Representative rotavirus isolates were selected for evaluating analytical performance, including Wa (G1P[8]), DS-1 (G2P[4]), AU-1 (G3P[9]), ST3 (G4P[6]), 69M (G8P[10]), 116E (G9P[11]), I-321 (G10P[11]), L26 (G12P[4]). Fecal samples tested previously positive for rotavirus by ELISA were provided from studies at the International Centre for Diarrhoeal Disease Research, Bangladesh, Kilimanjaro Christian Medical Centre, Tanzania, and Division of Consolidated Laboratory Services, Virginia. Bangladeshi samples were selected from a birth cohort study (2008 to 2009) in the Mirpur region of Dhaka (Mondal et al., 2012). Tanzanian samples were collected from inpatients with diarrhea from Kilimanjaro Christian Medical Centre and referral hospitals in Moshi from February 2008 to June 2009. More than 70% of the rotavirus positive samples were from children under age five. Virginia specimens were rotavirus positive diarrheal specimens collected during routine outbreak investigations (from February to April, 2011) by Division of Consolidated Laboratory Services, Virginia. All studies were approved by the University of Virginia, International Centre for Diarrhoeal Disease Research, Bangladesh and Kilimanjaro Christian Medical Centre institutional review boards.

Nucleic acid was extracted from fecal samples using the QuickGene RNA tissue kit SII (Fujifilm, Tokyo, Japan) as described previously (Liu et al., 2011).

Genotype specific primers and probes were designed in the variable regions of VP7 and VP4 and adapted or modified from published assays wherever feasible (Aladin et al., 2010; Gentsch et al., 1992; Gouvea et al., 1990; Iturriza-Gomara, Kang, and Gray, 2004) (Table 1). Oligonucleotides were synthesized by Integrated DNA Technologies (Coralville, IA) and Biosearch Technologies (Novato, CA). Real time RT-PCR was performed with AgPath-ID RT-PCR kit (Life Technologies, Carlsbad, CA) with a CFX system (BioRad, Hercules, CA). Three panels were formulated as Panel GI, including G1, G2, G9, and G12; Panel GII, including G3, G4, G8, G10; Panel P, including P[4], P[6], P[8], P[10], P[11]. An internal control assay targeting NSP3 was incorporated in both G panels with final primer and probe concentration at 200 nM and 100 nM, respectively (Zeng et al., 2008). The samples were denatured by incubation at 95°C for 5 min then on ice prior to mixing with the one step real time RT-PCR reagents. The cycling condition included reverse transcription at 45°C for 20 min, denaturation at 95°C for 10 min, 45 cycles of 95°C for 15 sec and 55°C for 1 min.

RT-PCR-Luminex assays were performed with QIAamp One Step RT-PCR kit (Qiagen, Valencia, CA) followed by luminex detection with a BioPlex system (BioRad) (Liu et al., 2011). Final concentrations of biotinylated and non-biotinylated primers were 300 nM and 200 nM, respectively, for all the genotype specific primer assays, except 150 nM and 100 nM for NSP3. Carboxylate microspheres were labeled with oligonucleotide probes with amino modifier, and luminex detection was performed as described previously (Liu et al., 2011). The cutoff set for positivity was two-fold Median Fluorescence Intensity above background (nuclease free water).

PCR amplicon was generated with consensus primers for VP7 (Beg9 and End9) and VP4 (con2 and con3) described previously (Gentsch et al., 1992; Gouvea et al., 1990), and sequenced by GENEWIZ (South Plainfield, NJ). Con2 and con3 for VP4 were modified slightly, forward primer 5’-TGGCTTCRCTCATTTATAGACA-3’, reverse primer 5’-ATTTCNGACCATTTATAWCC-3’.

Consensus VP7 or VP4 amplicon from a positive sample for each genotype was cloned, amplified and in vitro transcribed according to the previous protocol (Liu et al., 2011), with T7 RNA polymerase and SP6 RNA polymerase, respectively. The two RNA transcript products were mixed in equal molar concentration measured with Nanodrop (Bio-Rad) in 50 mM Tris, pH 8.0, 1mM EDTA, 100 mM NaCl, then denatured at 90°C for 3 min and hybridized by cooling down slowly to room temperature to generate double stranded templates. Non-denaturing 2% agarose gel electrophoresis was run to ensure the formation of double stranded RNA. For analytical performance, double stranded RNA transcripts were spiked into lysis buffer during extraction of fecal samples from healthy donors.

Correlation was tested by regression analysis using the analysis of variance (ANOVA) tests. Mixed infection rates were compared with Chi-Square test. All P values were two-tailed and values of <0.05 were considered statically significant.

Linearity and limit of detection were determined using in vitro transcripts of VP7 and VP4 sequences corresponding to the interrogated G/P-types. As shown in Table 2, Pearson coefficient varied between 0.98 and 1.00 for linearity. Limit of Detection, defined as the lowest concentration to achieve 100% detection in ten spiked samples, was 10⁶ copies of in vitro transcripts per gram of stool (equivalent to 200 copies per RT-PCR reaction prior to extraction). The average quantification cycle (Cq) values at the Limit of Detection were designated as the analytical cutoff used for further data analysis. Specificity of the assays was tested with 8 rotavirus isolates, each was positive for the expected G/P genotypes, and there were no false VP7 or VP4 detections. Furthermore, no cross-reaction was observed in clinical stool samples that were positive for adenovirus (n = 8), astrovirus (n = 3), norovirus GII (n = 4), diarrheagenic E. coli (EAEC, EPEC, ETEC, n = 12), Campylobacter (n = 3), Cryptosporidium (n = 3), Giardia lamblia (n = 6).

Forty Bangladeshi samples were selected to evaluate the real time assays because they were genotyped previously with a conventional gel-electrophoresis based RT-PCR method (Gentsch et al., 1992; Gouvea et al., 1990). Ninety-five percent of conventional results were confirmed by real time RT-PCR (Table 3). The four unconfirmed results included two samples that were genotyped as G9 previously but were G12 with real time RT-PCR and sequencing, one sample that was genotyped as P[8] previously but non-typeable with the real time assay and failed to generate amplicon with consensus VP4 primers, and one sample that was identified as a mixed P type previously but genotyped as single P[8] by real time RT-PCR and sequencing. In addition, the real time assays detected mixed G-types and P-types in 30% and 25% of the samples, versus 10% and 7.5% by conventional methodology, respectively (P < 0.05). Overall multiple G or P types were identified in 33% of the samples. For mixed G or P type infections, the dominant type was evaluated by qPCR Cq (since efficiency was similar between the assays, effectively the dominant type was that with the lowest Cq). Fourteen samples were subjected to amplicon sequencing with consensus VP7 and VP4 primers and revealed 100% concordance with the dominant types (labeled with asterisks in Table 3) identified by real time RT-PCR. The minority types in these samples were amplified with one genotype specific primer combined with the corresponding consensus VP7/VP4 primer, followed by sequencing. All results were confirmed except for two samples where the amount of amplification product was too low to produce reliable sequencing data.

Samples from Tanzania and Virginia were selected randomly without previous genotyping information. Amplicons were generated with consensus primers for VP7 and VP4 and were sequenced as the gold-standard (Table 3). The dominant genotypes detected with real time RT-PCR assays were 100% consistent with sequencing results on 49 Tanzanian and Virginian samples. One of 11 Virginia samples had mixed G-types and 3 had mixed P-types. G12P[8] was found to be the predominant genotype. Among 38 rotavirus positive samples from Tanzania, 50% were G1P[8], 11% G12P[6], 8% G1P[6], 8% G8P[6], 5% G2P[4], 5% G9P[8], 5% G3 (P-type non typeable), 3% G12P[8], and 5% P[8] with mixed G1 and G3.

Quantitation was used to determine the G and P type combination in mixed infections. The copy number of each genotype in a sample was calculated based on the linear regression derived from the linearity experiment. Then the relative abundance of each G or P genotype in a mixed infection could be plotted as exampled in Figure 1. In this sample, the predominant type would most likely be G9P[8] followed by G9P[4], while the minority types could presumptively be G2P[6] and G2P[4], or some other combinations. In addition, the Cq value of the predominant type in each sample was plotted against NSP3 Cq value in the corresponding sample (Figure 2). A tight linear correlation was obtained with the set of samples tested (n = 89; R² = 0.877, P < 0.001).

RT-PCR-Luminex showed 87% concordance with real time RT-PCR results (Table 3). Discrepancies were exclusively due to samples with low viral load, usually minority types that were not identified by RT-PCR-Luminex assay. Receiver Operating Characteristic analysis showed that RT-PCR-Luminex often lost detection for the genotypes with a Cq value after 35.2.