Ethics Statement

The SCOPE study was approved by the University of California San Francisco Committee on Human Research and the University of British Columbia/ Providence Health Care Research Ethics Board. All participants provided written informed consent.

Study Cohort

Subjects were identified from the University of California, San Francisco SCOPE cohort, a clinic-based ongoing prospective cohort of adults with chronic HIV-1 infection who are followed at either San Francisco General Hospital or the San Francisco VA Medical Center. Both treatment-naïve and treatment-experienced patients are enrolled. As previously described [16], SCOPE was enriched for treatment-experience patients who were initiating a raltegravir-based regimen. At the time of our current study, a total of 73 subjects enrolled in the SCOPE cohort had initiated a raltegravir-based regimen, 24 of whom failed to achieve an undetectable viral load. Eight of these 24 patients developed at least one primary or secondary raltegravir-associated DRMs in plasma by either/both Sanger and 454 “deep” sequencing. Out of the eight patients with DRMs, five of them developed at least one of the three major raltegravir resistance pathways (Y143R/C/H, Q148H/R/K and/or N155H/S), had longitudinal paired plasma and PBMC samples available, and had samples that could be successfully amplified and sequenced. Among these five patients, four (3180, 3242, 3501 and 3508) were raltegravir-naïve at the start of the study. These four subjects were the focus of this study.

Sanger Sequencing

HIV-1 RNA or DNA from plasma or PBMC (approximately 5×10⁶ cells per sample) was extracted by NucliSENS easyMag (bioMérieux). For the RNA samples, full-length HIV-1 int (HXB2 nucleic acid coordinate 4230–5093, in at least three replicates) and the protease and reverse transcriptase portion of HIV-1 pol (HXB2 nucleic acid coordinate 2253-3749 or 3269, single reaction) were amplified with reverse transcription and first-round PCR using SuperScript III One-Step RT-PCR System with PlatinumTaq (Invitrogen) followed by “nested-PCR.” To obtain total HIV-1 DNA from the PBMC samples, reverse transcription was omitted before PCR (see Table S1 for primers used). Sanger sequencing was performed on ABI 3730 DNA Sequencer using BigDye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems). Alignment and base-calling for the resulting chromatograms were done by in-house software RECall [22]. All Sanger sequences have been deposited into GenBank (accession numbers JQ666887-JQ667036).

454 “Deep” Sequencing

454 “deep” sequencing was performed with Genome Sequencer FLX System Standard Kit with a read length of 250 bases according to the supplied protocol (454 Life Sciences, Roche). The forward 454 primer covered HIV-1 integrase (int) amino acids position 83–152 (210 bases), and the reverse primer covered position 117–192 (228 bases) (Table S1). Samples were processed in triplicates as mentioned above, and were pooled for deep sequencing to ensue random sampling. All 454 “deep” sequencing raw and processed data are available from the authors upon request.

“Deep” Sequencing Data Processing and Analysis

In order to gain insight into potential inadequate sampling of the plasma and/or PBMC populations, we attempted to estimate the input copy number of HIV-1 genomic templates in 454 “deep” sequencing reactions with a newly reported “deep” sequencing input copy quantification method [23]. In this method, each HIV-1 genomic template would receive a random “primer ID,” thus allowing input number quantification at the end of the “deep” sequencing pipeline. Three pairs of representative samples were selected which showed discordance in plasma and PBMC (subject 3180 day 177, subject 3501 day 226 and subject 3508 day 197).

As for the main dataset, resulting 454 “deep” sequences were aligned with our in-house 454 data-processing pipeline. Results were analyzed with (a) no sequence exclusion criterion, and (b) exclusion of sequences with an arbitrary cut-off of less-than-or-equal-to two reads within a sample. The “number of reads” was defined as the total number of occurrences detected per unique sequence within a sample.

In the first part of our analysis, prevalence of each raltegravir-associated DRM in each sample was calculated. Primary raltegravir-associated DRMs were defined to be E92Q, Y143R/C/H, Q148H/R/K, and N155H/S (HXB2 int amino acid coordinates), according to the International AIDS Society guidelines [11]. Secondary raltegravir-associated DRMs were defined to be E92V, Q95K, T97A, F121Y, E138A/K, G140A/C/S, S147G, V151A/I/L, M154I/L, E157Q, and G163K/R, and linkages examined were T97A and Y143C/R, E138A/K and Q148H/K/R, G140S and Q148H/K/R, and G163K/R and N155H (consensus B definitions), according to Stanford University HIV-1 Drug Resistance Database [13]. Since the forward and reverse 454 primers resulted in double coverage at amino acid position 117–152, DRM prevalence values reported were adjusted to include results from both primer directions.

In the second part of our analysis, 454 “deep” sequences from each PBMC sample were computationally matched with sequences from its paired plasma sample in order to quantify the fractions of sequences within one compartment (PBMC) which have matching sequences in its corresponding compartment (plasma). When a match was found, the sequence pairs were removed from the analysis pool. A Python programming script was used to perform the matching process. A “roll-in” step was included for quality control purpose: Sequences that matches in all positions except Ns and out-of-frame indels were considered “ambiguous match” to account for the intrinsic PCR and 454 “deep” sequencing technical errors, and the counts of these “ambiguous” sequences were “rolled-into” the most prevalent sequence of the group.

All phylogenetic analyses in this study were performed with ClustalX using the neighbor joining approach.

Prolonged and Substantial Discordance in the Prevalence of Raltegravir-associated DRMs between Plasma and PBMC

The baseline pre-raltegravir characteristics of the four subjects studied are outlined in Table 1. Paired plasma and PBMC samples were collected pre- and post- baseline longitudinally for up to 497 days post-therapy. All samples were subjected to bulk Sanger sequencing and 454 “deep” sequencing. Sequences with less-than-or-equal-to two reads prevalence were excluded from analysis to adjust for PCR errors, resulting in an average read depth of 2483 (minimum: 715, maximum: 7119).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Patients characteristics (N = 4).

https://doi.org/10.1371/journal.pone.0046181.t001

We observed substantial differences in the resistance genotypes of the viral populations in the plasma and PBMC compartments. Table 2 summarizes the prevalence of major raltegravir-associated DRMs in each sample, revealed by 454 “deep” sequencing. Strikingly, prevalence of DRMs was consistently higher in plasma than in PBMC. A pictorial example of the observed trend is given in Figure 1. As shown in this figure, 454 “deep” sequencing revealed that in subject 3501, prevalence of all DRMs detected in PBMC were below 15% even at 338 days post-therapy. In contrast, prevalence of plasma DRMs increased rapidly to 69% (Q148H and G140S) 54 days post-therapy. The same trend for discordance was observed in secondary raltegravir-associated mutations T97A, E138A/K, S147G, V151I, M154I and G163R (Table 2 and Table S2, individual DRMs; Table S3, linked DRMs). Due to the higher prevalence of DRMs in plasma, it was not surprising that they were consistently detected earlier in plasma than in PBMC samples with multiple replicates of Sanger sequencing (Table S4a and S4b).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Prolonged and substantial discordance in the prevalence of DRMs in plasma and PBMC.

A pictorial representation of the prolonged and substantial discordance in the prevalence of DRMs in plasma and PBMC, taken from subject 3501 (see Table 2). An average of 3415 sequences (minimum 1663, maximum 5988) were sampled at each time point by 454 “deep” sequencing. Raltegravir-associated DRMs (A) E138A, (B) E138K, (C) G140S and (D) Q148H were detected at a higher prevalence in plasma than in PBMC. Squares (▪) and solid lines represent plasma samples. Triangles (▴) and dashed lines represent PBMC samples. Diamonds (♦) and dotted lines represent the subject’s log viral load.

https://doi.org/10.1371/journal.pone.0046181.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Prolonged and substantial discordance in the prevalence of drug resistance mutations in plasma versus PBMC. Percentages represent the prevalence of the indicated resistance mutations revealed by 454 “deep” sequencing.

https://doi.org/10.1371/journal.pone.0046181.t002

Discordance in plasma and PBMC continued even as the raltegravir-containing therapy terminated. Plasma DRMs prevalence declined in subjects 3180 and 3242 as they stopped receiving raltegravir-containing regimens on day 177 and 224, respectively (Table 2, superscript ^a), which was consistent to previous reports [17]. Note, however, that in subject 3180, the prevalence of Q148H and G140S in PBMC increased from 1% to 6% while the plasma prevalence decreased from 100% to 69%. Post-raltegravir paired plasma and PBMC samples were only available after 150 days of therapy termination for Subject 3242, and DRMs prevalence in both compartments were at 0%.

Similar discordance in plasma and PBMC was also observed in subject 3508, who developed multiple raltegravir-resistance pathways. At 83 days post-therapy, Q148R and N155H were detected in plasma and PBMC at 72% versus 33% and 12% versus 0% respectively. Note 115 days later, the prevalence of Q148R and N155H in both compartments dropped to 0% while Y143R took over and reached 100% in plasma yet remained at 0% in PBMCs.

The above analysis was also repeated without any sequence exclusion criterion, resulting in an average “deep” sequencing read depth of 3272 (minimum:1230, maximum: 8687) and yielded similar discordance patterns (results not shown). Protease and/or reverse transcriptase inhibitors (PI, NRTI and NNRTI) associated DRM profiles were also examined with Sanger sequencing, and were relatively concordant between plasma and PBMC in these heavily treatment-experienced individuals (Table S5).

Profound and Substantial Discordance in Absolute Sequence Identities between Plasma and PBMC

Another way to compare viral populations in plasma and PBMC is to compare the prevalence of sequences that share absolute sequence homology in the two compartments. For each unique “deep” sequence obtained within a PBMC sample, we computationally searched for a matching sequence in its paired plasma sample. Our findings are graphically represented in Figure 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Absolute sequence homology comparison revealed 21% to 99% of sequences derived from PBMC samples had no identical match in their plasma counterparts.

An absolute sequence homology comparison revealed prolonged and substantial differences between plasma and PBMC viral populations. (A) Subject 3180, (B) Subject 3242, (C) Subject 3501, and (D) Subject 3508. 454 “deep” sequencing sampled an average of 3345 sequences from each sample. The forward 454 primer covered HIV-1 integrase (int) amino acids at approximately position 83–152 (210 bases), and the reverse primer covered approximately position 117–192 (228 bases). Within each sample, identical sequences were grouped, and the prevalence of each group of unique sequences was calculated. Next, a pair of plasma and PBMC collected from the same subject at the same time point was compared. Each group of identical sequences within a PBMC sample was compared with all the sequences derived from the plasma. The result was graphed on individual horizontal bars in this figure, in which each bar represents 100% of the total viral population sampled. When a match was found (defined by 100% nucleotide similarity and a minimum overlap of 73 bases and a maximum overlap of 228 bases), the corresponding sections in the horizontal bars were shaded white and a vertical/slanted line was drawn to link the two identical groups from the two compartments. Solid grey shading represents sequences that found no match. Note that the white sections represent groups of identical sequences, but in contrast the grey sections represent all sequences that found no match and thus represent a heterogeneous mixture of sequences.

https://doi.org/10.1371/journal.pone.0046181.g002

Due to the inclusion of a roll-in process outlined in the Materials and Methods section as a quality control process, we first started with the raw dataset without any sequence exclusion criterion, which had a raw average read depth of 3345 (minimum: 1233, maximum: 8688). The minimum/maximum read depths in this analysis were different from the first part of the study because here we only included samples that had paired PBMC/plasma samples. The average number of unique reads before roll-in was 816 (minimum: 326, maximum: 1558). After roll-in, the average number of unique reads decreased to 662 (minimum: 254, maximum; 1306).

We again observed that plasma and PBMC exhibited drastically different viral populations. As shown in Figures 2A, B, C, D, 21% to 99% of sequences derived from PBMC samples had no identical match in their plasma counterparts, whereas 14% to 100% of sequences derived from plasma samples had no identical match in their PBMC counterparts (grey shading within horizontal bars, Figure 2A, B, C, D). At least some of these sequences that had no match contained DRMs. For example, in Subject 3180 on day 177 post-therapy, 22% of the plasma viral sequences had no match in the PBMC population (Figure 2A, grey shading), however the plasma DRM prevalence at this time point approximated 100% (Table 2). This implied most variants among the 22% of plasma sequences which did not match with any PBMC sequence contained DRMs.

In cases in which identical sequences were found, the prevalence of such identical sequences was very different in the two compartments (Figure 2A, white shades). For example, in Subject 3180 on day 177 post-therapy, 77% of the plasma virus shared identical sequences and contained the G140S/Q148H mutations (Figure 2A, day 177 white shading), but the same sequence could only be found in 1% of the counterpart PBMC population.

The above analysis was repeated with (a) reversing the match direction (plasma to PBMC), and (b) imposing an exclusion criterion of less-than-or-equal-to two reads in the initial dataset. Both resulted in similar observations as reported in Figure 2 (results not shown). Table S6 summarizes read depths and number of unique reads in all 454 “deep” sequencing analysis performed.

In another attempt to confirm findings in Figure 2, we reanalyzed the dataset by defining “discordance” as the “fraction prevalence of sequences within a sample (PMBC/plasma) that has an ‘exact’ or ‘ambiguous’ match in its corresponding compartment (plasma/PBMC).” We started with no sequence exclusion criterion and calculated the fraction sum of sequences within one sample that found an “exact match” (two sequences differing in single-base insertions and Ns did not count as a match) in its corresponding compartment, or an “ambiguous match” (two sequences differing in single-base insertions and Ns were counted as a match) in its corresponding compartment, allowing for repeated matches. This alternative approach to interpret the dataset also showed discordance in the plasma versus PBMC viral populations: We observed that 0–76% of PBMC sequences found an “exact match” in plasma; 2–89% found an “ambiguous match”; whereas 0–90% of plasma sequences found an “exact match” in PBMC, 4–98% found an “ambiguous match.” The match was performed in both directions (plasma-to-PBMC and PBMC-to-plasma) and similar discordance was observed.

Furthermore, in attempt to provide a potential explanation for the observed discordance, we hypothesized that a low plasma viral load and/or CD4 count prevented the rapid development of equilibrium between the plasma and PBMC viruses, leading to increased discordance in the two compartments. To address this hypothesis, we asked if a statistically significant association existed between “discordance” quantified as fraction prevalence as defined above versus “patient viral load” or “CD4 count on the date of sample collection.” Spearman correlation tests did not indicate any significant association (p>0.05) between patient viral load and fraction prevalence of finding a match. The same test was repeated with patient CD4 count; no significant association was observed.

Technical Validation: Sampling Depth and Contamination Check

Inadequate sampling of the starting HIV-1 genomic templates may bias the observed viral species diversity within a sample in “deep” sequencing studies and could result in potentially misleading results. To reduce this, we undertook triplicate sampling at all time points in our longitudinal analyses. In addition, to directly assess the effects of “oversampling”, we performed an independent 454 “deep” sequencing run using a novel “primer ID” assay [23] which tags individual starting molecules with a unique barcode as outlined in the Materials and Methods section (minimum reads 1771; maximum reads 3261). Importantly, discordance in DRMs prevalence found in plasma versus PBMC was reproducible with or without the primer ID (results not shown). Following Jabara’s method, we used a cut-off of accepting at least three reads per primer ID. Our results indicate that the three RNA samples started with 253, 230 and 402 HIV-1 RNA templates (patient viral load 4.4, 5.68 and 4.69 log per mL respectively); whereas 224 and 202 HIV-1 DNA templates were picked up in two of the three samples and one sample failed sequencing. Using a more conservative cut-off than Jabara et al. - requiring at least four reads before a primer ID was called "real" - the same three RNA samples started with 210, 122, and 311 HIV-1 RNA templates and the two DNA samples that passed started with 54 and 49 copies. These results suggest that there was indeed “oversampling” in the 454 analyses of both plasma HIV-1 RNA and PBMC HIV-1 DNA, but that the effective input copy numbers should be high enough to reflect some degree of inherent variability in both the PBMCs and plasma.

Finally, to rule out potential sample mixup and contamination, we performed phylogenetic analyses on all sequences generated in this study (Figure S1, int Sanger sequences; Figure S2, int 454 “deep” sequences in which the top-most prevalent sequences in PBMC samples and their matching sequences in plasma samples were included). No contamination was found. Importantly, our phylogenetic analysis showed that the most prevalent sequences derived from paired plasma/PBMC did not generally share the closest genetic distances among sequences derived from the same individual, supporting our claim that plasma and PBMC had different viral population makeup.