All 3,988 serum samples were tested by EIA, and 209 seroreactive samples were further tested by Western blot: 140 (3.5%) were HIV-1 seropositive, and 69 (1.7%) had indeterminate Western blot patterns. Samples from infected persons who have not yet seroconverted can display indeterminate Western blot patterns. If we were to assume that all 69 persons with indeterminate Western blot results were seroconverting at the time of sample collection, then the upper limit for the frequency of HIV infections within these early samples was still low at 5.2%. No HIV-2 EIA-positive specimens were confirmed by HIV-2 Western blot analysis.

The serum samples had been stored at –20°C for almost 2 decades, a temperature at which RNases are still active. Because we were concerned that the quality of the RNA would be compromised, we attempted to amplify three relatively short gene regions: the p17 region of gag (474 bp) and two env fragments, C2V3C3 (400 bp) and gp41 (460 bp). All serum specimens could not be amplified, most likely because of the poor quality of the samples. A comparison of intrasubtype diversity was determined for our 50 C2V3C3 sequences and for 181 DRC C2V3C3 sequences from 1997 (1, GenBank, http://www.ncbi.nlm.nih.gov/) with known subtype assignments (Table 1). Unclassifiable sequences were omitted from this analysis. We found the intrasubtype variability was already high in the mid-1980s (9.6%–18.7%) yet was significantly greater for each 1997 env subtype with sufficient numbers for comparisons (A, C, D, E, F1, G). No significant change was found in the frequency of each subtype (Table 1).

Figure 1 represents a neighbor-joining tree containing the following classifiable and unclassifiable C2V3C3 sequences: 56 of our early sequences from Kinshasa, 197 from the DRC collected in 1997 (subtype classifications taken from the V3–V5 sequences published in GenBank), and subtype-specific reference sequences. Some sequences that were difficult to classify in trees containing just our 1980s Kinshasa sequences and subtype-specific reference strains (data not shown) were easier to classify in the context of the larger combined Zaire/DRC phylogenetic tree. At least one of our early strains clustered with subtypes A, B/D, C, D, F (F1), G, H, J, and K, and the circulating recombinant form (CRF)01-AE clustered with 1997 CRF01 sequences. Also, some of our unclassifiable sequences clustered with 1997 unclassifiable DRC strains. One major clade contained subtype A sequences; however, at least four other distinct lineages branched independently in the tree, yet also contained 1997 sequences previously identified as subtype A viruses. One of these lineages that contained 1997 sequences previously designated as subtype A or unclassifiable also contained CRF01-AE strains in the apical portion of the cluster. Three additional unique lineages are present in the tree composed of unclassifiable sequences, perhaps representing new subtypes. Unclassifiable 1997 sequences also branched between the F1 and F2 lineages along with one of our 1980s strains, which suggests a continuum of diversity within subtype F strains instead of distinct sub-subtypes.

Even the node connecting the subtype B and D sequences was no longer distinct because of sequences falling basal to the two lineages, which made the subtype B strains appear to be part of a larger subtype D group. The 1997 sequence labeled D.3 in the tree [97DC.KS26 (accession no. AJ404096)] possibly contained a recombinant breakpoint, which caused it to branch very deep along the D lineage. However, our sequence 30884 also fell outside the node connecting subtypes B and D. In fact, this strain significantly clustered within the node connecting subtypes B and D in p17, C2V3C3, and gp41, yet did not significantly cluster with either subtype. A combined Zaire/DRC consensus sequence was constructed and used as an outgroup in this tree. The branch representing this sequence was very short and was located close to the center of the tree.

To determine an estimate of potentially "pure" subtypes, we analyzed 66 of our mid-1980s HIV strains from which we were able to amplify and sequence two or three gene regions: 53 (80.3%) viruses were amplified in all three gene regions (p17, C2V3C3, and gp41). Figure 2 shows the viruses with concordant phylogenies (possibly pure subtypes) in the sequenced gene regions as follows: A (27%), C (3%), D (18%), F1 (2%), G (8%), K (3%), and our unusual B/D virus (2%). The remaining 37% of the strains represented recombinant viruses; 32% of these appeared to be unique recombinants, except where indicated in Table 2, and 5% were CRF01_AE, a mosaic lineage containing predominantly subtype A and a unique env lineage called E. Not only was this circulating recombinant form present by the mid-1980s, but only env subtypes A, D, and G were more prevalent. Some of the discrepant gene regions possibly indicate the presence of dual/multiple infections, and the sets of specific gene region primers selectively amplified different strains. However, since our hospital worker population does not represent a high-risk group, the prevalence of infections in Kinshasa was low, and recombinant viruses are generally rapidly observed following dual infections (20), we henceforth refer to samples with discordant gene regions as recombinant viruses.

Despite a low frequency of HIV infections in Zaire in the mid-1980s compared to the level of infections currently observed in Africa, we found a remarkably high diversity of HIV-1 strains, with gene regions representing all the group M clades as well as unclassifiable regions. Intrasubtype diversity was already high among our samples, indicating that the HIV epidemic in Kinshasa was already mature by the mid-1980s. No significant change was seen in the frequency of group M subtypes between our early Kinshasa strains and a set of DRC samples collected in 1997 (1), despite political and social upheavals from 1986 to 1997.

We hypothesized that if the HIV epidemic originated in central Africa, phylogenetic analysis of our 1984/1986 Kinshasa and 1997 DRC C2V3C3 sequences might provide a more detailed history of HIV-1 evolution. We found a spectrum of diversity both within and between currently recognized HIV-1 group M subtypes and sub-subtypes (Figure 1). Some lineages, thought to represent discrete subtypes (or sub-subtypes) within the current global epidemic, no longer appeared distinct when analyzed in the context of the large diversity within Zaire/DRC sequences, i.e., multiple, distinct C2V3C3 subtype A lineages; new lineages containing unclassifiable strains; a continuum of diversity within and between the F1 and F2 sub-subtypes; and deep, unclassifiable branches. These findings demonstrate that the HIV diversity in the mid-1980s in Kinshasa was far more complex than in strains currently found in other parts of the world.

Another finding was the placement of C2V3C3 sequences of CRF01-AE in the apical branches of a lineage containing subtype A and unclassifiable sequences as basal branches. CRF01-AE is generally believed to be a recombinant virus with gag and pol genes sharing a common ancestry with subtype A, while the env and vpu genes are derived from a currently unknown parental strain, subtype E. However, others hypothesize that the unique subtype E portion of envelope is not a result of recombination (21) but may be because a higher evolutionary rate in the env and vpu genes make it unrecognizable as subtype A. Our data may be the first demonstration of subtype E env sequences grouping within a cluster of subtype A and previously unclassifiable sequences.

A combined Zaire/DRC consensus sequence used as an outgroup was very short and located near the center of the tree, close to the expected location of the hypothetical ancestor of the HIV-1 group M subtypes. This finding, along with the complexity of the combined phylogenetic tree, supports the theory that the group M subtypes evolved from a common ancestor, which may have originated in this part of Africa.

Since less than 1/10 of the genome was sequenced and analyzed in the present study, and recombination can occur anywhere along the genome, 37% recombinant viruses is very likely a significant underestimate of the actual frequency of chimeric strains cocirculating in Kinshasa by the mid-1980s. In fact, most HIV-1 strains may have contained recombinant genomes by that time. Unfortunately, our data do not allow us to discriminate whether this level of recombination through superinfection intensified the fitness of HIV strains at that time or merely contributed to viral diversity.

Although the precise ancestry of HIV-1 is still uncertain, it appears to have a zoonotic origin (8,9). Ecologic factors that would have allowed human exposure to a natural nonhuman primate host carrying a precursor virus to HIV were therefore instrumental in introducing the virus into humans. This fact suggests that zoonotic introductions of HIV may have occurred in isolated populations but went unnoticed as long as the recipients remained isolated (11) (Figure 3A). However, with increasing movements from rural areas to cities, which occurred in sub-Saharan Africa during the 1960s and 1970s, such isolation became, or was becoming, increasingly rare. The number of sub-Saharan African cities with >500,000 inhabitants rapidly increased from 3 in 1960 to 28 by 1980 (23). From 1965 to 1985, the proportion of the total population living in urban areas of central Africa rose from 21% to 35% and in western Africa from 17% to 29%.

Transmission of HIV would have amplified among high-risk populations, such as commercial sex workers, where HIV superinfections generating recombinant viruses could have resulted (Figure 3B). These recombinants could then have been spread to other, lower risk persons, such as men who visit female sex workers, their spouses, other heterosexual partners, and persons at highest risk for sexually transmitted diseases (24). Other transmission routes could also have helped create and spread recombinant viruses. For example, in 1986, studies from Kinshasa showed a strong association between receiving medical injections and HIV seropositivity among HIV-seropositive infants born to seronegative mothers (25), among hospitalized children 2–14 years of age (26), and among healthcare workers (27). Demand for blood transfusions was also high because of malaria-associated anemia, pregnancy-related complications, and sickle-cell anemia (28,29); at this time, blood was generally not screened for HIV infection (30). Therefore, many factors associated with population migration and urbanization, patterns of sexual activity, and medical practices could have played roles in producing and spreading high numbers of recombinant viruses in Kinshasa by the mid-1980s.

When HIV subtypes were initially genetically characterized in the early 1990s, the first identified viruses were assumed to represent pure subtypes, and viruses found afterward were compared to these prototypic strains. However, our data demonstrate that substantial intersubtype recombination had already occurred at the time when HIV-1 viruses were initially classified (Figure 3C). With so many recombinant viruses present at the cusp of the global HIV epidemic, at least some of the recombinant viruses in central Africa were likely classified as pure subtypes after being exported from Africa and establishing regional epidemics in other parts of the world. This realization could affect our current understanding of the range of diversity within the HIV epidemic (http://hiv-web.lanl.gov/content/hiv-db/REVIEWS/PEETERS2000/Peeters.html); of immunogen selection for vaccine design (31,32), especially in light of superinfections involving multiple subtypes (33); and of mathematical models based on molecular clocks and nonrecombinant strains (22,34,35) for dating the introduction of HIV into humans.