Control cohorts consisting of nonvaccinated individuals with symptomatic and asymptomatic primary infections were established to screen for the significance of potential observed differences in viral genotype. Control cohorts were also used to compare plasma viremia in vaccinated and nonvaccinated individuals. Case controls were selected and matched by age (plus or minus 5 years), gender, presumed risk of exposure (male homosexual, heterosexual, or intravenous drug user), and the year of primary infection. Controls were not matched for geographic location; however, geographic location within the United States generally does not correspond to clustering patterns in viral phylogenetic analyses (76). One phylogenetic cluster of viral sequences from two infected vaccine cases and two controls was observed in this study. For these four cases, we requested information concerning the city where the study subject was residing (see Results).

Neighbor-joining phylogenetic reconstructions with bootstrap resampling (41, 56) were generated with the PHYLIP programs dnadist, dnaboot, and neighbor, with a Kimura two-parameter distance matrix which had a transition-to-transversion ratio of 1.3 (Phylogeny Inference Package, version 3.5c; http://evolution.genetics.washington.edu/phylip.html) (42, 43, 44, 154). Bootstrap resampling was done with 100 replicates (63). Qualitatively similar trees were generated with the program FastDNAml, version 1.0 (http://central.pasteur.fr:80/docs/doc-gensoft/fastDNAml), with three randomizations of input order but using only one sequence per person (40, 124).

To screen for potential contamination of product DNA (88), all PCR-amplified viral sequences included in the study were cross-compared through phylogenetic analysis, by screening within- and between-subject Hamming distances, and by comparing them with sequences in the viral subsection of GenBank by using BLAST (2). Evidence of potential problems is the near identity of viral sequences to those of laboratory strains or to those derived from epidemiologically unlinked individuals that are distinct from other viral sequences obtained from the putative source (81, 91).

To search for signature patterns that might be indicative of distinct characteristics among the viral sequences from the infected vaccinees, we first created sequence sets comprised of a single sequence per person from the infected vaccine recipients and from the control group. The MN and SF2 rgp120 vaccine recipients were considered separately for this analysis, with only those who had three or more vaccinations prior to infection being included. The control set of viral sequences was derived from nonvaccinated, asymptomatic individuals with a documented infection, matched for time of primary infection, route of transmission, and relative risk of infection. All of gp120 was scanned for potentially interesting amino acid signatures. The program VESPA was used to search for differences in the most common amino acid in a given position between the infected vaccine cases and controls (77, 79). A minimum frequency change of 0.5 in the most common amino acid was required for selection as a signature site. Power tests for sample sizes of 7 for the query data set and 31 for the background data set indicate that differences of 0.5 can yield a P value of <0.01 with Fisher’s exact test. The program ENTROPY was used to determine if there was greater variation in the sequences of the viruses from the infected vaccinees for any position in the alignment relative to the control sequences (80). Potential asparagine (N)-linked glycosylation sites were considered as both unlinked amino acids and as units, where an N of an N-X-T/S sequon was distinguished from other asparagine residues (93, 118, 119).

The first immunization usually elicited a weak, often undetectable IgG antibody response. Subsequent booster immunizations caused rapid increases in anti-gp120 titers in most individuals; in only one case (C20, an SF-2 rgp120 recipient) was there an almost complete failure to develop an antibody response to the immunogen. In contrast, several individuals developed a very strong anti-rgp120 response to each booster immunization, exemplified by MN rgp120 recipients C11, C24, and C26, whose peak anti-MN rgp120 midpoint titers approached those found in long-term nonprogressors (Table ​(Table2)2) (12). However, the antibody response to rgp120 was transient. Measurements of the exponential decay of anti-gp120 antibodies in several of the best-responding individuals showed that the half-lives were typically about 40 to 60 days, although there was some variation among individuals and among the responses to repeat booster immunizations in the same individual (data not shown). Anti-rgp120 responses sustained between booster immunizations were noted in most individuals, but these were relatively low (titers of 1:1,000 to 1:5,000). Samples from vaccinated, uninfected individuals were not available for this analysis. However, an independent analysis confirmed that antibody titers in subjects infected despite vaccination were not significantly different from those in noninfected subjects from the same original study (57). Thus, it is highly improbable that those individuals who became infected with HIV-1 were clustered among the poorer responders to vaccination. Consistent with this, the peak anti-immunogen rgp120 titers in the infected vaccine recipients were distributed over a 200-fold range (Table ​(Table2).2).

Some individuals became infected fairly soon after receiving booster immunizations, when their antibody titers were close to maximum—for example, C17 (an MN rgp120 recipient) and C16 (an SF-2 rgp120 recipient) (Fig. ​(Fig.2).2). Other individuals became infected after their antibody responses had decayed significantly from the last booster peak—for example, C24 (an MN rgp120 recipient) and C21 (an SF-2 rgp120 recipient). It was notable that HIV-1 infection was associated with a very rapid increase in anti-gp120 antibody titers in several individuals (C10, C11, C13, C21, C24, and C26). We presume that this phenomenon represents an anamnestic response to the antigens expressed by the infecting virus strain. Other individuals developed an antibody response to the infecting strain more gradually (Fig. ​(Fig.22).

In all but one vaccinee (C16, an SF-2 rgp120 recipient), there was a significantly stronger (on average, 5.0-fold) response to the rgp120 used for immunization than to the heterologous rgp120 (i.e., anti-MN rgp120 titers were usually higher than anti-SF-2 rgp120 titers in MN rgp120 recipients, and vice versa) (Table ​(Table2).2). In contrast, the anti-MN and anti-SF-2 rgp120 titers in two unrelated cohorts of HIV-1-infected people were very similar, the mean ratio of the antibody titers to the two proteins being 1.4. The same ratio of anti-rgp120 titers (1.3 to 1.4) was found in sera from two individuals, pC04 and pC05, who were the putative transmitters of HIV-1 to individuals C04 and C05, respectively (Table ​(Table2).2). Furthermore, the anti-rgp120 titers in the vaccinees postinfection were less dependent on the test rgp120 than they were prior to infection (Fig. ​(Fig.2).2). One interpretation of these data is that although both MN and SF-2 rgp120s were cloned from clade B strains, a significant fraction of the antibody response to these proteins is to type-restricted epitopes, probably within the variable loops. In contrast, in HIV-1-infected people, the infection-induced anti-gp120 response is probably more broadly directed, leading to increased cross-reactivity with both rgp120 proteins. Alternatively, infection of the vaccinees by HIV-1 strains genetically equidistant from both MN and SF-2 (see below) might account for the broadening of antibody binding. However, it is clear that some cross-reactive anti-rgp120 binding antibodies were raised in response to both MN and SF-2 rgp120 immunizations (Table ​(Table22).

Several explanations for the failure of the MN and SF-2 rgp120 vaccines to induce antibodies able to neutralize the infecting strains of HIV-1 seem feasible. One is that the antibody response to the immunogens may be directed mostly at nonneutralizing epitopes. Alternatively, the immunogens may induce primary-virus neutralizing antibodies at an inadequate titer. Another possibility is that the infecting strains of HIV-1 are unusually resistant to neutralization; primary viruses have a wide spectrum of sensitivity to neutralization by HIV-1-positive sera, MAbs, and soluble CD4-based reagents (25, 27, 93, 98, 146). Therefore, it is possible that the rgp120 vaccines allowed transmission of only relatively resistant viruses.

To address the latter hypothesis, we tested the sensitivity of HIV-1 strains isolated from infected vaccine recipients to neutralization by a panel of test reagents. The panel comprised three human MAbs (IgG1b12, 2G12, and 2F5) and a tetrameric form of the CD4-IgG molecule (CD4IgG2) shown in independent studies to have the broadest and most potent primary-virus neutralizing activity yet described (23, 35, 144). We also included a commercially available human MAb (447/52-D) against the third variable (V3) region of gp120 (55), although this neutralizes very few primary isolates (35, 108). Except for 447/52-D (which was in limited supply), this reagent panel was tested against eight HIV-1 strains isolated from individuals who received at least three immunizations with MN or SF-2 rgp120s (MN recipients C07, C10, C11, and C17; SF-2 recipients C13, C16, C20, and C21). As controls, we selected an isolate from a placebo recipient (C18), five strains isolated from nonvaccinated individuals with acute HIV-1 infection, and three strains from individuals who had had a partial vaccination course, receiving only one or two immunizations with MN or SF-2 rgp120s (C04, C06, and C09). In view of the limited immune response to the immunogen in these individuals before infection (Fig. ​(Fig.2),2), the viruses isolated from C04, C06, and C09 provide a control group for the viruses isolated from those participants who received a complete vaccination course. Also included in the experiment were viruses isolated from four of the putative transmitters of HIV-1 to four of the infected vaccinees (pC04, pC13, pC16, and pC18). Furthermore, historical data on the neutralization sensitivity of 12 other clade B primary strains isolated from non-acute-phase individuals were also available for comparison (144). The sequences of all the HIV-1 strains tested, including the historical controls, were randomly distributed in a phylogenetic tree analysis, indicating that there was no selection bias (data not shown).

Overall, compared to the control groups, there were few differences in the neutralization sensitivities of the group of viruses isolated from those infected vaccinees who received a complete course of vaccinations (Table ​(Table3).3). A similar conclusion was reached when the 90% infectious dose values for the groups of isolated viruses were compared (data not shown). One of the test MAbs (2G12) was noticeably less effective against the control group comprising the viruses isolated from the placebo subjects, the infected subjects who received a partial vaccination course, and the acutely infected subjects than it was against the other vaccinee isolates and the historic control isolates (Table ​(Table3).3). Whether this truly represents a relative insensitivity of viruses isolated from control subjects to neutralization by this MAb is uncertain. It is not likely that the rgp120 vaccines have selected for 2G12-sensitive viruses, although this possibility exists. The small number of isolates in this study limits the significance of any conclusions that can be drawn about putative vaccine-induced selection pressures on viral phenotype.

Recent studies have shown that the levels of HIV-1 virion-associated RNA sustained in the plasma following acute infection are more predictive of the rate of disease progression than is the peak of viremia (92, 102, 129, 154). We compared the levels of HIV-1 RNA found in the plasma 9 to 12 months after infection for the group of vaccinees who had received a complete course of immunization with those of the partial vaccination and placebo groups and found no significant difference in plasma viral burden (P = 0.08, Wilcoxon rank sum test). The lowest level of plasma HIV-1 virion-associated RNA was recorded for C12, a placebo recipient (Fig. ​(Fig.2).2). No significant difference in the viral burdens of those individuals immunized with MN rgp120 and those immunized with SF-2 rgp120 was noted (P = 1.0, Wilcoxon rank sum test) (Fig. ​(Fig.5).5). Individuals with a high viral burden in plasma following primary infection have been shown to progress more rapidly to AIDS (102, 122). Whether prior immunization with HIV-1 rgp120 will alter the rate of clinical progression in the vaccinees, however, is beyond the scope of the present study.

We also compared the plasma RNA levels of the infected vaccinees at 9 to 12 months postseroconversion with those of controls derived from several cohorts of nonvaccinated individuals monitored through the acute stage of HIV-1 infection (Fig. ​(Fig.5).5). Controls were selected from among individuals who had not been given antiretroviral therapy, who were sampled in the comparable time frame of 9 to 12 months postinfection, and who were matched for age, presumed risk of exposure, and the year of primary infection. Also included in the control set were three placebo vaccination cases, two from the 401 study and one from an earlier vaccine trial (57). The placebo recipients as a set had the lowest viral burdens among the controls. One of the infected vaccinees (C20) had received antiretroviral therapy (zidovudine [AZT]) prior to 9 months postinfection, as indicated in Fig. ​Fig.5;5; this individual still had a relatively high viral RNA level (28,000 RNA copies/ml) and was retained in the analysis (Fig. ​(Fig.2).2). Nonparametric statistics were used for the comparison, because the distribution of RNA levels in the control group did not fit a normal or log-normal distribution by a Kolmogorov-Smirnov test for normality, probably due to differences in sample collection procedures among the different cohorts (Fig. ​(Fig.55).

The highest viral RNA values were found among the nonplacebo control cases (Fig. ​(Fig.5),5), but a Wilcoxon rank statistic did not indicate that the infected vaccine cases had significantly lower values than the controls (P = 0.12). The overall median viral burdens for the infected vaccine cases and the controls were comparable (8,500 and 9,325 RNA copies/ml, respectively). A similar conclusion was reached in an independent study of the infected vaccinees and controls (57). The median plasma RNA value for the infected vaccine recipient group and the controls was intermediary relative to the results from other cohorts. A median HIV-1 plasma RNA burden of 13,020 copies/ml was found for a group of 180 homosexual or bisexual men enrolled in the Pittsburgh MACS study between April 1984 and March 1985 (102); a median value of 5,871 copies/ml was found in a study of 165 hemophiliacs 12 to 36 months after the presumed date of infection (122); and Garcia et al. found a median viral burden of 9,331 copies/ml for asymptomatic individuals at a median time of 48 months after detection of infection (52). We did note that there were fewer high-viral-load burdens among the infected rgp120 recipients than among the highest quartile in the matched control group (Fig. ​(Fig.5).5). However, artifacts due to differences in sample collection protocols at different clinical sites could contribute to this observation, and its overall significance is unclear.

The viral burden analyses should be considered with several qualifications in mind, all of which apply in general to comparisons of viral loads between cohorts. First, several different cohorts were included among the controls, and two trials contributed to the rgp120 vaccinee cohort. Differing methods of sample storage and collection (e.g., serum versus plasma), or perhaps even differences in the underlying biology (e.g., intravenous drug use versus sexual transmission), may subtly influence comparisons, because we noted distinctive distributions of RNA values in the different control cohorts (Fig. ​(Fig.5).5). Second, the distribution of viral burdens for a cohort may be influenced by the age of the subjects (122) or by the frequency of acute HIV-1 infection in enrollees who undergo symptomatic primary infection. These individuals tend to have higher sustained RNA levels (21, 52, 62, 67, 92, 126) and may be more likely to be enrolled in certain natural history cohorts. Inspection of control subjects’ clinical records, however, indicated that there was no recorded bias for symptomatic acute infection or age among our control set. Third, the intrinsic variability of the plasma RNA assay and within-subject variation together add significant noise to the data. It has been proposed that sustained changes in viral RNA levels within a patient must be at least 3-fold (log 0.5) (131) or 2.6-fold (log 0.41) (125) to be considered indicative of a biologically relevant change. Differences between the median values of the vaccinee and control cohorts were less pronounced than this (Fig. ​(Fig.55).

Three sites meeting the minimum criteria to be considered as potential signature sites (see Materials and Methods) were identified and are indicated in Fig. ​Fig.6.6. The first site, which was found to be distinctive for SF2 vaccinees but not for MN vaccinees, was located in the signal peptide, so it is unlikely to be biologically meaningful. The second site was found in the C1 domain of gp120. The viral sequences from the MN and SF2 rgp120 data sets both had changes at this position relative to the control set (Fig. ​(Fig.6).6). Both the MN and SF2 vaccine strains carry a threonine at this position. Comparison of the viral sequences from the MN and the SF2 rgp120 data sets, as well as the control data set, for the presence or absence of a threonine at this position indicated a potentially interesting difference: 0 of 9 MN rgp120 recipients, 1 of 7 SF2 rgp120 recipients, and 7 of 11 controls carried a threonine at this position (P = 0.0025 when comparing SF2 and MN rgp120 vaccinee strains combined to the controls; Fisher’s exact test). However, only 24 of 81 sequences of the B subtype in the Los Alamos database were found to have a threonine at this site, dramatically reducing the significance of the above finding (P = 0.19 when comparing SF2 and MN to the database sequences; Fisher’s exact test). Alanine, not threonine, was the most common amino acid at this position in the B subtype sequences in the database (found in 34 of the 81 sequences) and was also the most common amino acid at this position in the vaccinees’ viral sequences. The biological significance of these observations is unclear.

The third potentially interesting signature site was an N-linked glycosylation site (defined as an N-X-S/T sequon) at the base of the V2 loop. It was present in only 2 of 7 sequences from SF2 rgp120 recipients but was conserved among the matched-control viral sequences (26 of 31) and in the clade B sequences in the database (71 of 103 sequences) (117). It is, however, also conserved among MN rgp120 recipient sequences (eight of nine), so it is not obvious that the variation observed at this site in the SF2 rgp120 recipients is of any biological relevance.

We thank the AVEG for conducting the MN and SF-2 rgp120 clinical trials and for the provision of samples. We thank W. Weibull (TORCH Study), J. Phair, A. Saah, R. Detels, C. Rinaldo, Jr., A. Muñoz, L. Jacobson (all of MACS), S. Buchbinder (San Francisco City Clinic Cohort), D. Vlahov (ALIVE study), J. Geodert (National Cancer Institute), and G. Woody (Philadelphia VA HIVNET study) for providing data and specimens for control subjects. We are grateful to Genentech, Inc., and Chiron, Inc., for supplying samples of the rgp120 immunogens. We thank Hermann Katinger, Dennis Burton, George Lewis, and Paul Maddon for the gifts of MAbs and CD4IgG2. We also thank the staff at each CHIP and AVEG site that contributed to this study: K. Sheridan, A. Pomales, J. Leu, S.-W. Poon, and J. Binley of The Aaron Diamond AIDS Research Center of The Rockefeller University; A. Ploss, B. Hertzfeld, P. Otto, S. Wu, and T. Myers of Northwestern University Medical School; D. Wolpert, C. Macken, J. Bingham, and A. Halpern of Los Alamos National Laboratory; N. Jones and A. Shea of Harvard University School of Medicine; G. Shaw, P. Smith, M. Mulligan, S. J. Prince, R. Kulhavy, R. Casteel, S. Ghosh, Z. Moldoveanu, S. Duncan, M. Smith, J. Decker, B. Lambert, S. Campbell-Hill, S. Morrison, B. Jian, L. Li, M. Mosteller-Barnum, K. Agnew, L. Castell, and H. Nguyen of The University of Alabama at Birmingham School of Medicine; P. Wright, P. Spearman, D. Karzon, M. A. Harbison, L. Wagner, K. Rybzyck, K. Crumbo, M. Braeuner, D. Owens, G. Rees, F. Robinson, L. Horton, I. Kuli-Zade, R. Smith, R. Cornell, and J. Keltner of Vanderbilt University School of Medicine; G. Gorse, R. Belshe, D. Kennedy, S. Frey, H. Isreal, C. Berry, B. Reed, T. Spitz, T. Pacatti, G. Patel, M. Mandava, T. Grant, and K. Feurer of St. Louis University School of Medicine; J. Lambert, D. Schwartz, C. Hilton, A. Funkhouser, and M. Johnson of Johns Hopkins University School of Medicine; M. Keefer, T. Evans, W. Bonnez, R. Reichman, L. Demeter, S. Erb, M. A. Pugliese, and J. Nichols of the University of Rochester School of Medicine and Dentistry; L. Corey, M. J. McElrath, D. Berger, H. Stacey, and L. Burke of University of Washington at Seattle School of Medicine; T. Matthews, D. Montefiori, K. Weinhold, C. McDanal, T. Greenwell, D. Davison, D. Woodford, T. Harding, A. Pilgrim, and A. Gaitin of Duke University Medical Center; P. Fast, A. Schultz, J. McNamara, B. Savarese, S. Wescott, M. C. Walker, B. Mathieson, W. Rida, D. Lawrence, and N. Ketter of the National Institute of Allergy and Infectious Diseases; and P. Barr, M. Wolff, D. Stablein, C. Smith, D. Brown, N. Lomax, and T. Voss of EMMES Corporation.

This work was supported by the National Institutes of Health under grant AI45218, the Correlates of HIV-1 Immune Protection Contract.