In this study, we provide conservative molecular evidence for internal HIV-1 transmission in a large proportion of Senegalese HIV-1 concordant couples. The majority of these couples reported the male partner as the source of the intracouple transmission, who in turn most frequently reported to have acquired HIV through an occasional sexual relationship. These data suggest that HIV-1 is frequently transmitted within married couples in Senegal with the husband as the most likely index partner. Our findings are in line with previous reports showing that married women in Africa have a high risk for acquiring HIV as a result of the extramarital relationships of their husbands 
. Our data also corroborate studies proposing concurrency as an important risk factor for onward HIV-1 transmission to the stable partner 
. Indeed, in our study, several husbands in linked couples who reported an occasional sexual relationship as source of infection had their first positive HIV-1 test many years after marriage (Table S1
), suggesting that their occasional relationships concurred with marriage. Under such conditions, HIV can rapidly spread from the occasional relationship to the wife as a result of the highly contagious acute infection of the husband 
, as well as from the often undiagnosed and/or undisclosed HIV status of the husband during this period 
The high proportion of linked transmissions (74%) in our cohort is in agreement with previous studies of HIV-1 transmission in African married couples. Two large HIV prevention trials in cohorts of HIV-1 discordant couples found genetically linked transmissions in 72% and 76% of the cases, respectively 
. A prospective study of HIV-1 discordant couples from Zambia found 87% of linked transmissions 
. Nevertheless, considerable numbers of HIV-1 acquisitions in these and our studies were unlinked to the partner's virus, thus emphasizing the need for molecular confirmation of HIV-1 transmission before embarking on clinical trials or studies of risk of HIV-1 transmission.
Our data are not in agreement with recent meta-analyses of HIV-1 discordant couple cohorts estimating large proportions of wife-to-husband transmissions in African married couples 
. A possible reason for this discrepancy could be the existence of an inclusion bias in HIV-1 discordant couple cohorts resulting in an overrepresentation of couples with HIV-1 infected wives. Indeed, in our Senegalese cohort we also counted more HIV-1 discordant couples with HIV-1 infected wives than with HIV-1 infected husbands (author's unpublished data and ref. 
). In fact, we noted that HIV-1 infected husbands frequently declined to participate because of fear of disclosing their HIV status to their wives (author's unpublished data). HIV-1 infected wives felt no stigma preventing them from participating because they often had an already disclosed HIV-1 infection originating from a previous marriage. In addition, HIV-1 discordant couples with HIV-1 infected husbands probably evolve more rapidly to a HIV-1 concordant status, exactly because of the husbands' frequent undisclosed and acute infection status, and as such they are not available for inclusion in HIV-1 discordant couple cohorts. At least the latter factor can be expected not to influence HIV-1 concordant couple cohorts, given that they can be enrolled any time after intracouple transmission has taken place. It is possible that HIV-1 infected men who are the index partners in HIV-1 concordant couples feel similar stigma preventing them from participating like those in HIV-1 discordant couples. However we expect this to be less given that both partners are now HIV-1 positive. In any case, such bias would not have influenced our conclusions as it would have led to an underestimation of the frequency of husband-to-wife transmissions in our Senegalese study population. Together, HIV-1 concordant couple cohorts with molecularly confirmed internal transmission probably allow a more trustworthy estimation of the relative frequencies of male-to-female and female-to-male transmission than HIV-1 discordant couple cohorts.
An important limitation of HIV-1 concordant couple studies however is that they have to rely on interview data to determine the direction of the intracouple transmission. To date, the order of infection events cannot easily be deduced from viral sequence information 
. Interview data may be subject to reporting error and bias, especially when they concern sensitive matter like sexual behavior. Indeed, one wonders to what extent the high proportion of reported male-to-female transmissions in our cohort could reflect women's resistance to report extramarital relationships. On the other hand, several observations support the reliability of the data that were used in our study. First, the interview data corresponded completely with the results of the molecular analyses: all couples with genetically linked viruses reported internal HIV transmission and all couples with genetically unlinked viruses reported external sources of HIV infection. Second, none of the couples reported conflicting sources of HIV infection for both partners. And finally, a number of clinical parameters of HIV-1 infection like the time since self-reported first positive HIV test, antiretroviral therapy status and duration, CD4 count, and the proportion gp41 ambiguous base pairs corroborated the reported order of infection events. Couples for whom information on the suspected source of infection was not available for both partners were excluded from the analyses. However it is unlikely that this has influenced our conclusions given that it affected only 6 couples and that their partially reported sources of infection were very comparable to the total study population (Table S1
The application of phylogenetic analysis to confirm HIV-1 transmission events is well established, and it is frequently used for providing evidence in legal cases 
, HIV prevention trials 
, or specific epidemiological or molecular studies 
. The choice of an appropriate HIV-1 gene is crucial as it should match the genetic variability of the target population 
. For instance, the rapid genetic diversification of the variable env
gp120 V2–V3 region makes it ideal for studying recent transmission events. On the other hand, the low genetic variability of HIV-1 pol
makes it more suitable for establishing older transmissions. Assuming that our Senegalese cohort of HIV-1 concordant couples contained recent as well as older HIV-1 transmissions, we decided to use env
gp41, a gene with a genetic variability in between that of env
gp120 and pol
. Previous studies successfully used gp41 in HIV-1 subtyping and linkage analyses of similarly diverse datasets 
. The use of gp41 sequences in our study population showed high specificity and sensitivity relative to the interview data. Our assays lacked sensitivity to prove genetic linkage for 6 couples, 5 of whom reported internal HIV-1 transmission. These couples yielded undetermined molecular results, i.e. the gp41 sequences clustered in the phylogenetic tree but without meeting the conservative bootstrap and distance criteria. Relaxation of the bootstrap and distance criteria could increase the sensitivity without affecting the specificity. Alternatively, more conserved HIV-1 gag
genes or the use of clonal or deep sequencing techniques could help establish genetic linkage in these couples, but this was beyond the scope of this study.
Phylogenetic analyses cannot exclude that two linked sequences are connected through a third intermediate sequence not included in the analysis. This possibility could frequently occur in populations of e.g. men who have sex with men where promiscuous behavior has been shown to result in large clusters of linked viruses with uncertainty of the origins and directionality of the underlying transmissions 
. However, such interconnectivity is unlikely in our study population of well-defined heterosexual partnerships originating from a large metropolitan area and characterized by much less promiscuous behavior than men who have sex with men. Therefore, the majority of the observed phylogenetic clusters in our study likely represent genuine intracouple transmissions. Another consideration that should be taken into account in phylogenetic studies is HIV-1 superinfection, i.e. the infection of an already HIV-1 positive individual with a second HIV-1 strain 
. For instance, it could be possible that for some linked couples in our study both partners initially acquired HIV-1 from outside their relationship and that a subsequent superinfection from one partner to the other was picked up by our assays as a linked transmission. Detection of HIV-1 superinfection can be complex and requires clonal or deep sequencing techniques which was beyond the scope of our study. Nevertheless, given that the genetic analyses in our study were completely supported by the interview data, and that frequencies of HIV-1 superinfection are considered to be low in low-risk populations with low background HIV-1 prevalences 
, superinfection likely has not confounded our observations much.
The majority of the couples enrolled in our study lived in the Dakar metropolitan area, and they reported a wide range of occupations and levels of education characteristic of those of an African suburban population. Therefore, we should be careful with extrapolating the observed patterns of HIV-1 transmission in this study population to populations living in more rural areas or in other African countries. Future studies should investigate HIV-1 transmission in HIV-1 concordant couple cohorts enrolled from other African populations with different socio-epidemiological factors.
In summary, we found that a majority of HIV-1 concordant couples in Dakar, Senegal showed genetically linked HIV-1 transmission with the husband as likely index partner. Our data emphasize the risk of married women for acquiring HIV-1 as a result of the occasional sexual relationships of their husbands. Understanding the origins and dynamics of HIV-1 infection in married couples in Africa will be crucial for the future planning of successful HIV prevention programs.