Comparing Novel Low-Input Total RNA-Seq Methods

ABSTRACT

Until recently, performing RNA-sequencing (RNA-seq) at low input often required specialized library preparation kits. The SMARTer® Stranded Total RNA-Seq Kit v3—Pico Input Mammalian workflow offered by Takara Bio is a popular option for generating high-quality libraries from low-input material. Recently, two RNA-seq library preparation kits, the Watchmaker RNA Library Prep Kit with Polaris Depletion by Watchmaker Genomics and the sparQ RNA-Seq HMR Kit by Quantabio, that offer a larger input range became available. These kits may offer a more accessible and affordable workflow for low-input RNA-seq; therefore, to determine their suitability for low-input applications, a comparative evaluation of the kits was performed using the market standard Takara Pico kit for benchmarking. To do this, two replicates of three total RNA samples of varied quality were prepared at three inputs (250 pg, 1 ng, and 10 ng) using each workflow. Paired-end 50 bp sequencing was performed on a NovaSeq 6000 to obtain 13 million reads per library. While both novel kits generated next-generation sequencing libraries at all inputs tested, neither workflow was specifically tailored for low inputs. We observed variability in library diversity, proportion of duplicate reads, types of transcripts detected, sensitivity of detection, and proportion of nuclear rRNA reads. This suggests that further optimization is required to obtain high-quality libraries from 250 pg to 10 ng input amounts using the novel workflows.

INTRODUCTION

Over the past decade, RNA-sequencing (RNA-seq) has become routine in molecular biology research. The increased accessibility of next-generation sequencing (NGS) technology has been driven, in part, by the development of commercially available library preparation kits. Historically, RNA-seq library preparation kits have been constrained by the quality and integrity of available input material.¹ Standard total RNA-seq library preparation protocols require a large amount of input material that is generally intact. RNA-seq library generation from input materials of insufficient quantity or integrity can result in high duplication rates, decreased library complexity, insufficient coverage, and, in extreme cases, outright failure.² These outcomes have been particularly prohibitory to research that relies on the use of challenging samples. Clinical research samples, for example, are often precious, with no option to re-isolate the RNA. Other samples may only be available from formalin-fixed paraffin-embedded (FFPE) tissues, which are notoriously degraded. As a result, researchers have had to accept inevitable trade-offs in data quality to proceed with sequencing low-quantity or poor-quality samples.

Over time, vendors have sought to provide solutions for these challenges. Advancements such as unique molecular identifiers (UMIs) and ribonuclease H–based ribosomal RNA (rRNA) depletion methods have helped mitigate the negative effects seen in low-input libraries, and vendors have developed library preparation products specifically tailored for low-input applications.^3,4,5 The SMARTer Stranded Total RNA-Seq Kit v3—Pico Input Mammalian (Takara Bio, Cat. No. 634487, Kusatsu, Shiga, Japan) workflow is a strand-specific total RNA library preparation kit for low-input mammalian samples that utilizes a proprietary switching mechanism at the 5’ end of the RNA template (SMART) method for complementary DNA (cDNA) synthesis.⁴ The SMART method incorporates adapters during cDNA synthesis, avoiding the need for adapter ligation steps, which are increasingly inefficient at low input. The SMARTer Pico kit also employs a novel probe-based rRNA depletion method. In this method, R-Probes target species-specific abundant rRNA regions in the cDNA after cDNA synthesis and tag them for depletion, unlike other methods that deplete rRNA before beginning cDNA synthesis. As rRNA transcripts comprise approximately 90% of a total RNA sample, initial rRNA depletion methods are prone to leaving insufficient material for cDNA synthesis when starting with low-input amounts.^1,2,3 The SMARTer Pico kit allows generation of high-quality libraries with as little as 250 pg input RNA, and it is regarded as the market standard for low-input applications.

Unsurprisingly, the SMARTer Pico kit’s superior low-input capabilities come at a cost. Specialized library preparation kits, including the SMARTer Pico kit, tend to be more expensive per reaction when compared with the current market standards for mid- to high-input workflows. Furthermore, the SMARTer Pico kit protocol is also more labor-intensive than most standard workflows. For example, the popular KAPA RNA HyperPrep Kit with RiboErase (Roche Molecular Systems, Cat. No. KK8561, Pleasanton, California, USA) workflow for standard input RNA-seq can be completed in approximately 4 hours and can be fully automated. Meanwhile, the SMARTer Pico kit workflow requires a minimum of 7.5 hours to complete and cannot be fully automated.

Recently, several new strand-specific total RNA library preparation kits with larger input ranges have become commercially available. They include the Watchmaker RNA Library Prep Kit with Polaris Depletion (Watchmaker Genomics, Cat. No. BK002-096, Boulder, Colorado, USA) and the sparQ RNA-Seq HMR Kit (Quantabio, PN 95216-096, Beverly, Massachusetts, USA). The novel kits and the Takara SMARTer Pico kit are comparable in cost when all required components are considered. The Quantabio and Watchmaker kits claim to produce high-quality libraries from 1 ng to 1 µg of input RNA. The workflows for both kits are fully automatable and can be completed in approximately 4.5 hours. The flexible input ranges and shorter workflows offered by the Quantabio and Watchmaker kits make them an attractive alternative to other low-input library preparation kits with similar price points.

This study systematically compared three commercially available strand-specific total RNA library preparation kits to determine if any of the novel workflows are suitable for low-input applications when compared with the current market standard. Using human RNA samples of varying integrity, the kits were used to prepare libraries at different input amounts. This allowed comparisons for a range of conditions to be made and provided deeper insight into appropriate use for each kit. Results were evaluated on many factors, including duplication rate, diversity of library composition, proportion of nuclear rRNA, types of transcripts detected, sensitivity of detection, and proportion of uniquely aligned reads.

METHODS AND MATERIALS

Study design

To assess the performance of each kit, three human RNA samples were chosen and aliquoted. Each sample was designated low, medium, or high quality based on RNA integrity number (RIN). Samples were named RIN_low, RIN_med, and RIN_high accordingly. RINs were as follows: RIN_low (3.3), RIN_med (7.0), and RIN_high (10.0). For each kit, the three samples were prepared in technical duplicate using 250 pg, 1 ng, and 10 ng of input RNA for a total of 18 libraries per chemistry. Each kit was compared across varied input quantities and qualities. The experimental design is reported graphically in Figure 1.

Figure 1. Schematic of study design. RIN_low, RIN_med, and RIN_high were processed using three different total RNA library preparation kits. Samples were prepared in technical duplicate from three different input amounts with each kit. All libraries were sequenced on the NovaSeq 6000 at 2 x 50 bp for a minimum 13 million paired-end reads.

Input RNA preparation

Input RNA was derived from the lymphoblastoid cell line EBVLCL-nM1 (Supp. File 1). RIN_low and RIN_med were prepared using the AllPrep DNA/RNA FFPE Kit (Qiagen, Cat. No. 80234, Hilden, Germany) according to the manufacturer’s instructions (Qiagen HB-0373-006), with the following modifications to the protocol: skip steps 1-4 to account for dry pellet storage. RIN_high was prepared with the RNeasy Plus Universal Kit (Qiagen, Cat. No. 73404) according to the manufacturer’s instructions (Qiagen HB-0391-003), with the following modifications to the protocol: in step 13, after discarding flowthrough, perform the optional DNase digest (Qiagen, Cat. No. 79254) according to the supplementary protocol (Qiagen HB-0456-004). Samples were assessed for RIN by the Agilent Bioanalyzer RNA 6000 Nano Chip (Agilent Technologies, PN 5067-1511, Santa Clara, California, USA) as presented in Figure 2A. Aliquots of each sample were stored at −80˚C until use.

Figure 2. Input RNA and final library quality assessments. A) Samples were assessed for RNA integrity number (RIN) using the Agilent Bioanalyzer RNA 6000 Nano Chip. Bioanalyzer traces for each sample are shown. Labels above each trace indicate the sample name and assessed RIN value. B) Bioanalyzer traces of select libraries are shown. For each kit, the highest-quality library that was generated is shown before additional cleanups. Labels above each trace indicate the kit name and input amount used to generate the library; the RIN score is also provided. C) Bioanalyzer traces of select libraries are shown. For each kit, replicate A generated from 250 pg of sample RIN_low (RIN 3.3) is shown. Labels above each trace indicate the kit name and input amount used to generate the library; the RIN score is also provided. All library traces not shown can be found in Supplemental Files 2-7.

Protocol normalization

For each kit, web conferences or email communications were arranged with the vendor before library generation. The standard protocols were reviewed in detail, and appropriate modifications for the low-input application were discussed. A single technician prepared all libraries according to the manufacturer’s protocol with the recommended modifications.

Library construction

All libraries generated by a single chemistry were prepared simultaneously. Additional thermocyclers were utilized to accommodate distinct fragmentation conditions and polymerase chain reaction (PCR) cycle numbers. Library preparation conditions are summarized in Table 1. Quality and quantity of the finished libraries were assessed using a combination of the Agilent DNA High Sensitivity Chip (Agilent Technologies, PN 5067-4626, Santa Clara, California, USA) and QuantiFluor dsDNA System (Promega Corp., Cat. No. E2670, Madison, WI, USA). For each chemistry, kit specifications and protocol details are briefly described below.

Table 1. Library preparation conditions.

Kit	Input range supported	Fragmentation temperature	Fragmentation time		Amplification PCR cycles		Additional cleanups performed
SMARTer Pico	250 pg to 10 ng	94˚C	RIN_low	3 minutes	250 pg	16	N/A
			RIN_med	4 minutes	1 ng	14
			RIN_high	4 minutes	10 ng	12
Watchmaker	1 ng to 1000 ng	85˚C	RIN_low	2 minutes	250 pg	21	0.75x
			RIN_med	3 minutes	1 ng	21
			RIN_high	5 minutes	10 ng	17
sparQ	1 ng to 1000 ng	94˚C	RIN_low	1 minute	250 pg	17	0.75x
			RIN_med	2 minutes	1 ng	17
			RIN_high	2 minutes	10 ng	17

Takara Bio’s SMARTer Stranded Total RNA-seq Kit v3—Pico Input Mammalian

The recommended input range for this kit is 250 pg to 10 ng of mammalian total RNA (Takara Bio, protocol v.120523, Kusatsu, Shiga, Japan). The kit is suitable for degraded or intact RNA, including material extracted from FFPE samples. A DV200 of >50% is recommended for optimal assay performance. Libraries were prepared according to the manufacturer’s protocol with the following modifications. Steps C-F: KAPA Pure Beads (Roche Molecular Systems, PN 07983271001, Pleasanton, California, USA) were used for library purification and size selection. Step F1: 90 µL of KAPA Pure Beads were used for all samples. Step F13: 18 µL of KAPA Pure Beads were used. Libraries were prepared over 2 days using the 4˚C incubation in step A6 as a stopping point. Fragmentation at 94˚C was performed for 3 minutes for RIN_low, 4 minutes for RIN_med, and 4 minutes for RIN_high. Libraries were prepared using Takara Bio’s SMARTer RNA Unique Dual Index Kit—96U Set B (Takara Bio, Cat. No. 634457, Kusatsu, Shiga, Japan). Library indexing PCR was performed using 5 cycles for all samples. Library amplification PCR was performed using 16 cycles for 250 pg input libraries, 14 cycles for 1 ng input libraries, and 12 cycles for 10 ng input libraries.

Watchmaker Genomics’s RNA Library Prep Kit with Polaris Depletion

The recommended input range for this kit is 1 ng to 1000 ng of human, mouse, or rat total RNA. The kit is suitable for degraded or intact RNA, including material extracted from FFPE samples. Libraries were prepared according to the manufacturer’s protocol (Watchmaker Genomics, PTD-26 WMUG210 v.1.0.0722, Boulder, Colorado, USA) with the following modifications. Step A7.4: 3.33 µL of diluted adapter was added to each tube. Libraries were prepared over 2 days using the optional stopping point after the post-ligation cleanup. Fragmentation at 85˚C was performed for 2 minutes for RIN_low, 3 minutes for RIN_med, and 5 minutes for RIN_high. Libraries were prepared using xGen Stubby Adapter-UDI Primers (Integrated DNA Technologies, Cat. No. 10005921, Newark, New Jersey, USA). Adapters were diluted to 0.2 µM based on Table A3 in the manufacturer protocol. The optional second post-ligation cleanup was not performed. Library amplification PCR was performed using 21 cycles for 250 pg and 1 ng input libraries and 17 cycles for 10 ng input libraries. The Bioanalyzer results revealed adapter dimer present in many samples. A single-sided size selection at 0.75× volume ratio was performed on all libraries using KAPA Pure Beads to remove the excess adapter dimer.

Quantabio’s sparQ RNA-seq HMR Kit

The recommended input range for this kit is 1 ng to 1000 ng of human, mouse, or rat total RNA. The kit is suitable for degraded or intact RNA, including material extracted from FFPE samples. A minimum of 10 ng input is recommended for degraded samples for optimal performance. Libraries were prepared in 1 day according to the manufacturer’s protocol (Quantabio 95216 / IFU-136.1 REV 02) with the following modifications. Step 25: 3.33 µL of diluted adapter was added to each sample. Fragmentation at 94˚C was performed for 1 minute for RIN_low, 2 minutes for RIN_med, and 2 minutes for RIN_high. Libraries were prepared using Quantabio’s sparQ UDI Adapters (Quantabio, PN 95211-096, Beverly, Massachusetts, USA). Adapters were serially diluted (1:200) based on Table 7 in the manufacturer protocol. Library amplification PCR was performed using 17 cycles for all inputs. The Bioanalyzer results revealed adapter dimer in many samples. To avoid loss, libraries were pooled according to the 200-1000 bp region Bioanalyzer quantifications to be as equimolar as possible. A single-sided size selection at 0.75× volume ratio was performed on the pool using KAPA Pure Beads to remove the excess adapter dimer.

Sequencing

Using an S2 100-cycle sequencing kit (v1.5), 2 x 50 bp paired-end sequencing was performed on an Illumina NovaSeq 6000 sequencer (Illumina Inc., San Diego, California, USA) to a minimum depth of 13 million reads per library. Base calling was done by Illumina RTA3, and output was de-multiplexed and converted to fastq format with Illumina Bcl2fastq v2.20.0.

Alignment and data processing

RNA-seq data were processed and analyzed as follows. Briefly, all fastqs were down-sampled using seqtk⁶ to the lowest number of reads of any sample (13,340,949 reads). Latent Illumina adapter sequences were identified and removed from input 50-mer RNA-seq data using Cutadapt⁷ implemented within TrimGalore.⁸ Additionally, 14 bp (UMIs, linker, and adapter) were removed from the 5’ end of Takara R2 reads using the --clip_R2 14 option, to eliminate any UMI-related differences between the kits. Adapter-free RNA-seq reads were aligned to the hg38 reference genome sequence using STAR⁹ in 2-pass mapping mode, also quantifying count reads per gene. Salmon quant (1.10.0) was run on the trimmed fastqs with an index built from hg38, with flags -l A and --validateMappings. The data for the gene body coverage figure were generated with RSeQC (5.0.1), using the geneBody_coverage.py on the genes listed within GENCODE.v38.bed. To assess diversity in predicted future library complexity, the preseq¹⁰ (3.2.0) lc_extrap function was used on each of the aligned and sorted bams, using the “MAX_INSERT_SIZE” calculated by GATK’s¹¹ (4.3.0.0) CollectInsertSizeMetrics as the seg_len. Reads per gene from STAR were fed into R¹² using the Bioconductor¹³ package edgeR’s¹⁴ readDGE, and transcript per million (TPM) data were added from Salmon with the tximport function from the tximport¹⁵ (1.26.1) R package. For the gene type comparisons, gene count data in the SummarizedExperiment object were annotated with gene type information from the GRCh38 GENCODE gtf file (version 33; Ensembl 99).

The total deduplicated percentage metric from fastQC (0.12.1)¹⁶ was used as the basis for the percentage duplicate reads comparison. The duplication fraction was modeled with meta regression within glmmTMB,¹⁷ with an interaction between kit and input, and an interaction between binned RIN and read. To compare extrapolated library complexity between kits, the expected distinct number of reads—up to 200 million—was modeled with a negative binomial model (family = nbinom2) within glmmTMB, with an interaction between kit and input, and binned RIN as a fixed effect, with an offset of the log of total reads. The number of expected distinct reads and total reads were both divided by 1 x 10⁶ during modeling. The uniquely_mapped_percent metric (reads that map to only one location) from STAR was used to compare percentage unique alignment between kits. This metric was modeled using a beta regression model within glmmTMB, with an interaction between kit and input, and binned RIN as a fixed effect. For the nuclear rRNA comparison, the percentage of all RNAs that are nuclear (non-mitochondrial) rRNA was modeled using a beta regression model within glmmTMB, with an interaction between kit and input, and binned RIN as a fixed effect. A ziformula of input only was used for zeros in the data that otherwise are not modeled appropriately. Similarly, the percentage of mitochondrial reads is modeled with a beta regression model within glmmTMB, with an interaction between kit and input, and binned RIN as a fixed effect. Each of the top five most prevalent RNA subtypes in these samples from GENCODE’s “gene_type” were tested using a beta regression model within glmmTMB, with an interaction between kit and input, and binned RIN as a fixed effect. Within each sample, each gene was classified as either “lowly expressed” (0.1-1 TPM) or “highly expressed” (≥1 TPM). To determine the sensitivity of each kit, the glmmTMB package was used to generate a negative binomial (family = nbinom1) model for each expression level category. The number of genes for each sample was modeled with an interaction between kit and input, with binned RIN as a fixed effect. The log of the total number of genes per sample was used as an offset.

The estimated marginal means were computed using the emmeans function from the emmeans¹⁸ (1.8.9) package, which provided the odds ratios and p values for each comparison, with results given on the response scale. RNA subtype comparison p values were adjusted for false discovery rate together; all other p values were adjusted using the Tukey method.

RESULTS

This study evaluated the performance of three commercially available RNA-seq library preparation kits. The kits reviewed were the SMARTer RNA-Seq Kit v3 for Pico Input (Takara Bio), the Watchmaker RNA Library Prep Kit with Polaris Depletion (Watchmaker Genomics), and the sparQ RNA-Seq HMR Kit (Quantabio). Each kit integrates rRNA depletion and produces NGS libraries that retain strand-specificity. The Takara kit utilizes a probe-based rRNA depletion method in which species-specific probes that target rRNA regions are hybridized to the cDNA, tagging them for subsequent enzymatic digestion. This method requires a bead cleanup prior to proceeding. The Watchmaker kit utilizes a similar probe-based rRNA depletion method in which species-specific probes target and tag rRNA transcripts for a subsequent enzymatic digestion. However, this digestion is performed on the input RNA prior to cDNA synthesis, and a bead cleanup is required prior to proceeding. The rRNA depletion method utilized by the Quantabio kit is proprietary and is not described in the manufacturer’s documentation. The method combines rRNA depletion and fragmentation into one step and does not require a bead cleanup prior to proceeding with cDNA synthesis.

To assess the performance of each kit, we used RNA of varying quality derived from the same lymphoblastoid cell line. The cohort was composed of one highly degraded (RIN_low), one partially degraded (RIN_med), and one intact (RIN_high) sample (Figure 2A). All library preparations utilized a negative control to eliminate confounders introduced by possible contamination. All negative control libraries showed no significant amplification and therefore were not sequenced. As shown in Figure 1, each chemistry was used to prepare the three samples in technical duplicate for three input amounts: 250 pg, 1 ng, and 10 ng. All libraries were prepared, assessed for quantity and sizing, and pooled for sequencing by a single technician. In total, 54 libraries were sequenced across two NovaSeq6000 S2 flow cells.

Each kit was first evaluated on whether it was able to successfully produce sequencing libraries at all three input amounts from degraded and intact input material (Figure 2B-C). Each chemistry successfully produced library material of sufficient quantity and appropriate sizing from all samples at each input level. However, the Takara kit was the only chemistry that generated libraries ready for sequencing without extra bead cleanups. All libraries that were generated by the Watchmaker and Quantabio kits contained excessive (>1%) adapter dimer, which necessitated additional bead-based cleanups before sequencing. Additional bead cleanups may lower the amount of total library material and introduce batch effects or bias. Additionally, excessive adapter dimer may indicate an imbalance of adapters:library insert material during ligation that could impact the diversity of the library. With further optimization, the correct component concentrations may eliminate this issue.

Prior to all subsequent data processing, fastq files were randomly sub-sampled to the lowest number of reads obtained per library (approximately 13 million). Additionally, we opted to trim the UMI data featured in the Takara libraries to allow for more equitable comparisons between each chemistry. To better quantify the differences observed between kits, a combination of odds ratio [OR] and p value was used. Cut-offs were defined as p < 0.05 and [OR] not within 0.9-1.2 range.

To assess the technical biases associated with each kit, duplicate reads were identified and expressed as a percentage of total reads obtained for each library (Figure 3A). All datasets contained at least 40% duplicate reads, with some containing as many as 93% duplicate reads. For each chemistry, the most duplication occurred at 250 pg input and the least at 10 ng input. The duplication rates of the Watchmaker libraries were higher than those of the Takara libraries by a large magnitude at all inputs (250 pg: [OR] 3.77, 1 ng: [OR] 4.764, 10 ng: [OR] 2.09). The duplication rates of the Watchmaker libraries were also higher than those of the Quantabio libraries at all inputs (250 pg: [OR] 2.046, 1 ng: [OR] 1.846, 10 ng: [OR] 2.089). We found the variation between the duplication rates of the Takara and Quantabio libraries to be less pronounced within input groups. At 250 pg and 1 ng input, the duplication rates of the Takara libraries were lower than those of the Quantabio libraries (250 pg: [OR] 0.542, 1 ng: [OR] 0.387). At 10 ng, however, the duplication rates between the Takara libraries and the Quantabio libraries were comparable ([OR] 0.999). Libraries were generated from identical sources for each kit, and the native duplication levels were assumed to be relatively unchanged between aliquots. Thus, the observed deviations were assumed to be chemistry specific.

Figure 3. Assessment of library diversity. A) Duplicate reads expressed as a percentage of the total reads obtained are shown for each library. Input amount is indicated along the top of the plot. Library preparation kit is indicated by color, and abbreviations are indicated along the horizontal axis. Sample integrity is indicated by shape. RIN_low libraries are shown as circles, RIN_med as triangles, and RIN_high libraries as squares. The vertical axis is scaled for better resolution. B) Preseq complexity curves extrapolated to 200 million reads are shown. The gray-shaded region extends up to 13 million reads to indicate where the extrapolation begins. The gray dashed line shows y = x for reference. Each line represents a single library. Sample integrity is indicated at the top of the plot. Input amount is indicated down the right side of the plot. Kit is indicated by color. Kit abbreviations: QB = Quantabio sparQ kit, TK = Takara SMARTer Pico kit, WM = Watchmaker RNA kit.

High duplication rates can be indicative of low-complexity libraries. As the sequencing depth of an RNA-seq library increases, progressively greater sequencing reads are required to obtain new information. Further sequencing would eventually stop yielding new information, and that threshold indicates the overall diversity of the library. To assess the diversity of the dataset, preseq (3.2.0) was used to generate library complexity curves, which were then extrapolated to 200 million reads (Figure 3B). Based on the fraction of unique reads present in the initial 13 million reads, the curves visualize the theoretical yield of additional sequencing and the point at which the threshold is met for each library.¹⁰ The Takara kit produced libraries with greater complexity than the Watchmaker kit and Quantabio kit at all three inputs (Watchmaker—250 pg: [OR] 6.535, 1 ng: [OR] 7.194, 10 ng: [OR] 4.385; Quantabio—250 pg: [OR] 6.670, 1 ng: [OR] 5.786, 10 ng: [OR] 1.656). As expected, the magnitude of differences in complexity between the Takara libraries and the other libraries became less extreme at 10 ng input. The Watchmaker and Quantabio kits produced libraries of comparable complexity at 250 pg input ([OR] 1.022). However, at 1 ng and 10 ng input, the Quantabio libraries had greater complexity than the Watchmaker libraries (1 ng: [OR] 1.243, 10 ng: [OR] 2.645).

To further assess differences between chemistries, libraries were aligned to the human reference genome hg38 with STAR,⁹ and the percentage of uniquely mapped reads obtained by each library was compared (Figure 4A). The Watchmaker libraries had higher unique alignment rates than the Takara libraries at all three inputs (250 pg: [OR] 3.471, 1 ng: [OR] 3.884, 10 ng: [OR] 4.704) and higher unique alignment rates than the Quantabio libraries at 250 pg and 1 ng input (250 pg: [OR] 7.951, 1 ng: [OR] 2.855). At 10 ng input, the unique alignment rates between the Watchmaker and Quantabio libraries were comparable ([OR] 1.237, p = 0.217). The Takara libraries had greater unique alignment than the Quantabio libraries at 250 pg input ([OR] 2.291), but at 1 ng and 10 ng, the Takara libraries had lower unique alignment than the Quantabio libraries (1 ng: [OR] 0.735, 10 ng: [OR] 0.263). A gene body coverage plot was generated to assess for kit-specific coverage biases (Figure 4B). We observed the libraries separate out by kit clearly on the 3’ end. The Quantabio libraries have the most even coverage overall, displaying a slight 3’ bias, while the Watchmaker and Takara libraries both display a more obvious 5’ bias. However, we found all libraries to show adequate coverage across the gene body with no notable dropouts observed.

Figure 4. Assessment of alignment quality and gene body coverage. A) Uniquely aligned reads mapped with STAR, expressed as a percentage of total reads obtained, are shown for each library. Each boxplot represents the six libraries generated by the specified kit from the specified input amount. Input amount is indicated along the top of the plot. Library preparation kit is indicated by color, and abbreviations are indicated along the horizontal axis. Sample integrity is indicated by shape. RIN_low libraries are shown as circles, RIN_med libraries as triangles, and RIN_high libraries as squares. The vertical axis is scaled for better resolution. B) The percentage of coverage plotted against gene body percentile (5’ to 3’) is shown for each library. Each line represents an individual library. Library preparation kit is indicated by color. Kit abbreviations: QB = Quantabio sparQ kit, TK = Takara SMARTer Pico kit, WM = Watchmaker RNA kit.

In RNA-seq libraries, poor or incomplete depletion of rRNA is a common cause of multi-mapping. Effective depletion of rRNA is critical to producing useful and informative sequencing libraries. Each kit we reviewed utilized a different method for rRNA depletion. To quantify the performance of each method, non-mitochondrial ribosomal mapping reads were identified and expressed as a percentage of the total reads obtained (Figure 5B). Regardless of input, we observed that the Takara libraries had higher percentages of nuclear rRNA mapping reads than the Watchmaker libraries (250 pg: [OR] 19.119, 1 ng: [OR] 20.039, 10 ng: [OR] 21.156) and the Quantabio libraries (250 pg: [OR] 14.492, 1 ng: [OR] 38.461, 10 ng: [OR] 9.433). We found the percentage of nuclear rRNA mapping reads between the Quantabio and Watchmaker libraries to be comparable at all inputs (250 pg: [OR] 1.328; p = 0.87, 1 ng [OR] 0.524; p = 0.33, 10 ng: [OR] 2.247; p = 0.08).

Figure 5. Assessment of transcripts detected. A) Reads categorized by RNA subtype and expressed as a percentage of total reads obtained are shown for each library. Sample integrity is indicated at the top of the plot. Input amount is indicated down the right side of the plot. Library preparation kit is indicated by abbreviations on the left of each row. Color indicates the RNA category. Category abbreviations: IG_C_gene = constant chain immunoglobin gene, lncRNA = long non-coding RNA, misc_RNA = miscellaneous non-coding RNA that cannot be classified, Mt_rRNA = mitochondrial rRNA, other = reads not fitting the categories shown. B) Nuclear rRNA mapping reads expressed as a percentage of total reads obtained are shown for each library. Each boxplot represents the six libraries generated by the specified kit from the specified input amount. Input amount is indicated along the top of the plot. Library preparation kit is indicated by color, and abbreviations are indicated along the horizontal axis. Sample integrity is indicated by shape. RIN_low libraries are shown as circles, RIN_med libraries as triangles, and RIN_high libraries as squares. The vertical axis is scaled for better resolution. C) For each library, the number of lowly expressed (0.1-1 TPM) and highly expressed (>1 TPM) genes detected is shown. Each bar represents the sum of genes detected in both technical replicates of the specified library. Shading indicates the expression level. Dark shading indicates genes detected with >1 TPM, and light shading indicates genes detected with 0.1-1 TPM. Library preparation kit is indicated by color. Input amount is indicated along the top of the plot. Sample integrity is indicated along the horizontal axis. Kit abbreviations: QB = Quantabio sparQ kit, TK = Takara SMARTer Pico kit, WM = Watchmaker RNA kit.

To evaluate differences in library contents between kits, reads were categorized by RNA sub-type using gene type information from GENCODE’s GRCh38 (version 33; Ensembl 99) annotation gtf file and plotted as a percentage of total reads obtained (Figure 5A). All samples show a minimum of 87% alignment to protein-coding genes, with the Quantabio libraries showing higher mapping to protein-coding genes than the Watchmaker (250 pg: [OR] 1.307, 1 ng: [OR] 1.313, 10 ng: [OR] 1.229) and Takara libraries (250 pg: [OR] 1.506, 1 ng: [OR] 1.497, 10 ng: [OR] 1.451) at all inputs. However, the Quantabio libraries had lower mappings than the Takara and Watchmaker libraries to constant chain immunoglobin genes at 250 pg and 1 ng input (Takara—250 pg: [OR] 0.593, 1 ng: [OR] 0.608; Watchmaker—250 pg: [OR] 0.583, 1 ng: [OR] 0.582), and miscellaneous RNAs at all inputs (Takara—250 pg: [OR] 0.661, 1 ng: [OR] 0.721, 10 ng: [OR] 0.837; Watchmaker—250 pg: [OR] 0.434, 1 ng: [OR] 0.443, 10 ng: [OR] 0.469). All other RNA sub-types were represented similarly between libraries, except for mitochondrial RNA (mtRNA), which ranged from 0.000386% to 2.84% of total reads obtained (Supplemental Figure 1). When we broke this down by kit, we found the Takara libraries had higher percentages of mtRNA mapping reads than the Watchmaker libraries (250pg: [OR] 195.326, 1ng: [OR] 247.931, 10ng: [OR] 456.296) and the Quantabio libraries (250 pg: [OR] 140.918, 1 ng: [OR] 173.331, 10 ng: [OR] 127.748) at all inputs. The percentage of mtRNA mapping reads was comparable between the Quantabio and Watchmaker libraries at 250 pg ([OR] 1.386, p = 0.676) and 1ng inputs ([OR] 1.430, p = 0.665); however, the Quantabio libraries had a greater percentage of mtRNA mapping reads than the Watchmaker libraries at 10 ng input ([OR] 3.572).

Supplemental Figure 1. Reads mapping to mitochondrial RNA. Mitochondrial rRNA or tRNA mapping reads expressed as a percentage of total reads obtained are shown for each library. Each boxplot represents the six libraries generated by the specified kit from the specified input amount. Input amount is indicated along the top of the plot. Library preparation kit is indicated by color, and abbreviations are indicated along the horizontal axis. Sample integrity is indicated by shape. RIN_low libraries are shown as circles, RIN_med libraries as triangles, and RIN_high libraries as squares. The vertical axis is scaled for better resolution. Kit abbreviations: QB = Quantabio sparQ kit, TK = Takara SMARTer Pico kit, WM = Watchmaker RNA kit.

Total RNA-seq is often selected because it can quantify non-polyadenylated RNAs of interest, such as small nuclear RNAs, micro RNAs, or alternative splice variants, and these RNA populations are often lowly expressed. When selecting a library preparation kit, it is important to consider the sensitivity required for the intended application. To evaluate the sensitivity of each kit, we compared the number of highly expressed genes (>1 TPM), and lowly expressed genes (0.1-1 TPM) that were detected in each library (Figure 5C). At all inputs, we found there were more lowly expressed genes detected in the Takara dataset than in the Quantabio (250 pg: [OR] 5.903, 1 ng: [OR] 2.919, 10 ng: [OR] 1.295) and Watchmaker datasets (250 pg: [OR] 5.434, 1 ng: [OR] 3.859, 10 ng: [OR] 1.630). Fewer highly expressed genes were detected in the Takara libraries than in the Watchmaker libraries at 1 ng and 10 ng input (1 ng: [OR] 0.894, 10 ng: [OR] 0.800). At 250 pg input, we found the highly expressed genes detected in the Takara and Watchmaker libraries to be comparable ([OR] 1.132). Similarly, fewer highly expressed genes were detected in the Takara libraries than in the Quantabio libraries at 10 ng input ([OR] 0.798). At 250 pg and 1 ng input, we found the highly expressed genes detected in the Takara and Quantabio libraries to be comparable (250 pg: 1.018, 1 ng: 1.193). At 250 pg input, we found the lowly expressed genes detected in the Quantabio and Watchmaker libraries were comparable ([OR] 0.920). However, at 1 ng and 10 ng input, more lowly expressed genes were detected in the Quantabio libraries than in the Watchmaker libraries (1 ng: [OR] 1.322, 10 ng: [OR] 1.258). Regardless of input, we found the highly expressed genes detected in the Watchmaker and Quantabio libraries to be comparable (250 pg: [OR] 1.154, 1 ng: [OR] 1.067, 10 ng: [OR] 1.001).

This study evaluated three total RNA library preparation kits across different input amounts using samples of varied integrity. As a result, we were able to compare each kit’s performance as a function of input amount as well as a function of input integrity (RIN score). However, we found input integrity to have little meaningful impact when drawing comparisons between each kit’s performance (Supp. File 8). While each kit’s performance typically improved as input integrity increased, the overall trends observed for any specific metric remained consistent regardless of input material integrity.

DISCUSSION

As advancements are made and new companies enter the NGS marketplace, the catalog of commercially available total RNA-seq library preparation kits continues to expand. This has left investigators with a staggering selection of resources for performing RNA-seq, each with unique strengths and weaknesses. Previously, similar studies have compared RNA-seq library preparation kits at low inputs, but this was the first to include Watchmaker’s RNA Library Prep Kit with Polaris Depletion and Quantabio’s sparQ RNA-Seq HMR Kit.

Overall, Takara’s SMARTer Pico kit produced libraries with the greatest consistency between input groups. The Takara libraries were more complex and had lower duplication rates than libraries generated with the Quantabio and Watchmaker kits. Comparisons of the genes detected within each library suggested that the Takara kit had greater sensitivity than the Watchmaker and Quantabio kits when detecting lowly expressed genes, and it was the only kit tested that generated libraries suitable for sequencing without additional bead cleanups.

However, the Takara libraries had fewer uniquely aligned reads than the Quantabio and Watchmaker libraries regardless of input. This can likely be attributed to the higher proportions of repetitive rRNA and mtRNA mapping reads that were observed in the Takara libraries. These findings suggested that the Takara ZapR ribosomal depletion method may not be as robust as the Watchmaker Polaris and Quantabio sparQ depletion methods.

Documentation for the Takara kit states that ZapR specifically targets mammalian rRNA (18s and 28s) and human mitochondrial rRNA (12s and 16s). Documentation for the Watchmaker kit states that Polaris targets cytoplasmic rRNAs (28s, 18s, 5.8s, and 5s), mitochondrial rRNAs (12s and 16s), 45S ETS and ITS rRNAs, and various globin RNAs. The Watchmaker documentation specifies that the 45S ETS and ITS rRNA probes and the globin probes are designed only for humans, suggesting that the other probes are designed for human, mouse, and rat samples. Due to the proprietary nature, documentation for the Quantabio kit does not specify the targets utilized for depletion. However, we confirmed that the Quantabio sparQ depletion method’s targets include cytoplasmic rRNAs (5s, 5.8s, 18s, and 28s), mitochondrial rRNAs (12s and 16s), and globin (Quantabio Field Applications, personal communication). Species was not specified, suggesting the probes are designed for human, mouse, and rat samples. Based on these specifications, Takara ZapR’s depletion method has the fewest overall rRNA and mtRNA targets, which may help explain the variation in library content observed between kits. Regardless of the breadth of their depletion targets, all methods depleted ribosomal and mitochondrial rRNA transcripts to reasonable levels for most RNA-seq applications.

While each of the novel kits successfully produced sequencing libraries from all samples, the differences that we observed in duplication rate, library complexity, and sensitivity of detection led us to conclude that neither Watchmaker’s or Quantabio’s kit was better suited to low-input applications than Takara’s SMARTer Pico kit.

There were several notable limitations to this study. Most importantly, the specifications for the Watchmaker and Quantabio kits recommended a minimum input of 1 ng total RNA, as opposed to the minimum 250 pg input recommended for the Takara kit. Thus, it was unsurprising that the greatest variability between chemistries occurred at 250 pg input. There were extra single-sided cleanups performed on the Quantabio and Watchmaker libraries before sequencing, but this was not required for the Takara libraries. This introduced a potential source of bias that may have contributed to the lower complexity we observed in the Quantabio and Watchmaker libraries. Additionally, the libraries tested in this study were all prepared by a single technician within a core laboratory setting. As such, the results obtained may not be representative of the “average user.” It should also be noted that since conducting this evaluation, the protocols for the Quantabio and Watchmaker kits have been updated with several new guidelines for preparing low-input samples including reduced PCR cycles. Similarly, Takara Bio has recently announced the SMART-Seq Total RNA Pico input with UMIs (ZapR Mammalian) kit, which is intended to replace the SMARTer Stranded Total RNA-Seq Kit v3—Pico Input Mammalian kit that was tested in this study.

Although the Quantabio and Watchmaker kits were not best suited to low-input applications, their overall performance greatly improved at the 10 ng input level. The uniquely flexible input ranges that these two kits support make them a great potential resource for projects that fall just short of the input that is required for standard workflows, or as general use kits for labs looking to handle a wide range of RNA inputs with a single protocol. We plan to repeat this study with additional library preparation chemistries across an expanded range of inputs to better assess how the Watchmaker and Quantabio kits perform at higher input levels.