The island of Hispaniola, shared by Haiti and the Dominican Republic, included in 2016 around 217,000 people living with the human immunodeficiency virus-type 1 (HIV-1), the etiologic agent of the acquired immunodeficiency syndrome (AIDS) (UNAIDS, 2013, 2017). Initially, AIDS cases recognized in Haiti (Pape et al., 1983) and the Dominican Republic (Koenig et al., 1987) mostly involved men who had sex with men, but these countries now have generalized epidemics predominantly driven by heterosexual sex (Figueroa, 2008). Despite the high HIV prevalence rates reported for the general population in Haiti (2.1%: 1.9–2.3%) and the Dominican Republic (1.0%: 0.7–1.4%) in 2016, significant declines of the HIV incidence rates were observed in both countries over the last decade (UNAIDS, 2013, 2017).

The HIV-1 epidemic in Hispaniola is dominated by subtype B (Nadai et al., 2009; Myers et al., 2012; Lopez et al., 2015). Genetic evidence suggests that the subtype B epidemic in Hispaniola, as in many other Caribbean countries, is mainly driven by the transmission of multiple early divergent non-pandemic subtype B lineages (designated “B_CAR”), although the worldwide disseminated “B_PANDEMIC” lineage also circulates (Cabello et al., 2014, 2015; Divino et al., 2016; Bello et al., 2018). The early dissemination dynamics of the HIV-1 subtype B epidemics between Haiti and the Dominican Republic remain largely unknown. Previous evolutionary analyses pointed that subtype B probably entered in the Americas through Haiti around the mid-1960s and then moved to other countries (Gilbert et al., 2007; Junqueira et al., 2011; Worobey et al., 2016), but no or very few (n < 15) subtype B sequences from the Dominican Republic were included in those studies. Another study with a large number of HIV-1 sequences from the Dominican Republic has shown that subtype B virus circulates in this country since the early 1960s (Lopez et al., 2015), a result comparable to the estimated age of the Haitian epidemic. This study, however, may have traced the age of the Haitian subtype B ancestor rather than of the Dominican one, given that it assumed that all subtype B infections in the Dominican Republic resulted from a single introduction and that the hypothesis of multiple independent viral introductions from Haiti was not formally tested.

The objective of this work was to reconstruct the early spatiotemporal dynamics of dissemination and demographic history of non-pandemic HIV-1 B_CAR lineages on the island of Hispaniola and to estimate the current degree of geographic compartmentalization of the B_CAR epidemic between Haiti and the Dominican Republic.

All new HIV-1 pol sequences from Haiti obtained here (n = 127) were confirmed as non-recombinant subtype B by ML phylogenetic analysis (Supplementary Figure S1) and RDP recombination analysis (data not shown). The HIV-1 subtype B Haitian sequences were combined with viral strains representative of the B_CAR diversity in different Caribbean islands (n = 200) and of the B_PANDEMIC diversity in the US and France (n = 300) (Supplementary Table S2) as previously characterized (Cabello et al., 2014). The ML phylogenetic analysis revealed that most subtype B Haitian sequences (n = 91, 72%) were intermixed among basal non-pandemic B_CAR lineages, whereas the remaining ones (n = 36, 28%) branched within the well-supported (aLRT = 0.89) B_PANDEMIC clade (Figure 1). Analysis of the epidemiological characteristics of Haitian subjects showed that both B_CAR and B_PANDEMIC viral lineages circulated among males and females of different age groups (Supplementary Table S3). No significant differences were observed in the frequency of subtype B lineages according to gender (p = 0.82) or age group (p = 0.45), although subjects infected with B_CAR strains had a relative younger mean age (36.0 years) as compared with those infected with the B_PANDEMIC clade (40.1 years) (p = 0.03).

A closer inspection of the phylogenetic relationship among B_CAR sequences confirm that sequences from Haiti and the Dominican Republic were highly intermixed with each other and are usually basal to sequences from other Caribbean islands (Supplementary Figure S2). Most sequences from other well-sampled islands like Jamaica and Trinidad and Tobago, by contrast, branched in country-specific subclades that were nested within the Hispaniola B_CAR diversity. Very few sequences from Hispaniola (<1%) branched within Trinidadian and Jamaican clusters, confirming that most B_CAR infections in Hispaniola resulted from internal viral dissemination and not from re-introductions of viral strains from other Caribbean islands. To investigate the phylogeographic structure of the HIV-1 epidemic within Hispaniola, B_CAR pol sequences from Haiti here identified (n = 91) were combined with Haitian (n = 12) and Dominican (n = 123) B_CAR pol sequences identified in a previous study (Cabello et al., 2014) and analyzed using BaTS. Analyses of population subdivision rejected the null hypothesis of panmixis (i.e., complete intermixing of sequences from Haiti and the Dominican Republic) (Supplementary Table S4), demonstrating that despite frequent viral intermixing between both countries, the geographic subdivision of the HIV-1 B_CAR sequences from Hispaniola was greater than expected by chance.

The same dataset of B_CAR pol sequences from Haiti (n = 103) and the Dominican Republic (n = 123) was then subjected to Bayesian phylogeographic analyses. Reconstruction of the spatiotemporal dissemination dynamic traced the T_MRCA for the HIV-1 B_CAR epidemic in Hispaniola at 1967 (95% HPD: 1961–1972), but failed to uncover its precise epicenter (Figure 2). After combining six independent Bayesian MCMC runs, the root location of the HIV-1 B_CAR ancestor in Hispaniola was traced with equal probability [posterior state probability (PSP) = 0.50] to Haiti and to the Dominican Republic. The difficulty to trace the location of the B_CAR root into a single country was also evidenced when the results obtained from independent Bayesian MCMC runs were visualized separately (Supplementary Table S5). Quantification of B_CAR flux between countries using Markov jump counts support a mean of 18 viral migrations from Haiti to the Dominican Republic and 9 viral migration events from the Dominican Republic to Haiti. These viral migrations started in the early 1970s and were homogenously distributed between the mid-1970s and the mid-2000s (Supplementary Figure S3). The Bayesian phylogeographic analysis also revealed 20 country-specific (14 Dominican and 6 Haitian) B_CAR monophyletic subclades with relative high node support [posterior probability (PP) > 0.70] (Figure 2). The Dominican B_CAR subclades mostly arose between the mid-1970s and the early 1980s (Supplementary Table S6) and together comprised 55% (n = 68) of B_CAR sequences from the country.

HIV-1 B_CAR sequences from Haiti (n = 103) and the Dominican Republic (n = 123) were finally used to reconstruct the demographic history of this viral epidemic in Hispaniola. Reconstruction of population dynamics with the BSKL coalescent-based model suggested that the B_CAR epidemic in Hispaniola experienced an initial phase of fast exponential growth until the beginning of the 1990s, followed by a stabilization of the Ne (Figure 3A). The BSKG model, however, supported a longer exponential growth phase until the late 1990s, followed by an epidemic decline that extended until the most recent coalescent event around the mid-2000s (Figure 3B). The UNAIDS epidemiological data, in agreement with the BSKG model, indicate a growth in the number of new HIV infections in Hispaniola until the early 2000s and a subsequent phase of decline extending until the most recent date (Figure 3C). To estimate the growth rate of the B_CAR epidemic at the initial phase, three parametric coalescent models (logistic, exponential, and expansion) were compared. According to the best-fit logistic growth coalescent model (Supplementary Table S7), the mean growth rate of the B_CAR epidemic during the first decades of expansion in Hispaniola was 0.50 (95% HPD: 0.37–0.65).

This study confirms that the HIV-1 subtype B epidemic in Haiti is mostly driven by dissemination of early divergent non-pandemic B_CAR lineages (Cabello et al., 2014). The relative prevalence of B_CAR lineages among HIV-1 subtype B infections in Haiti (72%) closely matches that estimated in the neighboring Dominican Republic (74%) (Cabello et al., 2014). Together, Haiti and the Dominican Republic host the largest population (~160,000 people) of B_CAR-infected individuals in the Americas. Our results reveal a high degree of phylogenetic intermixing of the HIV-1 B_CAR sequences from Haiti and the Dominican Republic, consistent with the historical intense cross-border population mobility (Cohen, 2006; Rojas et al., 2011), but also support some level of geographic structure within Hispaniola. We estimate that nearly 55% of B_CAR infections in the Dominican Republic probably occurred within local transmission networks.

Previous evolutionary analyses (Gilbert et al., 2007; Junqueira et al., 2011; Worobey et al., 2016) support that the HIV-1 subtype B arrived in Haiti earlier than in any other American country. Here, we tested the hypothesis of the Haitian origin of subtype B epidemic by using for the first time a geographically balanced HIV-1 B_CAR dataset of Haitian (n = 103) and Dominican (n = 123) sequences. Our Bayesian phylogeographic analysis traced the root location of the B_CAR epidemic into Haiti and the Dominican Republic with the same posterior probability. This result revealed the complexity to uncover the geographic origin of the subtype B epidemic within Hispaniola by using only genetic data, probably due to the continuous cross-border viral movements between countries since the very early epidemic times. Thus, integration of non-genetic information (such as HIV incidence and prevalence data and human flows) into phylogeographic inference (Graf et al., 2015) would be probably indispensable to resolve the precise location of the HIV-1 B_CAR ancestor within Hispaniola.

Although phylogeographic analyses conducted here support the Haitian or Dominican origin of subtype B with the same probability, some epidemiological and historical data favor the Haitian origin hypothesis. The first AIDS cases were recognized in Haiti (1978–1979) a few years before those of the Dominican Republic (1983) and, by the early 1980s, HIV seroprevalence among Haitians was higher than among Dominicans (Pape et al., 1983; Koenig et al., 1987). Furthermore, the estimated T_MRCA of the B_CAR ancestor at around the late 1960s coincides with the return of Haitian professionals from the Democratic Republic of Congo (Gilbert et al., 2007), a country with an established HIV epidemic by that time (Worobey et al., 2008). By contrast, we found no similar historical link supporting relevant human flows between the Dominican Republic and the Democratic Republic of Congo during the 1960s.

Some epidemiological studies suggested that HIV-1 transmission between populations in Haiti and the Dominican Republic was uncommon during the early years and that tourists were the most likely source of first virus transmissions to Dominicans (Pape et al., 1983; Koenig et al., 1987). Our phylogeographic analysis, however, supports that HIV-1 B_CAR strains have been disseminated between Haiti and the Dominican Republic since the early 1970s and that several B_CAR transmission networks were already established in the Dominican Republic by mid-1970s, nearly a decade before the first AIDS cases were officially recognized in the country. The overall time scale here obtained for the B_CAR epidemic is fully consistent with that recovered in previous studies (Gilbert et al., 2007; Junqueira et al., 2011; Cabello et al., 2014; Worobey et al., 2016), supporting the reliability of our T_MRCA estimates. These results clearly indicate that HIV-1 B_CAR strains have been disseminated between Haitian and Dominican populations quite frequently since the early 1970s onwards.

While the BSKL model supports a stabilization of the B_CAR epidemic in Hispaniola from the early 1990s onwards, the BSKG reconstruction supports a sustained expansion until the late 1990s and a subsequent decline until the most recent coalescent event. The pattern here recovered by the BSKG model is consistent with the reported decline in HIV incidence in Haiti and the Dominican Republic since the late 1990s (UNAIDS, 2013, 2017), which is likely partially driven by changes in sexual behavior since the mid-1990s (Halperin et al., 2009). This finding is in agreement with previous studies that described that the BSKG model may uncover some aspects of the population history undetected by other Bayesian models (Gill et al., 2013; Mir et al., 2018). The mean growth rate estimated here for the B_CAR epidemic in Hispaniola during the first decades (0.50 year⁻¹) is similar to those estimated for B_CAR and B_PANDEMIC lineages spreading in American countries with generalized heterosexual epidemics (0.35–0.45 year⁻¹) (Cabello et al., 2014; Mendoza et al., 2014; Mir et al., 2015; Bello et al., 2018).

A drawback to consider in our study is the relative small size of our sample. According to the UNAIDS, the number of people living with HIV was estimated at around 150,000 in Haiti and 67,000 in the Dominican Republic in 2017. Assuming that 70–75% of those infections probably correspond to B_CAR viruses, a very small fraction (<1%) of B_CAR-infected people living in those countries was included in our study. This low sampling density does not provide adequate power to assess HIV-1 clustering in generalized epidemics and could produce misleading results (Novitsky et al., 2014). The second limitation of our study is the lack of meta-data (such as city of origin, age, sex, or potential risk behavior) for most of the HIV-infected individuals included in our analysis, avoiding the identification of trends between individuals linked within the same local cluster. Finally, it is unclear whether our sample truly represents the whole diversity of HIV-1 B_CAR in Haiti because most Haitian individuals here analyzed were from the capital city (Port-au-Prince).

In summary, this study highlights that the HIV-1 epidemic in Haiti is mainly driven by dissemination of early divergent non-pandemic B_CAR strains. Our findings revealed that the HIV-1 B_CAR epidemics in Haiti and the Dominican Republic are highly connected by intensive bidirectional viral dispersal since the early 1970s and that local B_CAR transmission was already established in both countries when the first AIDS cases were officially recognized. Despite the use of a geographically balanced B_CAR dataset, probabilistic Bayesian phylogeographic models cannot uncover the root of the subtype B epidemic in Hispaniola by using only genetic sequence information. Our findings support that both national and bi-national coordinated prevention measures are necessary to further control the HIV-1 dissemination in Haiti and the Dominican Republic.

All HIV-1 sequences were deposited in the GenBank database (accession numbers MK639799–MK639925).

Comité National d’Ethique du Ministère de la Santé Publique et de la Population de Haiti (13/07/2009).

GS, GB, VL, and MN conceived and designed the study. GS, MO, JB, RC, and BL collected the samples and performed the HIV sequence amplification and genotyping. GB and IA performed the phylogenetic and phylodynamics inferences. MN performed the statistical analyses. GS, GB, IA, and VL wrote the manuscript. All authors analyzed the data and discussed and reviewed the manuscript.