High-Throughput Miniaturized 16S rRNA Amplicon Library Preparation Reduces Costs while Preserving Microbiome Integrity

Existing methods for automating amplicon-based library preparation methods using the EpMotion robots have brought down consumable and reagents costs to $3.44 per sample when scaling to 384 samples. To determine if cost savings could be realized by miniaturizing the PCR volume and gDNA input, we evaluated the quality and reproducibility of assessing microbial communities at the reduced reaction volumes compared to full-scale reactions. We compared the following four common microbial communities of increasing diversity: molecular-grade water (negative control), the commercially available Zymo mock community (ZMC), seawater, and marine sediment in four replicates across eight different input volumes of gDNA. Success rates did not vary across PCR volumes when rarified to 1,000 reads, indicating that miniaturizing the reaction did not cause issues of insufficient materials or reactants to obtain quality libraries for sequencing. Intriguingly, when we applied a higher read threshold by rarefying at 10,000 reads, success was significantly improved with lower PCR volumes of 5 µl from 21% to 46% (P < 0.05) and further improved to 69% at a 2.5-µl reaction volume (P < 0.001) (see Fig. S2a and b in the supplemental material). Decreasing the PCR volumes did not affect alpha (Fig. 2a) or beta (Fig. 2b) diversity (P > 0.05, Kruskal-Wallis).

Decreases in within-sample alpha diversity (in complex samples) and increases in between-sample beta diversity occurred when gDNA input volumes were less than 0.05 µl, with the most drastic effects at 0.005 µl; we thus opted for gDNA input of 0.2 µl (Fig. 2c and d). Sample volume-related changes could be due to subsampling effects or background contamination. Accurate and reproducible composition of the positive-control samples (ZMC) was obtained across all PCR volumes and with gDNA volumes over 0.005 µl (Fig. 3).

To further validate the miniaturized library preparation method across diverse sample types, we sequenced microbiome samples from five body sites of 46 individuals of the marine fish Scomber japonicus (Pacific chub mackerel) across seven months of sampling (Fig. 4a). Sample exclusion testing indicated a read cutoff of 1,362 reads (Fig. S3). Alpha diversity was highly correlated when samples were processed through the two different methods, for richness and Shannon diversity, respectively (P < 0.0001, r² = 0.9534; P < 0.0001, r² = 0.962) (Fig. 4b and c). Interestingly, the slopes for both richness and Shannon diversity indicate a slightly higher alpha-diversity estimate for samples processed with the EpMotion than with the Echo 550. Both weighted (Fig. 4d) and unweighted (Fig. 4e) UniFrac distances for technical variation (PCR methods) across all body sites with enough replicates were significantly smaller than the total variation within a body site (Mann-Whitney, P < 0.0001).

Automation is revolutionizing sample acquisition, sample processing, and data analysis in many fields of science. While sequencing costs have dramatically decreased over time, the cost of commercial next-generation sequencing (NGS) library preparation has largely remained constant, at around $10 to 20 per sample.

We were able to reduce costs by an additional 58% to $1.42 per sample when using a miniaturized 5-µl reaction volume or $1.34 per sample with a 2.5-µl reaction volume (Fig. 1a). Since the cost reduction for going from 5 µl to 2.5 µl is minimal, and concerns for evaporation increase with lower volumes, we consider the 5-µl reaction volume to be the optimal compromise. The costs savings are primarily a result of a decrease in use of PCR mastermix ($0.933 for 25 µl versus $0.14 for 5 µl) and tips ($0.532 for 25 µl versus $0.175 for 5 µl) (Fig. 1b), which when scaling to 50,000 samples a year can result in annual savings of $17,847 and $39,655, respectively (Fig. S1a and b). Other significant cost savings when miniaturizing the reaction include primer costs ($0.166 per sample), plate consumables ($0.108 per sample), PCR cleanup and pooling ($0.406 per sample), and gel quality control (QC) ($0.152 per sample) (Fig. 1). In full, a total cost reduction of $1.02 from miniaturizing the library preparation 5-fold results in an annual cost savings of $101,000 for 50,000 samples and $202,000 for the 100,000 samples typical of our laboratory (Fig. S1c). Since it is common to multiplex 768 samples on a single Illumina sequencing lane, we calculated the library preparation cost savings per run to be $1,549, or slightly more than a MiSeq 2 × 150-bp sequencing run according to current pricing available on the Illumina website (as of August 2018).

Our results demonstrate that the miniaturized PCR protocol maintains data quality with respect to traditional methods while achieving substantial cost advantages (more than 10-fold relative to commercial kits) and producing much less plastic tip waste. The use of an acoustic liquid handler also permitted dramatic decreases in sample volume usage, which may be important when using precious historical samples (5) with limited volumes or running large numbers of assays per sample. While we performed this miniaturization using an instrument produced by Labcyte (San Jose, CA, USA), similar results may be achievable with the instruments from EDC Biosystems (Fremont, CA, USA), although we have not evaluated performance on the EDC Biosystems instrument. Likewise, other nanoliter-dispensing robots would be able to achieve similar volume miniaturization but not obtain the significant environmental and cost savings from the elimination of tips. Overall sample success may actually be improved when performing library preparation reactions at lower volumes. Higher success at lower volumes could be explained by higher proportions of amplicon going into the library pool or improved efficiency of enzymatic reactions at lower volumes (6–8). Alpha- and beta-diversity measures were highly correlated across methods. While we did not explicitly test well-to-well contamination, other studies have demonstrated that contamination may be lower when using acoustic liquid handlers over traditional pipette-based liquid handlers for molecular assays while also having higher dispensing accuracy (9). Finally, while we demonstrate the utility of this method for the 16S rRNA gene, this method could easily be applied to any amplicon based library preparation method. We therefore recommend the new protocol for large-scale studies.

Although the acoustic liquid handling instrument on which it relies is expensive, considerable cost savings can be achieved, particularly in the context of a core facility or in a high-throughput laboratory. Further advances in low-volume liquid handling will be expected to make this technology available to more individual research groups. The ability to perform amplicon sequencing in much higher throughput makes accessible large-scale ecological and epidemiological applications that are presently cost-prohibitive.

Four sample types of varied microbial diversity were chosen to first optimize and compare sequencing success. These samples included a marine sediment sample from the Scripps Institution of Oceanography (SIO) pier, filtered sea water from the San Diego bay, the ZymoBIOMICS DNA mock community (ZMC) (catalog no. D6305; Zymo Research), and molecular-grade water (catalog no. W4502; Sigma-Aldrich) as a negative control. To collect samples representing true ecological diversity, 46 Scomber japonicus mackerels were collected across 16 sampling events from the end of the SIO pier (32.867, −117.257) from 28 January 2017 to 4 August 2018. Fish were caught using hook and line at sunset, euthanized, and then immediately wrapped in foil and stored on dry ice and then −80°C prior to dissection. Upon dissection, ∼50 mg of hindgut and pyloric caeca was placed into extraction tubes while the gill, skin, and digesta were swabbed (catalog no. 806-wc; Puritan) for processing by DNA extraction.

Samples were processed using the standard 16S rRNA Earth Microbiome Protocol (earthmicrobiome.org). Specifically, gDNA was extracted using single tubes, followed by DNA magnetic bead cleanup with the Mo Bio PowerMag kit (catalog no. 27000-4-KF), which improves the limits of detection for low-biomass samples (10). Additional positive and negative DNA controls were included so that sample exclusions based on read counts could be calculated (10).

Extracted gDNA was then PCR amplified in triplicate reactions for 35 cycles using the EMP 16S V4 515f/806rB barcoded primers (11, 12) at a 25-µl (n = 32) volume or miniaturized 5-µl (n = 32) or 2.5-µl (n = 32) reaction volumes with eight volumes of gDNA (5 µl, 3 µl, 1 µl, 0.5 µl, 0.3 µl, 0.1 µl, 0.05 µl, and 0.005 µl) in quadruplicate from the four communities (negative controls, ZMC, seawater, and sediment) normalized to 3 ng ul⁻¹ DNA. Multiple body sites, including the gill, skin, digesta, gastrointestinal (GI) tract, and pyloric caeca from 46 S. japonicus individuals spanning seven months of sampling were amplified in standard 25-µl-volume PCR setup with 1 µl gDNA input using the EpMotion 5075 versus the optimized 5-µl PCR setup using the Echo 550 and 200 nl gDNA input. Positive gDNA controls of Vibrio fischeri and ZMC were also included.

Sample sequencing and data processing. Amplicons were quantified using a PicoGreen assay (Thermo Fisher Scientific, San Diego, CA), and then 2 µl of each sample amplicon library was equally pooled into a single tube and then sequenced on a MiSeq 2 × 150-bp v2 kit. Raw sequencing files were uploaded and processed with Qiita (qiita.ucsd.edu) and QIIME 1.9.1 (13), with the first read trimmed to 150 bp and processed through Deblur (14), a de novo sOTU picking method. A phylogenetic tree of the 16S sub-operational taxonomic unit (sOTU) single-sequence tags was created using SEPP (15). The minimum rarefaction depth for experiment 1 was 1,000 reads to exclude most negative controls (Fig. S4), whereas the depths for experiment 2 were empirically determined by calculating the read counts at which 90% of the reads from the DNA extraction positive controls map back to the positive controls (10). Alpha diversity was calculated using measures of richness and Shannon entropy (16), while beta diversity was calculated using either Bray-Curtis (17) or weighted and unweighted UniFrac (18, 19) distance and visualized in EMPeror (20). Alpha- and beta-diversity statistical significance was tested using a Kruskal-Wallis test (21) and Mann-Whitney test (22). Correlations of alpha diversity between the two methods of the fish samples was performed using Spearman’s rho correlation (23, 24).

Data availability. All protocols associated with the Echo-550 scripts have been uploaded and are publicly available on protocols.io. The links can be found here: https://www.protocols.io/view/high-throughput-miniaturized-16s-rrna-amplicon-lib-u2ieyce. All raw and processed 16 amplicon sequencing data and metadata from the first PCR validation experiment (Qiita study identification [ID] 11432; https://qiita.ucsd.edu/study/description/11432; EBI study accession no. ERP109772) and the second validation experiment, including the full fish mucosal time series data set (Qiita study ID 11721; https://qiita.ucsd.edu/study/description/11721; EBI study accession no. ERP109537), are publicly available in Qiita and EBI.