The Genetic Bottleneck in Vertical Transmission of Subtype C HIV-1 Is Not Driven by Selection of Especially Neutralization-Resistant Virus from the Maternal Viral Population

Abstract

Subtype C human immunodeficiency virus type 1 (HIV-1C) continues to cause the majority of new cases of mother-to-child transmission (MTCT), and yet there are limited data on HIV-1C transmission. We amplified env from plasma RNA for 19 HIV-1C MTCT pairs, 10 transmitting in utero (IU) and 9 transmitting intrapartum (IP). There was a strong genetic bottleneck between all mother-infant pairs, with a majority of transmission events involving the transmission of a single virus. env genes of viruses transmitted to infants IP, but not IU, encoded Env proteins that were shorter and had fewer putative N-linked glycosylation sites in the V1-V5 region than matched maternal sequences. Viruses pseudotyped with env clones representative of each maternal and infant population were tested for neutralization sensitivity. The 50% inhibitory concentration of autologous serum was similar against both transmitted (infant) and nontransmitted (maternal) viruses in a paired analysis. Mother and infant Env proteins were also similar in sensitivity to soluble CD4, to a panel of monoclonal antibodies, and to heterologous HIV-1C sera. In addition, there was no difference in the breadth or potency of neutralizing antibodies between sera from 50 nontransmitting and 23 IU and 23 IP transmitting HIV-1C-infected women against four Env proteins from heterologous viruses. Thus, while a strong genetic bottleneck was detected during MCTC, with viruses of shorter and fewer glycosylation sites in env present in IP transmission, our data do not support this bottleneck being driven by selective resistance to antibodies.

INTRODUCTION

The biological mechanisms involved in human immunodeficiency virus type 1 (HIV-1) transmission remain largely unclear. A genetic bottleneck has been routinely observed in both horizontal and vertical transmission, although findings on characteristics of these transmitted viruses are conflicting and may depend on the mode of transmission or subtype of the infecting virus. Owing to the availability of matched donor-recipient pairs and the relatively well-defined timing of transmission, mother-to-child transmission (MTCT) of HIV-1 is a tractable setting in which to study this bottleneck and determine the viral characteristics and/or immune responses associated with transmission, with the potential to suggest mechanisms.

Correlations between HIV-1 transmission, variable loop length and number of putative N-linked glycosylation (PNG) sites encoded in the HIV env gene have been reported in some studies (9, 10, 33, 58, 59, 60) but not in others (9, 32, 47). In horizontal transmission, acutely infected subjects were found to have shorter variable loops and fewer PNG sites encoded in env compared to subjects with chronic HIV-1 infection for subtypes A and C, but not subtype B (10, 32, 34). In vertical transmission, one study of an HIV-1 subtype CRF_AE-infected cohort found no difference in sequence length or PNG sites, while in other studies analyzing multiple subtypes there were fewer PNG sites in transmitted viruses (47, 59). Shorter variable loops and fewer PNG sites have separately been shown to correlate with increased viral fitness (42) and greater neutralizing antibody sensitivity (22). One study suggested HIV-1C viruses newly transmitted from mother-to-child were more fit, had significantly fewer PNG sites, and were more resistant to autologous maternal serum than nontransmitted viruses (60). For studies that analyzed vertical transmission stratified by timing, viral populations have been reported to have different properties if transmitted in utero (IU) or intrapartum (IP) (3, 11, 30).

There is no consensus on the role neutralizing antibodies may play in MTCT. Animal studies have demonstrated that neutralizing antibodies elicited by a simian immunodeficiency virus (SIV) vaccine can at least slow disease progression (56), while direct administration of antibodies matched to the challenge virus can block transmission (16, 44). Studies of natural MCTC have yielded conflicting results (2, 3, 7, 20, 26, 31, 32, 48), although possibly for identifiable reasons. The breadth of the neutralizing antibody response may depend on the subtype of HIV-1 being studied (5, 10), and neutralizing antibody levels may be associated with the timing of transmission (3). A comprehensive picture of the effect of neutralizing antibodies on MTCT is difficult to obtain because of small sample sizes and different subtypes and methods involved in the reports. Thus, larger studies of relevant subtypes accounting for transmission timing are needed to better understand the transmission mechanisms. There is also uncertainty about the role of antibodies in superinfection (6, 54).

In the present study we analyzed HIV-1 subtype C env genes from 19 mother-infant pairs: 10 transmitting IU and 9 transmitting IP. We confirmed the strong genetic bottleneck associated with vertical transmission. Compared to the maternal viral population, viruses transmitted IP tended to have shorter variable loops and fewer PNG sites than viruses transmitted IU. However, we also found that env pseudotyped viruses from transmitted infant variants and maternal variants were not significantly different in overall sensitivity to neutralization by monoclonal antibodies or heterologous or autologous serum. These data argue against a model in which the transmitted virus is differentially sensitive or resistant to neutralization compared to the mother's virus population.

MATERIALS AND METHODS

Ethics statement.

Plasma samples were collected as part of the Malaria and HIV-1 in Pregnancy (MHP) prospective cohort (17, 29, 30, 38, 39). The MHP study was approved by both the Malawi College of Medicine Research Committee and the Institutional Review Board at the University of North Carolina at Chapel Hill. Informed, written consent was obtained for all participants.

Study participants.

Women and their newborn infants were all documented to have received single-dose nevirapine according to the HIVNET 012 protocol (19). There was no association between malaria coinfection and HIV-1 transmission in the MHP study (40), but women transmitting IU had higher rates of syphilis coinfection than all other groups (39).

Plasma and serum were isolated from blood collected at labor-ward admission from the women prior to nevirapine administration. Plasma was also stored from the umbilical cord and infant heel-pricks at three time points: within 48 h of birth, at 6 weeks of age, and at 12 weeks of age. HIV-1 transmission from mother to infant was categorized by timing according to the method of Bryson et al. (8) as follows: infants who were determined to be HIV-1 DNA negative by real-time PCR (35) at 0 and 6 weeks were defined as nontransmitters, infants who were HIV-1 DNA positive at birth were defined as IU infections; and infants who were HIV-1 DNA negative at birth and DNA positive at 6 weeks were defined as IP infections. Given this definition of IP transmission, infants could have been infected late IU, IP, or during early breast-feeding. Based on the availability of the plasma, the ability to obtain PCR products, and the confirmation of a phylogenetic linkage of maternal and infant sequences, 10/35 IU and 9/39 IP transmission pairs were included in the present study. Within these pairs, eight IP and three IU pairs had additional plasma from their 12-week postpartum visit available. A sample from 6 weeks was also available for one additional IU transmission pair and was analyzed. Of these women, one who transmitted IU (subject 1468) was infected with malaria, and one who transmitted IU (subject 1629) and two who transmitted IP (subjects 312 and 2038) were infected with syphilis.

Single genome amplification.

Viral RNA was isolated from 140 μl of plasma by using a QIAmp viral RNA minikit (Qiagen, Germantown, MD). When <140 μl of plasma was available, the entire volume was used. We used the single-genome amplification (SGA) protocol (14, 41, 51) as developed for the env gene (46) and described for subtype C (1). Briefly, cDNA was generated by using Superscript III reverse transcriptase and an oligo(dT) primer, followed by RNase H treatment (Invitrogen, Carlsbad, CA). The env gene was amplified by nested PCR from the dilution of cDNA that resulted in ca. 30% positive PCRs. These conditions ensure that >80% of the amplifications are initiated with a single template, and it eliminates artifactual PCR-mediated recombination, which has been observed when PCR amplifications are initiated with multiple templates (46). PCR products were directly sequenced. Double peaks in the chromatograms were considered evidence for the presence of multiple templates in the reaction, and the sequences from these amplicons were discarded, as were sequences with deletions larger than 100 nucleotides compared to the consensus sequence of the individual.

Phylogenetic analysis.

Sequences were manually edited using Seaview (15), and the boundaries of the V1-V5 region were determined from the protein alignment in the HIV Sequence Compendium 2009 (28). A neighbor-joining tree, including sequences from all pairs, was constructed to confirm epidemiological linkage between the pairs and the absence of cross-sample contamination (see Fig. S1 in the supplemental material). Matched maternal and infant sequences were most closely related to each other. All sequences clustered with HIV-1 subtype C (data not shown). Epidemiologically linked sequences were aligned by using MAFFT version 5.8 according to the L-INS-i method (23). The model of evolution was determined by FindModel (www.hiv.lanl.gov) for each pair, with GTR plus gamma often being the best fit, and a maximum-likelihood phylogenetic tree was constructed without gap-stripping using the Mobyle platform for the PHYML program (21). Trees were resampled 100 times, and bootstrap values greater than 70 were considered significant. Unrelated subtype C outgroups were used to root each tree (data not shown for space).

We estimated the average evolutionary divergence within each sample using the Kimura two-parameter method in MEGA4 (25, 55). The rate variation among sites was modeled with a gamma distribution (shape parameter = 0.5). All positions containing alignment gaps and missing data were eliminated only in pairwise sequence comparisons (pairwise deletion option) to retain the most data for comparison.

Positive selection was assessed by using the Datamonkey webserver of the HyPhy Package (43). Sequences from IP infant 1100 could not be analyzed because of the absence of three unique sequences within the 15 amplicons. Sequences that appeared to be from recombinants were excluded from the analysis, and distinct lineages within an infant with more than three sequences were analyzed separately. Codon-optimized alignments were created with HIValign (http://www.hiv.lanl.gov) and uploaded to the Datamonkey server. Evolutionary models were chosen by the automatic model selection tool (most alignments fit the F81 model) and then analyzed by the Parris method (49) for an overall test of selection. There was not enough diversity in the infant sequences for an accurate measure by codon-specific selection analysis.

Evidence of recombination was initially identified by using the Highlighter tool and visual inspection. Where recombination was suspected, the single breakpoint recombination (SBP) tool on the Datamonkey webserver was used to confirm recombination, and the genetic algorithms for recombination detection (GARD) tool was used to identify breakpoints (27). Sequences from longitudinal infant time points were analyzed together when available.

Pseudotyped virus.

env genes amplified with the upstream primer EnvA* (CACCGGCTTAGGCATCTCCTATGGCAGGAAGAA) and the downstream primer EnvN (CTGCCAATCAGGGAAGTAGCCTTGTGT) were cloned into the pcDNA3.1 vector (Invitrogen). env amplicons for cloning were chosen to represent the diversity of each viral population, assessed using the maximum-likelihood phylogenetic trees described above. Pseudotyped virus stocks were made according to previously published protocols (37) for each env gene clone, as well as with plasmids containing SF162, TV-1, MW965, and JRCSF env genes. MW965 encodes a highly neutralization-sensitive subtype C Env protein (40, 50), while JRCSF encodes an Env protein that is subtype B and only moderately neutralization sensitive. The env expression plasmid and Δenv pNLCH backbone (a derivative of pNL4-3) plus luciferase (J. Jeffries, unpublished data) were cotransfected into 293T cells using the FuGENE 6 transfection reagent (Roche Applied Science, Indianapolis, IN). The supernatants were harvested after 48 h.

Serum and monoclonal antibodies.

Maternal serum was available from the hospital admission for testing autologous neutralization sensitivity. Soluble CD4 (R&D Systems, Minneapolis, MN), a panel of broadly neutralizing monoclonal antibodies (IgG1b12, 2G12, 2F5, and 4E10), and heterologous sera (HIVIG, BB8, BB28, BB47, and BB68) from HIV subtype C-infected individuals (18) were also used to test general neutralization sensitivity of the Env proteins.

Viruses pseudotyped with SF162, TV-1, MW965.26, and JRCSF were used to test heterologous neutralizing antibody activity in maternal sera. All available serum samples from transmitting mothers were used (at least 23 IU and 23 IP; more were used for some env genes, but there was not sufficient volume of all sera to test all four Env proteins), while serum samples from twice the number of the transmitting group (n = 50) were used from the nontransmitting group of mothers.

Neutralizing antibody assay.

Single-round neutralizing antibody assays were performed as described previously (37). Neutralization was quantified by a reduction in luciferase reporter gene expression after infection of TZM-bl cells (NIH AIDS Research and Reference Reagent program catalog no. 8129). Titers were reported as the antibody concentration or reciprocal serum dilution with a 50% reduction in relative luminescence units. Values were interpolated by using five-parameter curve fitting.

Glycosylation site comparison.

Sequences were translated by using the Translate program (www.hiv.lanl.gov) and aligned with MAFFT version 5.8 using the L-INS-i method (23); putative glycosylation sites were identified by using the N-Glycosite tool (61). The fraction of maternal sequences with each glycosylation site was subtracted from the fraction of infant sequences with that same site to calculate the fractional difference between mothers and infants. Some variability in the location of each glycosylation site was allowed, particularly within the variable regions (i.e., if each sequence from a subject had a single PNG site within a 2- to 3-amino-acid range of the alignment, this was considered an equivalent site for the calculation).

Statistical methods.

Differences in average sequence length and the number of glycosylation sites between transmission pairs were calculated both overall and within strata defined by transmission route, using linear generalized estimating equations (GEE; SAS 9.2; SAS, Inc., Cary, NC), with standard errors (SE) calculated using robust variance estimators (based on exchangeable correlation matrices). GEE estimates were calculated using the putative transmitted variants (n = 25 sequences) only. To confirm model-based results, a resampling-based approach was also used both to calculate transmission pair differences and to account for clustering between pairs. Here, mean PNG or mean nucleotide length was first calculated for each cluster of maternal sequences. Each maternal cluster was then sampled 10,000 times with replacements, and the differences between the cluster mean and the sampled observation were calculated. These differences were summed across maternal clusters, resulting in a normal distribution of summed differences centered at zero. For both mean PNG and nucleotide length, the corresponding transmitted variant value for each paired infant was subtracted from the maternal mean, summed across pairs, and compared to the 95% confidence intervals of the bootstrapped maternal distribution overall and for IP and IU pairs separately.

Differences in the average 50% titers between maternal and infant clones were compared by using a Wilcoxon signed-rank test (STATA 8.2; STATA, College Station, TX). In addition, for a regression-based analysis (SAS 9.2), the distributions of 50% inhibitory titers were examined to determine the variability between pairs and the overall sampling distribution. Ultimately, values were categorized by using the median number of micrograms added (monoclonal antibodies) or a 1/dilution (polyclonal antibodies) for 50% inhibition of infectivity as cutoff points, since this allowed us to examine the relationship between mother and infant clones for the broadest panel of antibodies (59). Using these binary variables, we modeled the probability of increased resistance to neutralization using GEE-based logistic regression with exchangeable correlation matrices to estimate odds ratios and 95% confidence intervals comparing resistance among mother versus infant clones, both overall and by the timing of the HIV transmission. Thus, odds ratios that contrast maternal and infant clones >1 are associated with increased resistance to neutralization for maternal clones, whereas odds ratios <1 are associated with decreased resistance to neutralization.

As a sensitivity analysis to confirm that sequences from our small sample of mothers were similar to those from other chronically infected individuals, we compared IU and IP maternal sequences to a large data set of sequences from subjects chronically infected with subtype C HIV-1 (>1,000 sequences; from 69 subjects with approximately 20 sequences per subject [L. Ping et al., unpublished data]). For each population, we sampled 10 sequences with replacements, calculated a mean for each population (maternal or other chronic) and variance, and then calculated a difference in means and variance between each population. This was repeated 10,000 times for each parameter (nucleotide length and PNG site number) to obtain a bootstrapped SE estimate around the difference in means used to calculate 95% confidence intervals. All estimates were calculated using SAS version 9.2.

Standard statistical tests were performed with Prism 4 for Macintosh (version 4.0a). Correlations between number of glycosylation sites and loop length were tested by using a two-tailed Pearson coefficient model. Differences in median titers of serum from nontransmitting and IU- and IP-transmitting participants to SF162, TV-1, MW965, and JRCSF pseudoviruses were assessed by using the Kruskal-Wallis test.

Nucleotide sequence accession numbers.

The sequences were submitted to GenBank under accession numbers JN108036 through JN108760.

RESULTS

Subject characteristics.

We used single-genome amplification to amplify the entire env gene (∼2,600 bp) and sequenced the V1-V5 region from HIV-1 viral genomes isolated from plasma for 19 MTCT pairs. Ten infants were HIV positive at birth and classified as being infected IU, and nine infants were HIV negative at birth but positive at 6 weeks and were classified as being infected IP. We generated a median of 15 env gene amplicons per subject (range, 4 to 24). The characteristics of these mother-infant pairs are shown in Table 1.

Table 1.

Participant characteristics

MHP study ID	Maternal viral load (log10)	Maternal CD4 count	Transmission	No. of maternal sequences	Maternal pairwise comparisona (SD)	No. of first positive infant sequences	First infant pairwise comparison (SD)	No. of infant sequences at wk 12	Pairwise comparison at wk 12 (SD)
1468	4.6	274	IU	13	5.7 (0.9)	17	0.0 (0)	14	0.2 (0.2)
1551	4.7	360	IU	13	7.8 (1.1)	12	0.4 (0.2)	16	2.5 (0.5)
1585	4.4	529	IU	9	0.70 (0.3)	14	0.1 (0.1)	ND	ND
1629	4.8	342	IU	17	5.4 (0.9)	19	0.2 (0.1)	ND	ND
1851	NDb	761	IU	14	3.2 (0.9)	16	1.2 (0.4)	ND	ND
2199	3.8	476	IU	4	1.8 (0.4)	15	0.1 (0.1)	12	0.1 (0.1)
2444	4.8	511	IU	18	4.8 (0.8)	11	0.3 (0.2)	16c	0.1 (0.1)
2570	ND	65	IU	15	5.0 (0.9)	12	1.1 (0.3)	ND	ND
2797	ND	399	IU	19	9.4 (1.6)	21	0.2 (0.1)	ND	ND
3321	ND	220	IU	9	6.2 (1.1)	11	0.3 (0.2)	ND	ND
312	5.3	180	IP	18	7.3 (1.4)	18	0.7 (0.3)	14	1.0 (0.3)
819	4.4	891	IP	5	3.3 (0.9)	16	1.1 (0.3)	23	2.3 (0.7)
874	4.9	228	IP	9	4.5 (0.007)	16	0.1 (0.1)	15	0.5 (0.2)
1100	5.4	127	IP	16	12 (1.6)	16	0 (0)	ND	ND
1846	5.7	44	IP	19	1.8 (0.4)	24	0 (0)	12	1.8 (0.4)
1945	5.5	157	IP	14	3.6 (0.7)	19	0.2 (0.1)	17	0 (0)
2038	4.4	156	IP	8	7.1 (1.3)	21	4.4 (0.9)	23	0.1 (0.1)
2684	4.7	122	IP	15	6.3 (1.0)	14	0.5 (0.2)	11	1.2 (0.3)
2909	5.3	1,092	IP	19	5.0 (0.9)	12	0 (0)	21	1.2 (0.3)

^aCalculated with MEGA 4.1 0.5 gamma distribution.

^bND, no data.

^cSix-week sample.

Complexity of transmitted virus in vertical transmission.

Using pairwise diversity comparisons, we quantified the genetic bottleneck between maternal and infant viral populations and found a significant reduction in diversity (P = 0.0001) (Table 1); representative phylogenetic trees are shown in Fig. 1. Using both the diversity calculations and the phylogenetic trees, we determined that viral populations in 80% of the infants infected IU (8 of 10) and 56% of the infants infected IP (5 of 9) were homogeneous (defined here as <0.4% diversity with negligible branching in trees) and represented infection seeded by a single variant (see Fig. 1A and B for examples). We define the “transmitted variant” as the variant(s) that passed from mother to child and replicated in the infant to a level detectable by our sequence sampling methods. Additional variants may have passed into the recipients but failed to establish an infection. It should also be noted that all women received single-dose nevirapine, which could have reduced the number of variants circulating in women during IP transmission, thus possibly masking an even greater difference in the proportion of multiple-variant transmissions. Based on this analysis we conclude that a single variant initiated the observed viral population in most of the IU and more than 50% of the IP transmission events.

Fig. 1.

Representative maximum-likelihood phylogenetic trees of sequences from the V1-V5 region of env. Pink circles denote maternal sequences, blue squares denote the first-positive infant sequences, and green diamonds denote infant sequences at 12 weeks for transmission pairs 1468 (A), 3321 (B), 8151 (C), 2038 (D), and 312 (E) (a star marks the putative second parent of recombined sequences). M-X (maternal), B-X (infant at birth), and 6wk-X (infant at 6 weeks) env genes were cloned, and the data for sensitivity to neutralizing antibody of the pseudotyped virus are found in Tables S1 and S2 in the supplemental material. Branches containing unrelated outgroups were deleted for space.

The remaining six infants (two IU and four IP) were infected with multiple maternal variants. In two IU-infected (infants 1851 and 2570) and one IP-infected (infant 2038) infants, variants segregated to two distinct branches in the phylogenetic trees, indicating the transmission of multiple variants (Fig. 1C and D). For three additional IP pairs (infants 312, 2684, and 819), there was significant branching and low bootstrap values for nodes including infant sequences, a finding consistent with the transmission of multiple variants. Visual inspection of the alignments using the Los Alamos Highlighter tool (http://www.hiv.lanl.gov) suggested recombination of the major variant with an undetected transmitted variant. Recombination was supported by the Datamonkey SBP tool in all three alignments, and breakpoints were identified by GARD in the sequences from infants 312 and 819 (27). The absence of detection of the putative second parent is likely due to limited sampling of the infant viral populations. This interpretation is supported by longitudinal sequences from infant 312, where the second parent of a recombined sequence at 6 weeks was detected at 12 weeks (Fig. 1E, the lineage marked with a star; see also Fig. S4 in the supplemental material). Although this could represent a second transmission event, given the diversity in the maternal sequences and the probability that variants of <20% abundance could be missed in our sampling, an increase in abundance over time in the infant seems more likely than the transmission of the matching parent sequence by chance. Thus, phylogenetic trees and recombination analysis provide evidence for multiple variants transmitted in 2 of 10 IU and 4 of 9 IP MTCT pairs.

Features of transmitted env sequence.

We next examined Env protein characteristics encoded in the env gene sequence. We stratified the mother-infant pairs according to the timing of the infant HIV-1 infection and analyzed the differences in average sequence length with a GEE model. In this analysis a single sequence was used for each transmitted virus. Infant env sequences were on average significantly shorter than matched maternal env sequences for IP transmission pairs (P = 0.008), with statistical support also obtained for a shorter V1-V2 domain (P = 0.012). Although the infant env sequences of the IU transmission pairs were often different in length from the maternal sequences, they were not consistently longer or shorter (P = 0.11). We also compared the most abundant transmitted infant env sequence to the matched maternal population using a resampling approach. IP infant env sequences tended to be shorter than maternal env sequences but fell outside the resampling distribution only for the C3-V4 region (Fig. 2A). IU infant sequences tended to be longer but were within the 95th percentile resampling range (Fig. 2A). For this data set we also noted that the mean length of the maternal env sequences of mothers who transmitted IP was longer than the mean from IU-transmitting mothers (mean, 1,013 versus 1,010 nucleotides, respectively). In order to exclude the possibility that the IP mothers included in the present study represented a biased sample, we compared the average env length of the mothers who transmitted IU and IP against a random selection of HIV-1 subtype C env genes. Neither the IU nor the IP mean maternal env sequence length was statistically different from either the randomly selected chronic subtype C samples or each other (P = 0.6, 0.4, and 0.7, respectively). We previously noted in this cohort that the mothers who transmitted IP were more likely to have lower CD4 cell counts compared to the mothers who transmitted IU (30), but we found no correlation between CD4 count and maternal env sequence length (data not shown). Therefore, we conclude that the mean length of the env V1-V5 region of the transmitted variant is on average shorter than the mean of the maternal env population for IP but not for IU transmissions.

Fig. 2.

Infant sequence characteristics compared to resampling distribution of each maternal population. The bar represents the 95% confidence interval of the summed differences for each maternal population between 10,000 random resamples of each population and the actual mean for nucleotide length (A) and the number of putative N-linked glycosylation sites (B). Diamonds and squares indicate summed differences between infant and source maternal populations.

We next analyzed the number of encoded PNG sites in each sequence, as identified using the N-Glycosite program (61). Using the GEE model, we found the mean number of PNG sites in V1-V5 in IP infant env sequences was significantly less than matched maternal env sequences (P = 0.012), and the effect was most pronounced in V1-V2 (P = 0.006). No difference was seen in IU pairs. We also analyzed this difference using the resampling approach, and again IP-transmitted sequences encoded fewer glycosylation sites than the maternal sequences (Fig. 2B), with both the V1-V2 and the V4 variable regions contributing to this effect. In the IU-transmitted sequences there was variation in the encoded glycosylation sites between maternal and infant env genes, fewer in V1-V2 and greater in C3-V4 (Fig. 2B), but not a significant difference in number of total N-linked glycosylation sites. Thus, there appears to be selection for viruses with fewer glycosylation sites in infants infected by IP, but not IU, transmission.

Early env gene sequence evolution in the infant.

Twelve-week longitudinal samples were available for 8 of 9 IP-infected infants and for 3 of 10 IU-infected infants; an available 6-week sample for IU-infected infant 2444 was also analyzed. Sequences from these samples were compared to sequences from the first positive sample to examine early changes within each transmitted infant population.

There was statistical evidence of positive selection even with the sparse amount of sequence evolution that occurred over this time period in two of the IP-infected infants (infants 819 and 2909) and one IU-infected infant (infants 1551) (data not shown). Additional infants had a small number of fixed mutations that, while statistically insignificant, suggest selection, with the majority of fixed mutations found in C3-V4. Furthermore, in one IU-infected (infant 1551) and five IP-infected (infants 312, 874, 1846, 2684, and 2909) infant sequences there were clustered mutations or different nonsynonymous mutations at a single amino acid codon and/or glycosylation site (see the Highlighter plot in Fig. S3 in the supplemental material). These types of mutations are consistent with the types of patterns seen with CTL escape or for a replication advantage associated with the loss of a glycosylation site.

In three infants (infants 1551, 2038, and 2909) we observed changes in abundance of variants within a population from selective outgrowth, a possible second transmission event through breast-feeding, or both. Because a second transmission event is statistically unlikely, these shifts imply a changing selective advantage during early infection, whether in relative fitness or escape from developing immune pressure.

Sensitivity of transmitted and maternal Env proteins to neutralization by heterologous antibodies.

We tested full-length env clones from each maternal and infant sample to determine whether transmitted viruses had different sensitivities to neutralizing monoclonal antibodies and heterologous neutralizing antibodies compared to nontransmitted maternal variants. This was done for the 10 IU and 6 IP transmission pairs with stored maternal serum available. Pseudotypes were made with cloned env genes that represented the diversity of the maternal viral population (based on phylogenetic tree topology [Fig. 1 and see Fig. S2 and S3 in the supplemental material]; n = 50 clones in total) and infant/transmitted sequences, including multiple variants when more than one variant was transmitted (n = 20).

Consistent with previous studies of subtype C primary isolates (4, 32), the pseudotyped viruses displayed some sensitivity to 4E10 and soluble CD4 but were resistant to 2G12, 2F5, and IgG1b12. There were no significant differences in neutralizing antibody titers between transmitted and nontransmitted variants (analyses summarized in Table 2). Stratified by timing, however, variants transmitted IU tended to have increased resistance to 4E10 compared to the values for the average of the maternal Env proteins (Table 2), however driven by only a subset of cases (see Table S1 in the supplemental material).

Table 2.

Signed-rank z-scores comparing 50% neutralization sensitivities between paired maternal and infant clones^a

Serum	Signed-rank z-score
	All pairs	IU	IP
sCD4	−1.50	−1.17	−0.73
4E10	1.71	2.19	0.31
HIVIG	−0.49	−1.84	0.94
BB8	1.24	−0.15	2.11
BB28	1.03	2.09	−0.73
BB47	0.72	1.48	−0.31
BB68	−0.23	0.76	−1.94
Autol	0.97	0.89	0.52

^aA z-score of >|1.96| indicates two-tailed significance at P = 0.05. A positive z-score indicates that the infant sequences were more resistant to neutralization.

The sensitivity to heterologous subtype B and C sera was also tested. In most cases, the transmitted/infant env clone was similarly sensitive to the range of values obtained with viruses generated with the maternal env clones, and there was no pattern of higher or lower sensitivity when considering the absolute titer (Table 2).

Sensitivity of transmitted and maternal Env proteins to autologous antibodies.

We next compared the sensitivity to neutralization by concurrent maternal antibodies. Due to limited available serum volume not all clones could be tested. As shown in Tables S1 and S2 in the supplemental material, all of the transmitted pseudotyped viruses demonstrated sensitivity to maternal serum, and at least one maternal clone from each woman was sensitive to neutralization. However, there was no consistent pattern of difference within pairs overall or when stratified by transmission status (Table 2, Autol). Thus, in this data set we found no evidence for selection of a neutralization sensitive or resistance variant during vertical transmission.

Neither the number of PNG sites nor the loop length was significantly associated with neutralization sensitivity. Longer loop length was associated with greater resistance to neutralization by the monoclonal antibody 4E10 over the whole V1-V5 region (P = 0.006 [data not shown]), though with only R² = 0.11. gp41 was not sequenced; thus, we could not examine potential linked changes in the 4E10 epitope, although other regions or structures can impact epitope recognition (13).

Maternal heterologous neutralizing antibody titers.

We also tested maternal sera available from additional study participants who transmitted IU (n = 23) and IP (n = 23), along with a subset of nontransmitting women (n = 50), for neutralizing antibody activity against viruses pseudotyped with heterologous subtype C and B Env proteins. Median neutralizing antibody titers were similar in sera from the three groups of women to all Envs (Fig. 3). We compared NT, IU, and IP women by the number of participants with an IC50 of 40 or greater to multiple pseudotyped viruses using a chi-squared analysis to assess breadth, and there were no differences between the groups. Thus, we did not find a difference in neutralization activity or breadth of response to heterologous Env proteins between transmitting and nontransmitting maternal sera.

Fig. 3.

Heterologous neutralizing antibody titers of maternal sera by transmission status: NT, nontransmitting; IU, in utero transmission; IP, intrapartum transmission. The 50% neutralization titer was calculated against virus pseudotyped with env genes: SF162 (subtype B) (A), TV-1 (subtype C) (B), MW965 (subtype C) (C), and JRCSF (subtype B) (D). The lines represent the mean 50% inhibitory concentration (IC₅₀).

DISCUSSION

The characteristics and mechanisms of MTCT of HIV-1 are conflicting in the current literature, with no studies to date obtaining a full analysis of the genetic sequence, susceptibility of transmitted and nontransmitted virus to antibody neutralization, and breadth and potency of maternal serum all within the same cohort. We analyzed these properties of HIV-1 subtype C transmission using 10 pairs with IU MTCT and nine pairs with IP MTCT.

In our previous study using samples from this cohort, we examined the diversity of env V1-V2 using the heteroduplex tracking assay in 25 IU and 23 IP mother-infant pairs, as well as 32 nontransmitting women. We concluded that on average fewer variants are transmitted to infants infected IU than IP (30). In the three infants (infants 819, 2038, and 2684) previously reported with two variants at an abundance detectable by our level of SGA sampling (∼20%), the current analysis showed a heterogeneous infant population.

A severe genetic bottleneck during transmission was confirmed in all pairs and, in a majority of transmission cases, the infants harbored a single viral variant. We also found evidence for the transmission of multiple viral variants in 2 of 10 IU-infected infants and 4 of 9 IP-infected infants, 6 of 19 overall. Previous studies of MTCT, using similar methods, found a heterogeneous infant viral population in 7 of 13 infants infected with subtype CRF01_AE (4 of 6 infected IU and 3 of 7 infected IP) and in 3 of 17 infants infected with subtype A (unknown transmission timing) (47, 57). Thus, our data for subtype C HIV-1 transmission are in line with previous studies of other subtypes in the frequency of multiple-variant transmission in MTCT, and overall with the conclusion that in a majority of cases the infant is infected with a single variant.

HIV-1 subtype C transmitted IP had shorter variable loops and fewer PNG sites than the maternal populations, but no overall difference was seen in viruses transmitted IU. Previous work by Samleerat et al. found no differences in these characteristics in 11 IP pairs or 6 IU pairs infected with subtype CRF01_AE (47). However, Wu et al. and Zhang et al. reported fewer PNG sites in mostly IP-transmitted subtype A and C viruses, respectively, and an association between number of PNG sites and variable loop length (59, 60). Our detection of viruses with fewer glycosylation sites in infants infected IP is also consistent with the detection of underglycosylated viruses in heterosexual transmission (9, 10, 32, 34) (Ping et al., unpublished). Fully glycosylated virus could be selected by some mechanism within the placental environment, may not replicate as rapidly in the new host (60), or may not be transmitted as efficiently as underglycosylated virus via a mucosal route. This also suggests IP may be more similar in mechanism than IU to heterosexual transmission.

We previously found IP, but not IU, transmission was associated with the disruption of the placental barrier (29). A priori, mixing of maternal and infant blood suggests random exposure to maternal blood variants. Although the number of variants transmitted IP does trend higher (30) (Table 1), if extensive mixing of maternal and infant blood was common the transmission rates would likely be higher.

We saw evidence for early sequence evolution in the infant. Replication after sexual transmission revealed the stochastic incorporation of neutral mutations within the viral population (24). Beyond these stochastic sequence changes we were able to detect mutations becoming fixed and also examples of clustered mutations. This suggests ongoing selective pressure possibly to repair fortuitously fixed deleterious mutations by reversion or with compensatory mutations or to select for certain replication properties in the new environment. We also observed evidence for recombination, which requires dually infected cells and is plausible given the high viral loads in infected infants (51). These features of early sequence evolution are consistent with the known features of HIV replication after transmission.

Neutralizing antibodies are of intense interest for their potential role in blocking transmission. It has been reported that in heterosexual transmission the transmitted variant is more neutralization sensitive to autologous antibodies than other donor viral Env proteins (10). Autologous antibodies passed during pregnancy and in breast milk could have two effects: neutralizing antibodies in the mother could potentially block viruses from passing to the infant, or the presence of neutralizing antibodies in the infant could select for viruses that were especially resistant to neutralization from among the viral population in the mother.

In comparing the breadth and potency of serum for heterologous neutralizing antibodies from transmitting mothers to nontransmitting mothers, we found no difference in titers to any of the four Env proteins tested, nor did we see evidence for the serum of nontransmitting mothers to have greater breadth of activity to these Env proteins than the transmitting mothers. Although this is in contrast to previous results in subtype B-infected women (12), a recent study has yielded results similar to ours in that the neutralization breadth of maternal antibodies measured in infant samples did not correlate with protection from infection (36). Higher breadth of maternal neutralizing antibodies, at least for women infected with non-B subtypes, has thus shown no correlation with protection from MTCT.

Autologous neutralizing antibodies would seem to have the greatest potential to impact transmission, given that they are against the very viral population the infant could be exposed to. In a chronically infected individual, virus is generally resistant to the contemporaneous autologous antibodies, i.e., the host is always catching up to the rapidly escaping virus (45). This may not be true for all replicating virus, however, since studies report autologous neutralizing activity in at least 50% of chronically infected women with multiple subtypes (59) and up to 100% of women infected with subtype C (60) (see Tables S1 and S2 in the supplemental material).

One could hypothesize that the transmitted/founder viruses will be escape variants, resistant to any autologous neutralizing activity. It has been reported that a virus that grew out in the presence of maternal antibodies in cell culture was more similar to the transmitted infant virus than the bulk maternal population, interpreted to indicate that a neutralization escape variant was transmitted (12). In that study, many of the infant isolates that could not be neutralized by maternal serum came from mothers who lacked autologous neutralizing antibodies, likely providing a second test of the lack of maternal neutralizing antibodies rather than testing the relative sensitivity of the transmitted virus. In cases where both maternal and infant isolates could be neutralized by the maternal serum, the relative neutralization titers were not reported to be significantly different. Also, the potential for PCR-mediated recombination makes the reported genetic relationship uncertain (46). Other studies have also reported that the transmitted variant is neutralization resistant relative to the mother's viral population (59, 60). Although these studies took approaches similar to that used in the present study, there are significant differences. Even though in most cases the complexity of the transmitted virus was consistent with a single variant being transmitted, several clones were generated and analyzed from each infant sample; however, since they were all derived from a single transmitted virus, they cannot be considered independent observations for the purpose of analysis. In addition, viral sequences from the mother were generated from peripheral blood mononuclear cell (PBMC) DNA. Discordance in drug resistance markers and env sequences (52, 53) demonstrates the overall turnover of viral sequences in PBMC DNA to be slower than in the replicating viral RNA population. Given the sequential nature of neutralization escape that has been seen in the blood (45), one might observe a difference in neutralization sensitivity to concurrent antibodies in viral sequences derived from PBMC simply because of the difference in the rates of turnover in these different compartments. In contrast, when we compared the neutralization sensitivity of viral sequences present as RNA in maternal blood plasma to the transmitted infant sequence, we found no statistically significant difference in neutralization sensitivity to the autologous maternal antibodies (Table 2). Thus, the transmitted virus is characterized as having neutralization properties similar to those of the replicating maternal viral sequences.

In summary, we found that in this cohort, a single variant of HIV-1 subtype C was detected most often in infants after MTCT, although examples of multiple-variant transmission were also detected. We also found that on average viruses with shorter sequence lengths in env and fewer encoded glycosylation sites in V1-V5 compared to the matched maternal sequences were transmitted to infants infected IP, but not IU. Neutralization sensitivity to autologous maternal serum did not differ in a statistically significant way when comparing maternal and infant viruses, nor did sensitivity to a broad panel of monoclonal antibodies or heterologous sera. In conclusion, we did not find evidence that the transmitted virus in vertical transmission is selectively resistant to maternal antibodies compared to the contemporaneous maternal viral population, nor is the transmitted variant distinctly sensitive or resistant to heterologous antibodies.