Ambiguous nucleotide calls from population-based sequencing of HIV-1 are a marker for viral diversity and the age of infection.

BACKGROUND: The time passed since the infection of a human immunodeficiency virus (HIV)-infected individual (the age of infection) is an important but often only poorly known quantity. We assessed whether the fraction of ambiguous nucleotides obtained from bulk sequencing as done for genotypic resistance testing can serve as a proxy of this parameter.
METHODS: We correlated the age of infection and the fraction of ambiguous nucleotides in partial pol sequences of HIV-1 sampled before initiation of antiretroviral therapy (ART). Three groups of Swiss HIV Cohort Study participants were analyzed, for whom the age of infection was estimated on the basis of Bayesian back calculation (n = 3,307), seroconversion (n = 366), or diagnoses of primary HIV infection (n = 130). In addition, we studied 124 patients for whom longitudinal genotypic resistance testing was performed while they were still ART-naïve.
RESULTS: We found that the fraction of ambiguous nucleotides increased with the age of infection with a rate of .2% per year within the first 8 years but thereafter with a decreasing rate. We show that this pattern is consistent with population-genetic models for realistic parameters. Finally, we show that, in this highly representative population, a fraction of ambiguous nucleotides of >.5% provides strong evidence against a recent infection event <1 year prior to sampling (negative predictive value, 98.7%).
CONCLUSIONS: These findings show that the fraction of ambiguous nucleotides is a useful marker for the age of infection.

Human immunodeficiency virus type 1 (HIV-1) infections are initiated in most cases by a single virus [1], leading initially to a monomorphic viral population. Subsequently, viral diversity builds up gradually during HIV infection, first in a linear fashion but then at decreasing rates until a plateau is reached [2, 3]. In late-stage HIV infection, even decreases in viral diversity have been observed [3]. Thus, the diversity of the HIV population within an individual patient is potentially informative about the age of the infection, which is an important parameter because it allows an assessment of how far and how fast the infection has progressed. Seroconversion data are often lacking, and acute retroviral syndrome may have not occurred or may not have been recognized as such [4]. Thus, a method to estimate the infection duration based on viral sequences would be attractive, given the abundance of HIV sequence data from genotypic drug resistance tests.

Genotypic resistance tests use nucleotide sequences to infer to what degree different drugs may inhibit a given viral population of an HIV-infected individual. For economic reasons, these sequences are obtained by bulk sequencing; that is, the sequencing procedure is applied to a diverse sample of the HIV population. If the frequency of the most frequent nucleotide at a given position exceeds a threshold (typically around 80%), bulk sequencing returns the predominant nucleotide at this position. However, if this is not the case, then so-called ambiguous nucleotide calls are reported, implying that the patient harbors viral strains with different nucleotides at this locus. Thus, the fraction of ambiguous nucleotides is a measure of the degree of polymorphism of the HIV population within a patient, which in turn should scale with the age of infection. Here, we assess to what degree the proportion of ambiguous nucleotides correlates with the time elapsed between HIV infection and sampling for genotyping. To this end, we relate the fraction of ambiguous nucleotides to the age of infection derived with methods of different accuracy from 4 data sets.

MATERIALS AND METHODS

We used previously analyzed nucleotide sequences from patients included in the Swiss HIV Cohort Study (SHCS) drug resistance database (see Kouyos et al [5] and GenBank accession numbers therein for a random sample of sequences). The SHCS is a nationwide, prospective, clinic-based cohort study with continuous enrollment and semiannual study visits [6, 7]. The SHCS has been approved by the ethical committees of all participating institutions, and written informed consent has been obtained from the participants. The SHCS drug resistance database contains the results of 13,201 genotypic resistance tests from 9,231 patients performed by the 4 laboratories engaged in HIV resistance testing in Switzerland, stored in a central database developed and hosted by SmartGene (Zug, Switzerland; Integrated Database Network System, version 3.5.0) [8]. Resistance data stem from routine clinical testing (60% of tests) and from tests performed retrospectively from frozen repository plasma samples (40% of tests). Retrospective sequencing was performed systematically by analyzing the earliest plasma sample available for each patient. All laboratories perform population-based sequencing of the full protease gene and at minimum codons 28–225 of the reverse transcriptase gene by means of commercial assays (Viroseq version 1, PE Biosystems; Virsoseq version 2, Abbott AG; vircoTYPE HIV-1 assay, Virco Lab) and in-house methods [9].

Subtype B dominates the Swiss HIV-1 epidemic [10, 11], and we therefore focused on this subtype (n = 9,157 sequences). Because antiretroviral therapy strongly distorts viral diversity [12], we included only sequences from patients who were therapy-naïve at the time of sampling and for whom an independent estimate of infection time was available (3,307 sequences, each from a different patient). The year of infection was estimated as described elsewhere [13] as the median of patient-specific infection time estimates based on a Bayesian back calculation model incorporating the dates of the first positive or last negative HIV test results and CD4 counts as predictor variables. Note that this infection time estimate is independent of the viral sequence. We refer to this data set as the full data set. Furthermore, we considered 3 additional sets of patients, for whom times or time differences are known with better accuracy: (1) In the set of seroconverters (366 patients), patients were recruited to the cohort within 1 year after infection, based on documented negative and positive HIV test results no longer than 180 d apart [11]; (2) The longitudinal set contains 124 patients for whom sequences are available at 2 time points (at least 6 months apart). Although for this data set the time of infection is not more precise than for the large data set, the age difference of the 2 samples is known exactly; and (3) The Zurich Primary HIV Infection Study (ZPHI; ClinicalTrials.gov identifier, NCT00537966) set contains 130 patients with sequences obtained during acute infection (median time from infection, 41 d; interquartile range [IQR], 28–56 d) [14].

The relationship between the proportion of ambiguous positions (the dependent variable) and the time since infection was investigated using linear regression analysis. Because model residuals were not normally distributed, we performed a bootstrap analysis with 1,000 replicates to obtain 95% confidence intervals (CIs). We verified results by repeating all analyses on logit-transformed fractions by use of a generalized linear model for proportions. Both crude and adjusted analyses, using the patient's mode of HIV acquisition, ethnicity, sex, and age, were performed. In addition, a variable coding for the sequence-generating laboratory was included to account for assay- or laboratory-specific effects. All P values were 2-sided, and the level of statistical significance was set at .05.

We sought to identify a cutoff in the proportion of ambiguous sites to classify a patient's infection status into recent (infected for ≤1 year) or chronic (infected for >1 year). The classification performance of 2 categorizations of the proportion of ambiguous sites was evaluated with receiver operating characteristic (ROC) analyses. The first analysis included ambiguous sites as a categorical variable with 5 groups based on quintiles, whereas for the second analysis only 2 categories were included based on an a priori defined cutoff of ≤.5% or >.5% ambiguous positions.

RESULTS

In this study, we included HIV subtype B sequences of 3,307 patients. As shown in Table 1, the majority of individuals in our sample were male (n = 2,593; 78.4%) and of white ethnicity (n = 2,933; 88.7%). Of these 3,307 patients, 1,482 (45%) had acquired HIV through homosexual contacts, 862 (26.1%) through intravenous drug use, and 857 (25.9%) through heterosexual intercourse. The median year of infection was 1995 with an IQR of 1991–2000, and the median duration from infection to sampling for genotypic sequencing was 4.7 years (IQR, 3.3–6.9 years).

Table 1.

Summary of the Adjusted Least Squares Regression for the Full Data Set

	All patients		Patients infected ≤8 years
Measure	No. (%) of patients	Median (95% CI)	No. (%) of patients	Coefficient (95% CI)
Total no. of patients	3,307 (100)	–	2,729 (100)	–
Time since infection (per year)	–	.195 (.170–.221)a	–	.158 (.141–.175)a
Time since infection (quadratic term)	–	−.007 (−.008 to −.005)a	–	NA
Female sex	714 (21.6)	.034 (−.053 to .121)	542 (19.9)	.054 (−.047 to .159)
Age by quartile, years
25–30	949 (28.7)	Referent	862 (31.6)	Referent
33–35	789 (23.9)	.146 (.060–.233)a	605 (22.2)	.140 (.047–.233)a
38–41	781 (23.6)	.232 (.144–.319)a	613 (22.5)	.247 (.145–.338)a
45–54	788 (23.8)	.237 (.146–.328)a	649 (23.8)	.252 (.153–.348)a
Ethnicity
White	2,933 (88.7)	Referent	2,455 (90.0)	Referent
Black	63 (1.9)	.169 (−.060 to .399)	58 (2.1)	.080 (−.122 to .290)
Hispano-American	97 (2.9)	.325 (.141–.509)a	89 (3.3)	.343 (.140–.546)a
Asian	57 (1.7)	.090 (−.147 to .328)	55 (2.0)	.095 (−.109 to .290)
Unknown ethnicity	157 (4.7)	−.039 (−.200 to .122)	72 (2.6)	.001 (−.186 to .211)
Mode of HIV acquisition
Heterosexual risks	857 (25.9)	Referent	722 (26.5)	Referent
Intravenous drug use	862 (26.1)	.233 (.142–.323)a	595 (21.8)	.247 (.140–.357)a
Homosexual risks	1,482 (44.8)	−.066 (−.154 to .021)	1326 (48.6)	−.063 (−.154 to .031)
Unknown risk	106 (3.2)	.121 (−.062 to .305)	86 (3.2)	.186 (−.006 to .416)
Laboratory
Laboratory A	213 (6.4)	Referent	178 (6.5)	Referent
Laboratory B	1,450 (43.8)	.407 (.274–.540)a	1176 (43.1)	.357 (.246–.456)a
Laboratory C	319 (9.6)	.404 (.247–.562)a	264 (9.7)	.312 (.180–.430)a
Laboratory D	1,325 (40.1)	.787 (.653–.921)a	1111 (40.7)	.705 (.594–.803)a
Calendar year of sequencing, median (IQR)	2007 (2006–2008)	.119 (.101–.137)a	2007 (2006–2008)	.111 (.093–.127)a
HIV RNA load at time of GRT by 33 percentiles, log10 copies/mL
3.3–4.0	1,013 (30.6)	Referent	829 (30.4)	Referent
4.5–4.8	1,013 (30.6)	.097 (.017–.176)a	859 (31.5)	.043 (−.050 to .128)
5.2–5.7	1,011 (30.6)	.136 (.056–.217)a	853 (31.3)	.063 (−.020 to .151)
HIV RNA load missing	270 (8.2)	−.106 (−.235 to .023)	188 (6.9)	−.130 (−.273 to .026)
Constant term (1 unit = 1%)	–	−1.046 (−1.231 to −.860)	–	−.885 (−1.042 to −.732)

NOTE. The 95% confidence intervals (CIs) were estimated with bootstrapping over 1,000 replicates. Note that for the quadratic model (all patients) both the linear and the quadratic term significantly improve the fit, but not when only the first 8 years of infection were considered. GRT, genotypic resistance test; HIV, human immunodeficiency virus; IQR, interquartile range; NA, not applicable.

Statistically significant by the Wald test.

We found that for HIV-1 sequences sampled before initiation of antiretroviral therapy, the fraction of ambiguous nucleotides increases significantly with the age of infection. This relationship is shown in Figure 1A for the full data set (3,307 patients) for whom the age of infection has been estimated with the back calculation algorithm (see Materials and Methods and Taffe and May [13]). Figure 1A shows that the fraction of ambiguous sites grows linearly with time in the first few years of infection and then flattens off. This effect can be described by including time since infection as a linear and a quadratic term in an adjusted regression model with the fraction of ambiguous nucleotides as outcome: if all patients are taken into account, both the positive linear and the negative quadratic components significantly improve the quality of the fit (model with linear term, r2 = .14; model with additional quadratic term, r2 = .17) (Figure 1B; Table 1). Because the initial increase in the proportion of ambiguous positions is almost linear during the first phase of infection, we further restricted our analyses to patients with an HIV sequence obtained in this phase of linear increase, that is, within the first 8 years of infection (n = 2,729). These adjusted analyses revealed a yearly increase in the proportion of ambiguous sites of ∼.2 percentage points per year (Table 1), which was comparable with results from unadjusted regression analyses with yearly increases of .25% (95% CI, .23%–.28%) for the full sample and .19% (95% CI, .18%–.21%) for the sample restricted to patients with an infection duration of ≤8 years. Interestingly, Table 1 indicates that the mode of HIV acquisition significantly affects the level of diversity. In particular, intravenous drug users exhibit a larger fraction of ambiguous nucleotides than do both heterosexuals and men who have sex with men, which may suggest broader transmission bottlenecks for blood-transmitted than for sexually transmitted HIV.

Figure 1.

Relationship between the year of infection and the fraction of ambiguous nucleotides (f). A, Mean of f as a function of the age of infection, where data points have been binned according to the age of infection in years (n = 3,307 patients). The shaded area corresponds to the 95% confidence interval of the means of f. These confidence intervals have been determined by bootstrap (with 1,000 samples). The red point corresponds to the mean of f for the Zurich Primary HIV Infection Study data set (n = 130), for which all sequences stem from the first few months after the infection. The associated red line gives the 95% confidence interval of this mean. B, Quadratic and linear fit of the full data set. Note that the linear fit is restricted to ages of infection of ≤8 years. C, Linear fit of the different data sets. Only sequences obtained within the first 8 years after infection were considered. D, Distribution of the age of infection (in years) for different fractions of ambiguous nucleotides. The left plot depicts the density plot of the age of infection for the 5 quintiles of f. The right plot depicts, for each of the 5 quintiles of f, the 25%–75% percentiles (green lines) and 5%–95% percentiles (black lines) of the year of infection.

To further assess the validity of the relationship between ambiguity and age of infection, we examined the smaller seroconverter set, which consists of 366 individuals who were recruited to the cohort in the early phase of their infection and for whom therefore the time point of infection was known within a given time interval of 6 months, and the longitudinal set, which contains 124 patients with sequence samples obtained from 2 time-points at least 6 months apart from each other. In the latter data set, inclusion was restricted to those patients for whom both samples were obtained within the first 8 years of infection. For these patients, we correlated the length of the time interval between the 2 measurements with the increase of the fraction of ambiguous nucleotides. Figure 1C shows that both additional data sets yield an estimate for the increase of ambiguous nucleotides over the first 8 years after infection that is similar to the estimate yielded by the original data set. Finally, we used the ZPHI data set, which contains the most stringent infection time estimate, to test the diversity around the time point of infection. Figure 1A shows that the fraction of ambiguous nucleotides in the ZPHI set is consistent with that found in the other 3 data sets. Moreover, for the ZPHI set, we observed a highly significant correlation (Spearman ρ = 36; P < .001) between the fraction of ambiguous nucleotides in the pol gene and the diversity of clonal env sequences [14], which confirms that the fraction of ambiguous nucleotides is a good marker for viral diversity.

In order to explore whether the observed correlation between infection time and the fraction of ambiguous sites could be exploited for age classification of individual HIV sequences, we subdivided our data set into quintiles of the fraction of ambiguous nucleotides f and inferred the distribution of the time of infection for each of these quintiles. Figure 1D shows that the fraction of ambiguous nucleotides provides useful information mainly on the lower bound of the age of infection: whereas sequences with f < .68% (ie, the first two quintiles) (Table 2) may stem from the early phase of an infection, this is unlikely for sequences with a larger fraction of ambiguous nucleotides. However, it is important to note that although it is unlikely that the HIV sequence from a recent HIV infection has a large fraction of ambiguous nucleotides, this may occur if the HIV population is founded by >1 virus, in which case diversity is large from the beginning on. This effect can be clearly seen in the ZPHI data set: a substantial minority of 24 (18%) of 130 patients exhibit a fraction of ambiguous nucleotides >.68%, even though all patients’ samples have been sequenced in the first months of their infection. This can be explained by results of Keele et al [1], who found that 23% of HIV infections are founded by >1 virus and hence should exhibit a large diversity even during primary infection.

Table 2.

Summary of the Two Cutoff Methods: A Priori Defined Threshold of .5% and Cutoffs Based on Quintiles

Stratum number	Stratification cutoff, %	No. per stratum	No. (%) of patients infected ≤1 year	Comparison of strata a	Sensitivity, % b	Specificity, % c	Correctly Classified, %
Predefined
1	≤.5	1,117	184 (16.5)	1 vs 2	86.8	69.9	70.9
2	>.5	2,190	28 (1.3)	–	–	–	–
By quintiles
1	0–.15	679	146 (21.5)	1 vs 2, 3, 4, 5	68.8	82.8	81.9
2	.16–.67	649	44 (6.8)	1, 2 vs 3, 4, 5	89.6	63.2	64.9
3	.68–1.35	691	14 (2.0)	1, 2, 3 vs 4, 5	96.2	41.4	44.9
4	1.36–2.07	633	7 (1.1)	1, 2, 3, 4 vs 5	99.5	21.1	26.2
5	2.08–6.65	655	1 (.2)	–	–	–	–

Indicates how strata are collapsed for comparison; for example, strata 1 versus all remaining strata.

Proportion of patients infected ≤1 year whose human immunodeficiency virus (HIV) sequence had ≤.5% ambiguous positions.

Proportion of patients infected >1 year whose HIV sequence had >.5% ambiguous positions.

We explicitly evaluated 2 different stratifications for cutoffs to predict whether a patient was infected for <1 year at the time of sampling for the genotypic test. As shown in Table 2, the selection of a cutoff of ≥.5% ambiguous positions predicted recent infections quite well in our full data set of 3,307 patients. Of 212 patients with a recent infection at the time of genotypic testing, 184 had ≤.5% ambiguous positions in their HIV sequence (sensitivity, 86.8%), but the specificity, which is the proportion of chronically infected patients with >.5% ambiguous positions, was only 70% (ROC area under curve, 78.3%). Therefore, this cutoff of ≤.5% may not be very useful to identify recently infected patients, but it can very accurately discriminate against patients with a chronic infection: only 28 (1.3%) of 2,190 patients with >.5% ambiguous positions had an infection duration of ≤1 year at the time of genotyping. Thus, if the observed fraction of ambiguous positions is >.5%, then there is a probability of 98.7% (95% CI 98.2%–99.1%) that the genotypic test was performed >1 year after infection (negative predictive value). Stratification by quintile did not improve the classification performance but essentially confirmed the above finding: when quintiles 1 and 2 are collapsed into a single category and compared with the remaining strata (ie, sequences with <.68% vs those with ≥.68% ambiguous positions), sensitivity increased to 89.6% but specificity decreased to 63.2%. When testing the .5% cutoff in the sample of observed seroconverters (n = 366), the performance was very similar to that in the full sample. Sensitivity and specificity were 84.3% and 59.8%, respectively (data not shown). Overall, 79.8% of patients were correctly classified, and the area under the ROC curve was 72.0%.

The observed increase in ambiguous positions over the time of infection likely reflects the diversification of the HIV population after the genetic bottleneck at the infection event. In the simplest scenario, this diversification is controlled only by mutation and genetic drift (ie, random extinction of viral strains). We modeled this scenario by simulating the Wright-Fisher model (WFM) [15], starting from an initially uniform population (ie, with no initial diversity). As the WFM assumes neutrality, we focused on ambiguous nucleotides at 4-fold degenerate third-codon positions for which the assumption of neutrality is best justified. We found that the WFM can reproduce the temporal increase of the fraction of ambiguous sites for parameters that have been shown to reproduce neutral evolution in HIV (see Kouyos et al [16] and references therein) (Figure 2). This finding implies that, at least for 4-fold degenerate third-codon positions, the increasing diversity of the HIV population can be understood as the combined effect of mutation and genetic drift acting on a virus population, which is homogenous at the time point of infection.

Figure 2.

Temporal increase of the fraction of ambiguous nucleotides in the Wright-Fisher model (WFM) for a population size of 500 and a mutation rate of 3 × 10−5 mutations per generation (solid green line) and at 4-fold degenerate third-codon positions in the full data set (dashed black line). The curve for the WFM has been obtained by averaging over 104 runs of the model. N and m denote the effective population size and the mutation rate, respectively. The WFM describes discrete and nonoverlapping generations in a population with fixed size N. Every generation, each of the N genomes undergoes mutation with probability m. Then the N genomes for the next generation are determined from the gene pool by drawing every offspring genome with uniform probability from the N parental genomes. Note that the WFM assumes selective neutrality.

DISCUSSION

The most relevant clinical result of this study is that a large fraction of ambiguous nucleotides provides evidence against a recent infection event. In particular, we found that a threshold of .5% ambiguous nucleotides yields a negative predictive value of 98.7%. It is important to note that, by definition, the negative predictive value depends on the overall composition of the considered population at time of diagnosis. A strength of the sample studied here, however, is the high representativeness of the SHCS for the epidemic in Switzerland [6-8]. Thus, the composition of our sample is likely to be very similar to that of the population of HIV-1-infected individuals at the time of HIV infection diagnosis, and therefore the negative predictive value inferred here closely describes the situation in clinical practice. Finally, we find that although the frequency of ambiguous nucleotide calls differs significantly between laboratories (Table 1), the negative predictive value varies only marginally when calculated for samples from each laboratory separately (range, 98.5%–99.5%; χ2 test, P = .668).

In order to test the robustness of our findings, we additionally assessed the impact of HLA types, transmitted resistance, and subtype. HLA types may affect viral diversity, for instance, by stabilizing selection (and possibly by maintaining detrimental viral mutations, as may be the case for type HLA-B*57) or, in the opposite, by leading to increased viral replication and diversity by triggering only suboptimal cytotoxic T lymphocytes responses, for instance, in the presence of homozygous HLA alleles. Using available HLA data from 352 individuals, we considered the homozygosity of HLA-B (n = 32) and the following HLA-B haplotypes: HLA-B*57 (n = 23), HLA-B*27 (n = 25), HLA-B*5801 (n = 7), and HLA-B*35 (n = 80). Upon inclusion of these variables in the adjusted regression model (restricted to a time of ≤8 years), neither of these HLA-B alleles reached statistical significance. There was a nonsignificant trend for higher viral diversity among carriers of homozygous HLA-B alleles of .243 (95% CI -.054 to .551). Concerning HIV subtype, it should be noted that the present analysis focused on subtype B, because most patients in the SHCS have been infected with that subtype. We found, however, very similar results for other subtypes (Table 3), indicating that our methodology extends beyond subtype B. However, it is also clear from the limited data available for non-B subtypes that additional tests are required for these subtypes. Finally, transmitted resistance mutations (present in 10% of the included patients) did not seem to have an impact when included in an adjusted model analogous to that in Table 1 (with time restricted to ≤8 years), since the parameters were not statistically significant (data not shown). In summary, these 3 tests suggest that our findings are robust against HLA type, subtype, and transmitted resistance.

Table 3.

Unadjusted and Adjusted Estimates for the Growth Rates Over the First Eight Years of Infection and Summary of the Cutoff Method Based on the .5% Threshold for the Four Largest Non-B Subtype Groups in Switzerland

				Cutoff of <.5%
Subtype	No. (%) of recent infections	Univariable model, Coefficient (95% CI)	Multivariable model, Coefficient (95% CI)	Sensitivity, %	Specificity, %	ROC, %	NPP, %
A (n = 185)	11 (7.1)	.197 (.121–.276)	.193 (.105–.276)	63.6	63.6	63.6	95.6
CRF01_AE (n = 173)	10 (5.8)	.210 (.142–.285)	.141 (.064–.217)	100.0	59.5	79.8	100.0
CRF02_AG (n = 185)	15 (8.1)	.159 (.081–.240)	.110 (.098–.220)	80.0	51.8	65.9	96.7
C (n = 131)	6 (4.6)	.150 (.069–.230)	.104 (.024–.175)	100.0	66.4	83.2	100.0

NOTE. The unadjusted and adjusted estimates for the growth rates over the first 8 years of infection are analogous to Table 1. The summary of the cutoff method based on the .5% threshold is analogous to Table 2. CI, confidence interval; NPP, negative predictive value; ROC, receiver operating characteristic.

A potential limitation of our analysis is the accuracy of infection date estimates. We have confirmed the correlation between the age of an infection and the fraction of ambiguous base calls in several data sets with different estimation methods for infection duration. Moreover, we performed a sensitivity analysis to assess the impact of the uncertainty of infection date estimates from the back calculation model (data not shown). This was done by repeating the adjusted and unadjusted regression analyses on a random time point between the upper and the lower bound of the posterior infection date distribution [13]. Time trends were somewhat attenuated due to the introduction of additional variation with an increase of.13% per year (95% CI, .11%–.14%) for the unadjusted analysis and an increase of .11% per year (95% CI, .12%–.13%) for the adjusted analysis, but these point estimates still reached statistical significance.

We would like to emphasize that this study represents a proof of principle and that the details of the method should ideally be replicated and calibrated for each sequencing laboratory or method (especially in the light of our finding of significant differences between laboratories) (Table 1). Moreover, it is important to note that 10%–20% of recently infected patients do have a high viral diversity because they are infected with several strains. Therefore, the fraction of ambiguous nucleotides should be only one of several measurements used to decide whether a patient is recently infected.

Detection of ambiguous nucleotides is a byproduct of bulk sequencing. Here, we have shown that this byproduct carries important information on the age of the infection at sampling time. In particular, a large frequency of ambiguous nucleotides argues against a recent infection event. The qualitative pattern of an initially linearly increasing amount of diversity and subsequent saturation is consistent with the diversification pattern observed for the env gene [3]. Overall, our study highlights the usefulness of diversity measures as markers for infection age. This usefulness might further increase with the advent of array-based pyrosequencing, which allows the diversity of the HIV population to be analyzed in much more detail.

Supplementary Material

Supplementary materials are available at Clinical Infectious Diseases online (http://www.oxfordjournals.org/our_journals/cid/).

Supplementary materials consist of data provided by the author that are published to benefit the reader. The posted materials are not copyedited. The contents of all supplementary data are the sole responsibility of the authors. Questions or messages regarding errors should be addressed to the author.