Researchers at the Gladstone Institutes and the University of California, San Francisco are evaluating next-generation sequencing for its ability to detect early signs of drug resistance in HIV patients.
In an Illumina-sponsored webinar, Gordon Bentley, a senior research associate at the Gladstone Institute of Virology and Immunology, discussed his group's use of deep amplicon sequencing of the HIV reverse transcriptase gene to look for known mutations conferring resistance to antiretroviral drugs.
His team has been evaluating both Illumina's MiSeq platform and Roche's 454 GS FLX+ system to do viral haplotyping in the HIV-RT gene. "This approach requires long-read sequencing," Bentley said during the presentation.
In an initial comparison, the team created 48 amplicons from 48 HIV RNA specimens, covering 465 bases of the HIV-RT gene. The amplicons were indexed and pooled and sequenced on each system.
Using the GS FLX+, the team sequenced the entire 465-base target in its entirety in both the forward and reverse directions. Sequencing on the MiSeq covered the region with 2x250-base paired end sequencing with a 35-base overlap in the middle.
The group performed analysis of the data from both systems using CLC bio's Genomics Workbench software.
MiSeq generated around 5 million paired end reads of 465 bases, 91 percent of which were at Q30 or above. The GS FLX+ produced 720,000 reads that clustered at 485 bases with 82 percent at Q30 or above.
Twenty-three of the specimens came from four recently infected subjects that had participated in a pre-exposure prophylaxis, or PrEP, clinical trial of Truvada, a combination pill that contains drugs that prevent the virus from making copies of itself. Such drugs, known as nucleoside reverse transcriptase inhibitors, incorporate themselves into the virus by binding to the reverse transcriptase gene, which prevents HIV from multiplying.
However, said Bentley, "drug resistant mutations arise and they make it difficult for NRTI drugs to bind to the HIV-RT gene and become incorporated." There are known drug resistance sites within the 465-base region of the HIV-RT gene.
As part of the PrEP trial, the subjects that were sequenced had started the drug and then stopped as per the study protocol and were then tracked longitudinally.
In one subject, sequencing on both platforms detected a point mutation known to confer drug resistance that was also validated with PCR. Because the MiSeq generates so many more reads than the GS FLX+ there was "wide difference in read depth" of the mutation, Bentley said.
When the team evaluated the variant frequency over time, they found that 24 weeks post-infection, allele frequency was determined to be 2.47 percent, 2.49 percent, and 2.04 percent by GS FLX+, MiSeq, and PCR, respectively. "So, very concordant data," said Bentley. Going further out in time, however, GS FLX+ determined variant frequency to be 0.42 percent and the MiSeq determined it to be 0.53 percent.
Looking at variant frequencies in the remaining subjects, Bentley said that the platforms were all very concordant, calling variants down to 0.5 percent frequency.
However, as variant frequencies drop, the MiSeq may have an advantage over the GS FLX+, he said, simply because it generates so many more reads.
"At the very low end of the frequency range, one possible advantage of the MiSeq is that we are able to feel more confident about the variant frequencies obtained when we have many reads representing that variant," Bentley said.
For instance, when the team did viral haplotyping over time of one subject, they identified a very low frequency variant that was present in all time points, but only at 0.29 percent frequency by MiSeq and 0.26 percent frequency by GS FLX+. However, 312 MiSeq reads contained the variant, while only 26 GS FLX+ reads had the variant. "When the number of reads drops below 50, some people may start to get uncomfortable whether that number of reads can be trusted," he said. With so few read counts it is difficult to distinguish between true variants and sequencing error.
Additionally, Bentley found that the GS FLX+'s accuracy may be impacted by homopolymers. For instance, there is a known drug resistance site, k65r, that is flanked by a homopolymer stretch. Specimens with this subtype that were sequenced with the GS FLX+ had "numerous insertions and deletions surrounding the k65r codon in the reverse reads," Bentley said. However, the forward reads did not contain those insertions and deletions. Additionally, Bentley found no insertions and deletions in the MiSeq reads. He said he thinks the homopolymer regions is "responsible for the insertions and deletions," and makes it "difficult to get an accurate variant frequency result."
Comparing the cost of the two protocols, Bentley said that one run of 48 amplicons on the MiSeq cost around $1,500, or $31 per sample. Generating the data on the GS FLX+ cost around $6,000, or $120 per sample.
Going forward, Bentley said that in order to integrate HIV-RT sequencing clinically on the MiSeq, it would be necessary for consortiums to get together to do reproducibility experiments and studies to determine sensitivity and specificity.
Additionally, it will be critical to set "appropriate cut offs for the determination of what is a positive variant frequency result versus what would be considered a negative, or non-clinically relevant result." While both platforms demonstrated that drug resistant variants were present at low frequencies in individuals participating in PrEP, it is not clear at what level those variants are clinically relevant.
Other researchers looking to implement next-gen sequencing in identifying low-frequency HIV drug-resistant mutations are also struggling to determine at what level a variant becomes clinically significant.
Current approved methods for detecting drug-resistant mutations use Sanger sequencing, which does not reliably detect variants below around 10 to 20 percent frequency. For instance, in a study testing an assay developed by Roche to identify HIV drug-resistant mutations, researchers found variants that were missed by Siemen's TruGene assay, a US Food and Drug Administration-cleared assay that uses PCR and Sanger sequencing, because they were present at frequencies below what Sanger sequencing can reliably detect. However, it is unclear what clinical significance those variants have and what frequency level would be high enough to cause resistance (CSN 1/30/2013).