Relationship between the number of required ultra-deep pyrosequencing reads per position (Required coverage) and the detection limit of the frequency of a minority sequence variant (Detection threshold). The excess number of reads required to detect minor variants is a result of the Poisson model for handling potential sequencing errors. For example, ∼400 (rather than 100) reads are required to detect a minor variant present at about a 1% level in a non-homopolymeric region. Because homopolymeric regions are more likely to contain sequencing errors, higher numbers of reads are required to obtain similar sensitivities in these regions.

Sequence variants detected by ultra-deep pyrosequencing of eight clinical plasma samples including seven samples from antiretroviral-experienced patients (F56272, V10606, V11909, V47683, V63557, V65472, V9878) and one sample from an untreated patient (92_2664). Sequence variants were defined as differences from the consensus population-based sequence. The X-axis represents all 99 protease codon positions followed by the first 240 reverse transcriptase codon positions. Positions are demarcated at intervals of 20 codons. The Y-axis shows the frequency of each minor variant observed by ultra-deep pyrosequencing. Synonymous minor variants are shown in gray; non-synonymous variants are shown in red. The presence of more than one variant at the same position is indicated by superimposing variants of lower frequency onto variants with higher frequency. Drug-resistance mutations detected only by ultra-deep pyrosequencing are indicated with arrows. A horizontal line separates minor variants present at above and below a frequency of 20%. As noted in the text, 55 of 72 variants present in >20% of GS20 reads, but only nine of 392 variants present in <20% of GS20, were detected by conventional Sanger sequencing.