A high-throughput pipeline for DNA/RNA/small RNA purification from tissue samples for sequencing

Large-scale projects, including The Cancer Genome Atlas (TCGA), successfully used the column-based AllPrep-mirVana protocol to purify DNA and RNA, including miRNA, from various human tumor tissue samples [1–3]. This workflow requires two column-based kits that involve multiple washes and centrifugation steps. The need for centrifugation, plus the use of toxic chemicals like phenol and chloroform, makes such protocols challenging to automate with conventional liquid handlers. Other protocols for miRNA purification also use columns [4–6] or require specialized equipment [7]. The current authors previously reported an automated, high-throughput, solid-phase reversible immobilization (SPRI) bead-based protocol to purify DNA and RNA of high quality and yield, suitable for next-generation sequencing (NGS) applications [8,9]. This method, referred to as the Aline protocol, was optimized and automated using the EvoPure RNA Tissue Isolation kit . However, miRNA recovery is suboptimal in the Aline purification process due to the weaker binding of smaller RNA molecules to the beads. This necessitates a separate workflow for small RNA, including miRNA. Given the utility of miRNA profiling in omics studies [10–13], the existing purification protocol was modified to improve the recovery of miRNA such that a single total nucleic acid (TNA) workflow yields DNA, RNA and small RNA usable for library construction and sequencing. This TNA protocol, GRAB-ALL, uses a single kit and is fully automated on the Hamilton Microlab NIMBUS96 liquid handling system.

To validate these modifications, cell pellets were purified and tumor samples were embedded in an optimal cutting temperature (OCT) compound using the existing bead-based Aline protocol and GRAB-ALL. Libraries were prepared and sequenced and TNA yields and RNA expression profiles were compared. miRNA libraries were also generated to benchmark GRAB-ALL against the established AllPrep-miRVana workflow.

Materials & methods

Biological materials

Patient samples used in this study were obtained as part of the Personalized OncoGenomics project approved by the Research Ethics Board of BC Cancer (REB# H12-00137). Informed consent was obtained from all patients upon enrollment. Ten human tumor biopsy samples embedded in OCT were used to evaluate the purification protocols. HeLa S3 cells were obtained from the Cell Culture Company (MN, USA). Universal Human Reference total RNA was purchased from Stratagene (#740000, CA, USA). Human placental total RNA was purchased from Ambion (AM7950, Now a part of Life Technologies, CA, USA).

TNA purification

DNA/RNA was purified from fresh-frozen and OCT samples using the Aline EvoPure (R-907T, Aline Biosciences, MA, USA) protocol previously described [9] and GRAB-ALL. Reagents and steps in both protocols are outlined in Figure 1. A detailed standard operating procedure for GRAB-ALL is described in Supplementary File 1. Briefly, fresh-frozen or OCT-embedded tissue samples were lysed in 400 μl lysis buffer (RLT Plus buffer; #1053393, Qiagen, Hilden, Germany) with 20 mM Tris(2-carboxyethyl) phosphine hydrochloride overnight at room temperature with agitation. The next morning, 40 μl ammonium acetate and 680 μl binding buffer (80 μl binding beads + 600 μl 100% isopropyl alcohol [IPA]) were sequentially added to the lysate. After mixing and incubating for 10 min at room temperature, samples were placed on a magnet and allowed to clear. After the protein-containing supernatant was removed, the beads were washed once with 1 ml of wash buffer (1:1 mixture of EvoPure Wash Buffer 1 Concentrate and IPA) and three times with 800 μl 85% ethanol. The beads were then allowed to dry on a magnet for 10 min. TNA was eluted in 40 μl of DEPC-treated H₂O, divided into aliquots and frozen at -80°C.

TNA purification yield was determined by Qubit 4 Fluorometer (Q33426, Invitrogen, MA, USA). The yield and quality of purified total RNA were determined using Agilent RNA Nano LabChip (5067-1511, Agilent, CA, USA). An AllPrep DNA/RNA/miRNA Mini Kit (#80204, Qiagen) was used for DNA and RNA purification according to the manufacturer's protocol. Small RNA was further isolated from TNA using the MirVana miRNA Isolation Kit (AM1561, Thermo Fisher Scientific, MA, USA) according to the manufacturer's instructions (Figure 1, left).

Optimization to enhance small RNA recovery

To enhance small RNA binding to SPRI beads (hereafter referred to as the enhanced conditions or “Enh”), several changes were made to the washing and binding conditions of the Aline protocol: the amount of isopropanol added to the EvoPure Bind Buffer was increased from 320 μl to 600 μl per 80 μl beads; a 1:1 ratio of isopropanol was added to the EvoPure Wash Buffer Concentrate 1 instead of the amount recommended by the manufacturer; and 75% ethanol washes were replaced with 85% ethanol washes (Figure 1, starred). Further improvements to precipitate small RNA were tested, by adding to the tissue lysate either a commercial small RNA enhancing additive (miRNA Homogenate Additive from the miRVana kit), 3M sodium acetate or 7.5M ammonium acetate.

RNA library generation & sequencing

Strand-specific RNA-sequencing (RNA-Seq) libraries were constructed using an automated protocol previously described [9,14]. Briefly, TNA was treated with DNase I (Invitrogen 18068-015) to remove DNA. The NEBNext rRNA Depletion Kit (6310X, New England Biolabs, MA, USA) was used to remove ribosomal RNA from DNase-I treated total RNA. First-strand cDNA was synthesized using the Maxima H Minus First Strand cDNA Synthesis Kit (K1652; Thermo Fisher Scientific), which contained dUTP and Actinomycin D. Second-strand cDNA was synthesized using the NEBNext Ultra Directional Second Strand cDNA Synthesis Module (NEB E7550) and sheared using an LE220 sonicator (Covaris LLC, MA, USA). Purified cDNA underwent end-repair and phosphorylation followed by one round of bead-based purification with PCRClean DX beads (Aline C-1003-450). Adenylation (A-tailing) was performed using NEB Paired-end Sample Prep Premix Kit–A Tail (NEB E6876B-GSC) and Illumina paired-end (PE) adapters were ligated using the NEB Paired-End Sample Prep Premix Kit–Ligation (NEB E6877B-GSC). Bead purified, adapter-ligated cDNA was digested with USER enzyme (NEB M5505L) to remove dUTP-marked second-strand cDNA, followed by 13 cycles of PCR using the Illumina PE primer set and Phusion High-Fidelity DNA Polymerase (F-530XL; Thermo Fisher Scientific). PCR products were bead-purified twice and quantified using the Qubit 4 Fluorometer or Quant-iT dsDNA Hi Sensitivity assay (Invitrogen Q33120) to determine concentrations. The LabChip DNA High Sensitivity assay (760517/760568, Caliper Life Sciences, MA, USA) or Agilent High Sensitivity DNA assay was used to determine size distribution. An equimolar RNA library pool was sequenced on an Illumina HiSeq2500 with PE 75-base pair (bp) sequencing.

miRNA library generation & sequencing

miRNA libraries were prepared as previously described [14]. Briefly, total RNA purified using AllPrep-mirVana (#80204, Qiagen; AM1560, Invitrogen), or TNA purified using GRAB-ALL, was ligated to a 3′ adapter and incubated for 1 h at 22°C. Two rounds of bead clean-up were performed post-ligation. To ligate 5′ adapters, bead-purified RNA was added to a 5′ adapter ligation brew and incubated for 1 h at 37°C. First-strand cDNA was synthesized using the Maxima H Minus RT enzyme for 1 h at 44°C and size-selected using Aline SPRI beads to remove primers and nontarget products. Libraries were enriched with 15 cycles of PCR using PE and indexed primers. Quality control (QC) was performed using a LabChip DNA High Sensitivity Assay. Finally, equimolar library quantities were pooled and the pool was size-selected on a Blue Pippin gel electrophoresis system (SAG-BLU0001, Sage Science, MA, USA) to remove primer dimers and enrich for miRNA library products in the size range of 140–150 bp. miRNA libraries were sequenced on an Illumina NextSeq with single-end 75-base sequencing. Oligonucleotides used in miRNA library construction are listed in Table 1.

Sequence analysis

A total of 75 bp reads were down-sampled to 40 million reads per library for comparison. RNA-Seq reads were aligned using JAGuaR [15] to the hg19 reference genome and Ensembl 69 annotations. BioBloom Tools [16] was used to quantify ribosomal and mitochondrial reads. miRNA sequencing data were analyzed as single-end reads as previously described [1]. Down-sampling of RNA and miRNA sequences was performed using Picard [17]. Skewer [18] was used to remove adapter sequences, leading and trailing random hexamers before alignment. Alignment was done against hg19 with BWA 0.7.13 [19], and alignment coordinates were compared with annotated miRNAs in miRBase (V20) [20].

Unsupervised clustering

miRNA species detected in at least one library were used for clustering analysis (n = 1461). The read counts were log10-transformed and median-centered before ConsensusClusterPlus [21] was run with the following parameters: pItem = 0.8, pFeature = 0.8, reps = 1000, maxK = 12, clusterAlg = hc and distance = Pearson. All 0 values were assigned the value of 0.0001 before log transformation. The resulting cluster was visualized using ComplexHeatmap [22].

Statistical analysis

A two-tailed Student's t-test was used in the statistical comparison of two conditions. A p-value of < 0.05 was considered statistically significant. Statistical comparison of more than two conditions was performed using one-way analysis of variance (ANOVA) with Dunnett's multiple-comparison test. Power analysis was performed with G*Power 3.1 [23] using an expected effect size of 0.5 (i.e., moderate-to-large differences).

Results & discussion

Protocol modifications increased miRNA library yield

Small RNA species bind weakly to SPRI beads under conditions normally used to purify TNA. The mirVana kit employs a homogenate additive to improve small RNA binding to glass-fiber filters. Sodium acetate and ammonium acetate are routinely used to facilitate ethanol precipitation of RNA [24], due to their ability to neutralize the charges on nucleic acid backbones. Therefore, these three additives were tested to determine whether they could facilitate small RNA binding to SPRI beads.

To determine whether small-RNA-enhancing modifications made to the Aline protocol affected TNA or RNA yield, TNA was purified from 5 × 10⁵ HeLa S3 cells using the Aline protocol and each of the three variants of the modified protocol. Compared with the Aline protocol, modified versions reduced TNA yield by 25–40% on average (Figure 2A; one-way ANOVA; **p < 0.01, *p < 0.05; power = 0.65). The combination of ammonium acetate with the enhanced washing conditions resulted in the least amount of TNA loss (25%). A total of 400 ng of TNA from each condition was then treated with DNase I to measure the total RNA yield and extrapolate the theoretical maximum yield if all TNA was treated. The protocol modifications tested had no significant effect on RNA yield (Figure 2A; one-way ANOVA; power = 0.65).

To perform more in-depth miRNA QC, miRNA libraries were constructed to determine the effect of the protocol modifications at the sequencing level. Positive and negative controls for the library construction process were placental total RNA and DEPC-treated H₂O, respectively. Sequencing results detected the highest number of miRNA reads and miRNA species in libraries constructed with the AllPrep-mirVana purification protocol, followed by libraries constructed using the modified Aline purification protocol with ammonium acetate (Figure 2B). The original Aline protocol had a low miRNA yield, as expected. Altering the washing and binding conditions alone (Enh) did not consistently improve the number of miRNA reads and miRNA species recovered. Additional analysis revealed that miRNA purification with enhanced washing conditions plus ammonium acetate resulted in libraries with a high percentage of aligned reads and total miRNA reads (Figure 2C). Based on these data, the GRAB-ALL workflow was finalized using enhanced washing conditions with ammonium acetate.

Automated GRAB-ALL can be used for high-throughput TNA purification

To evaluate the suitability of GRAB-ALL for automation, GRAB-ALL was implemented on a Hamilton Microlab NIMBUS96 liquid handler. Ten OCT-embedded tumor specimens (Supplementary Table 1, four cores each) were lysed and divided into two sets of technical duplicates to undergo Aline or GRAB-ALL purification. HeLa S3 cells were used as a positive control for the purification process. GRAB-ALL purified less TNA compared with Aline (Figure 3A; p < 0.01, paired t-test; power = 0.82), where yield was reduced by 23% on average, consistent with previous results.

To determine RNA yield, 200 ng of TNA from technical duplicates were pooled and treated with 5U DNase I. The amount of RNA obtained was not significantly different between the Aline and GRAB-ALL purification protocols (Figure 3B; p > 0.05, paired t-test; power = 0.82). Aside from two samples where RNA concentrations were too low to measure RNA integrity number (RIN) reliably (Samples 3 and 7), RIN was not significantly different between samples purified using the Aline protocol and GRAB-ALL (Figure 3C; p > 0.05, paired t-test; power = 0.79). RNA size profiles were similar between the Aline and GRAB-ALL purification (Figure 3D). These results suggest that GRAB-ALL purification did not reduce RNA quality.

To further assess the effect of GRAB-ALL purification on RNA expression profiles, strand-specific RNA-Seq libraries were generated using ten pairs of TNA samples purified with Aline or GRAB-ALL. All libraries met library size (200–500 bp) and yield (>2 nM in 10 μl) thresholds for NGS (Supplemental Figure 1), with only two samples (4 and 7) having <200 nM of final library yield. Paired HeLa S3 and three OCT libraries were sequenced on HiSeq2500 with PE75 reads. Alignment metrics such as duplication rate and percentage of aligned reads were not significantly different between Aline and GRAB-ALL libraries (paired t-tests; Figure 3E; power = 0.66). Pairwise Pearson's correlation coefficients showed high (rho >0.99) correlations between paired libraries (Figure 3F). Taken together, these data suggest that the GRAB-ALL workflow did not diminish the quality of RNA-Seq data compared with the Aline workflow.

GRAB-ALL purifies miRNA suitable for analysis

The suitability of GRAB-ALL for miRNA profiling was next investigated by constructing miRNA libraries using nucleic acids purified from tumor samples. TNA purified from six OCT-embedded specimens and HeLa S3 cells using AllPrep-mirVana or GRAB-ALL was used in miRNA library construction. To determine the minimum amount of purified RNA required for successful library construction, 100 ng, 250 ng and 500 ng of total RNA input were tested (Supplementary Table 1). Titration of placental RNA at 100 ng, 250 ng, 500 ng and 1000 ng was performed to determine the effect of the input amount itself on miRNA library quality using both protocols. All libraries were pooled, sequenced and down-sampled to 1 million reads for subsequent comparisons. Despite all libraries having sufficient yield for sequencing (Supplementary Figure 2), a 100-ng input GRAB-ALL library of sample 6 yielded an exceptionally low (<2000) number of reads, indicative of pipetting or loading error. Therefore, this library and its matching pair were removed from subsequent analysis. Placental RNA titration showed that 100 ng total RNA input produced a lower percentage of miRNA reads than higher input amounts (Supplementary Figure 3), suggesting that the amount of total RNA input affects miRNA library quality. Compared with HeLa S3-derived libraries, OCT-derived tumor libraries showed higher variability in the percentage of miRNA reads and the number of miRNA species (Figure 4A), some of which may be attributed to biological differences. For instance, all libraries derived from sample 6 (adrenocortical carcinoma) contained relatively high numbers of miRNA species regardless of input amount and purification method. Libraries generated using AllPrep-mirVana purified RNA tended to have higher percentages of miRNA reads compared with GRAB-ALL (Figure 4A). Percentages of reads mapping to other small RNA species were similar between the two protocols (Supplementary Figure 4). Both protocols uniquely captured similar numbers of miRNA species (Figure 4B) except at the 100-ng input, where more miRNAs were uniquely detected in AllPrep-mirVana libraries than GRAB-ALL (p < 0.05, paired t-test; power = 0.63). These data suggest that AllPrep-mirVana is higher in sensitivity, especially at lower input amounts.

To determine concordance between miRNA libraries generated from GRAB-ALL and AllPrep-mirVana purifications, unsupervised clustering was performed using miRNAs that were detected in at least one library (Figure 4C; n = 1461 miRNAs). Apart from one library (sample 1, 100 ng), libraries clustered by source sample rather than purification method. Notably, samples 4 and 5, both of sarcoma origin, clustered together. The pairwise correlation between libraries was then determined using Pearson's correlation coefficient. miRNA expression between matched libraries was, in general, highly correlated (Figure 4D), further indicating that GRAB-ALL produces miRNA libraries of adequate quality. Taken together, these results suggest that GRAB-ALL purified sufficient miRNA for library construction and sequencing. For robust miRNA detection and profiling, miRNA library construction should start with sufficient total RNA as input, preferably 500 ng or more.

In this study, an automated workflow, potentially amenable to high throughput, to purify DNA, RNA and small RNA from fresh-frozen or OCT-embedded tissue samples was established. GRAB-ALL processes samples in automation-friendly 96-well plates and produces miRNA-rich TNA in less than 90 min of hands-on time. In contrast, the throughput of column-based purification methods is limited by manual processing efficiency. For instance, the AllPrep-mirVana protocol required 2 h to purify miRNA, where one technician can usually process 6–8 samples in parallel. Phenol-chloroform is not used in GRAB-ALL, making it safer for the user and the environment. Despite a reduction in TNA yield, the GRAB-ALL workflow purified RNA of sufficient yield and quality for library construction and transcriptome profiling. In our production environment where the minimum input required for PCR-free whole-genome sequencing (WGS), RNA-Seq and miRNA library construction were 250 ng TNA, 50 ng RNA and 250 ng RNA, respectively, nine out of the ten OCT-embedded tumor specimens used in this study would have had sufficient yield for WGS and RNA-Seq libraries from Aline or GRAB-ALL purification, while seven out of nine samples would have had sufficient yield for all three library types if purified with GRAB-ALL. While the sample size is limiting, these observations suggest that GRAB-ALL could be advantageous if small RNA libraries are required in addition to WGS and transcriptome libraries.

During library construction, variations in RNA purification yield and RNA-Seq library yield were observed. While some of these variations may be stochastic, preanalytical variables such as tissue type, biopsy methods and OCT embedding protocols could also have affected library yield [25]. Future work should examine the contribution of preanalytical variables on RNA purification and library yield.

GRAB-ALL was able to purify miRNA of sufficient quality and diversity for NGS, albeit with some reduction in sensitivity compared with AllPrep-mirVana. To ensure robust performance, sufficient nucleic acid purified using GRAB-ALL should be used as input in miRNA library construction. Caution should be taken when interpreting the results of low-input libraries.

Conclusion

Existing methods of small RNA purification usually require either a workflow separate from DNA/RNA isolation or involve extensive hands-on components that are difficult to automate. GRAB-ALL, an SPRI-bead-based protocol that copurifies small RNA with DNA and RNA was introduced in this work. For large-scale projects where column-based manual protocols become a bottleneck and are cost-prohibitive, and when miRNA libraries are required in addition to RNA and DNA libraries, GRAB-ALL could be a good solution.

Future perspective

Multi-omic analyses are becoming integral to many fundamental and clinical research projects, necessitating the development of sample-processing pipelines that preserve and purify multiple classes of molecules of interest. This study described a high-throughput workflow where DNA, RNA and small RNA can be copurified from the same piece of fresh-frozen or OCT-embedded material. Future work is required to determine whether proteins can be copurified using this protocol, which would allow proteomic profiling in addition to the genomic profiling described here.