A detailed understanding of the time course of viremia and antibody seroconversion during primary HIV infection is a prerequisite for optimizing screening and diagnostic algorithms [1–4], and may provide useful reference data for clinical trials, for example by providing a new basis for the stratification of patients in different study arms.
Because of the difficulty in obtaining blood samples representing early acute HIV infection from clinical patients, most patients do not come to medical attention until weeks to months after infection, we resorted to stored, frozen plasma collections from plasma donors, who unrelated to donating became infected with HIV, and were deferred from further donating. As plasma donors donate on average twice a week, and every donation is tested for HIV and held for 60 days before release, their archived samples provide a unique record of the infection from timepoints before viral exposure until seroconversion and beyond.
In order to construct a detailed timetable of the emergence of HIV RNA, HIV-1 (p24) antigen, HIV antibody and Western blot reactivity, we applied these assays to a plasma donor panel set consisting of 435 samples from 51 donors. A second, separate set of 145 samples from 44 plasma donors representing only preantibody seroconversion timepoints of infection was analysed in parallel with HIV-RNA and p24 antigen assays to develop a model that would allow the prediction of how much earlier quantitative RNA tests detect infection vis-à-vis p24 antigen reactivity.
Materials and methods
Anti-HIV seroconverting plasma donor panels
Plasma donations (600–800 ml) from source plasma donors were routinely collected at approximately twice weekly intervals and stored frozen at −20°C or less. After confirmation of anti-HIV or p24 antigen seroconversion, donors were permanently deferred from further donation, and all quarantined plasma donations, i.e. donations routinely held back from distribution for a 60-day waiting period, were retrieved from storage to construct panels containing sequentially drawn plasma samples from seroconverting donors. Each donation was rapidly thawed, aliquoted and the aliquots re-frozen at −20°C or less. Serial donation aliquots were coded and compiled into anonymized panels, not linked to individual donors. Records of each donor visit date, as well as the results of routine and research laboratory tests for each plasma aliquot, were entered in a computerized database. Panel samples were obtained from Alpha Therapeutic Corporation (Los Angeles, CA, USA) and Boston Biomedica (West Bridgewater, MA, USA).
We examined two sets of serial plasma donor panels. (A) The first set of 435 samples was from 51 donors who had seroconverted based on HIV antibody screening between June 1984 and October 1994. Samples included in the panels date back several weeks before antibody seroconversion and continue up to 6 months after the detection of antibodies to HIV. The median interval between donations was 5 days (range 1–204 days). (B) The second set of 145 samples represents serial donations from 44 plasma donors who became HIV positive when donating between June 1996 and March 1998, after the introduction of routine HIV p24 antigen screening. The samples were further characterized by screening with an HIV-RNA assay and were selected to represent only preantibody seroconversion timepoints. The present analysis included donations for which a quantitative HIV-1 reverse transcriptase–polymerase chain reaction (PCR) assay detected 100 HIV-1-RNA copies/ml or greater (analytical sensitivity of the assay) through the first peak of viremia and antibody assay reactivity. The median duration of intervals between donations was 4 days (range 1–10 days).
Assays for HIV-1 markers
Donor panels were tested in parallel with an array of commercial nucleic acid and serological HIV assays in strict accordance with the manufacturer's recommendations. Quantitative HIV-l-RNA PCR utilized either the Amplicor HIV Monitor kit (Roche Molecular Systems, Branchburg, NJ, USA) on a Perkin-Elmer GeneAmp 9600 PCR system (Applied Biosystems, Foster City, CA) or the SuperQuant assay developed by the National Genetics Institute (Los Angeles, CA, USA). HIV-1 p24 antigen was performed using a Food and Drug Administration (USA) approved enzyme immunoassay (EIA; Coulter Corporation, Hialeah, FL, USA) without an immune complex dissociation step. HIV antibodies were detected using both a second-generation indirect sandwich EIA (Genetic Systems Corp., Redmond, WA, USA) and a third-generation HIV-1/HIV-2 antigen sandwich EIA with established IgM sensitivity (Abbott no. 3A77; Abbott Diagnostics, Abbott Park, IL, USA). EIA-reactive samples were tested by Western blot (Cambridge Biotech, distributed by Ortho Diagnostics, Raritan, NJ, USA).
Modeling the relationship between HIV-1-RNA levels and p24 antigen signal-to-cutoff ratio in the preantibody period
This analysis was based on the ‘B’ set of plasma donor panels. Log values of HIV RNA and p24 antigen signal-to-cutoff ratios were plotted against the day of sample collection. The p24 antigen signal-to-cutoff ratio signifies the strength of the assay reactivity relative to a cutoff level, so that signal-to-cutoff ratios equal to or greater than 1 (equivalently, log10 p24 antigen signal-to-cutoff ratios equal to or greater than 0) are considered to be p24 antigen reactive. The first day of RNA detection greater than 100 copies/ml by the quantitative HIV-RNA assay was arbitrarily defined as day 0 of the timescale for each plasma donor. This choice has an impact on the calculation of random intercepts and their correlation in the model, but does not substantively affect analysis objectives, as we consider only the differences between the times of events. From each donor we included values for all ‘early’ samples, which were defined as those up to and including the first sample when log10 HIV RNA was greater than 4.45, leaving 97 out of 145 samples available for the model calculations. We decided on this cutoff after visual inspection of a graph plotting HIV-RNA levels over time; above this viral load cutoff there was no longer a consistent increase in RNA levels, indicating that viral replication rates were lower and were inconsistent with the dynamic viral load increase characterizing the initial ramp-up phase of viremia. From the data we developed a regression model of HIV-RNA levels and p24 antigen signal-to-cutoff ratios over time (see Statistical evaluation section). For each donor, the model allowed the prediction of (i) the HIV-RNA level at p24 antigen seroconversion, i.e. when the log10 p24 antigen signal-to-cutoff ratio equals 0; (ii) the time in days when p24 antigen seroconversion occurs, relative to the first possible day of RNA detection; and (iii) the time in days, relative to p24 antigen seroconversion, when HIV RNA reaches defined levels during the ramp-up phase of initial viremia. For the aforementioned predictions of RNA levels and time differences, 95% confidence intervals were calculated using a parametric bootstrap . Note that this calculation often extrapolates beyond available data (occasionally extrapolating forward to the predicted time when p24 antigen seroconversion occurs, often extrapolating backward to the predicted time when the HIV-RNA level equals 100, and always extrapolating backward to the predicted time when HIV RNA reaches defined levels less than 100). The model thus relies on the assumption of linearity (on the log scale). Also, the relatively small sample size (97 samples from 44 donors) limits the accuracy of the model in predicting ‘true’ viral replication rates during the ramp-up phase of viremia. Furthermore, the unequal representation of donors in the model (number of samples per donor ranged from one to four) could potentially introduce a bias, if there was a relationship between a donor's viral replication rate and their number of samples. Despite these limitations, the model provides a useful approximation of the relationships of HIV-RNA levels and p24 antigen signal-to-cutoff ratios over time.
Descriptive statistics and group comparisons (stratified Mann–Whitney and stratified Kruskal–Wallis tests) were calculated (SAS Institute, Cary, NC, USA). Statistical significance was assumed if P < 0.05. The duration of the assay-specific window periods was projected by fitting a parametric Markov model to the data as previously described . The appropriateness of the Markov model for these data was confirmed using a non-parametric semi-Markov model and a model that allowed for tracking (positive correlation among times between the sequence of events ). The relationships of HIV RNA and p24 antigen signal-to-cutoff ratios over time were approximated with a bivariate repeated measures linear regression model . For each donor, random intercepts and random slopes were included. The random slopes allow that each donor has a unique rate of increase or ‘doubling time’ for RNA and for p24 antigen. The model correlation structure includes a correlation between random intercepts and a correlation between random slopes (i.e. p24 antigen at day 0 is correlated with RNA at day 0, and the ‘doubling time’ for RNA is correlated with the ‘doubling time’ for p24 antigen). Model parameters were calculated using SAS statistical software (SAS Institute).
Stages of early HIV infection
Of 435 sequential samples from 51 plasma donors who seroconverted to HIV-1 antibody (`A’ panels), 113 samples were non-reactive in each of the assays employed; the remaining 322 samples were reactive in one or more assays. The pattern of assay reactivity in serial samples was highly consistent for each donor, which prompted us to propose the following laboratory stages of primary HIV-1 infection: stage I: HIV present in blood samples, only RNA assay positive; stage II: RNA and HIV-1 p24 antigen tests positive, antibody EIA non-reactive; stage III: RNA, HIV-1 antigen and HIV IgM-sensitive EIA reactive, but Western blot without HIV-1-specific bands; stage IV: as stage III, but in addition indeterminate Western blot pattern, i.e. the presence of HIV-1-specific Western blot bands that fail to meet interpretative criteria for reactive Western blot defined by the USA Food and Drug Administration as reactivity to two of the following three bands: p24, gp 41, gp 120/160; stage V: as stage IV, but reactive Western blot pattern, except lacking p31 (pol) reactivity; stage VI: as stage V, but full Western blot reactivity including a p31 band.
Fig. 1 shows how the reactive samples in aggregate represent all six stages, and provides details of assay reactivity for representative donor panels. Table 1 provides projections of the duration of each stage derived from fitting a parametric Markov model to the data. Notably, the projected duration of each of stages I–IV is relatively brief, lasting on average only 3–5 days, whereas stage V (positive Western blot without p31 band) is estimated to last 69.5 days on average. No endpoint was defined for the final stage VI, characterized by a full Western blot pattern. On the basis of previously published results, stage VI can be subdivided further into recent versus early chronic infection, as illustrated in Fig. 2.
Inherent in the proposed laboratory staging concept is the dependency of stage assignment on assay sensitivity. The less sensitive second-generation EIA became reactive on average one week after the IgM-sensitive EIA. Samples with indeterminate Western blot patterns (stage IV) reacted only infrequently with this EIA (nine out of 39 samples, 23%); higher reactivity rates were observed for samples classified as stage V (118 out of 143, 83%) and VI (46 out of 46, 100%).
Relationship between HIV stages and HIV-1-RNA levels
The HIV-1-RNA levels are not equal through the six stages (P < 0.001, stratified Kruskal–Wallis test). The median HIV-1-RNA levels in stage II tended to be on average 2 log10 higher than those in stage I (Fig. 2, and P < 0.001, stratified Mann–Whitney test). Further sequential pairwise comparisons of stages find the RNA levels consistent through stages II, III and IV, and decreasing in stages V and VI (Fig. 3., and P = 0.09 stage II versus III, P = 0.05 stage III versus IV, P < 0.001 stage IV versus V, and P < 0.001 stage V versus VI).
Model prediction of HIV-1 RNA and p24 antigen signal-to-cutoff ratios in the preantibody period
HIV-RNA and p24 antigen signal-to-cutoff ratios in the ‘B’ set of plasma donor panels, representing the initial phase of viremia, rise concurrently during 13 days from the first detection of RNA (Fig. 4a,b). As observed for the ‘A’ donor panels, median HIV-1-RNA levels in the 85 stage II samples from the ‘B’ panels were on average 2 log10 higher than those in the 60 stage I donor samples (P < 0.0001, stratified Mann–Whitney test). In the linear regression model, the slopes for log10 HIV-RNA and p24 antigen signal-to-cutoff ratios were calculated to be 0.35 and 0.23, respectively, suggesting an average increase in HIV viral load of 0.35 log10 (approximately 2.2-fold) per day, corresponding to a ‘doubling time’ of HIV-RNA copies/ml of 20.5 h [95% confidence interval (CI) 18.2–23.4 h] during this period of dynamic viral load expansion. The correlation of HIV-RNA and p24 antigen intercepts was 0.94 and the correlation of slopes was 0.82. The slope correlation of 0.82 between subjects supports the assumption of a concurrent increase of HIV-1 RNA and p24 antigen signal-to-cutoff ratio during this phase.
From the regression model shown in Fig. 4c, the point estimate for HIV-1-RNA concentration at p24 antigen seroconversion is 4.01 log10 (10 000) copies/ml, with an estimated 95% CI of 3.31–4.97 log10 (2000–93 000 copies/ml). Fig. 4c also illustrates the derivation of time intervals relative to p24 antigen seroconversion when HIV-RNA assays can be expected to detect virus in diluted (pooled) and undiluted samples.
Projection of window period closure and yield of p24 antigen, minipool and individual donation nucleic acid testing assays
Average window period reductions for p24 antigen testing (equivalent to the length of stage II in this study), as well as for minipool and individual donation nucleic acid testing, derived from the model calculations reported here, are shown in Table 2 relative to an IgM-sensitive HIV antibody EIA with an estimated window period of 22 days . These projections also assume that patients will present within defined stages at a rate consistent with the duration of each stage, whereas in different screening settings, the presence or absence of recent risk behavior or symptoms will probably influence the probability of presentation .
On the basis of our observation that HIV-1 assay reactivity progressed sequentially and highly consistently in seroconverting plasma donors, we propose a laboratory staging system of primary HIV infection that recognizes six stages, each defined by a unique pattern of assay reactivity (Fig. 2). Consistent with previous results , we found peak HIV viral load levels during antibody seroconversion (stage III), followed by a decline towards apparent steady state levels coinciding with Western blot maturation (stages V, VI; Fig. 2). In none of our cases did the HIV-RNA levels become undetectable during stage VI, although this has been reported in some cases when standard HIV-RNA assays were used . In these cases, HIV RNA was detectable using ultrasensitive methods . HIV-1 p24 antigen was detected in all samples during stages II and III, but gradually disappeared subsequently, consistent with the immune complexing of antigen by emerging antibodies to HIV [3,18,19]. The presence of Western blot bands defining HIV infection, but lacking p31 band reactivity (stage V), was indicative of antibody seroconversion within the previous 2–3 months, a finding confirmed in a recent study of 98 newly infected individuals identified in the Acute Infection and Early Disease Research Program network .
By providing estimates for infectious window periods associated with standard HIV-1 viral assays, in which the window period is defined as the time period between infectivity and the detectability of infection by the assays, the proposed staging system has direct application for blood donor screening purposes and prevention of the spread of HIV in other settings. On the basis of a presumed 22-day window period to the detection of HIV antibodies by IgM-sensitive EIA, and approximately 5 days each for the two preantibody stages characterized in this study, we have derived estimates of the marker-negative, potentially infectious window periods for p24 antigen and HIV RNA of approximately 17 and 12 days, respectively. These estimates can then be combined with incidence rate data to project the risks of the transfusion of window phase units in the context of different donor screening scenarios [21,22].
As the length of the window period is dependent on assay sensitivity, our 5-day window period estimate for HIV-RNA detection reflects the somewhat lower sensitivity of an earlier generation reverse transcriptase PCR assay. In order to project window period durations for newer ‘ultra-sensitive’ HIV-RNA tests, we developed an approach to relate the HIV-RNA copy number during the dynamic phase of initial viral expansion to evolving p24 antigen reactivity over time. Using a bivariate repeated measures linear regression model and linear regression analysis, we were then able to estimate the average HIV-RNA copy number at the timepoint of p24 antigen seroconversion, and to estimate HIV-RNA levels at earlier timepoints relative to the detectability of the p24 marker. The model predicts an average HIV-RNA level of 10 000 copies/ml (95% CI 2000–93 000 copies/ml) at the time p24 antigen becomes detectable, and an increase of 0.35 log10 HIV RNA per day. Assuming HIV-RNA assays with different detection thresholds, we project that an assay with 50 copies/ml sensitivity detects HIV approximately 7 days earlier than the p24 antigen test, and that the theoretical detection threshold of 1 copy/ml HIV would be reached approximately 11 days earlier.
Our model also allows window period projections for testing pooled samples with HIV-RNA tests, as is currently done in US blood donor screening . In this situation, virus present in a sample would be diluted in the pool, lowering the apparent sensitivity of the assay. If the pool contains 20 samples, as in the average so-called ‘minipool’ testing strategy employed in blood donor screening, an HIV-RNA assay with a detection threshold of 50 copies/ml would appear to have a detection threshold of 1000 copies/ml because of the 20-fold dilution of virus in the pool. In this scenario, the model would predict the detectability of HIV by the RNA assay approximately 4 days later than if the sample had been tested undiluted, or 3 instead of 7 days before p24 antigen reactivity on an undiluted sample. Two recent cases of HIV transmission by donations that tested non-reactive in pooled testing, but reactive by testing undiluted samples, confirm that such donations are infectious [24,25]. Our analysis allows the projection of the frequency of such donations on the basis of known incidence rates in donors, for use in policy deliberations regarding the value of implementing individual donation RNA testing in blood donor screening. Similar projections can be used to estimate the yield of RNA or p24 antigen screening in other settings, such as sexually transmitted disease clinics or vaccine trial populations (Table 2).
Finally, we believe the proposed staging system has applications in clinical practice, although several caveats need to be considered. First, our staging system relies on the use of an IgM-sensitive EIA for the detection of antibody seroconversion; delayed reactivity can be expected when a less sensitive EIA is used, as we noted in this study. Second, the short duration of stages I–IV would seem to limit their use in most practice settings, in which patients usually present past seroconversion. Finally, our observations in plasma donors may not be applicable to newly infected patients in general. In particular, patients who present with a highly symptomatic retroviral syndrome may have higher levels of viremia and delayed seroconversion compared with plasma donors who continued to present for donation, and therefore as a group may represent infected individuals with no or milder acute symptoms. With these reservations in mind, the proposed staging provides a reference framework for when to expect viral markers to turn positive in primary HIV infection, which could aid in resolving difficult or unusual diagnostic situations. Furthermore, the staging approach provides a new basis for the stratification of patients enrolled in clinical trials. Defining the laboratory stage at initial diagnosis and the beginning of therapeutic intervention may help answer a number of important questions: (i) Can very early treatment during the viremic preantibody seroconversion stages abort infection? (ii) If not, can early sustained highly active antiretroviral therapy or supervised treatment interruption protocols [14,26] result in the long-term downregulation of the viral setpoint with sustained clinical benefit? (3) Does the time of the initiation of therapy (i.e. stage) influence the response to early treatment? Therapeutic trials, utilizing the laboratory staging system of HIV infection described in this report, are currently under way.
The authors would like to thank Barbara Johnson for expert assistance with the preparation of the manuscript.
Sponsorship: This work was supported by National Institutes of Health grants NO1-HB-47114 and UO1-AI-41531, and by Centers for Disease Control and Prevention cooperative agreement U64-CCU-902948.
1. Mylonakis E, Paliou M, Lally M, Flanigan TP, Rich JD. Laboratory testing for infection with the human immunodeficiency virus: established and novel approaches. Am J Med
2. Nguyen KA, Busch MP. Evolving strategies for diagnosing human immunodeficiency virus infection. Am J Med
3. Busch MP, Satten GA. Time course of viremia and antibody seroconversion following human immunodeficiency virus exposure. Am J Med
:117–124; discussion 125–116.
4. Hecht FM, Busch MP, Rawal B, Webb M, Rosenberg E, Swanson M, et al
. Use of laboratory tests and clinical symptoms for identification of primary HIV infection. AIDS
5. Efron B, Tibshirani R. An introduction to the bootstrap.
Boca Raton: Chapman and Hall/CRC; 1998.
6. Satten GA, Sternberg MR. Fitting semi-Markov models to interval-censored data with unknown initiation times. Biometrics
7. Satten GA. Estimating the extent of tracking in interval-censored chain-of-events data. Biometrics
8. Searle SR, Casella G, McCulloch CE. Variance components
. New York: Wiley and Sons; 1992.
9. Janssen RS, Satten GA, Stramer SL,
Rawal BD, O'Brien TR, Weiblen BJ, et al. New testing strategy to detect early HIV-1 infection for use in incidence estimates and for clinical and prevention purposes. JAMA
10. Kahn JO, Walker BD. Acute human immunodeficiency virus type 1 infection. N Engl J Med
11. Schreiber GB, Glynn SA, Satten GA, Kong F, Wright D, Busch MP, et al
. HIV seroconverting donors delay their return: screening test implications. Transfusion
12. Dodd RY, IV, Notari EP, Stramer SL. Current prevalence and incidence of infectious disease markers and estimated window-period risk in the American Red Cross blood donor population. Transfusion
13. Holmberg SD. The estimated prevalence and incidence of HIV in 96 large US metropolitan areas. Am J Public Health
14. Lori F, Lisziewicz J. Structured treatment interruptions for the management of HIV infection. JAMA
15. Pilcher CD, McPhersons JT, Leone PA, Smurzynski M, Owen-O'Dowd J, Peace-Brewer AL, et al
. Real-time, universal screening for acute HIV infection in a routine HIV counseling and testing population. JAMA
16. Lindback S, Karlsson AC, Mittler J, Blaxhult A, Carlsson M, Briheim G, et al. Viral dynamics in primary HIV-1 infection. Karolinska Institutet Primary HIV Infection Study Group. AIDS
17. Garrett PE, MacKeen VL, Schumacher RT. Routine HIV RNA assays should not replace anti-HIV-1 Western Blot to confirm HIV infection. Transfusion
2002, 42 (Suppl.)
18. Cabezas T, Quiros E, Garcia F, Hernandez-Quero J, Bernal MC, Martinez MA, et al. Immune complex p24 antigen: a new prognostic marker in human immunodeficiency virus infection. Infection
19. Cao YZ, Valentine F, Hojvat S, Allain JP, Rubinstein P, Mirabile M, et al
. Detection of HIV antigen and specific antibodies to HIV core and envelope proteins in sera of patients with HIV infection. Blood
20. Hecht FM, Holte S, Busch MP, Hogan C, Little S, Schacker T, et al
. Absence of p31 band identifies persons with recent HIV seroconversion. Abstract presented at XIVth International AIDS Conference
. Barcelona, Spain, 7–12 July 2002 [Abstract TuPeC4885].
21. Busch M. Closing the windows on viral transmission by blood transfusion.
In: Stramer S, editor. Blood safety in the new millenium
. Bethesda, MD: AABB; 2000. pp. 33–54.
22. Glynn SA, Kleinman SH, Wright DJ, Busch MP. International application of the incidence rate/window period model. Transfusion
23. Busch MP, Kleinman SH, Jackson B, Stramer SL, Hewlett I, Preston S. Committee report. Nucleic acid amplification testing of blood donors for transfusion-transmitted infectious diseases: Report of the Interorganizational Task Force on Nucleic Acid Amplification Testing of Blood Donors. Transfusion
24. Ling AE, Robbins KE, Brown TM, Dunmire V, Thoe SY, Wong SY, et al. Failure of routine HIV-1 tests in a case involving transmission with preseroconversion blood components during the infectious window period. JAMA
25. Delwart E, Kalmin N, Jones S, Ladd D, Tobler L, Tsui R, Busch M. First case of HIV transmission by an RNA-screened blood donation.
Presented at the 9th Conference on Retroviruses and Opportunistic Infections
. Seattle WA, 24–28 February 2002 [Abstract 768-W (Poster)] http://www.retroconference.org/2002/
26. Altfeld M, Walker BD. Less is more? STI in acute and chronic HIV-1 infection. Nat Med