Nonrandom HIV-1 infection and double infection via direct and cell-mediated pathways

Abstract

Cells infected with two related retroviruses can generate heterozygous virions, which are the precursors of recombinant proviruses. Although many studies have focused on the frequencies and mechanisms of retroviral recombination, little is known about the dynamics of double infection. To examine this issue, viruses generated from two HIV-1 vectors containing different markers were mixed together, and were used to infect target cells. The numbers of cells expressing none, one, or both markers were measured and were used to calculate whether double infection occurred at frequencies expected from random infection events. We found that double infection occurred significantly more frequently than predicted from random distribution; increased rates of double infection were observed in both a T cell line and primary activated CD4⁺ T cells. In addition to direct virus infection, we also examined the nature of cell-mediated HIV-1 double infection. Increased double infection was observed in all experiments regardless of whether a cell line or primary human dendritic cells were used for capture and transmission of HIV-1. Therefore, our results indicate that HIV-1 double infection occurs more frequently than it would at random in both direct and cell-mediated HIV-1 infections. To our knowledge, this is the first direct evidence of nonrandom double infection in HIV-1. Frequent double HIV-1 infections in infected individuals would allow the generation of recombinant viruses that could then affect their pathogenesis and evolution.

One of the major difficulties in treating HIV-1 infection and generating an effective vaccine is the high rate of genetic variation in the viral population (1). This variation promotes the generation of viral strains that have become resistant to drugs used in treatment protocols or that have escaped the host immune response (1, 2). Three inherent features of HIV-1 replication contribute to the high variation: mutations generated by reverse transcriptase while copying the RNA genome into DNA (3–5), frequent recombination that further reassorts these mutations (6–8), and rapid turnover of infected cells in vivo, leading to large numbers of replication cycle in the course of infection (9, 10).

Retroviruses package two copies of viral RNA into one virion (11, 12) and frequent recombination occurs during reverse transcription between the two copackaged RNAs (13, 14). Recombination can only be observed when the two copackaged RNA molecules are genetically different (heterozygotes) (13). Thus, formation of heterozygotes is a prerequisite for frequent recombination. Heterozygous virions are exclusively generated from cells that are infected by more than one retrovirus. Therefore, the frequency of cells infected with more than one virus (double infection) has a large impact on the observed recombination events.

Currently, very little is known about the kinetics of double infection. If infection events are random, then the frequency of double infection can be calculated based on the frequencies of infection. For example, if two populations of viruses are present and each has a multiplicity of infection (moi) of 0.1 infectious unit per cell, then an estimated 1% of the cells would be infected by both viruses. However, the assumption that all infection events are random also implies that all target cells are equally susceptible to infection. If all cells are not equally susceptible to HIV-1 infection, then double infection would not be random.

In addition to infecting cells directly, HIV-1 can be transmitted via a cell-mediated pathway. It has been demonstrated that dendritic cells (DCs) can capture HIV-1 particles and later transmit the viruses to target cells (15, 16). The mechanisms of cell-mediated HIV-1 infection are currently being dissected, and some of the biological consequences of these events are unclear. For example, it is not known how many virions are transmitted from a DC to a target cell when these two cells interact. The number of virions transmitted can have a significant impact on the frequency of double infection.

In this study, we sought to directly probe the dynamics and nature of double infection. We observed that HIV-1 double infection is not random, but rather, it occurs at much higher than expected rates, both in cell-free direct virus infection and in cell-mediated infection. This is, to our knowledge, the first direct evidence of nonrandom double infection in retroviruses. Increased double infection also implies that recombinant HIV-1 is generated in infected individuals at rates higher than previously anticipated. These results have significant implications in understanding the in vivo dynamics of HIV-1 infection in patients.

Materials and Methods

Cells and Flow Cytometry Analyses. The 293T cells are a human embryonic kidney cell line containing simian virus 40 large T antigen (17, 18). Thp-1 cells used in these experiments are a transformed line of B cells (19) and Thp-1/DC-SIGN cells are transduced Thp-1 cells that express DC-specific intercellular adhesion molecule 3-grabbing nonintegrin (DC-SIGN) on the cell surface (15). Hut/CCR5 cells are derived from Hut78 cells, a human T cell line, and express the chemokine receptor CCR5 (20).

Primary immature DCs and CD4⁺ T cells were derived from peripheral blood mononuclear cells that were isolated from healthy donors and maintained as described (21). DCs were used in experiments 4–5 days postpurification. T cells were activated through T cell receptor by using anti-CD3 and anti-CD28 antibodies and maintained in IL-2-containing medium (21). This procedure generated >99% purity of CD4⁺ T cells as determined by flow cytometry analyses. Cells were used for infection 6–8 days after activation.

Transfections were performed by the calcium phosphate method (22). Harvested supernatant was clarified through a 0.45-μm filter to remove cellular debris. Phycoerythin-conjugated α-heat-stable antigen (HSA) Ab and allophycocyanin-conjugated α-CD4 (Hut/CCR5) or α-CD3 (primary T cells) Ab (BD Pharmingen) were used to stain cells, and cells were analyzed by flow cytometry. In cell-mediated infection protocols, infections were only measured in CD4⁺ (Hut/CCR5) or CD3⁺ (primary T cells) cell populations.

Statistical Analyses. Pearson's χ² test was used to analyze whether the odds ratio was significantly different from 1. The P value for statistical significance was set at 0.05. Data from parallel experiments were analyzed by using logistic regression to determine whether the odds ratios from direct infection and cell-mediated infection were significantly different. Simulations are detailed in Supporting Text, which is published as supporting information on the PNAS web site.

Results

HIV-1 Double Infection in a T Cell Line. To study the frequency of double infection, two HIV-1-derived vectors, HDV-eGFP (23) and HIV-IHSA, were used. Both vectors were derived from pNL4–3 with deletions in vpr, vif, vpu, and env; GFP or internal ribosomal entry site and mouse HSA were inserted in the nef reading frame. Viruses were generated by transiently transfecting 293T cells with HDV-eGFP or HIV-IHSA plasmid along with pIIINL(AD8)env, which expresses an Env protein derived from the AD8 strain of HIV-1 that uses CCR5 as a coreceptor (24). For convenience, HDV-eGFP- and HIV-IHSA-derived viruses are referred to as the GFP and HSA viruses in the text, respectively. Freshly harvested viruses were used to infect Hut/CCR5 cells; these cells were stained with antibodies 2–3 days postinfection and subjected to flow cytometry analyses.

A set of representative analyses is shown in Fig. 1. In these graphs, the x and y axes represented GFP and HSA expression, respectively. Depending on the levels of GFP and HSA expression, a cell could be in one of the four quadrants (labeled A–D in Fig. 1a). Cells in the A quadrant are positive for HSA but negative for GFP expression (HSA⁺/GFP^–). Cells in the B, C, and D quadrant are HSA⁺/GFP⁺, HSA^–/GFP^–, and HSA^–/GFP⁺, respectively. Mock-infected cells are shown in Fig. 1a and single infections are shown in Fig. 1 b and c. In the double infection experiments, the GFP virus and the HSA virus were mixed together and were then used to infect target cells (Fig. 1d); the distributions of the cell phenotypes were as follows: 13.7% of the cells were HSA⁺/GFP^– (A quadrant), 36.5% of the cells were HSA⁺/GFP⁺ (B quadrant), 20.7% of the cells were HSA^–/GFP^– (C quadrant), and 29.1% of the cells were HSA^–/GFP⁺ (D quadrant).

Fig. 1.

Flow cytometry analyses of HIV-1 vector infections in Hut/CCR5 cells. (a) Mock-infected cells. (b) Cells infected with the GFP virus. (c) Cells infected with the HSA virus. (d) Cells infected with a mixture of GFP and HSA viruses. The x and y axes are in log scale and represent GFP and HSA expression, respectively. Quadrants A–D (labeled in a) are cells with the phenotypes of HSA⁺/GFP^–, HSA⁺/GFP⁺, HSA^–/GFP^–, and HSA^–/GFP⁺, respectively. The percentage of cells conferring each phenotype is indicated in its respective quadrant. These designations are used for all flow cytometry analyses shown in Figs. 3 and 4.

To analyze whether double infection occurred randomly, we examined the odds ratios of these events. The numbers of cells or events measured by flow cytometry, instead of percentages of cells, were used to facilitate statistical calculations. The odds of HSA⁺ cells being GFP⁺ can be calculated by dividing the number of events in the B quadrant by the number of events in the A quadrant (B/A) (This formula is derived from [B/(B + A)]/{1 – [B/(B + A)]}.) The odds of HSA^– cells being GFP⁺ can be calculated by dividing the number of events in the D quadrant by the number of events in the C quadrant (D/C) (derived from [D/(C + D)]/{1 – [D/(C + D)]}). If double infection is random, then the frequency of HSA⁺ cells that are GFP⁺ and the frequency of HSA^– cells that are GFP⁺ is expected to be equal; therefore, the odds ratio, or (B/A)/(D/C), would be 1. If double infection occurs more frequently than expected from random events, then the frequency of HSA⁺ cells that are GFP⁺ (B/A) is expected to be higher than the frequency of HSA^– cells that are GFP⁺ (D/C); therefore, the expected odds ratio (B/A)/(D/C) would be >1. If double infection occurs less frequently than expected from random events, then the expected odds ratio (B/A)/(D/C) would be <1. The calculations shown here compared the frequency of GFP⁺ observed in HSA⁺ and HSA^– cell populations. However, the GFP and HSA viruses were used to simultaneously infect the target cells. The odds ratio of double infection could also be calculated by comparing the frequency of HSA⁺ observed in GFP⁺ and GFP^– cell populations, or (B/D)/(A/C). These two calculations yielded the same odds ratio mathematically: (B/A)/(D/C) = BC/AD = (B/D)/(A/C).

By using this approach, we calculated the odds ratio of double infection in such experiments. For example, in Fig. 1d, the numbers of cells in quadrants A–D were 4,269, 11,391, 6,480, and 9,077, respectively. This distribution yielded an odds ratio of 1.9, which was significantly different from the odds ratio expected from random double infection events (P < 10^–11). Table 1 shows six sets of representative data from >50 sets of independent experiments. In all 50 experiments, the odds ratios of double infection were significantly higher than random (P < 10^–6 to P < 10^–11). Therefore, double infection occurred more frequently than predicted from a random distribution.

Table 1. Double infection analyses using Hut/CCR5 cells.

Experiment	A H+/G-	B H+/G+	C H-/G-	D H-/G+	Odds ratio*
1	4,269	11,391	6,480	9,077	1.9
2	4,029	1,772	10,198	1,740	2.6
3	3,777	6,462	5,827	2,633	3.8
4	3,486	3,460	9,423	2,682	3.5
5	2,821	17,404	1,995	5,797	2.1
6	2,891	907	10,525	1,072	3.1

A—D correspond to quadrants shown in the flow cytometry analyses in Fig. 1 (the same designations are used in all tables).

^*All odds ratios were significantly >1 (P < 10^-11).

Possible Roles of Cell and Virus Aggregation. To determine whether the increased double infection was caused by possible experimental factors, we examined the roles of cell aggregation and virus aggregation. Aggregation of infected cells, such as an HSA^–/GFP⁺ cell and an HSA⁺/GFP^– cell, could have occurred and been misclassified as an HSA⁺/GFP⁺ during analyses. This possible cell aggregation would have artificially elevated the number of cells in the B quadrant. The forward- and side-scattering profiles during flow cytometry and our visual examination of the cells argue against cell aggregation. However, we performed experiments to examine the possibility of cell aggregation affecting our results. A GFP virus and an HSA virus were used to separately infect Hut/CCR5 cells. Two days postinfection, cells from these two groups of infected cells were mixed together, processed, and analyzed by flow cytometry. Our analyses demonstrated that only insignificant numbers of cells were detected in the B quadrant, indicating that the increased double infection observed was not caused by cell aggregation (data not shown).

It is also possible that virus aggregation could have occurred and affected our results. When the GFP and HSA viruses were mixed before infection, virions could have aggregated such that when infection occurred, more than one virus could enter a cell. To explore the role of virion aggregation on double infection, we infected Hut/CCR5 cells with one virus for 1 h, washed off the unbound virus, and infected the cells with the other virus. Cells that were simultaneously infected with both viruses, infected with the HSA virus first, and then infected with the GFP virus, or infected with the GFP virus first, and then with the HSA virus generated similar odds ratios, indicating that virus aggregation did not play a significant role in our measurements and conclusions (data not shown).

Kinetics of Virus Infection and the Effect of moi on Double Infection. It has been reported that retroviral infection can be blocked by negative factors such as Fv-1 in murine leukemia virus infection or Fv-1-like factors in HIV infection (25–27). These blocks can be ablated by infection with high doses of viruses. If such blocks exist in our experimental system, then successful infection would occur only in cells that are infected by multiple viruses, which would create a bias for increased double infection. Similarly, if our detection system allows only doubly infected cells to be measured, we would also have observed increased double infection. If only doubly or multiply infected cells are detected, then virus infection would not have single hit kinetics in our system.

Experiments to test the kinetics of infection in our system are shown in Fig. 2. Single-, double- or multiple-hit kinetics can be distinguished by determining the numbers of cells infected as a function of virus dilution. Hypothetical lines reflecting single and double hit kinetics (dotted red and blue lines, respectively) in which virus dilutions are shown on the x axes and moi (Fig. 2a) or percent of infected cell (Fig. 2b) is shown on the y axis. If the kinetics of virus infection are a double hit, the amounts of virus infection would decrease much more rapidly with virus dilution than if they are one hit. The amount of virus infection will decrease even more rapidly with virus dilution in multiple-hit kinetics than two-hit kinetics (not shown). To examine the kinetics of virus infection, we measured the effect of virus dilutions on the proportion of infected cells. We performed serial two-fold virus dilutions, infected target cells with these dilutions, and determined the number and percentage of cells infected. Data summarized from six sets of independent experiments are shown Fig. 2 a and b; in all experiments, the relationship between virus dilution and the proportion of infected cells (or moi) paralleled that predicted from single-hit kinetics.

Fig. 2.

Virus infection kinetics. The relationship between virus dilution and moi (a) and proportion (b) of infected cells. The y axis denotes moi (a) or proportion (b) of infected cells, and the x axis denotes virus dilutions with each mark indicating a 2-fold dilution. Dotted red and blue lines represent predicted values from single- and double-hit kinetics, respectively. Six sets of independent experiments are shown as solid lines. (c) The effect of moi on odds ratio of double infection. The x axis denotes moi, and the y axis denotes odds ratio. Representative data from two independent experiments are shown as yellow and purple squares.

By using the same virus dilution approach, we have also examined the effect of moi on odds ratio of double infection. We observed that nonrandom double infection occurred at all virus dilutions and moi values tested. Data from two sets of experiments (purple and yellow squares) are shown in Fig. 2c, in which the x axis denotes moi, and the y axis represents the odds ratio. The odds ratios of all data points were significantly higher than 1 (P < 10^–11).

HIV-1 Double Infection in Primary CD4⁺ T Cells. The experiments in the previous section were performed by using cells derived from a permanent T cell line. To determine whether the same phenomenon was present in primary cells, we performed experiments to investigate HIV-1 infection of activated human CD4⁺ T cells. By using the same protocols described for the Hut/CCR5 cell experiments, we used the GFP and HSA viruses to infect primary T cells and performed flow cytometry analyses. Representative analyses are illustrated in Fig. 3; in cells infected with both viruses (Fig. 3d), quadrants A–D had 1,520, 1,461, 14,331, and 1,683 cells, respectively, leading to an odds ratio for double infection of 8.2, which was significantly different from 1. Six independent sets of experiments are shown in Table 2. The odds ratios from all of these experiments were significantly >1(P < 10^–11). These results indicate that double infection in activated primary human CD4⁺ T cells also occurred at a frequency much greater than random.

Fig. 3.

Flow cytometry analyses of HIV-1 vector infection in activated primary CD4⁺ T cells. (a) Mock-infected cells. (b) Cells infected with the GFP virus. (c) Cells infected with the HSA virus. (d) Cells infected with a mixture of GFP and HSA viruses.

Table 2. Double infection analyses with CD4⁺ primary T cells.

Experiment	A H+/G-	B H+/G+	C H-/G-	D H-/G+	Odds ratio*
1	2,780	1,643	19,837	2,678	4.4
2	1,927	1,506	15,994	2,846	4.4
3	1,056	1,430	11,606	1,592	9.9
4	1,520	1,461	14,331	1,683	8.2
5	1,189	737	25,464	2,106	7.5
6	946	446	17,225	1,636	5.0

^*All odds ratios were significantly >1 (P < 10^-11).

Cell-Mediated HIV-1 Infection. In addition to direct infection by cell-free virions, HIV-1 can also be transmitted by a cell-mediated pathway (15, 16), in which DCs can capture HIV-1 and transmit the virus to T cells, resulting in a productive infection of these cells (15, 16). DC-SIGN was shown to play an important role in cell-mediated HIV-1 transmission (15). During cell-mediated transmission, it is not clear whether multiple viruses are passed from a DC to a T cell during each transmission event, and thus whether double infection would be random in cell-mediated transmission.

To investigate this question, we first examined cell-mediated HIV-1 transmission by using Thp-1 and Thp-1/DC-SIGN cell lines. It was previously shown that Thp-1/DC-SIGN cells can capture and transmit HIV-1, whereas Thp-1 cells cannot mediate HIV-1 transmission (15). We performed cell-mediated HIV-1 infection experiments by using the following protocol: Thp-1 or Thp-1/DC-SIGN cells were incubated with viruses for 1 h, unbound viruses were removed by washing the cells, target Hut/CCR5 cells were then added to Thp-1 or Thp-1/DC-SIGN cells, and cells were processed 2–3 days later and were analyzed by flow cytometry. Consistent with the previous reports, Thp-1/DC-SIGN cells, but not Thp-1 cells, were able to capture and transmit HIV-1 (data not shown). A representative analysis of infection mediated by Thp-1/DC-SIGN cells is shown in Fig. 4a; the numbers of cells in quadrants A–D were 2,258, 3,319, 13,536, and 3,476, respectively; the odds ratio for GFP infection in HSA⁺ and HSA^– cell populations was 5.7, which was significantly different from expected from random double infection. Data from three sets of experiments are shown in Table 3.

Fig. 4.

Flow cytometry analyses of cell-mediated HIV-1 vector infections. (a) Infection by using Thp-1/DC-SIGN as virus-capturing cells and Hut/CCR5 as target cells. (b) Infection by using primary DC as virus-capturing cells and primary CD4⁺ T cells as target cells.

Table 3. Analyses of cell-mediated infection.

Experiment	Cells*	A H+/G-	B H+/G+	C H-/G-	D H-/G+	Odds ratio†
1	Thp-Hut	2,623	3,976	10,604	3,957	4.1
2	Thp-Hut	4,202	5,639	24,396	6,656	4.9
3	Thp-Hut	2,258	3,319	13,536	3,476	5.7
1	Primary	387	311	16,043	923	14.0
2	Primary	934	790	14,120	1,069	11.2
3	Primary	1,294	375	52,723	2,881	5.3

^*Thp-Hut, Thp-1/DC-SIGN as virus-capturing cells and Hut/CCR5 as target cells; Primary, primary DCs as virus-capturing cells and T cells as target cells.

^†All odds ratios were significantly higher than 1 (P < 10^-11).

To investigate cell-mediated HIV-1 infection of primary cells, we performed capture experiments by using primary human DCs and activated human CD4⁺ T cells. A representative analysis is shown in Fig. 4b; in this example, the odds ratio for GFP virus infection of HSA⁺ and HSA^– cell populations was 11.2, which was again significantly different from expected from random double infection. Data from three sets of experiments are summarized in Table 3; in all of these experiments, double infection was greater than random.

In summary, our results indicate that HIV-1 double infection occurred significantly more frequently than it would at random in two distinct transmission pathways: direct cell-free infection and cell-mediated infection. In our experiments, we used both cultured cells and human blood-derived primary cells as targets and virus-transmitting cells. Regardless of the cells used, increased frequencies of double infection above those expected from a random distribution were observed.

Discussion

In the experiments presented here, we found that double infection occurred more frequently than predicted from random distribution. The possible mechanisms that cause such a phenomenon and the implications of the observed increased double infection are discussed below.

Modeling the Effects of Target-Cell Population. In order for double infection to occur randomly, the target cells must all be equally susceptible to virus infection. However, complete homogeneity is difficult to achieve in most biological systems; therefore, it is very likely that variation in susceptibility to virus infection exists in the target-cell population. To explain nonrandom double infection, we propose two hypotheses that are not mutually exclusive: the presence of noninfectable cells, and the presence of target-cell subpopulations that differ in susceptibility to infection. In the first hypothesis, the presence of noninfectable cells in the target-cell population would elevate the number of double-negative cells and thereby significantly change the odds ratio. A model depicting the relationship between moi and odds ratio in target-cell populations containing 0% (black), 5% (red), 10% (blue), and 20% (green) of noninfectable cells is shown in Fig. 5a. We did not simulate scenarios with >20% noninfectable cells because, in many experiments using the Hut/CCR5 target cells, >80% of the cells were infected. As shown here, the odds ratio is close to 1 at a lower moi and increases dramatically at a higher moi.

Fig. 5.

The effect of target cell subpopulations on virus infection. (a) Predicted and observed relationship between odds ratio and moi values in the presence of noninfectable cells. In these simulations, a fraction (≥ 80%) of the cells are equally susceptible to infection and there is a subpopulation of noninfectable cells. The noninfectable cell populations represented by the black, red, blue, and green lines are 0%, 5%, 10%, and 20%, respectively. (b) Predicted and observed relationship between odds ratio and moi values when the target cell population has variations in susceptibility to virus infection. In these simulations, target cells have five subpopulations with a gradient of susceptibility to virus infection. The susceptibility of the first subpopulation is 100%, and the susceptibility of the subsequent subpopulations are portions of the susceptibility of the prior population as defined by the gradient. Infection is assumed to be random within each subpopulation. Gradients for black, red, blue, and green lines are 100%, 70%, 60%, and 50%, respectively. For a and b, the y axis shows the odds ratio and the x axis shows the moi; yellow and purple blocks represent data obtained from two independent experiments shown in Fig. 2c. (c and d) Simulated infection kinetics as predicted by the noninfectable cell population hypothesis (c) and variation in susceptibility to the virus infection hypothesis (d). The y axis denotes the predicted moi values, and the x axis denotes virus dilution. Black, red, blue, and green lines in c and d have the same assumptions as those in a and b, respectively.

The second hypothesis proposes that within the target cells, there are subpopulations of cells with different susceptibility to infection. Under this hypothesis, even if infection is random in each subpopulation of cells, the sum of the entire population can be nonrandom. An example of population variation is shown in Table 4. In this hypothetical situation, there are three subpopulations of target cells, with 10,000 cells in each population, and infection is random within each population. It is assumed that there are two viruses (G and H), each of which infects 50%, 20%, and 5% of the cells in the first, second, and third cell populations, respectively. As shown in the Table 4, the odds ratio in each subpopulation is 1; however, the odds ratio of the entire population is 2.5, indicating that double infection occurs far more frequently than expected.

Table 4. Simulation of the effect of variation in target cell susceptibility to virus infection on odds ratio.

Subpopulation	No. of cells	A H+/G-	B H+/G+	C H-/G-	D H-/G+	Odds ratio*
1	10,000	2,500	2,500	2,500	2,500	1
2	10,000	1,600	400	6,400	1,600	1
3	10,000	475	25	9,025	475	1
Total	30,000	4,575	2,925	17,925	4,575	2.5

A simulated population containing equal proportions of cells with different susceptibility to infection (see text) was infected with a mixture of two viruses; each virus has an moi sufficient to infect half of subpopulation 1.

^*(B/A)/(D/C).

Simulated results generated from the second hypothesis are shown in Fig. 5b. These simulations assume the presence of five different subpopulations of cells and random infection within each subpopulation. In one scenario, the five subpopulations have a gradient of 70% susceptibility (red line); for example, the second subpopulation has 70% of the susceptibility of the first, and the third subpopulation has 70% of the second, and so forth. In other scenarios, the five subpopulations have a gradient of 60% (blue line), 50% (green line), or all five subpopulations have 100% susceptibility (black line). As indicated in the graph, the relationship between the odds ratio and the moi is distinct from that of the first hypothesis.

We compared these simulations with results generated by using the Hut/CCR5 target cells (Fig. 2c). Our data are not consistent with the prediction generated from the noninfectable cell population hypothesis (Fig. 5a); specifically, in these experiments, the odds ratio did not drop to close to 1 at lower moi values such as 0.125 or 0.25, as predicted by the first hypothesis. Although we cannot exclude the possibility of noninfectable cells in the target-cell population, nonrandom double infection cannot be explained solely by their presence. In contrast, our data are consistent with predictions generated from the second hypothesis (Fig. 5b), indicating that factors such as variations in susceptibility to virus infection in the target-cell population could have affected the odds ratio and generated the observed nonrandom double infection.

The simulations above are intended to establish that variation in the target-cell population can cause nonrandom double infection. Compared with Hut/CCR5 cells, primary cell populations should have far more heterogeneity in their susceptibility to HIV-1 infection, including the presence of a significant portion of noninfectable cells. Testing the aforementioned two hypotheses by using the primary target cells is difficult. First, although we observed nonrandom double infection in every donor tested (cells derived from at least 20 donors were tested), the odds ratio fluctuated greatly among activated CD4⁺ T cells derived from different donors. Second, it is technically difficult to achieve HIV-1 infection of high moi values in primary cells; in parallel experiments, virus stocks that generated moi values of 2 in Hut/CCR5 only yielded moi values of 0.2–0.3 in primary cells. These issues prevented us from successfully probing the nature of the varied susceptibility to HIV-1 infection in primary cells.

Reconciling the Paradox Between Single-Hit Kinetics of Virus Infection and Increased Double Infection. Retroviruses infect permissive cells with single-hit kinetics (28). In this study, we have shown HIV-1 infection kinetics consistent with that of a single hit; however, double infection occurs more frequently than anticipated from random events. These two observations may appear paradoxical; however, when the target-cell subpopulations varied in their susceptibility to infection, it was possible to have nonrandom double infection with infection kinetics approaching that of a single hit. For example, the kinetics of virus infection in the two aforementioned hypotheses are similar to single-hit kinetics (Fig. 5 c and d), and yet both hypotheses predict nonrandom infection (Fig. 5 a and b).

Nonrandom Virus Infection and Its Effect on Moi. The observed increased double infection implies that HIV-1 infection is nonrandom. However, the idea of a nonrandom virus infection challenges the concept of moi, which is based on the assumption that virus infection is random. As indicated in our simulations, although virus infection is nonrandom due to moderate variation of the target cells' susceptibility to virus infection, the kinetics of virus infection still mimic single-hit kinetics and parallel the predictions generated by the moi concept (Fig. 5 c and d). Therefore, moi remains a useful tool for estimation of infectious events.

Double Infection Through Cell-Mediated Infection. Electron microscopy (EM) analyses of human and other primate DCs that have captured HIV-1 or simian immunodeficiency virus have revealed the existence of cellular compartments, each containing multiple virions (refs. 29 and 30 and Q.D., K. Nagashima, M. Gignac, and W.-S.H., unpublished observations). If the virions from one of the cellular compartments were passed to a target cell during DC–T cell interaction, then multiple virions could be transmitted. Additionally, EM and other imaging analyses have revealed the presence of multiple virions between the contact points of a DC and a T cell during cell–cell interaction (30, 31). Thus, it is possible that the multiple noninternalized virions could also be transmitted from a single DC to a T cell. Our double infection data, together with the EM images, led us to hypothesize that multiple HIV-1 virions are transmitted from a DC to a T cell, thereby resulting in the increased double infection observed in our analyses. We have further tested this hypothesis by mixing target cells with one population of DCs that captured the GFP virus and a second population of DCs that captured the HSA virus, then analyzed the frequency of target-cell infection. Our data are consistent with the hypothesis that multiple viruses were transmitted by one DC (J.C., Q.D., and W.-S.H., unpublished observation).

Because different mechanisms were most likely involved during the different protocols, we examined whether the usage of different mechanisms might have affected the frequency of double infection. By using logistic regression, we compared the odds ratios generated from direct infection and cell-mediated infection performed in parallel. The data from seven independent experiments are summarized in Fig. 6. In all experiments, cell-mediated infection had higher odds ratios than those generated by direct infection. Logistic regression analyses indicated that the odds ratios from these two protocols were significantly different in six of the seven experiments. These data suggest that double infection occurs more frequently when HIV-1 is transmitted via a cell-mediated pathway than by direct infection.

Fig. 6.

Comparison of odds ratios among different HIV-1 infection protocols used. Seven sets of experiments performed in parallel are shown; each set of parallel experiment is shown in one color. Different infection protocols are indicated along the x axis, and the y axis represents the odds ratio.

Implication of Nonrandom HIV-1 Double Infection. Recombination plays an important role in the current AIDS epidemic. For example, many of the currently circulating primary strains of HIV-1 are actually intersubtype recombinants (32, 33). Additionally, it has been shown that multidrug-resistant virus strains can be generated by recombining the genomes of two viruses, each resistant to a single drug (34). The frequency of recombination in an infected individual can be affected by several factors: the number of doubly infected cells, the efficiency of heterozygote formation, and the rate of template switching during reverse transcription. In this report, we demonstrate that the frequency of double infection is much greater than expected from random events. Increased occurrence of doubly infected cells can have a direct impact on the frequency of observed recombination in the infected individuals. This finding can also have a very significant impact on the evolution of HIV-1.

In our study, viruses were added to cells to mimic HIV-1 transmission in vivo. Because relatively high amounts of virus were used in our experiments, one may question whether viral loads in infected individuals allow the occurrence of double infection. It has been demonstrated that patients have high viral loads during certain phases of HIV-1 infection (35). In addition, viral load detected from the circulating blood does not reflect the numbers of virions in tissues; for example, when infected cells release large bursts of viruses in lymphoid tissues, local virion concentrations higher than that of the peripheral blood can be generated. Interestingly, analyses of spleen samples from two patients revealed that 75–80% of HIV-1-infected cells harbored more than one provirus (36). This observation agrees with our study and lends further support to the hypothesis that double infection occurs far more frequently than it would at random in infected patients.