{"paper_id":"b1723f23-0f43-4d4e-8a0b-ca4ac53b944a","body_text":"Long-Read Deep Sequencing Reveals High Rates of Multilineage \nTransmission and Rapid Viral Population Changes in Acute HIV \nInfection \nJames I. Mullins1* \nWenjie Deng1^ \nElena E. Giorgi2 \nCraig A. Magaret2 \nMorgane Rolland3,4 \nTanmoy Bhattacharya5,6 \nDylan H. Westfall1^ \nAnna E.J. Yssel8^ \nRoger E. Bumgarner1 \nBen Murrell7 \nThumbi Ndung’u9,15-17 \nMerlin L. Robb3 \nRaabya Rossenkhan2 \nPaul T. Edlefsen2  \nKrista L. Dong16 \nLennie Chen1^ \nAsanda Gwashu-Nyangiri8 \nHong Zhao1^ \nRuwayhida Thebus8 \nFredrick Sawe10,11 \nSorachai Nitayaphan12 \nTalita York8 \nDavid Matten8 \nHugh Murrell8 \nAlec P. Pankow1^ \nMichal Juraska2 \nJames Ludwig2 \nJohn Hural2 \nMyron S. Cohen13 \nLawrence Corey2 \nM. Juliana McElrath2 \nPeter B. Gilbert2 \nCarolyn Williamson8,14 \n \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 2 \nAffiliations: \n1 Department of Microbiology, University of Washington, Seattle, WA, US \n2 Vaccine and Infectious Disease Division, Fred Hutchinson Cancer Center, Seattle, WA, US \n3 Henry M. Jackson Foundation for the Advancement of Military Medicine, Inc., Bethesda, MD, \nUS \n4 U.S. Military HIV Research Program, Walter Reed Army Institute of Research, Silver Spring, \nMD, US \n5 Theoretical Division, Los Alamos National Laboratory, Los Alamos, NM, US \n6 Santa Fe Institute, Santa Fe, NM, US  \n7 Department of Microbiology, Tumor and Cell Biology, Karolinska Institutet, Stockholm, SE \n8 Institute of Infectious Diseases and Molecular Medicine, Division of Medical Virology, \nUniversity of Cape Town, Cape Town, ZA \n9 African Health Research Institute, University of KwaZulu Natal, Durban, ZA \n10 Kenya Medical Research Institute, Walter Reed Army Institute of Research-Africa, Nairobi, KE \n11 Henry M. Jackson Foundation Medical Research International, Nairobi, KE \n12 Armed Forces Research Institute of Medical Sciences, Bangkok, TH \n13 University of North Carolina, Chapel Hill, NC, US \n14 National Health Laboratory Services, ZA \n15 HIV Pathogenesis Programme, The Doris Duke Medical Research Institute, University of \nKwaZulu-Natal, Durban, ZA \n16 Ragon Institute of Mass General Brigham, Massachusetts Institute of Technology and Harvard \nUniversity, Cambridge, MA, US \n17 Division of Infection and Immunity, University College London, London, UK \n \n^ Present addresses:  \n \nWD, DHW, LC: Vaccine and Infectious Disease Division, Fred Hutchinson Cancer Center, Seattle, \nWA, USA \nAEJY: Sand Technologies, Cape Town, ZA \nAPP: Department of Microbiology, Icahn School of Medicine at Mount Sinai, New York, NY, US \nHZ: Department of Obstetrics and Gynecology, University of Washington, Seattle, WA, US \n \n* Corresponding author:  \nDepartment of Microbiology, University of Washington School of Medicine, Seattle, WA, US. \nemail: jmullins@uw.edu  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 3 \nAbstract \nUnderstanding the selective forces acting up on HIV early in infection  is crucial  to design \nprevention strategies. By leveraging deep sequencing and the short diagnostic intervals of the \nFRESH and RV217 cohorts (median 4 days) between the last-negative and first-positive RNA tests, \nwe captured a  precise and early snapshot of acute HIV infection. The frequency of multiple \ntransmitted viruses of 38% in these as well as placebo recipients from the AMP trials was higher \nthan previously published, with the true frequency likely to be higher. The relative abundance of \nlineages fluctuated substantially  over time  in two-thirds of the multilineage infections, \ngenerating uncertainty in identifying the specific viruses that were transmitted and founding the \ninfection. Viral populations exhibited diversity and s election on the Gag and Env proteins at the \nearliest times examined, with sites inferred to be undergoing negative selection most evident . \nThese data may help explain vaccination failures and provide new targets for prevention.   \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 4 \n \nIntroduction \nRecent HIV prevention studies have shed new light on the essential components necessary for \nan efficacious vaccine 1-3. However, major challenges remain, including the high and increasing \ngenetic and antigenic diversity of HIV likely to be encountered at exposure. It is therefore crucial \nto understand characteristics of the acquired viruses  that expand exponentially during early \ninfection.  \nDespite a potentially large amount and diversity of HIV in the source (the donor), when a person \nis infected, a substantial genetic bottleneck  occurs, with only a single or very small subset  of \nviruses successfully establish ing ongoing infection 4-9. These variants may arise from random \n(stochastic) processes, higher fitness in the immunologically naïve host and initially encountered \ntarget cells, as well as  their ability to survive innate immune responses 10-13. A recent meta -\nanalysis found an overall probability of acquiring more than one viral lineage of 25%, and the \nprobability was higher for male-to-male transmission (30%) than for male-to-female transmission \n(21%)14.  \nTo gain a n in-depth understanding of early HIV populations  we targeted sequencing of 100 \nmolecules for each of two regions (totaling 5.5kb) of the HIV genome using a long-read, PacBio \nsingle-molecule-real-time (SMRT) platform with unique molecular identifiers (SMRT-UMI)15. \nSamples from four  prospective cohorts of adult males and female s were studied : RV21716, \nFemales Rising through Education, Support, and Health (FRESH)17,18, and individuals from  the \nplacebo arms in the two Antibody Mediated Protection (AMP) trials, from the Americas (HVTN \n704) and from southern Africa (HVTN 703) 1. The RV217 and FRESH cohort participants were \nsampled twice weekly for viral RNA in plasma whereas the AMP trial participants were sampled \nmonthly and then again within 1-2 weeks after the first detection of viral RNA. Viral populations \nwere then assessed for distinct lineages, diversification and lineage fluctuations, and selection \non the gag and env genes. This provided unprecedented insight into virus characteristics and \nlineage dynamics in early HIV infection , including  prior to the onset of  detectable adaptive \nimmune-driven selection. \nResults \nWe generated 104,157 si ngle-molecule, long -read viral sequences from  310 plasma samples \ntaken from 123 prospectively identified persons living with HIV (Table 1). Two regions of the viral \ngenome were sequenced: 2.5kb amplicons encompassing the gag gene and the first kb of the pol \ngene (GP fragments), and 3.0kb amplicons encompassing the rev, vpu, env and the first third of \nthe nef gene (REN fragments, also used for construction of Env pseudotype viruses in AMP trial \nstudies1) (Supplementary Fig. 1).  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 5 \nViral sequences were obtained from up to 5 time points (denoted by black dots along the viral \nload curves in Fig. 1) from the RV217 (N=29 participants) and FRESH (N=13) cohorts, over a period \nof 2-61 days from the estimated date of detectable HIV infection (EDDI). This date was taken to \nbe the center of bounds (COB) between the last negative and first viral RNA positive test dates , \nwith the COB  corresponding to approximately 7 days post -acquisition19. The same amplicons \nwere sequenced from 81 participants in  the placebo arms from the two AMP trial s1. These \nindividuals were screened for HIV monthly, and plasma virus from the first HIV diagnostic \ntimepoint was sequence d, plus 1-2 timepoints generally obtained 2-4 weeks later (Fig. 1). For \nthese cohorts, the EDDI were deri ved from HIV diagnostics and viral sequence data 2,20. Each \ncohort consisted of adults and included a total of 59 individuals assigned as female at birth with \nheterosexual transmission risk , and 64 individuals assigned as male at birth with primarily \nhomosexual transmission risk  (Table 1). All plasma samples were taken prior to  participants \ninitiating antiretroviral therapy, although 3 individuals from the HVTN 704 cohort reported taking \npre-exposure prophylaxis around the time of HIV acquisition (dotted lines in Fig 1F).  \nViral subtype analyses of the REN region revealed 45 acquisitions of subtype C, 38 with subtype \nB, 15 with CRF01AE, as well as smaller numbers of individuals acquiring subtypes A1, F, G, D \n(N=10) or inter-subtype recombinants (N=15) (Supplementary Table 1, Supplementary Fig. 2). In \n12 participants, subtype discordance was found between the two genomic regions sequenced, \n11 of which involved intersubtype recombinants. This was most evident in the RV217 cohort from \nKenya (in 5 of 13 participants), reflecting the long-standing presence of multiple subtypes in this \nregion of Africa. Similarly, cocirculation of subtype B and F1 viruses in Peru and Brazil was \nreflected in the large number (N=9 of 48) of B-F1 intersubtype recombinants.  \nMutational differences that were g ene- and subtype -specific were noted: gag genes were \nsignificantly less likely than env genes to have putative inactivating mutations in each of two time \nwindows (approximately 30 and 45 days EDDI, respectively; Supplementary Table s 2-4, \nSupplementary Fig. 3A-B; See Online Methods). Subtype C viruses had lower rates of intact genes \nin both Gag and Env than the subtype B and in Gag when compared to the combined group \n“other” (composed of other subtypes and recombinants) sequence datasets  (Supplementary \nTable 4A-B, Supplementary Fig. 3C-D). Surprisingly, despite a lower rate of intact genes, subtype \nC genes were not more likely to be hypermutated (Supplementary Fig. 3E-F; Supplementary \nDiscussion 1), which frequently results in the formation of stop codons. No correlation was found \nbetween viral loads and intact gene  proportions in either gene  (Supplementary Discussion 2). \nHowever, when sample datasets with no hypermutated sequences were excluded, there was a \nstrong trend for hypermutated sequences to decline with increasing viral loads: p=0.0006 and \n0.06 for gag at the first and second time window s, respectively, and p=0.005 and 0.02 for env.  \n(Supplementary Fig. 4). \nTwo approaches were taken t o discern whether an infection resulted from single o r multiple \nfounding lineages and to count lineages: Poisson Fitter21 (Supplementary Discussion 4) was used \nto examine sequences from the first available time point, and; an iterative method was devised \nthat included clustering of sequences from all time points into distinct phylogenetic  clades, \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 6 \nfollowed by  a manual review of pairwise distances to refine lineage designations  (See Online \nmethods and Supplementary Fig. 5). After removal of recombinants between lineages, pairwise \ndistance distributions typically clustered the sequences unambiguously  (Supplementary Fig. 6). \nThe results from the earliest  sample from all participants  was 100% concordant between the \nPoisson Fitter and phylogenetic/distance methods. We report results for the \nphylogenetic/distance method, given that it included sequences from all time points. \nThe frequency of multilineage acquisitions (MLA) was determined by considering both the GP \nand REN regions  and all time points sampled. These considerations as well as the deep \nsequencing performed resulted in a  higher than previously reported frequency of MLA (38% ) \ncompared to an authoritative prior meta analysis (25%)14. Results from GP and REN were typically \nconcordant. However, in 6 (13%) of the 47 individuals in which multiple lineages were identified, \nonly one region (4 in GP only and 2 in REN only) had evidence of an MLA , and in 3 of 47 (6%) a \nsecond lineage was only identified at the second time point. Delayed outgrowth of some lineages \n(noted in 4 cases in GP and 7 cases in REN), may contribute to this discordance and to the high \nMLA frequency. However, a lack of distinction between lineages in one region, or low viral loads \nand the consequent failure to achieve deep sequencing at the first time point was likely to be the \ncause of this discrepancy in some cases, as the median number of viral sequences recovered at \nthe first time point in these cases was somewhat lower (Supplementary Fig. 7).  \nIn addition to multiple lineages , 18 (24%) of the 76 individuals with otherwise single lineage \nacquisitions had evidence of lineages whose origin as transmitted or evolving in the new host \nwas uncertain due to a limited number of lineage -defining differences. These uncertain -origin-\nlineages (UOLs) were defined if detected at the first time point  and if they had 1-3 substitution \nmutations distinguishing them from a more common lineage (e.g., Supplementary Fig. 8). UOL \nsequences in these individuals represented a  median of 26.5% ( 95% CI  15-35%) of the total \nnumber of sequences from the first time point in otherwise single lineage acquisitions. In three \ncases, a UOL became the dominant lineage at later  time points. If indeed each  UOL actually \ncorresponded to a transmitted lineage, then 53% of the individuals we studied would have \nacquired multiple transmitted lineages (Table 1). \nThe representation of lineages  or recombinants over the first two months post COB fluctuated \nten-fold or more in 9 of 15  (60%) cases of multilineage acquisitions in the FRESH and RV217 \ncohorts (see FRESH participants 079, 267, 271 and 318 in Fig. 2, and RV217 participants 20337 , \n20502, 40061, 40265 and 40436 in Supplementary Fig. 9). Furthermore, initially minor variants \nor recombinants came to represent the major sequence population at a later time point in GP \nand/or REN in 6 of 15 (40%) cases (see FRESH participants 079, 271 and 318 in Fig 2, and RV217 \nparticipants 20337, 20502 and 40363 in and 318 in Fig. 2, and RV217 participants 20337, 20502, \n40061, 40265 and 40436 in Supplementary Fig. 9). Overall, these major lineage shifts were found \nin a total of 10 of 15 (67%) multilineage acquisitions in the FRESH and RV217 cohorts. \nAnalysis of pairwise maximum likelihood distances, as noted in previous studies12,22, showed that \nthe diversity of individual viral lineages often contracted relative to the first time point at either \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 7 \nthe second or third time point sampled, especially in the RV217 cohort (Supplementary Fig. 10), \nwhich on average was initially sampled earliest in infection  (Fig. 1). The diversity of the gag and \nenv lineages within each individual were evaluated by calculating Shannon entropies. Average \nentropy values were determined across both the gag and env genes and deduced viral proteins. \nSliding window displays of entropy variation comparing genes and proteins were similar \n(Supplementary Fig. 1 1) and for subsequent analyses only amino acid data will be discussed.  \nBecause averages can be substantially impacted by outliers within small datasets, we restricted \nthe following analyses to datasets comprised of at least 10 members in each lineage. For the \nsame reason , we employed a second measure , the proportion of lineages that had nonzero \nentropy at each position . The two types of entropy plots showed diversity in the same regions, \nbut with different relative amplitudes in some regions (Supplementary Fig. 11). Similar patterns \nwere observed when comparing entropies in viral Subtype C vs other “Not Subtype C” subtypes \n(Supplementary Figs. 12, 13); the number of samples of Non-C subtypes of sufficient lineage size \nwas judged to be too few to assess by individual subtype). \nIn addition to entropy we al so identified amino acid sites that were inferred to be undergoing \npervasive negative (purifying) or positive (diversifying) selection across the Gag and Env genes , \nagain pooling data from all cohorts and lineage s. Overall, 7 fold more sites were inferred to be \nundergoing purifying versus diversifying selection in Gag , a nd 1.5 fold more in Env \n(Supplementary Table 5). Next, we examined selection in the 6 individual coding regions within \nGag and after separating Env into 3 regions: gp120 and the extracellular and intracellular regions \nof gp41. In addition, we split the datasets into four time-based stages following COB/EDDI as \nnoted in Fig. 1 and Supplementary Table 6. The earliest ‘i’ stage was a period of exponential viral \nload increase. Stage ‘ii’ corresponded to a period of rapid decrease in viral load, ‘iii’ corresponded \nto a slower decrease in viral load and ‘iv’ the beginning of a period of approximately steady state \nviral load. Stage ‘iv’ included only AMP trial participants, had the least available data (Fig. 1), and \nthus will not be discussed further. Adaptive immune responses leading to positive selection and \nimmune escape are typically first detected in stage ‘ii’, with escape mutations first noted at one \nor very few sites within the viral proteome in stages ‘ii’ and “iii”23-25.  \nThe number of sites inferred to be undergoing negative selection increased through the 3 stages \nin nearly all Gag and Env coding regions. Positive selection was less consistent, with the highest \nlevels reached in the intracellular segment of gp41 within Env, p2 within Gag, and gp120 within \nEnv. The ratio of negative to positively sel ected sites was substantially higher in  the p24 than \nother coding regions  (Fig. 3A -B). Interestingly, this was not due to a higher level of negative \nselection, but rather an atypically low level of positive selection (Fig 3C). The greater negative to \npositive selection ratio in Gag vs. Env was associated with both lower levels of negatively selected \nsites and higher levels of positively selected sites (Supplementary Table 5). \nWe next assessed entropy and select ed sites along each coding region  and th rough post \nCOB/EDDI stages i-iii, combining data from all time points, participants and lineages. Entropy was \nevident within both Gag and Env  at the initial sampling times . Regions of higher entropy were \nfound near the N and C termini of p17 (matrix protein), in p2-p7(nucleocapsid)-p1 and the C-\nterminal portion of p6 in Gag (Fig 4A). These levels were substantially driven by changes in stage \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 8 \niii in p17, p7 and portions of p6 (Fig. 4C). Positive selection was most evident in stage iii in the N-\nterminal region of p17, p2, the N-terminal region of p7 and segments within p6, and in stage ii in \nregions of p17 and p7 (Fig. 4E). Interestingly, positive selection was focused on the N-terminal \nregion of p7, whereas negative selection peaked toward the C-terminal region of p7 (Fig. 4D). \nEntropy as well as selected sites were more distributed over Env, with several peaks of positive \nand negative selection. The positive selection noted in the cytoplasmic domain of gp41 (Fig. 3) \nwas associated with peaks in the central and C-terminal regions. \nIn a study of HIV subtype C infection of donor -recipient pairs, the envelope genes in recipients \nwere found to have shorter variable loop lengths and fewer potential N-linked glycosylation sites \n(PNGS)10 although this was less consistent for donor -recipient pairs infected with HIV Subtype \nB26,27. Here too, s ubtype C viruses  had variable regions that were shorter than other subtypes, \nand shorter than those found in the LANL database , the latter of which have sequences from \nviruses from acute as well as chronic infection and AIDS (Supplementary Table 7). Also consistent \nwith prior studies, subtype B viruses had the longest variable loops and were similar to or longer \nthan those found in the LANL database . A previous study of the RV217 cohort failed to \ndemonstrate a change in PNGS over the first 6 months of infection 28. Here too, t he median \nnumber of PNGS did not vary through stages i-iii (Supplementary Figure 14), although subtype C \nviruses had on average one fewer site vs Subtype B . We also tallied the sites within and nearby \nNLGS sequons that experienced positive and negative selection. Neither sequons nor the 9 amino \nacid regions centered on sequons had a higher level of selected sites compared to random \nexpectations, although sequons and immediately N-terminal and C -terminal amino acids  \nappeared to be enriched for negatively selected sites (Supplementary Fig. 15) and a significant \ndifference between the ratio of T vs S was noted when comparing positive and negative selection \nlevels in the 3rd position (p = 0.03941, OR = 3.33, 95% CI = (1.1, 11.4), 2-way exact Fisher test). \n \nDiscussion \nThis study offers an unparalleled view into the dynamics of HIV viral populations during early \ninfection and provides a glimpse into the processes that underly the establishment and early \nmaintenance of HIV infection. Our advanced methods detected a higher  frequency of \ntransmission of multiple distinct virus lineages compared to previous studies14. A major strength \nof our study derives from  sampling: larger numbers of sequences, i.e., sampling deeper  to \nidentify minor variants;  over larger stretches of the genome , which allows greater opportunity \nfor detecting distinguishing mutations, and ; over multiple time points , which increased the \nnumber of sequences evaluated and allowed for detection of potentially late appearing lineages. \nThis also allowed opportunities to view dynamic changes in variant detection and representation.  \nMassive virus population expansion occurs during the approximately 7 -day eclipse phase of \ninfection prior to detectability and continuing during initial acute infection19. Multiple forces may \nshape these emerging virus populations very early in infection, including: the stochastic process \nof variants entering cells with different division rates; inherent variant replication rate; purifying \nselection for more rapid growth, including possible loss of mutations selected for immune escape \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 9 \nin the prior host that also caused a loss in replication fitness , and; diversifying selection for the \nemergence of escape mutants from adaptive immunity11,12,29-31.  \nThe significant increase in the percentage of multilineage acquisitions from approximately 25% \n(95% CI of 21 -29%) in a prior meta-analysis14 to a minimum 38% (95% CI of 30-47%) reported \nhere, was largely due to the detection of minor variant lineages , and there was no evident bias  \nassociated with sex . However, 38% is very likely an underestimate due to multiple factors: 1) \nRapid and large changes in lineage representa tion. We found major shifts in lineage \nrepresentation over the first ~2 months of infection in two-thirds of the multilineage acquisitions \nin the RV217 and FRESH cohorts. These two cohorts were sampled at high temporal resolution \nand the intensive sampling provided a uniquely detailed view of viral dynamics, offering insights \nthat were previously inaccessible due to less frequent sampling . However, in one case (RV217 \nparticipant 40061 from Thailand ), a previous study using short -read length sequences from the \nsame individual found that a minor variant at day 14 was dominant between days 19 and 28 and \nnot detected at day 4232 (Supplementary Fig. 15C). In both the previous and current studies, this \nlineage represented about 25% of the sequences at day 14 . but in the current study no time \npoints were observed when this lineage dominated ( Supplementary Fig. 15A-B). This illustrates \nthat even with the closely spaced sampling we employed, major transient shifts in virus \npopulations may fail to be observed , including the failure to detect variants that at other times \nmay appear to be the founder of infection. 2) Shallow sampling relative to the in vivo population. \nAny feasible scale of virus population sampling is very shallow relative to the in vivo population, \nand we found that variants that only appeared subsequent to the first time point tended to have \nlower numbers of sequences recovered at the first time point  compared to the individuals in \nwhich multiple lineages  were detected at the first time point . We estimate that the average \nnumber of sequences we obtained using the SMRT-UMI method is likely to detect 99-100% of all \nvariants with a frequency of at least 5% whereas when using Sanger SGA to obtain 20 sequences, \nvariants present at 5% are missed 36% of the time . HIV sequences from the  RV217 and FRESH \nparticipants studied here were previously determined using Sanger SGA and in the cas e of the \nThailand cohort, 5/16 (3 1%) of in dividuals were found to have MLA 19, compared to 50% in the \ncurrent study. The same percentage of partici pants from the Kenyan cohort ( 2/13, 15%) were \nfound to have MLA by both Sanger  SGA19 and here using SMRT -UMI. In the case of the FRESH \ncohort, only one of the 13 individuals (8%) were found to have MLA by Sanger SGA, compared to \n5/13 (38%) here  (Ndung’u et al, unpublished). 3) The presence of lineages of uncertain orig in \n(UOL), as identified here in 27% of what we have characterized as single linea ge acquisitions, \ncould increase the tally of multilineage acquisitions to as high as 53% (95% CI 46 -62%) in our \ndatasets.  UOL may represent early evolution or transmission of multiple lineages from a newly \ninfected donor with a near homo geneous viral population. 4) A large fraction of  transmissions \noccur when the transmitter is in acute infection33,34 when the infecting population often has little \ndiversity. Hence, not all multilineage transmissions would be discernable by viral genetic analysis \nin these cases. 5) We studied 5.5kb of the 9kb viral genome, whereas inclusion of the remaining \n40% of the genome may have identified additional lineages. \nRecognizing that most individuals are sampled only once early in infection and precise staging \n(e.g., Fiebig35 or other staging21,36,37) is often not be possible, nua nce is appropriate in labelling \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 10 \nvirus lineages. In the current study  of RV217 and FRESH cohort individuals , for which  the \nthreshold of RNA positivity is known within a few days, the lineages observed were in the earliest \nrecognizable stages following HIV acquisition. We propose that the largely homogeneous variant \npopulations observed very early in infection should nonetheless be considered representatives \nof a transmitted lineage, rather than definitively characterized as the actual  transmitted \nvariant(s). As such, the common term “transmitted/founder” 25 or T/F lacks precision, in part \nbecause it combine s two different properties that are not necessarily linked :  a “transmitted” \nvariant does not necessarily indicate the “founder” of infection over the long term , and there is \nno clear definition implied of how long after infection a “transmitted” lineage is discernible . \nSimiarly,  “founder”38 can be misleading  since in some cases the major variant(s), i.e., the \npresumed founder, appears different depending on the precise timing of sampling . Another \nrationale for use of “lineage” comes from the observation that entropy and selection on the virus \npopulation was observed at the very earliest times of infection (i.e., stage ‘i’). As time of infection \nprogresses, detection of transmitted variants and even lineages becomes harder because of \nselection (both positive and negative) and recombination, and our inability to discern complex \nrecombination patterns.  From this conceptual advancement we propose the use of the term \n“transmitted founder lineages” (TFL) to describe virus populations detected early in infection. \nTwo observations were made that may be associated with viral subtype. First, subtype C virus  \npopulations were found to harbor significantly more defective gag and env genes than other \nsubtypes, but this was not d ue to higher rates of APOBEC-mediated hypermutation which \nfrequently leads  to the formation of stop codons. This suggests that the fidelity of reverse \ntranscriptase (RT) is lower in subtype C viruses, although very limited in vitro studies comparing \nthe fidelity of subtype B and C viruses have not shown differences39. Given the variability in the \nlevels of defective viruses in the infections we studied, a comprehensive study that examines \nmultiple representatives of each of the subtypes of interest is warranted. Second, atypically low \nlevels of MLA were found in the RV217 Kenyan female cohort. While the reason for this is unclear, \nit cannot be attributed simply to sex, as the Kenyan as well as  individuals from the FRESH and \nHVTN 703 cohorts were assigned as female at birth. Some differences in mode of transmission in \nthe Kenyan vs. the FRESH and HVTN 703 cohorts ( e.g., receptive vaginal vs. anal intercourse) \ncannot be ruled out. However, there was a difference in subtypes – nearly all acquisitions were \nwith subtype C in the FRESH and HVTN 703 cohorts whereas 11/13 individuals in the Kenyan \ncohort were subtype A1 or recombinants involving subtype A1.  \nGlycosylation can negatively affect the rate of protein folding40 and importantly, can impact viral \ninfectivity and provide a shield against antibody recognition 41-43. Thus, selection for the loss of \nPNGS may occur in the absence of neutralizing antibody responses in acute infection, as a more \naccessible and compact envelope protein  may increase viral replication fitness.  Consistent with \nthis hypothesis was the finding that negative selection was associated with a higher prevalence \nof S over T at the 3 rd position of sequons, preserving the sequon, and a higher prevalence of T \nover S in sequons that were lost as a result of positive selection at this site. Since glycosylation is \nup to 40 times more likely to occur when T is in the 3rd position44,45, both of the differences noted \nfavor unglycosylated sequons, and increased viral fitness in the absence of adaptive immunity. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 11 \nOverall, we found MLA in at least 37% (CI 26-50) of male-female transmissions and 39% (CI 28-\n51%) of male-to-male transmissions, both potentially higher and with less sex differences than \nreported in the previous meta-analysis (21%, 14-31 and 30%, 22-40%, respectively)14. \nThe potential implications of this study for HIV prevention are threefold. Given that we consider \nit probable that substantial numbers of variants are transmitted  in a large fraction of \ntransmissions, what we observe may only be the transient winners of the race to detectability. \nThis may help explain the difficulty in making effective HIV vaccines  beyond the problem of the \nincreasing the global diversity of HIV 46,47, since a diverse transmitting population has a greater \nlikelihood of harboring variants capable of escaping blockades the vaccines produce. This is \nconsistent with the findings from large scale HIV vaccine trials in which infection with viruses \nmore dissimilar to the vaccine are most likely to result in virus infection48-50. \nSecond, sites of strong selection, in particular purifying selection, have been a focus for vaccine \nantigen designs for the major viral proteins Gag, Pol and Env51-53. The strong conservation of p24, \nin particular  (and reinforc ed in this study) , has made this protein a target for research and \ndevelopment of a highly potent antiretroviral drug54,55 with particularly conserved regions of p24 \nused as vaccine candidates 56-61. Interventions that effectively target conserved regions of the \nviral proteome may enhance vaccine efficacy due to their strict structural requirements for viral \ninfectivity62. This study identified additional regions of Gag proteins to be undergoing negative \nselection, e.g., the central region of p17 and C-terminus of p7, and thus potential vaccine targets. \nThird, the intracellular region C -terminal to the transmembrane sequence, p2, the N-terminal \nregion of p7 and much of p6 wer e relatively strong target s of positive selection in the current \nstudy. As immunologically dominant epitopes can act as decoys that prevent immune recognition \nof vulnerable features of the viral proteome 63-66, vaccines that target subdominant, conserved \nfeatures of the viral proteome 52,53,65,67-69 critical to viral fitness 62,70,71 may benefit from the \nomission of these coding regions. \nOnline Methods  \nStudy Subjects and specimens  \nIndividuals included in this study were derived from 4 prospective cohorts of HIV acquisition \n(Table 1). The RV217 study included a cohort of males and transgender females (TGF) from \nThailand and a cohort of females from Kenya16. The Females Rising through Education, Support, \nand Health (FRESH)17 cohort included females from KwaZulu-Natal, South Africa. Individuals were \nprospectively identified in acute HIV infection by twice-a-week plasma RNA testing with the date \nof detectable acquisition taken to be the center of bounds (COB) between the last negative and \nfirst positive HIV RNA test 20. A subset of individuals from these cohorts were chosen for study \nbased on plasma specimen availability and samples were chosen to match the predicted \nestimated date of detectable infection (EDDI) in individuals in the AMP trials. Plasma samples \nused for viral genome sequencing  were taken from multiple  time points  prior to initiation of \nantiretroviral therapy (ART).  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 12 \nTwo cohorts corresponded to the control arms receiving placebo only in the Antibody Mediated \nPrevention (AMP) clinical trials HVTN 703/HPTN081 (females from southern Africa, abbreviated \nHVTN 703 ) and HVTN 704 /HPTN085 (MSM and transgender females from the Americas and \nSwitzerland, abbreviated HVTN 704)1. AMP trial participants were monitored monthly and often \nretested 1-2 weeks following an initial viral RNA positive finding. At least the first 2 RNA positive \nvisits were chosen for sequencing. Estimates of their EDDI37 were derived using both clinical \ndiagnostic and viral sequencing data corresponding to a preliminary dataset of the sequences \nreported hered20 (see below).  \nSMRT-UMI sequencing \nThe Pacific Biosciences single molecule real -time (SMRT) platform was used to sequence 2.5kb \nPCR amplicons encompassing the HIV gag and part of the pol gene (GP region), and 3kb amplicons \nfrom rev through env and a portion of the nef gene (REN region) ( S1 Fig), with samples split \nbetween the University of Washington and University of Cape Town laboratories. Each sequence \nwas derived from individual cDNA templates amplified by single genome amplification (SGA) and \ntagged with unique index adaptors for sequen cing or, in most cases, tagged with unique \nmolecular identifiers (UMI)  during cDNA synthesis and amplified in bulk ( SMRT-UMI)15. Both \nprotocols resulted in accurate single-molecule sequencing and viral RNA template quantitation15. \nSequencing was performed  using the Pacific BioSciences Sequel and Sequel IIe instruments \nfollowed by demultiplexing and processing into consensus sequences for each viral template  \nusing the PORPIDpipeline 15 for amplicons amplified using the SMRT-UMI approach or with a \nsimplified pipeline ( https://github.com/MullinsLab/sga_index_consensus) to demultiplex and \ngenerate a consensus from SGA. For samples with viral loads above 20,000 HIV RNA copies/ml of \nplasma, the concentration of amplifiable target amplicons were initially estimated using end -\npoint dilution (EPD) nested PCR and the Quality tool  (https://quality.fredhutch.org)72 with 3 \nreplicates each at 5-6 dilutions estimated to reach an endpoint based on clinical viral loads.  \nWestfall et a l15 estimated PCR and sequencing error rates using this method paired with the \nPORPIDpipeline software to be less than 8.6 x 10-8 per base, or less than one error in every ~3700 \nREN sequences. To minimize recombination during PCR, no more than 25 amplifiable copies of \ncDNA was added to each reaction.  PCR recombination  was observed if the number of cDNA \nmolecules in a PCR was above 100. \nGiven an inherently large standard error in the quantitation measurements and with a target of \nobtaining at least 100 sequences per sample, 200, and in later experiments 250 templates, were \ntargeted for amplification. Amplifiable templates in samples with viral load s between 1,000 & \n20,000 copies/ml were not quantified by EPD PCR but rather the cDNA (derived from up to 1mL \nof plasma) was divided into 20 nested PCR reactions and only those PCR that were positive \nfollowing a gel screen were processed using the SMRT-UMI protocol. SGA was performed for any \nsample with a viral load below 1,000 copies/ml of plasma. \nThe SMRT-UMI laboratory protocol resulted in carryover of a small amount of UMI -containing \ncDNA primers. This had the effect of  a small number of  cDNA primers being used as primers in \nsubsequent PCR steps, thus artificially inflating the number of templates recovered. Sequences \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 13 \nderived from these carryover cDNA primers typically had very small family sizes (a “family” \ncorresponds to a  group of sequences, each containing the same UMI). Following a series of \nexperiments to quantify cDNA carryover we conservatively “cleaned” the datasets by removal of \nsequences with small family sizes belo w a threshold corresponding to 15% of the total number \nof reads (or 20% for the HVTN 703 REN datasets). This approach removed some of the “real” \nsequences but removed nearly all of the sequences derived from cDNA carryover.  Sequences \nwere checked for contamination against a laboratory database before nucleotide and amino acid \nalignment using BLAST73. Final sequences were deposited in GenBank after collapsing identical \nsequences, with Accession numbers as follows: FRESH GP ( PV963961 - PV968620), FRESH REN \n(PV968621 – PV972048), RV217 GP (PX001706 – PX010774),  RV217 REN (PX010775 - PX020890), \nHVTN 703  GP ( PX030312 – PX031976, PX038076 – PX042058), HVTN 703 REN ( PX072574 – \nPX077815), HVTN 704 GP (PX154055 – PX161466), and HVTN 704 REN  (PX172337- PX183870). \nViral subtypes were determined using the Recombinant Identification Program (RIP, 74, \nhttps://www.hiv.lanl.gov/content/sequence/RIP/RIP.html). When subtype assignments were \nunclear, the R EGA HIV subtyping tool ( https://www.genomedetective.com/app/typingtool/hiv) \nand the National Center for Biotechnology Information (NCBI) genotyping tool \n(https://www.ncbi.nlm.nih.gov/projects/genotyping/formpage.cgi) were also employed.  \nSequence alignments \nNucleotide alignments were initially generated using the MUSCLE algorithm75 version 3.8.31. To \nreduce the number of taxa in alignments, identical sequences were collapsed using a python \nscript https://github.com/MullinsLab/sequence_collapsing. Three rounds of manual review and \nrefinement were then conducted by three different experienced scientists. In the first round of \nreview, codon position was used to assist placement of gaps, but otherwise alignments were not \ncodon-optimized but rather optimized for sequence homology. Subsequently, a second scientist \nreviewed the refined alignments and any changes noted and discussed with the primary \nreviewer. Finally, a third scientist assessed the alignments and any further edits noted, with each \nalignment finalized in consultation with the primary and secondary reviewers. \nLineage assignments \nLineages were assigned to each sequence within an individual using Poisson Fitter21 with \nsequences from the first available time point , and an iterative diversity/phylogenetic approach \nusing sequences pooled from all time points . In the iterative appro ach, maximum likelihood \nphylogenetic trees and highlighter and match plots were generated for GP and REN regions using \nan in-house pipeline (https://github.com/MullinsLab/phylobook_pipeline) and displayed within \nthe Phylobook tool76. Within Phylobook, clustering algorithms were used to help assign lineages \nalong with manual review and selection. After several iterations and reviews, variants with a \ncluster of 4 or more shared nucleotide positions were found to categorize sequences into distinct \nlineages, judged to have been derived by the transmission and outgrowth of distinct variants. \nVariants with scattered unique changes were not characterized as being transmitted. The origin \nof lineages detected at the first time point and differing by 1 -3 nucleotides were judged to be \nuncertain (i.e., uncertain origin lineages, UOL). For the comparison of distance distributions, UOL \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 14 \nsequences were included within the most closely related lineage(s). Not counted in the \nidentification of UOL were positions at which more than 2 prevalent nucleotide states were \nfound, as this is likely indicative of diversifying selection.  \nRecombinant sequences were identified by manual inspection as having at least 2 nucleotide \nchanges matching another lineage. As the structures of many recombinants were quite complex, \nno effort was made to separate them into distinct lineages. To assess lineage assignments, \nhypermutated sequences were removed using an R -script based on the LANL Hypermut tool77, \nand then maximum likelihood pairwise distance distributions generated within and between all \nlineages using PhyML v3.3 78 within the DIVEIN suite (https://divein.fredhutch.org)79. Distance \ndistributions from within- and between-lineages were compared and any values that overlapped \nthe two distributions re-evaluated in Phylobook for proper lineage assignment. Following initial \nlineage assignment, a second experienced reviewer assessed assignments and any discrepancies \ndiscussed and assignments amended if needed. PhyML within DIVEIN was also used for \nphylogenetic tree generation using the GTR substitution model, an estimated gamma distribution \nand optimization for both topology and branch lengths. \nStatistical Analyses \nDifferences between genes and time points in the frequency of intact genes were assessed using \nboth a parametric method, which allowed us to combine all variable into a single model, and a \nnon-parametric method, by running a Wilcoxo n test comparing each variable within each \nsubgroup (gene/time point/subtype) and then applying an FDR threshold of 0.2 to correct for \nmultiple testing ( Supplemental Table 4). For the parametric method,  we ran a random effects \ngeneralized linear model (GLM) with the logit transformed frequency as dependent variable , \ngene (either gag or env) and subtype (categorized as C, B or “other” for all other subtypes  and \nincluding all intersubtype recombinants ) as independent variable s, and participant ID  as a \nrandom effect.  To make days since COB comparable across cohorts , we chose times since \nCOB/EDDI to be as uniform as possible across participants, with each study participant \nrepresented only once for each model run. As such, we identified  two time points (“windows”) \nwith mean times since COB/EDDI of 30 and 45 days, respectively, selected as follows: 1) Sequence \ndatasets from approximately 30 days post COB/EDDI, including all first time point samples from \nV703 and V704,  the second time point for FRESH, and the third time point for RV217, as for both \nthe FRESH and RV217 the first time point was on average sampled around two weeks earlier than \nthe first time point in the AMP cohorts  (Supplementary Table 3). 2) Sequence datasets from \napproximately 45 days post COB/EDDI, including  the second time point for V703 and V704, t he \nthird for FRESH and the fourth for RV217. Associations between viral loads and fraction of intact \nsequences and hypermutated sequences were assessed using the Kendall correlation coefficient \nand test. These analyses were computed in R (version 4.2.1) using packages lme4, \nLaplacesDemon, and pbkrtest. Confidence intervals were calculated using the binomial (Wilson \nscore interval) method . The power to detect minor variants by frequency and number of \nsampled sequences was calculated using the binomial function (pbinom) in R Studio. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 15 \nMeasurement of entropy and selection  \nThe entropy found in each lineage, with all timepoints combined as well as during each of the i-\niv stages, were calculated using the R script https://github.com/MiguelMSandin/DNA-alignment-\nentropy modified to perform calculations in batch . Selection was similarly calculated using the \nFUBAR algorithm  modified to  perform calculations in batch  \n(https://github.com/MullinsLab/FUBAR_in_batch). For both entropy and selection analyses, data \nwas then summed at each nucleotide or amino acid positio n, with sequences placed in register \nfrom an inter-participant alignment. Python- and R-scripts were used for d ata processing and \nrendered in Prism (GraphPad Software, Inc). \nAcknowledgements \nWe gratefully acknowledge Kim Wong for technical assistance and the dedicated participation of \nthe many individuals in the FRESH, RV217 and AMP trials cohorts, and the UKZN HIV Pathogenesis \nProgramme laboratory staff without whom none of this work would be possible. This work was \nsupported by the Bill and Melinda Gates Foundation (Investment Record ID INV -016189 to JIM; \nWe thank Dr. Thandi Onami for critical guidance ) and by Public Health Service Grants (UM1 \nAI068614, to the HIV Vaccine Trials Network [HVTN]; UM1 AI068635, to the HVTN Statistical Data \nManagement Center, and ; UM1 AI068618, to the HVTN Laboratory Center) from the National \nInstitute of Allergy and Infectious Diseases (NIAID) of the National Institutes of Health (NIH). The \ncontent is solely the responsibility of the authors and does not necessarily represent the o fficial \nviews of the NIH. The FRESH program  was supported in part by grants from the Bill & Melinda \nGates Foundation, the International AIDS Vaccine Initiative (UKZNRSA1001), the Harvard \nUniversity CFAR grant (P30 AI060354), the Witten Family Foundation, Dan and Marjorie Sullivan, \nthe Mark and Lisa Schwartz Foundation, Ursula Brunner, AIDS Healthcare Foundation, and the \nHoward Hughes Medical Institute. TN was further supported through the Sub -Saharan African \nNetwork for TB/HIV Research Excellence (SA NTHE) which is funded by the Science for Africa \nFoundation to the Developing Excellence in Leadership, Training and Science in Africa (DELTAS \nAfrica) program [Del -22-007] with support from Wellcome Trust and the UK Foreign, \nCommonwealth & Development Offic e and is part of the EDCPT2 program  supported by the \nEuropean Union; the Bill & Melinda Gates Foundation [INV -033558]; and Gilead Sciences Inc. \n[19275]. MR, MLR, FS, and SN received support from a cooperative agreement (W81XWH -11-2-\n0174) between the Henry M. Jackson Foundation for the Advancement of Military Medicine and \nthe US Department of Defense.  \n \nEthical statement \nThe work described here complied with all relevant ethical regulations. The Institutional Review \nBoards/Ethic Committees of participating clinical research sites (CRS) approved the studies, \nwhich were conducted under the oversight of the NIAID Data Safety Monitoring Board 1. Viral \ngenome sequencing at the University of Cape Town was approved by the UCT Human Research \nEthics Committee (HREC reference no. 176/2017) and was considered exempt at the University \nof Washington. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 16 \n \nReferences \n1. Corey, L. , et al.  Two Randomized Trials of Neutralizing Antibodies to Prevent HIV -1 \nAcquisition. N Engl J Med 384, 1003–1014 (2021). \n2. Gilbert, P.B., et al.  Neutralization titer biomarker for antibody -mediated prevention of \nHIV-1 acquisition. Nat Med 28, 1924–1932 (2022). \n3. Seaton, K.E. , et al.  Pharmacokinetic serum concentrations of VRC01 correlate with \nprevention of HIV-1 acquisition. EBioMedicine 93, 104590 (2023). \n4. Zhu, T., et al. Genotypic and phenotypic characterization of HIV -1 patients with primary \ninfection. Science 261, 1179–1181 (1993). \n5. Delwart, E.L., Sheppard, H.W., Walker, B.D., Goudsmit, J. & Mullins, J.I. Human \nimmunodeficiency virus type 1 evolution in vivo tracked by DNA heteroduplex mobility \nassays. Journal of virology 68, 6672–6683 (1994). \n6. Delwart, E., et al. Homogeneous quasispecies in 16 out of 17 individuals during very early \nHIV-1 primary infection. AIDS (London, England) 16, 189–195 (2002). \n7. Gottlieb, G.S., et al. HIV-1 variation before seroconversion in men who have sex with men: \nanalysis of acute/early HIV infection in the multicenter AIDS cohort study. J Infect Dis 197, \n1011–1015 (2008). \n8. Keele, B.F., et al. Identification and characterization of transmitted and early founder virus \nenvelopes in primary HIV-1 infection. Proc Natl Acad Sci U S A 105, 7552–7557 (2008). \n9. Abrahams, M.R. , et al.  Quantitating the multiplicity of infection with human \nimmunodeficiency virus type 1 subtype C reveals a non -poisson distribution of \ntransmitted variants. Journal of virology 83, 3556–3567 (2009). \n10. Derdeyn, C.A. , et al.  Envelope-constrained neutralization -sensitive HIV -1 after \nheterosexual transmission. Science 303, 2019–2022 (2004). \n11. Herbeck, J.T., et al. Human immunodeficiency virus type 1 env evolves toward ancestral \nstates upon transmission to a new host. Journal of virology 80, 1637–1644 (2006). \n12. Herbeck, J.T., et al.  Demographic processes affect HIV -1 evolution in primary infection \nbefore the onset of selective processes. Journal of virology 85, 7523–7534 (2011). \n13. Carlson, J.M. , et al.  HIV transmission. Selection bias at the heterosexual HIV -1 \ntransmission bottleneck. Science 345, 1254031 (2014). \n14. Baxter, J., et al. Inferring the multiplicity of founder variants initiating HIV -1 infection: a \nsystematic review and individual patient data meta -analysis. Lancet Microbe  4, e102 –\ne112 (2023). \n15. Westfall, D.H., et al.  Optimized SMRT-UMI protocol produces highly accurate sequence \ndatasets from diverse populations —Application to HIV -1 quasispecies. Virus Evolution  \n10(2024). \n16. Robb, M.L., et al. Prospective Study of Acute HIV -1 Infection in Adults in East Africa and \nThailand. N Engl J Med 374, 2120–2130 (2016). \n17. Dong, K.L., et al. Detection and treatment of Fiebig stage I HIV-1 infection in young at-risk \nwomen in South Africa: a prospective cohort study. Lancet HIV 5, e35–e44 (2018). \n18. Ndung'u, T., Dong, K.L., Kwon, D.S. & Walker, B.D. A FRESH approach: Combining basic \nscience and social good. Sci Immunol 3(2018). \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 17 \n19. Rolland, M., et al. Molecular dating and viral load growth rates suggested that the eclipse \nphase lasted about a week in HIV -1 infected adults in East Africa and Thailand. PLoS \npathogens 16, e1008179 (2020). \n20. Rossenkhan, R., et al. Bayesian estimation of HIV acquisition dates for prevention trials. \nmBio 16, e0188125 (2025). \n21. Giorgi, E.E., et al.  Estimating time since infection in early homogeneous HIV -1 samples \nusing a poisson model. BMC bioinformatics 11, 532 (2010). \n22. Learn, G.H. , et al.  Virus population homogenization following acute human \nimmunodeficiency virus type 1 infection. Journal of virology 76, 11953–11959 (2002). \n23. Liu, Y., et al. Selection on the human immunodeficiency virus type 1 proteome following \nprimary infection. Journal of virology 80, 9519–9529 (2006). \n24. Liu, Y., Mullins, J.I. & Mittler, J.E. Waiting times for the appearance of cytotoxic T -\nlymphocyte escape mutants in chronic HIV-1 infection. Virology 347, 140–146 (2006). \n25. Goonetilleke, N., et al. The first T cell response to transmitted/founder virus contributes \nto the control of acute viremia in HIV -1 infection. The Journal of experimental medicine  \n206, 1253–1272 (2009). \n26. Frost, S.D. , et al.  Characterization of human immunodeficiency virus type 1 (HIV -1) \nenvelope variation and neutralizing antibody responses during transmission of HIV -1 \nsubtype B. Journal of virology 79, 6523–6527 (2005). \n27. Liu, Y. , et al.  Env length and N -linked glycosylation following transmission of human \nimmunodeficiency virus Type 1 subtype B viruses. Virology 374, 229–233 (2008). \n28. Li, Y., et al. Limited Evidence for a Relationship between HIV -1 Glycan Shield Features in \nEarly Infection and the Development of Neutralization Breadth. Journal of virology  95, \ne0079721 (2021). \n29. Leslie, A.J., et al.  HIV evolution: CTL escape mutation and reversion after transmission. \nNat Med 10, 282–289 (2004). \n30. Boutwell, C.L., Rolland, M.M., Herbeck, J.T., Mullins, J.I. & Allen, T.M. Viral evolution and \nescape during acute HIV-1 infection. J Infect Dis 202 Suppl 2, S309–314 (2010). \n31. Druelle, V. & Neher, R.A. Reversions to consensus are positively selected in HIV-1 and bias \nsubstitution rate estimates. Virus Evol 9, veac118 (2023). \n32. Kijak, G.H. , et al.  Rare HIV -1 transmitted/founder lineages identified by deep viral \nsequencing contribute to rapid shifts in dominant quasispecies during acute and early \ninfection. PLoS pathogens 13, e1006510 (2017). \n33. Pilcher, C.D., et al. Brief but efficient: acute HIV infection and the sexual transmission of \nHIV. J Infect Dis 189, 1785–1792 (2004). \n34. Wawer, M.J., et al. Rates of HIV-1 transmission per coital act, by stage of HIV-1 infection, \nin Rakai, Uganda. J Infect Dis 191, 1403–1409 (2005). \n35. Fiebig, E.W. , et al.  Dynamics of HIV viremia and antibody seroconversion in plasma \ndonors: implications for diagnosis and staging of primary HIV infection. AIDS (London, \nEngland) 17, 1871–1879 (2003). \n36. Pilcher, C.D., et al. A generalizable method for estimating duration of HIV infections using \nclinical testing history and HIV test results. AIDS (London, England) 33, 1231–1240 (2019). \n37. Facente, S.N., et al.  Estimated dates of detectable infection (EDDIs) as an improvement \nupon Fiebig staging for HIV infection dating. Epidemiol Infect 148, e53 (2020). \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 18 \n38. Lukashov, V.V. & Goudsmit, J. Founder virus population related to route of virus \ntransmission: a determinant of intrahost human immunodeficiency virus type 1 \nevolution? Journal of virology 71, 2023–2030 (1997). \n39. Xu, H.T., et al.  Comparative biochemical analysis of recombinant reverse transcriptase \nenzymes of HIV-1 subtype B and subtype C. Retrovirology 7, 80 (2010). \n40. Shental-Bechor, D. & Levy, Y. Folding of glycoproteins: toward understanding the \nbiophysics of the glycosylation code. Curr Opin Struct Biol 19, 524–533 (2009). \n41. Wang, W., et al. A systematic study of the N-glycosylation sites of HIV-1 envelope protein \non infectivity and antibody-mediated neutralization. Retrovirology 10, 14 (2013). \n42. Wei, X., et al. Antibody neutralization and escape by HIV-1. Nature 422, 307–312 (2003). \n43. Wagh, K. , et al.  Completeness of HIV -1 Envelope Glycan Shield at Transmission \nDetermines Neutralization Breadth. Cell reports 25, 893–908 e897 (2018). \n44. Bause, E. & Legler, G. The role of the hydroxy amino acid in the triplet sequence Asn-Xaa-\nThr(Ser) for the N -glycosylation step during glycoprotein biosynthesis. The Biochemical \njournal 195, 639–644 (1981). \n45. Kasturi, L., Chen, H. & Shakin -Eshleman, S.H. Regulation of N -linked core glycosylation: \nuse of a site-directed mutagenesis approach to identify Asn-Xaa-Ser/Thr sequons that are \npoor oligosaccharide acceptors. The Biochemical journal 323 ( Pt 2), 415–419 (1997). \n46. Rademeyer, C., et al. Features of Recently Transmitted HIV-1 Clade C Viruses that Impact \nAntibody Recognition: Implications for Active and Passive Immunization. PLoS pathogens \n12, e1005742 (2016). \n47. Wieczorek, L. , et al.  Evolution of HIV -1 envelope towards reduced neutralization \nsensitivity, as demonstrated by contemporary HIV -1 subtype B from the United States. \nPLoS pathogens 19, e1011780 (2023). \n48. Rolland, M., et al. Genetic impact of vaccination on breakthrough HIV -1 sequences from \nthe STEP trial. Nat Med 17, 366–371 (2011). \n49. Rolland, M., et al. Increased HIV-1 vaccine efficacy against viruses with genetic signatures \nin Env V2. Nature 490, 417–420 (2012). \n50. deCamp, A.C. , et al.  Sieve analysis of breakthrough HIV -1 sequences in HVTN 505 \nidentifies vaccine pressure targeting the CD4 binding site of Env -gp120. PloS one  12, \ne0185959 (2017). \n51. Altfeld, M. & Allen, T.M. Hitting HIV where it hurts: an alternative approach to HIV vaccine \ndesign. Trends Immunol 27, 504–510 (2006). \n52. Letourneau, S., et al. Design and pre-clinical evaluation of a universal HIV-1 vaccine. PloS \none 2, e984 (2007). \n53. Rolland, M., Nickle, D.C. & Mullins, J.I. HIV -1 group M conserved elements vaccine. PLoS \npathogens 3, e157 (2007). \n54. Bekker, L.G., et al. Twice-Yearly Lenacapavir or Daily F/TAF for HIV Prevention in Cisgender \nWomen. N Engl J Med 391, 1179–1192 (2024). \n55. Kelley, C.F., et al. Twice-Yearly Lenacapavir for HIV Prevention in Men and Gender-Diverse \nPersons. N Engl J Med 392, 1261–1276 (2025). \n56. Niu, L., et al.  Preclinical evaluation of HIV -1 therapeutic ex vivo dendritic cell vaccines \nexpressing consensus Gag antigens and conserved Gag epitopes. Vaccine 29, 2110–2119 \n(2011). \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 19 \n57. Kulkarni, V., et al. HIV-1 p24(gag) derived conserved element DNA vaccine increases the \nbreadth of immune response in mice. PloS one 8, e60245 (2013). \n58. Kulkarni, V., et al.  Altered response hierarchy and increased T -cell breadth upon HIV -1 \nconserved element DNA vaccination in macaques. PloS one 9, e86254 (2014). \n59. Jacobson, J.M., et al. The immunogenicity of an HIV-1 gag conserved element DNA vaccine \nin people with HIV and receiving antiretroviral therapy. AIDS (London, England) (2023). \n60. Hu, X. , et al.  HIV Env conserved element DNA vaccine alters immunodominance in \nmacaques. Human vaccines & immunotherapeutics 13, 2859–2871 (2017). \n61. Kalams, S.A., et al. Focusing HIV-1 Gag T-cell Responses to Highly Conserved Regions by \nDNA Vaccination in HVTN 119. JCI Insight (2024). \n62. Manocheewa, S., Swain, J.V., Lanxon-Cookson, E., Rolland, M. & Mullins, J.I. Fitness costs \nof mutations at the HIV-1 capsid hexamerization interface. PloS one 8, e66065 (2013). \n63. Liu, Y., et al. Evolution of human immunodeficiency virus type 1 cytotoxic T -lymphocyte \nepitopes: fitness-balanced escape. Journal of virology 81, 12179–12188 (2007). \n64. Christie, N.M. , et al.  Viral fitness implications of variation within an immunodominant \nCD8+ T-cell epitope of HIV-1. Virology 388, 137–146 (2009). \n65. Hancock, G., et al. Identification of effective subdominant anti -HIV-1 CD8+ T cells within \nentire post -infection and post -vaccination immune responses. PLoS pathogens  11, \ne1004658 (2015). \n66. Mothe, B., et al. A human immune data-informed vaccine concept elicits strong and broad \nT-cell specificities associated with HIV -1 control in mice and macaques. Journal of \ntranslational medicine 13, 60 (2015). \n67. Hanke, T. Conserved immunogens in prime-boost strategies for the next-generation HIV-\n1 vaccines. Expert Opin Biol Ther 14, 601–616 (2014). \n68. Liu, Y. , et al.  Conserved HIV -1 epitopes continuously elicit subdominant cytotoxic T -\nlymphocyte responses. J Infect Dis 200, 1825–1833 (2009). \n69. Frahm, N. , et al.  Control of human immunodeficiency virus replication by cytotoxic T \nlymphocytes targeting subdominant epitopes. Nat Immunol 7, 173–178 (2006). \n70. Martinez-Picado, J., et al. Fitness cost of escape mutations in p24 Gag in association with \ncontrol of human immunodeficiency virus type 1. Journal of virology  80, 3617 –3623 \n(2006). \n71. Troyer, R.M. , et al.  Variable fitness impact of HIV -1 escape mutations to cytotoxic T \nlymphocyte (CTL) response. PLoS pathogens 5, e1000365 (2009). \n72. Rodrigo, A.G., Goracke, P.C., Rowhanian, K. & Mullins, J.I. Quantitation of target \nmolecules from polymerase chain reaction -based limiting dilution assays. AIDS Res Hum \nRetroviruses 13, 737–742 (1997). \n73. Altschul, S.F., Gish, W., Miller, W., Myers, E.W. & Lipman, D.J. Basic local alignment search \ntool. J Mol Biol 215, 403–410 (1990). \n74. Siepel, A.C., Halpern, A.L., Macken, C. & Korber, B.T.M. A computer program designed to \nscreen rapidly for HIV type 1 intersubtype recombinant sequences. AIDS Research and \nHuman Retroviruses 11, 1413–1416 (1995). \n75. Edgar, R.C. MUSCLE: multiple sequence alignment with high accuracy and high \nthroughput. Nucleic acids research 32, 1792–1797 (2004). \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 20 \n76. Furlong, J.C., Darley, P.D., Deng, W., Mullins, J.I. & Bumgarner, R.E. Phylobook: a tool for \ndisplay, clade annotation and extraction of sequences from molecular phylogenies. \nBiotechniques 76, 263–274 (2024). \n77. Rose, P.P. & Korber, B.T. Detecting hypermutations in viral sequences with an emphasis \non G --> A hypermutation. Bioinformatics 16, 400–401 (2000). \n78. Guindon, S. , et al.  New algorithms and methods to estimate maximum -likelihood \nphylogenies: assessing the performance of PhyML 3.0. Systematic biology 59, 307–321 \n(2010). \n79. Deng, W. , et al.  DIVEIN: a web server to analyze phylogenies, sequence divergence, \ndiversity, and informative sites. Biotechniques 48, 405–408 (2010). \n \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 21 \n \nTable 1. Participants, HIV sequences and multilineage acquisitions \nCohort Country Sex \nNumber of \nSamples \n(Participants) \nTotal \nnumber of \nSequences \nGP sequences: \nSample median \n(Range) \nREN sequences: \nSample median \n(Range)2 \nMultilineage \nacquisitions \n[UOL %]3 \nRV217 Thailand Male, \nTGF 66 (16) 25,621 196 (4-802) 137 (0-464) 8/16 (50%) \n[11/16, 69%] \nRV217 Kenya Female 44 (13) 14,192 146 (13-958) 112 (0-508) 2/13 (15%) \n[3/13, 23%] \nFRESH S. Africa Female 36 (13) 13,117 191 (25-1014) 116 (5-356) 5/13 (38%) \n[6/13, 46%] \nHVTN \n7031 \nSouthern \nAfrica Female 62 (33) 16,419 126 (6-526) 83 (3-493) 15/33 (45%) \n[19/33, 58%] \nHVTN \n7041 \nAmericas, \nCH \n~99% \nMale \nor TGF \n102 (48) 34,808 138 (15-588) 127 (4-1143) 17/48 (35%) \n[26/48, 54%] \nTOTAL   310 (123) 104,157 154 (4-1014) 117 (0-1143) 47/123 (38%) \n[65/123, 53%] \n1From the trial control arms (placebo recipients) only. 2One REN sample from one member of each of the RV217 \ncohorts was PCR negative and thus no sequences were obtained. 3Fractions and percentages in brackets include \nlineages of uncertain origin – UOL - as potential transmitted lineages, differing from a major lineage by 1-3 \nmutations. CH = Switzerland. TGF = Transgender female. \n \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 22 \n \nFigure Legends \nFig. 1. Viral load and sampling times. The name of each cohort is listed at the top of each panel. \nThe FRESH cohort from South Africa is shown in panel A, and the AMP trial HVTN 703 control \ncohort is in panel B. The RV217 study included samples derived from two countries, Thailand (TH, \npanel C) and Kenya (KE, panel E). AMP trial participants from the HVTN 704 control cohort were \nthe largest group studied and split over two panels ( D, F). REN viral subtypes are indicated by \ncolored lines, with subtypes listed in the key at the bottom of the figu re. Inter -subtype \nrecombinants are also indicated. Sample times from which HIV sequences were obtained are \nindicated as black dots on the curves in each panel. For the FRESH and RV217 cohorts, days post \nCOB (Center of Bounds; corresponding to the day midway between the last HIV negative test and \nthe first RNA positive test) is plotted along the x -axis. For the HVTN 703 and HVTN 704 cohorts, \nwhich were sampled less frequently, the estimated dates of detectable infection (EDDI) were \nderived from HIV diagnosti cs and viral sequence data as described 2,20. For some analyses, \nsequence datasets were divided into four time frames (i -iv) see panel A), as demarcated by thin \nblue vertical lines (see S2 Table). \nFig 2. Plasma HIV RNA viral load and lineage frequencies over time. Data from the 5 participants \nfrom the FRESH cohort with multiple transmitted lineages are shown with stacked panels \nshowing GP (A, C) and REN data (B, D), respectively. Key abbreviations: VL, plasma HIV RNA; HM, \nhypermutated sequences; MxL(n), lineage designations with numbers (n) assigned in decreasing \norder of the abundance of the lineages at the 1 st time point; “x” is used instead of a given stage \nof infection since data from all time points are shown; MxL(n.m), UOL sequences closely related \nto MxL(n), numbered (m) in order of frequency at the first time point; REC, recombinants. \nFig 3. Inferred selection in Gag and Env coding regions as a function of stage of acute/early HIV \ninfection. The top p anels show the ratio  of ne gatively and positively selected sites across \nseparate coding regions of Gag (A) and Env (B), respectively. Panels C and D show the frequencies \nof negatively and positively selected sites separately. \nFig 4. Entropy and selection within Gag. Plots of the proportion of lineages with nonzero entropy \n(A) and (C) and selection (B, D and E) derived from an alignment from all lineages having at least \n10 members and plotted in sliding windows of 10 amino acids . Identity of c oding regions are \nindicated within bars above panel C and shaded with different colors. Panels A and B show data \nfrom all time points combined. Panel C shows nonzero entropy, panel D negative selection and E \npositive selection from each of the three stages of acute/early HIV infection . Dat a included \nderives from all lineages with at least 10 members. Stage i, red lines; ii, blue; iii, black. \nFig 5. Entropy and selection within Env. Plots of the proportion of lineages with nonzero entropy \n(A) and (C) and selection (B, D and E) derived from an alignment from all lineages having at least \n10 members and plotted in sliding windows of 10 amino acids. Identity of c oding regions are \nindicated within bars above panel C and shaded with different colors. Panels A and B show data \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n 23 \nfrom all time points combined. Panel C shows nonzero entropy, panel D negative selection and E \npositive selection from each of the three stages of acute/early HIV infection . Dat a included \nderives from all lineages with at least 10 members. Stage i, red lines; ii, blue; iii, black. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\n0 20 40 60 80 100 120 140 160 180\n1\n2\n3\n4\n5\n6\n7\n8\n9\nDays post COB\nViral load (Log10 copies/ml)\nFRESH\n0 20 40 60 80 100 120 140 160 180\n0\n1\n2\n3\n4\n5\n6\n7\n8\n9\nDays post COB\nViral load (Log10 copies/ml)\nRV217 (TH)\n0 20 40 60 80 100 120 140 160 180\n1\n2\n3\n4\n5\n6\n7\n8\n9\nDays post COB\nViral load (Log10 copies/ml)\nRV217 (KE)\n0 20 40 60 80 100 120 140 160 180\n0\n1\n2\n3\n4\n5\n6\n7\n8\n9\nDays post EDDI\nViral load (Log10 copies/ml)\nHVTN 703 (C)\n0 20 40 60 80 100 120 140 160 180\n0\n1\n2\n3\n4\n5\n6\n7\n8\n9\nDays post EDDI\nViral load (Log10 copies/ml)\nHVTN 704 (C) - 2\n0 20 40 60 80 100 120 140 160 180\n0\n1\n2\n3\n4\n5\n6\n7\n8\n9\nDays post EDDI\nViral load (Log10 copies/ml)\nHVTN 704 (C) - 1\nA1 B D\nA1/B\nC CRF01AE\nGF2 A1/C\nF1\nA1/D F1/B\nFigure 1\nA\nC\nE\nB\nD\nF\ni ii iii iv\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\nFigure 2\nA\nB\nC\nD\nVL\n0 20 40 60\n0.001\n0.01\n0.1\n1\nLineage Fequency\n079 GP\n0 20 40 60\n102\n103\n104\n105\n106\n107\n108\n186 GP\n0 20 40 60\n102\n103\n104\n105\n106\n107\n267 GP\nLog Plasma RNA copies/ml\n0 20 40 60\n0.001\n0.01\n0.1\n1\nDays post COB\nLineage Fequency\n079 REN\n0 20 40 60\n102\n103\n104\n105\n106\n107\n108\nDays post COB\n186 REN\n0 20 40 60\n102\n103\n104\n105\n106\n107\nDays post COB\n267 REN\nLog Plasma RNA copies/ml\n0 20 40 60\n0.001\n0.01\n0.1\n1\nLineage Fequency\n271 GP\n0 20 40 60\n102\n103\n104\n105\n106\n107\n108\n318 GP\n0 20 40 60\n0.001\n0.01\n0.1\n1\n102\n103\n104\n105\n106\n107\nDays post COB\nLineage Fequency\n271 REN\n0 20 40 60\n102\n103\n104\n105\n106\n107\n108\nDays post COB\n318 REN\nLog Plasma RNA copies/ml\nHM\nMxL1\nMxL1.1\nMxL1.2\nMxL2\nMxL3\nRec\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\nFigure 3\nA\nC\nB\nD\n1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23\n0.0001\n0.001\n0.01\n0.1\n1\n10\n100\nNeagative Selection\nPositive SelectionNonzero Selection\ni ii iii i ii iii i ii iii i ii iii i ii iii\np17 p24 p2 p7 p1 p6\ni ii iii\nα < β (Positive Selection)\nα > β (Negative Selection)\ni ii iii i ii iii i ii iii\ngp120 gp41\nextracellular\ngp41 \nintracellular\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\nFigure 4\nA\nB\nC\nD\nE\n0.00\n0.05\n0.10\n0.15\n0.20\n0.25Nonzero Entropy\n50 100 150 200 250 300 350 400 450 500 550 600\n0.00\n0.01\n0.02\n0.03\n0.04\nAlignment position (AA)\nNonzero Selection\nα<β (Diversifying selection)\nα>β (Purifying selection)\n0.00\n0.05\n0.10\n0.15\n0.20\n0.25\nNonzero Entropy\nStage: i ii iii\n0.00\n0.01\n0.02\n0.03\n0.04\nNonzero Selection\nα > β (Negative Selection)\n50 100 150 200 250 300 350 400 450 500 550 600\n0.00\n0.01\n0.02\n0.03\n0.04\nAlignment position (AA)\nNonzero Selection\nα < β (Positive Selection)\np17 p24 p2 p7 p1 p6\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint \n\nFigure 5\nA\nB\nC\nD\nE\n0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400\n0.00\n0.01\n0.02\n0.03\n0.04\nAlignment positions\nNonzero Selection\nα<β (Diversifying selection)\nα>β (Purifying selection)\n0.00\n0.05\n0.10\n0.15\n0.20\n0.25\nNonzero Entropy\nStage: i ii iii\n0.00\n0.01\n0.02\n0.03\n0.04\nNonzero Selection\nα > β (Negative Selection)\n100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400\n0.00\n0.01\n0.02\n0.03\n0.04\nAlignment position (AA)\nNonzero Selection\nα < β (Positive Selection)\n0.00\n0.05\n0.10\n0.15\n0.20\n0.25\nNonzero Entropy\nSP gp120 gp41v1 v2 v3 v4 tmv5\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted October 15, 2025. ; https://doi.org/10.1101/2025.10.15.682663doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}