Abstract Body

Background: Bench-top next-generation sequencing (NGS) has become increasingly popular for HIV applications within the past five years. While “454”-based methods have dominated HIV research, the Illumina MiSeq is gaining popularity based on cost, ease of use, read depth and accuracy through homopolymer regions. Methodology: A total of 514 plasma samples were run on two NGS platforms, and sequencing data from the HIV V3 region were analyzed. Stored firstround PCR products were obtained from the samples which were originally sequenced on a Roche/454 GS-FLX. The samples underwent further amplification with adaptors appropriate for the Illumina MiSeq. Barcoded samples were sequenced in multiplexed runs of 96 samples on both instruments. Samples were excluded if they had <750 reads by GS-FLX or <10,000 by MiSeq. CXCR4-using tropism was assigned if ≥2% of sequences had a geno2pheno false-positive rate <3.5 - parameters which were previously optimized for the 454, though not the MiSeq. Results: Both platforms were well correlated in terms of the proportion of X4 reads per sample (slope= 0.91, R2=0.90). Median read depth was ~89,000 by MiSeq and ~2500 by GS-FLX. For tropism classifications, the platforms were 89% concordant (459 of 514 samples, Kappa=0.78). Of the 55 samples with discordant classifications, 62% were X4 by MiSeq but R5 by 454, indicating potential over-sensitivity for minor variants by MiSeq. The median difference in %X4 reads between platforms was 0.2% higher by MiSeq (interquartile range [IQR]: -1.1%-1.3%). Where discordant, samples differed by an absolute value of 2.5% X4 (IQR: 1.3%-8.3%). We also noted systematic low-level carry-forward contamination on the MiSeq platform: on average, a minimum of 0.04% of the total number of recovered reads (standard deviation 0.02%) were obvious contaminants (non-V3 sequences carried over from previous runs). Contaminant sequences persist in subsequent runs, appearing as many as 7 runs after they were originally processed. Conclusions: The MiSeq and GS-FLX systems were well correlated with each other and gave similar tropism classifications in 89% of samples. Where discordant, MiSeq tended to call more samples as X4. Laboratories using MiSeq platforms should account for carry-over contamination where possible in their data processing.