High-throughput multi-parallel enteropathogen quantification via nano-liter qPCR

Quantitative molecular diagnostic methods, such as qPCR, can effectively detect pathogen-specific nucleic acid sequences. However, costs associated with multi-pathogen quantitative molecular diagnostics hinder their widespread use. Nano-liter qPCR (nL-qPCR) is a miniaturized tool for quantification of multiple targets in large numbers of samples based on assay parallelization on a single chip, with potentially significant cost-savings due to rapid throughput and reduced reagent volumes. We evaluated a suite of novel and published assays to detect 17 enteric pathogens using a commercially available nL-qPCR technology. Assay efficiencies ranged from 88-98% (mean 91%) and were reproducible across four operators at two separate facilities. When applied to complex fecal material, assays were sensitive and selective (99.8% of DNA amplified were genes from the target organism). Detection limits were 1-2 orders of magnitude higher for nL-qPCR than an existing enteric TaqMan Array Card (TAC), due to nanofluidic volumes. Compared to the TAC, nL-qPCR displayed 97% (95% CI 0.96, 0.98) negative percent agreement and 63% (95% CI 0.60, 0.66) overall positive percent agreement. Positive percent agreement was 90% for target concentrations above the nL-qPCR detection limits. nL-qPCR assays showed an underestimation bias of 0.34 log10 copies/gram of stool [IQR -0.41, -0.28] compared with the enteric TAC. Higher detection limits, inherent to nL-qPCR, do not hinder detection of clinically relevant pathogen concentrations. With 12 times higher throughput for a sixth of the per-sample cost of the enteric TAC, the nL-qPCR chip described here is a viable alternative for enteropathogen quantification for studies where other technologies are cost-prohibitive.


Assay design 79
We selected bacterial, protozoan, and helminthic enteropathogens identified as 80 contributing to diarrheal disease in children across twelve countries (33, 34). We 81 computationally designed and screened 175,000 candidate primer pairs to target 16 virulence 82 genes using methods described previously (32). Briefly, amino acid sequences corresponding to 83 all non-redundant members of each target gene's protein family (Pfam v 27.0) were clustered 84 based on percent pairwise identity using BLASTp all-vs-all search (35). We downloaded 85 corresponding nucleotide sequences from NCBI, and DNA oligonucleotide primers were 86 designed to target conserved DNA-level sequence motifs in sequence clusters containing the 87 target virulence protein for each pathogen. A Python script directed the software primer3 (36) 88 to develop thousands of candidate primer pairs for each target gene, which were then screened 89 in silico against other non-target clusters within the same protein family. Up to eight assays per 90 target gene were selected for laboratory screening. We included an additional 10 published 91 assays (see Table 1) to assess the suitability for inclusion of previously validated assays 92 robotically dispensed onto nL-qPCR chips using TakaraBio's SmartChip ™ platform. In a separate 100 plate, samples were added to additional master mix and robotically dispensed onto chips. 101 Duplicate chips were run, using the standard TakaraBio protocol: 95°C for 3 min, then 40 cycles 102 of (95°C for 60s, 60°C for 70s). We excluded assays that reproducibly displayed fluorescence in 103 the negative control (PCR grade water) prior to cycle 28, failed to amplify standards at 100 104 copies per well, or had PCR efficiencies less than 85%. Final assays were selected based on 105 optimal performance characteristics as described below. Each chip contained a minimum of 106 two negative (no-template) controls for each assay. 107

Analytical performance characteristics 109
Analytical performance was evaluated in accordance with the Minimum Information for 110 Publication of Quantitative Real-Time PCR Experiments (MIQE) guidelines (42). Assay 111 efficiencies were evaluated with a pool of synthetic DNA standards, described above. Standards 112 were 10-fold serially diluted (10 to 10 6 copies/reaction). Standard curves were run on a 113 minimum of 15 chips over two instruments at separate facilities (Fremont, CA and East Lansing, 114 MI) and with two different operators at each location. Efficiencies were calculated according to 115 Rutledge and Côté (43); mean efficiency over all runs is reported along with coefficient of 116 variation (CV). Limit of detection (LOD) was determined with pooled synthetic DNA standards 117 spiked into extracted DNA from 10 fecal samples to a final concentration of 10, 100, and 1000 118 copies/reaction; then each sample was run in duplicate on two separate chips. The mean cycle 119 quantification (C q ) value (i.e. the cycle at which sufficient copies of target DNA have been made 120 to produce a fluorescent signal detectable by the instrument) was calculated for duplicate 121 assays on a single chip, and all results under the C q cutoff of 30 were determined positive. A 122 total of 20 positive samples (10 samples x 2 chips) per target at each concentration were 123 assayed and LOD is defined as the lowest concentration which 95% were positively detected 124 (i.e. where 19 of the 20 were detected). 125 Inter-assay precision (reproducibility) was assessed across the standard curves used for 126 efficiency calculations measured over 15-20 chips, using different lots of master mix, different 127 batches of oligonucleotide primers, and 4 different operators at 2 separate facilities. We report 128 the mean CV on calculated copy numbers over all points on the standard curve as well as the 129 range. Intra-assay precision (repeatability) was measured within-chip and between chips. 130 Within-chip precision was evaluated in three samples in which extracted DNA from fecal 131 samples was mixed with positive controls at high (10 5 copies/reaction) and low (100 132 copies/reaction) concentrations and assayed 20 times on a single chip: we report the CV of 133 calculated copy number across the 20 replicates. Between-chip precision was evaluated in 223 134 fecal samples collected from a cohort of Bangladeshi children that tested positive for at least 135 one pathogen (C q < 30) plus 18 additional fecal samples into which we added positive controls. 136 Replicates for each sample were run on two chips and the CV of calculated template copies was 137 determined across all four replicates. We report mean CV of calculated template copies over all 138 samples, as well as the number of unique samples included in the calculation of the mean. 139 Sensitivity and specificity were evaluated using four pools of DNA standards, spiked into 140 extracted DNA from 40 pathogen-free fecal samples. For each pathogen 10 samples contained 141 the target at low concentration (100 copies/reaction), 10 samples at medium concentration 142 (10x the LOD) and 10 samples at high concentration (100x the LOD); an additional 10 samples 143 had no target. Sensitivity and specificity were determined based on positive or negative 144 detection in these 40 samples. In order to further verify assay specificity, we sequenced PCR 145 amplicons obtained from running 96 child fecal samples on the nL-qPCR chip. The Seq-Ready ™ 146 TE MultiSample FLEX protocol, PCR clean-up, and DNA quantification prior to sequencing were 147 done in accordance with TakaraBio's standard procedures, as described previously (Atshemyan, 148 2017; Firtina, 2017). The resulting paired-end Illumina MiSeq reads were quality filtered and 149 only sequences that were the expected target gene amplicon length (+/-3 bp) were 150 maintained. We verified the intended target (organism and gene) by conducting a nucleotide 151 BLAST search (35) on each unique sequence. We retained the top hit(s), defined as the highest 152 sequence identity with the lowest E value. 153 154

Sample collection 155
To test the performance of nL-qPCR chip and against the performance of enteric TAC in 156 epidemiology-relevant samples, we utilized 254 fecal samples from children in rural 157 Bangladesh. Children were between 10 and 18 months old and enrolled in the WASH Benefits 158 randomized controlled trail (44-47). Split samples from these children had been previously 159 assayed for pathogens with the TAC technology (manuscript in preparation). were compared with a pairwise Wilcoxon rank sum test, using the Benjamini-Hochberg 177 procedure to account for multiple comparisons (50). Sensitivity and specificity were calculated 178 using the epi.test function from the epiR package (51). Positive percent agreement and 179 negative percent agreement were calculated in the same manner and are reported with this 180 alternative nomenclature as recommended when no known reference standard is used (52). 181 Exact binomial 95% confidence limits on sensitivity and specificity were calculated according to 182 Collett (53). Unweighted Cohen's Kappa was calculated using the epi.kappa function with 183 confidence intervals calculated according to Rothman (54). Bias in calculated log 10 copy 184 numbers per gram of stool (corrected for extraction and PCR efficiency by normalizing to the 185 positive control PhHV spike-in) was evaluated according to Bland Table S1. 193

Analytical performance 196
The mean efficiency for each assay, based on the evaluation of standard curves run on 15-197 20 chips, ranged from 88-98% (mean 91%) with a coefficient of variation of 6.3% [IQR 5.3,198 7.3]( Table 2). The linearity over all assays on all chips was 0.990 [IQR 0.987, 0.992] and 199 detection limits were between 10-100 copies/100nL reaction, which corresponds to 8x10 5 -200 8x10 6 copies/g of stool (Table 2). Within-chip repeatability was assessed in ten replicates on a 201 single chip: synthetic DNA in high (10 5 copies/reaction) and low (10 2 copies/reaction)    Table 2, Repeatability). The highest variability was again seen at 211 the lowest concentrations ( Figure S2). 212 Assays were reproducible across two instruments and four operators, again with an inverse 213 relationship noted between variance and concentration (Table S2). At concentrations one or 214 more orders of magnitude above the detection limit, coefficient of variation on calculated copy 215 number ranged from 17 -44% (Table 2). Coefficient of variation at the limit of detection ranged 216 from 29% to 115% for pathogen virulence and marker genes, the highest of which was 217 analogous to 17 ± 20 copies detected. The highest variance (319%) observed was for the total 218 bacterial (16S rRNA) assay at the detection limit of 10 copies/reaction. Between-chip variance 219 was similar to variance across two instruments and four operators (p = 0.99) but both were 220 significantly higher than within-chip variance (p < 0.0001, pairwise Wilcoxon rank sum test). 221 Coefficients of variation of the magnitudes observed are not biologically relevant when 222 analyzing pathogen quantities on the log 10 scale, as is the normal procedure. 223 Analytical sensitivity ranged from 98-100% and specificity from 90-100% (Table 2)

Clinical performance 236
We analyzed 254 fecal samples collected from children in Bangladesh on both the nL-237 qPCR chip and the TAC to compare performance. Overall percent agreement was 90% for the 238 >4500 reactions and negative percent agreement was 97% (95% CI 0.96, 0.98)(Cohen's Kappa = 239 0.66 (95% CI 0.63 -0.69)). Positive percent agreement was highly dependent on concentration 240 of the target gene. At concentrations above nL-qPCR detection limits (>10 7 copies/g stool) 241 positive percent agreement was 90%; this dropped to 62% for concentrations near the nL-qPCR 242 detection limits (10 5 -10 7 copies/ g stool) and fell to 8% for concentrations below 10 5 copies/g 243 stool. In instances where both methods detected the presence of target genes, nL-qPCR assays 244 displayed a median underestimation bias of -0.34 log 10 copies [IQR -0.41, -0.28] (see Table 3 245 and Figure S3 for individual assay statistics). 246 Reactions detected by TAC but not by nL-qPCR were typically below nL-qPCR detection The primary difference in performance we observed was that a majority of the nL-qPCR 280 assays had detection limits 1-2 orders of magnitude higher than TAC. This was due to reactions In studies where quantitation is required at lower concentrations than were achieved in 299 this study, pre-amplification can be performed as described by Ishii et. al. (28). In addition, pre-300 printing primers directly onto chips, similar to the TAC spotting procedure, can reduce 301 detection limits by nearly 50%. However, a major advantage of the nL-qPCR SmartChip™ is the 302 flexibility of the platform. Therefore, if a research team does opt to pre-print primers onto 303 chips, we suggest also maintaining a stock of unprinted chips on-hand. The current 304 configuration of the chip was designed with large-scale epidemiology studies in mind, thus 305 increased throughput was prioritized over the inclusion of a higher number of assays. However, 306 researchers wishing to focus on a smaller set of targets can evaluate more samples per plate 307 (further reducing per-sample costs), or the number of samples can be reduced to accommodate 308 an increased number of assay targets. In large-scale studies, replicating analysis for 309 questionable samples is often necessary (e.g. when replicates give discordant results). 310 Unprinted chips allow for an operator to run a limited suite of sample/assay pairs that need to 311 be reanalyzed: for example, 384 samples with questionable results in the initial run from a large 312 study could be analyzed against a minimal suite of 12 assays on a specially designed chip at the 313 end of the study. This facilitates the resolution of discordant results and minimizes missing 314 values in the final dataset, which will maximize statistical power in the analysis stage. 315

316
Unprinted nL-qPCR chips also allow end-users to substitute assays from the ones that 317 we publish here, with appropriate assay validation. This evaluation included 10 pre-published 318 assays that operate at similar PCR conditions, and found they performed well in nL format, 319 suggesting end users have flexibility in re-designing the chip. We further show that seven 320 primer pairs previously validated using TaqMan with probe-based dyes had excellent specificity 321 among 96 fecal samples when utilized with SYBR Green intercalating dye instead. These results 322 suggest the additional reagent costs associated with probes is not necessary to achieve high 323 specificity and is consistent with other findings that have reported equal or superior specificity 324 with SYBR Green compared to TaqMan chemistry (59, 60). for absolute quantification of total bacteria due to the fact that general bacterial contamination 346 (via 16S rRNA) was detected in almost half of the no-template control samples, albeit at 347 concentrations near the detection limit. To ensure potential low-concentration contamination 348 is identified, we strongly recommend incorporation of replicates when using this technology or 349 more stringent C q filtering (e.g. C q 28 or lower). Another limitation of the current configuration 350 is the omission of viral enteric pathogen targets. The primary aim for this study was to validate 351 the nL-qPCR technology for bacterial and parasitic targets, and we expect that future iterations 352 of the chip will include viral targets, which could be combined with a reverse-transcriptase 353 protocol for the study of RNA as well as DNA viruses. 354 In conclusion, we found the nL-qPCR pathogen chip to be an acceptable alternative to        Table 2. Analytical performance of the nL-qPCR pathogen chip. 660 Table 3. Bias estimates by assay on calculated log 10 copy number per gram of stool for nL-qPCR 662 compared to TAC. The table includes 529 reactions that were concordant for detection on both 663 platforms. 664 Figure S1. Within-chip intra-assay precision among samples containing a mixture of synthetic 666 DNA standards and extracted DNA from 3 separate fecal samples. Each sample was assayed 10 667 times on a single chip. Synthetic DNA was added at either (a) 100 or (b) 10 5 copies/reaction. qPCR and TAC. The blue shaded area represents the mean bias and its 95% confidence interval. 680 The upper (green) and lower (red) limits of agreement and their corresponding 95% confidence 681 intervals are also shown. 682