distance was ~20 % larger when using Maxima H- compared to

SuperScript II (16.69 to 13.64 scores) (Figure 7 A). Using oligo(dT) priming, Maxima H-

separated clusters ~40 % further than SuperScript II (18.23 and 12.97 scores, respectively)

(Figure 7 B).

In summary, the high-throughput experiment demonstrated the impact of RTase on

the outcome of a typical single cell study as well as validated our previous data achieved using

synthetic RNA template. The better performing enzyme achieved higher positive rate and

signals as well as contributed to better separation of two populations of cells.

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

Figure 7 Better performing RTase separates biologically different clusters further apart. Using Maxima H-

clusters of control and diseased cells separate better compared to SuperScript II. Accounted variance by the PCs

is indicated in brackets. Cluster centroid is shown as color-adjusted cross.

Mixture of RT primers

PC1 (72.32 %)

PC2

(9.03 %)

-31 -28 -25 -22 -19 -16 -13 -10 -7

-7

-4

-1

16.69

13.64

Oligo(dT)

PC1 (68.49 %)

PC2

(11.55 %)

-28 -24 -20 -16 -12 -8 -4

-6

-2

18.23

12.97

Maxima H- ALS

Maxima H- Controls

SuperScript II ALS

SuperScript II Controls

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

DISCUSSION

In this study, we systematically evaluated the performance of 11 commercial RTases with

focus on RNA-based single-cell quantitative protocols. Key findings include that: (i) accuracy

of RT is generally high over wide range of template concentrations regardless of RTase and

priming strategy; (ii) the sensitivity of majority of tested RTases is sufficient and comparable

for more-abundant templates, but only few maintain sensitivity with low-abundant template;

(iii) the high sensitivity of these RTases is accompanied with high reproducibility and therefore

are preferable for single-cell applications; (iv) RTase performance is dependent on priming

strategy; (v) usage of better RTase significantly improves the gene detection rate from single

cells and improves definition of biologically distinct single-cell populations.

For single-cell expression profiling, the ability of RTase to capture transcripts present

in low concentrations is of prime importance. Using a set of spike-in molecules enabled us to

test this parameter over a wide range of concentrations (22 to 2822 molecules per RT

reaction). With exception of eAMV, the reaction positivity rate of the RTases does not change

considerably over the medium-to-high concentration range (>88 transcripts per reaction;

Figure 2 A), but differs substantially for low-abundant transcripts (Figure 2 B). Considering that

genes are expressed on median between ~3 and ~100 copies per cell

(~100 copies for

protein-coding, ~10 for splicing regulators and only ~3 for transcription factors), our results

show that the choice of RTase represents a strong variable in the sensitivity of single-cell

quantitative experiments. As the detection of low-abundant transcripts may be affected by

sampling noise, the increased number of replicates we used for lowest-abundant transcript (n

= 20) should minimize this factor and capture the trends in relative RTase performance.

Importantly, the influence of sampling noise is insignificant over 35 template copies

, which

highlights the poor performance of the AMV-based enzyme (Figure 2 A). Poor AMV

performance has been noted before

11,15–17

and has been attributed to its dimeric structure

Comparing priming strategies, we observed increased positivity rate (Figure 2 B) and

increased yield (Figure 4 A,C) when using RT primers mix compared to oligo(dT) only.

However, two confounding factors may contribute to this conclusion. First, polyA-based

priming of ERCC Spike-in is known to underestimate real oligo(dT) priming efficiency

as the

ERCC Spike-ins have unnaturally short polyA tails (20-30 bp)

. Secondly, different

concentration of RT primers mixture and oligo(dT) primers only was used throughout our study

(to mimic their typical usage in RT-qPCR and RNA-Seq studies) which may contribute to the

lower positivity rate as well

19,34

. Therefore, although different outcomes using two priming

strategies are apparent, we caution against the simplified interpretation that the RT primers

mix always outperforms oligo(dT) primers

6,35

The ability of RTases to reverse transcribe mRNA templates with the same efficiency

irrespective of template concentration was tested on a range from 22 to 282,200 ERCC Spike-

in molecules per RT. All enzymes retained linear performance across tested range (Figure 3,

Accuracy – linearity, Supp. data 1 tab), demonstrating that RTase can reliably reflect RNA

content for templates of varying abundancy

, with particular enzyme-specific efficiency (see

Yield, Figure 4). The variability of RT yield however increases with decreasing template

concentration (Reproducibility tab, Supp. data 1)

. Although all RTases performed with

substantial accuracy, better performing RTases showed better efficiency, were more

reproducible and sensitive at single-cell level, which makes them more suitable candidates for

single-cell applications. Some early reports proposed that RT efficiency is gene- or template

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

concentration-dependent

6,15,34,35

. Our findings are however in accordance with more recent

reports

8,11

that suggest RT is comparably efficient for different assays and template

concentration.

The RT yield showed large variation across tested enzymes and was priming strategy-

dependent (Figure 4). More pronounced differences between yields of various enzymes with

RT primers mixture suggest that especially SuperScript IV and Maxima H- can markedly

benefit from increased number of priming locations

. Multiple priming locations in combination

with strand displacement activity, lack of RNase H function, high processivity and strong

template binding affinity enable synthesis of more than just one cDNA copy from one RNA

transcript, thus yields > 100 %.

Differences in RT efficiency between single-cell and bulk templates were minor and

relatively consistent with exception of SuperScript II and eAMV enzymes (Figure 4). The worse

performance observed with SuperScript II and eAMV on the single-cell template could be

expected, as the RNA amount was below the recommended template range by the

manufacturers. Considering this, it is surprising that several popular single-cell RNA-Seq

protocols (CEL-Seq2

, Smart-Seq2

) are based on SuperScript II, suggesting there is a

potential for their improvement. Apart the discussed high and low efficient RTases, the cDNA

synthesis yield was typically in the range 50 – 80 %, which is in line with earlier studies

8,15–

17,23,34

. Only contradictory result is reported by Levesque-Sergerie et al. on SuperScript II and

III RTases

, which may be attributed to imprecise preparation of qPCR standard curves as

discussed by Miranda & Steward

Our study also reveals the relationship between performance of particular RTases and

their intrinsic biochemical properties (Table 1, Figure 5). The best-performing RTases,

SuperScript IV and Maxima H-, were thermostable, allowing to utilize higher reaction

temperature in the protocols. Previous reports have suggested that destabilization of

secondary RNA structures at elevated temperature leads to more frequent primer

hybridization and stable reverse transcription

6,36

. Our results support these hypotheses.

Interestingly, the majority of RTases employed protocols with the pre-incubation step aiming

to increase the efficiency of primer binding. The exception in our selection was iScript and

SensiScript. Both enzymes achieved relatively lower performance demonstrating the

importance of the pre-incubation step. However, if loss of sensitivity is not an issue, simplified

pipetting protocol, reduced possibility for contamination and errors may be advantageous

factors. RNase H activity was not observed to have any profound effect on the reaction

outcome, as previously reported

11,17

. The important factor for selection of RTase is also its

price. This is relevant especially for single cell RNA-Seq studies, where thousands of cells are

typically analyzed and price for RT is an important part of the budget. From this perspective,

Maxima H- may be a recommendable choice for many researchers for its high performance

and low price (second lowest in our comparison, Table 1). Notably, Maxima H- possesses also

the terminal transferase activity utilized in some RNA-Seq protocols

8–10

(mcSCRB-Seq, STRT-

Seq, SMARTer, Smart-Seq2).

Typically, high-throughput experimental results require reduction of their

multidimensional composition. PCA, the commonly used method, reduces dimensionality by

searching for the largest portions of variation in the data set. Observed variation, however,

arises not only from the biology of the experiment but also from technical factors of the

measurement, including RT

. In our model experiment, we hypothesized that higher

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

sensitivity will enhance separation of biologically distinct single-cell clusters by reducing the

frequency of missing values and sampling noise

. Indeed, the number of positive reactions

per assay as well as the expression levels were increased using Maxima H- compared to

SuperScript II regardless of the priming strategy (Figure 6). The improvement is also reflected

by increase in the cluster distances, as hypothesized (Figure 7). To our knowledge, this is the

first study that demonstrated the direct effect of RT on separation of distinct groups of single

cells in multidimensional expression analysis.

The aim of our study was to evaluate the current state of commercially available

RTases for the growing field of single cell transcriptomics. We showed substantial differences

in the performance of RTases highlighting the importance of their selection. For the first time,

we also demonstrated the impact of RTase on the outcome of a typical single cell study. We

believe that this study will initiate follow-up efforts to characterize other aspects of the RT

reaction and will lead to the improvements of existing workflows.

Acknowledgments

We would like to acknowledge the funding sources for this work: P303-19-02046S,

CZ.1.05/1.1.00/02.0109 and RVO 86652036.

REFERENCES

1. Arezi B, Hogrefe H. Novel mutations in Moloney Murine Leukemia Virus reverse

transcriptase increase thermostability through tighter binding to template-primer.

Nucleic Acids Res. 2009;37(2):473-481. doi:10.1093/nar/gkn952

2. Álvarez M, Menéndez-Arias L. Temperature effects on the fidelity of a thermostable

HIV-1 reverse transcriptase. FEBS J. 2014;281(1):342-351. doi:10.1111/febs.12605

3. Baranauskas A, Paliksa S, Alzbutas G, et al. Generation and characterization of new

highly thermostable and processive M-MuLV reverse transcriptase variants. Protein

Eng Des Sel. 2012;25(10):657-668. doi:10.1093/protein/gzs034

4. Macosko EZ, Basu A, Satija R, et al. Highly parallel genome-wide expression profiling

of individual cells using nanoliter droplets. Cell. 2015;161(5):1202-1214.

doi:10.1016/j.cell.2015.05.002

5. Mohr S, Ghanem E, Smith W, et al. Thermostable group II intron reverse transcriptase

fusion proteins and their use in cDNA synthesis and next-generation RNA sequencing.

Rna. 2013;19(7):958-970. doi:10.1261/rna.039743.113

6. Ståhlberg A, Håkansson J, Xian X, Semb H, Kubista M. Properties of the Reverse

Transcription Reaction in mRNA Quantification. Clin Chem. 2004;50(3):509-515.

doi:10.1373/clinchem.2003.026161

7. Deprez RHL, Fijnvandraat AC, Ruijter JM, Moorman AFM. Sensitivity and accuracy of

quantitative real-time polymerase chain reaction using SYBR green I depends on cDNA

synthesis conditions 10.1016/S0003-2697(02)00021-0 : Analytical Biochemistry |

ScienceDirect.com. 2002;307:63-69.

http://www.sciencedirect.com/science/article/pii/S0003269702000210.

8. Svensson V, Natarajan KN, Ly LH, et al. Power analysis of single-cell RNA-sequencing

experiments. Nat Methods. 2017;14(4):381-387. doi:10.1038/nmeth.4220

9. Bagnoli JW, Ziegenhain C, Janjic A, et al. Sensitive and powerful single-cell RNA

sequencing using mcSCRB-seq. Nat Commun. 2018;9(1):2937. doi:10.1038/s41467-

018-05347-6

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

10. Ziegenhain C, Vieth B, Parekh S, et al. Comparative Analysis of Single-Cell RNA

Sequencing Methods. Mol Cell. 2017;65(4):631-643.e4.

doi:10.1016/j.molcel.2017.01.023

11. Sieber MW, Recknagel P, Glaser F, et al. Substantial performance discrepancies

among commercially available kits for reverse transcription quantitative polymerase

chain reaction: A systematic comparative investigator-driven approach. Anal Biochem.

2010;401(2):303-311. doi:10.1016/j.ab.2010.03.007

12. Androvic P, Valihrach L, Elling J, Sjoback R, Kubista M. Two-tailed RT-qPCR: a novel

method for highly accurate miRNA quantification. Nucleic Acids Res. 2017;45(15):e144.

doi:10.1093/nar/gkx588

13. Chen C, Ridzon DA, Broomer AJ, et al. Real-time quantification of microRNAs by stem-

loop RT-PCR. Nucleic Acids Res. 2005;33(20):1-9. doi:10.1093/nar/gni178

14. Raymond CK, Roberts BS, Garrett-Engele P, Lim LP, Johnson JM. Simple, quantitative

primer-extension PCR assay for direct monitoring of microRNAs and short-interfering

RNAs. RNA. 2005;11(11):1737-1744. doi:10.1261/rna.2148705

15. Ståhlberg A. Comparison of Reverse Transcriptases in Gene Expression Analysis. Clin

Chem. 2004;50(9):1678-1680. doi:10.1373/clinchem.2004.035469

16. Okello JBA, Rodriguez L, Poinar D, et al. Quantitative assessment of the sensitivity of

various commercial reverse transcriptases based on armored HIV RNA. PLoS One.

2010;5(11). doi:10.1371/journal.pone.0013931

17. Lindén J, Ranta J, Pohjanvirta R. Bayesian modeling of reproducibility and robustness

of RNA reverse transcription and quantitative real-time polymerase chain reaction. Anal

Biochem. 2012;428(1):81-91. doi:10.1016/j.ab.2012.06.010

18. Levesque-Sergerie JP, Duquette M, Thibault C, Delbecchi L, Bissonnette N. Detection

limits of several commercial reverse transcriptase enzymes: Impact on the low- and

high-abundance transcript levels assessed by quantitative RT-PCR. BMC Mol Biol.

2007;8. doi:10.1186/1471-2199-8-93

19. Miranda JA, Steward GF. Variables influencing the efficiency and interpretation of

reverse transcription quantitative PCR (RT-qPCR): An empirical study using

Bacteriophage MS2. J Virol Methods. 2017;241:1-10.

doi:10.1016/j.jviromet.2016.12.002

20. Picelli S, Faridani OR, Björklund AK, Winberg G, Sagasser S, Sandberg R. Full-length

RNA-seq from single cells using Smart-seq2. Nat Protoc. 2014;9(1):171-181.

doi:10.1038/nprot.2014.006

21. Soumillon M, Cacchiarelli D, Semrau S. Characterization of directed differentiation by

high-throughput single-cell RNA-Seq. bioRxiv. 2014.

doi:http://dx.doi.org/10.1101/003236

22. Ståhlberg A, Bengtsson M. Single-cell gene expression profiling using reverse

transcription quantitative real-time PCR. Methods. 2010;50(4):282-288.

doi:10.1016/j.ymeth.2010.01.002

23. Bustin S, Dhillon HS, Kirvell S, et al. Variability of the reverse transcription step:

Practical implications. Clin Chem. 2015;61(1):202-212.

doi:10.1373/clinchem.2014.230615

24. External RNA Controls Consortium. Proposed methods for testing and selecting the

ERCC external RNA controls. BMC Genomics. 2005;6(1):150. doi:10.1186/1471-2164-

6-150

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint

25. Applied_Biosystems. ERCC RNA Spike-In Control Mixes User Guide. Thermo Fish.

2016;(4456740):1-28. papers2://publication/uuid/6D4FA07D-342A-49FA-A131-

90E92E6A6C0B.

26. Rusnakova V, Honsa P, Dzamba D, Ståhlberg A, Kubista M, Anderova M.

Heterogeneity of Astrocytes: From Development to Injury - Single Cell Gene

Expression. PLoS One. 2013;8(8). doi:10.1371/journal.pone.0069734

27. Hashimshony T, Senderovich N, Avital G, et al. CEL-Seq2: Sensitive highly-multiplexed

single-cell RNA-Seq. Genome Biol. 2016;17(1):1-7. doi:10.1186/s13059-016-0938-8

28. Kohl M. Introduction to Statistical Data Analysis with R. London; 2015.

www.sylvania.com. Accessed April 25, 2019.

29. Bergkvist A, Rusnakova V, Sindelka R, et al. Gene expression profiling - Clusters of

possibilities. Methods. 2010;50(4):323-335. doi:10.1016/j.ymeth.2010.01.009

30. Ringnér M. What is principal component analysis? Nat Biotechnol. 2008;26(3):303-304.

doi:10.1038/nbt0308-303

31. Marinov GK, Williams BA, McCue K, et al. From single-cell to cell-pool transcriptomes:

Stochasticity in gene expression and RNA splicing. Genome Res. 2014;24(3):496-510.

doi:10.1101/gr.161034.113

32. Ståhlberg A, Kubista M. The workflow of single-cell expression profiling using

quantitative real-time PCR. Expert Rev Mol Diagn. 2014;14(3):323-331.

doi:10.1586/14737159.2014.901154

33. Konishi A, Yasukawa K, Inouye K. Improving the thermal stability of avian

myeloblastosis virus reverse transcriptase α-subunit by site-directed mutagenesis.

Biotechnol Lett. 2012;34(7):1209-1215. doi:10.1007/s10529-012-0904-9

34. Schwaber J, Andersen S, Nielsen L. Shedding light : The importance of reverse

transcription efficiency standards in data interpretation. 2019;17(December 2018):1-7.

doi:10.1016/j.bdq.2018.12.002

35. Bustin SA, Nolan T. Pitfalls of quantitative real- time reverse-transcription polymerase

chain reaction. J Biomol Tech. 2004;15(3):155-166. doi:15/3/155 [pii]

36. Nolan T, Hands RE, Bustin SA. Quantification of mRNA using real-time RT-PCR. Nat

Protoc. 2006;1(3):1559-1582. doi:10.1038/nprot.2006.236

.CC-BY-NC-ND 4.0 International licenseunder a

certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available

The copyright holder for this preprint (which was notthis version posted May 7, 2019. ; https://doi.org/10.1101/629097doi: bioRxiv preprint