Serology shows uneven geographical distribution of HIV-2

HIV-2 serological data of the period 1985–91 meeting our criteria were available for 30 West African cities (listed in Materials and Methods). We applied Bayesian estimation to infer HIV-2 prevalence rates from the sample-based frequencies of HIV-2 seropositives (see Materials and Methods). HIV-2 prevalence was very unevenly distributed, but spots of high prevalence existed in both western and eastern parts of West Africa. Fig 1 depicts the prevalence data and the historical (pre-1950) range of Cercocebus atys atys according to primatological sources [51–53].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. HIV-2 seroprevalence in adults without specific risk factors in West Africa, 1985–91.

Complete information on HIV-2 prevalence rates is in S1 Dataset. The checkered area indicates the historical (pre-1950) range of sooty mangabeys [51–53].

https://doi.org/10.1371/journal.pone.0166805.g001

Historical male circumcision prevalence has an uneven geographical distribution

To research historical patterns of MC, we first consulted the Ethnologue database and ethnographic maps [54–56]. We then retrieved MC information from major ethnographic atlases and reviews [32,38–40,57,58], and filled in their MC information gaps with primary ethnographic articles and fieldwork in Guinea-Bissau. In total, we covered 218 ethnic groups encompassing almost all West Africa (see Materials and Methods).

MC was far less common, and geographically more patchy in most early (1890–1920) ethnographic accounts than it is in current times. After applying our assumptions, as explained in Materials and Methods, MC was found to be traditionally not practiced or practiced by a minority of men in 72 (33.0%) groups. These results broadly agree with major ethnographic atlases and reviews [32,38–40,57,58], although our coverage of MC permitted to fill gaps in these works. The results are condensed in S2 Dataset, and a concise explanation of the frequency of MC patterns, and its distribution over major ethnic families is in section a) of S1 Text.

Our fieldwork in Guinea-Bissau determined that a pattern of MC in late age persists to these days among non-Muslim groups: among the Balanta, MC was normally done after age 40, though at present it starts to be done between 20 and 30 years of age [59]; among the Manjako and the Felupe MC is done in collective rituals held each ~20 years, which causes many men to remain uncircumcised through adulthood (a pattern seen also among the Jola of the neighboring Casamance region, and among the Yowaba of northern Benin).

Robust correlation between HIV-2 seroprevalence and historical MC prevalence

We used interpolation to estimate MC frequency for each ethnic group over time, between early 20^th century and nowadays (keeping in mind that a major wave of MC adoption occurred for many ethnic groups). Through Latin Hypercube Sampling (LHS) of the parameter values, 1,000 different calculations of estimates of MC frequency per ethnic group were obtained (see Materials and Methods). MC frequencies per city were then compiled based on tables of ethnic partitioning of population (TEPP) derived from censuses and surveys (see Materials and Methods). Fig 2 shows these results for the 30 cities, in 1950 and 1988, under the default model (COLRST-LOGIS).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Estimated MC frequencies for the West African cities.

Estimated MC frequencies in 1950 (A) and 1988 (B) for the West African cities. Boxes indicate the interquartile range (IQR) of the frequencies; median is indicated by a horizontal line within the box, and whiskers extend to the farthest values that are not more than 1.5 times the IQR away from the box. Cities of the same country appear with the same color. Country colors: Côte d'Ivoire, red; Liberia, dark green; Sierra Leone, green; Guinea, dark cyan; Guinea-Bissau, magenta; Senegal, brown; Gambia, gray; Cape Verde, dark gray; Mauritania, yellow; Mali, chocolate; Burkina Faso, tomato; Ghana, violet-red; Togo, blue; Benin, dodger blue; Nigeria, deep sky blue; Niger, light gray.

https://doi.org/10.1371/journal.pone.0166805.g002

The correlation between HIV-2 prevalence in 1985–91 and estimated MC in 1988, for our default COLRST-LOGIS model, is substantial and significant (median Spearman rho = -0.560, interquartile range (IQR) = -0.562–-0.560, p≤0.0022 for all comparisons).

We performed two sensitivity analyses. In the first, we used the different models for interpolating, and three different time points. The results were broadly similar across models and time points (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. The correlation between HIV-2 prevalence and estimated MC rate at various times and for three models of interpolation.

https://doi.org/10.1371/journal.pone.0166805.t001

The independence of the results from the time of MC calculation within the 1950–88 period is not surprising because, even though MC frequency increased rapidly in many ethnic groups during that period, the relative rankings of the ethnic groups (and hence of cities) in MC frequency changed only moderately. Reassuringly, the correlation was similar across all models of MC interpolation.

We then focused on our COLRST-LOGIS default model and performed different calculations with specific subsets of cities (Table 2). First, and following the method of Bongaarts et al. (1989) for HIV-1 [30], we tested the correlation with only the capital/main city of each country included. Second, we tested it also with all main or major cities included.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. The correlation between HIV-2 prevalence and estimated MC rate for different subsets of cities.

https://doi.org/10.1371/journal.pone.0166805.t002

In Central Africa, Kinshasa was the main hub of dissemination of HIV-1 group M [15,60–62]. During the colonial period, Abidjan was, like Kinshasa, the center of a major transportation network, had a strongly male-biased sex ratio, and rampant commercial sex work [63,64]. Both HIV-2 groups A and B are common in Abidjan [50], and these groups and groups G, H, and I, appear to have originated in southwestern Côte d'Ivoire [3,7]. It is thus tempting to hypothesize that Abidjan was the main hub of dissemination of HIV-2. If so, the cities connected to Abidjan by sea, railway, or with known important migration ties to it provide a natural set of cities with relatively homogeneous exposure to HIV-2 exportation from the main hub. We thus calculated the correlation including only these cities, with MC estimated for 1960. The results of these various tests are in Table 2.

The correlations remain significant in all tests done with 20 or more cities (p-values ≤0.013), and in the range -0.46–-0.59. Significance was lost for the smallest dataset with only the main cities; however the correlation was of the same magnitude and the trend was conserved (p-values ≤0.09).

MC frequency was lower in cities that were probable earlier epicenters based on phylogeographical and serological evidence

Both HIV-2 groups A and B likely emerged before 1950 [10,11], and the same applies to HIV-1 groups M and O [12–16], suggesting that the probability of HIV epidemic emergence was higher in early 20^th century, which may have been related to genital ulcer disease epidemics, injection campaigns, lower historical rates of MC or other factors [61,62,65–67]. The main cities at or near the sooty mangabey range (Dakar, Banjul, Bissau, Conakry, Freetown, Monrovia, Abidjan) were probably exposed to the arrival of rural migrants carrying SIVsmm/HIV-2 infections with the potential to generate an epidemic strain. We thus calculated MC frequencies in these cities in 1930. The results are shown in Fig 3.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. The estimated MC frequencies in 8 large cities at or near the sooty mangabey range in 1930.

Cities of the same country appear with the same color. The country colors are the same as in Fig 2 (see Fig 2 legend).

https://doi.org/10.1371/journal.pone.0166805.g003

The MC rates were markedly lower in the cities of Côte d'Ivoire and Guinea-Bissau, the countries that both serological data [23–27] and phylogeographical analysis [10] suggest were the earliest epicenters.

The special case of Bissau and the Guinea-Bissau independence war

By the time of the independence war (1963–74), Bissau, with a population near 22,000 in 1960 and 68,000 in 1970 [68,69], received the influx of tens of thousands of Portuguese soldiers (17,000–32,000 in the country throughout the period 1965–74 [70], many of whom frequented Bissau). Many soldiers had contacts with local women and their contacts with high-turnover CSWs were common [24,26]. Therefore, during the war, Bissau had an extremely male-biased sex ratio, rife CSW and, since Europeans are almost universally uncircumcised, a much lowered MC rate (lower than the ones shown in Figs 2 and 3, because these were calculated without considering Europeans).

MC was not traditionally common in the rural region where epidemic HIV-2 crossed from simians to humans

Our ethnographic literature survey allowed us to map MC throughout West Africa, including in the area inhabited by sooty mangabeys (S2 Dataset). Since phylogenetic analyses of SIVsmm from multiple sites determined that the approximate location of cross-species transmission of five HIV-2 groups (A, B, G, H, and I) was probably southwestern Côte d'Ivoire [3,7], we aimed to compare this region with others of the sooty mangabey range with respect to traditional MC practices.

Focusing on the sooty mangabey range and on MC practices in early 20^th century, ethnic groups with absent, uncommon, or not general MC occupy two widely separated areas, one in northern Guinea-Bissau and adjacent Senegal and another in northeast Liberia and southwest Côte d'Ivoire. The latter area contains four ethnic groups in which MC was traditionally absent and still rare in the 1930s, and one group with MC not generally practiced by the 1950s. These groups occupy a contiguous area between the rivers Cess (Liberia) and San Pedro (Côte d'Ivoire) extending northwards up to Tai (Côte d'Ivoire) [54–56,71–76] (Fig 4). The exact ethnic groups involved are described in section b) of S1 Text. This area, which is less than 10% of the sooty mangabey range, encompasses the location where SIVsmm strains phylogenetically close to HIV-2 groups A, B, G, H, and I were found [3,7]. Therefore, the majority of known HIV-2 strains, including the two main ones, may have originated from a sub-area of the sooty mangabey area in which, exceptionally, MC was traditionally absent in the local native ethnic groups [71–74,76].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. The spread of MC practices at or near the sooty mangabey range in early 20^th century.

The information supporting this map is in S2 Dataset, with additional explanations in S1 Text. The checkered area indicates the historical (pre-1950) range of sooty mangabeys [51–53].

https://doi.org/10.1371/journal.pone.0166805.g004

HIV-2 prevalence data

We obtained HIV-2 seroprevalence data for 30 cities of 16 West African countries: Abidjan, Bouaké, Man, Bondoukou, Korhogo, Dabou (Côte d'Ivoire), Monrovia, Buchanan (Liberia), Freetown (Sierra Leone), Conakry (Guinea), Bissau, Bafatá (Guinea-Bissau), Dakar, Saint Louis, Kaolack, Ziguinchor (Senegal), Banjul, Highway (Gambia), Praia (Cape Verde), Nouakchott (Mauritania), Bamako (Mali), Ouagadougou, Bobo-Dioulasso, Banfora (Burkina Faso), Accra (Ghana), Lomé (Togo), Cotonou, Porto Novo (Benin), Lagos (Nigeria), Niamey (Niger). The data are described in detail in S1 Dataset.

In compiling HIV-2 seroprevalence data (S1 Dataset), we considered only surveys of adults from the general population (pregnant women, blood donors, professional groups) without known specific risk factors for HIV transmission (excluding data on CSW, hospitalized patients, or patients known to be infected by any STD, HIV or TB) [17–22]. To cover the different countries in an epidemiologically unbiased way, we focused on urban samples, from cities or towns over 20,000 inhabitants at the time of the collection, with more than 200 samples per city (summing up all collections for the city), and with the city well identified. We made a few exceptions from these rules to correct for biases and fill data gaps (described in S1 Dataset). Because low observed frequencies are subject to considerable stochastic error (in particular, zero observed infections might conceal low non-zero seroprevalence), we used Bayesian computation (as implemented in the binom R package (https://cran.r-project.org/web/packages/binom/ [100])) to calculate the posterior distribution and confidence interval of true prevalence values based on the Jeffreys interval method [101], and used the expected value of the distribution as estimated prevalence in the correlation analyses. For Nigeria we considered only the city of Lagos, because seroprevalence was zero in most other locations, reflecting a much later arrival of HIV-2 to this country compared with other countries.

Estimating male circumcision frequency per ethnic group, both current and in early 20^th century

All West African ethnic groups described in the Ethnologue database [54,55] with more than 10,000 people (in the 1990s) were considered. In addition, Kriol groups that do not have a traditional ethnic origin but are numerically important in cities (e.g., Americo-Liberians, Creole-Krio of British West Africa, Brazilian-Beninese, Cape Verdeans) were investigated. Together, our analysis was based on 218 ethno-linguistic and Kriol groups. Published ethnic maps [56] were used as a source for their geographic distribution.

For information about current MC rates, we gathered all DHS (period 2005–2015) [41] from the relevant countries, thus obtaining reliable estimates of MC frequency for each country and, for most countries, for the ethnic groups within the countries. We complemented this information with biomedical papers that reported MC frequencies within the relevant regions. One of us (MPT) performed fieldwork in Guinea-Bissau, based on focus-group discussions and semi-structured interviews (see S1 Appendix).

For information about past MC rates, we used a modified version of the Ethnographic Atlas (EA; Gray (1999) [38]), which expanded on the early work of Murdock (1967) [32] as a starting point for MC data. The EA lacked MC information for many traditional ethnic groups, and for all Kriol groups,which we recovered from major ethnographic reviews [39,40,57,58], early published primary ethnographic field reports, and our fieldwork (S1 Appendix) to complete MC information on the period 1880–1950.

Of the 218 ethno-linguistic and Kriol groups, for 193 (88.5%) we were able to compile information about MC practices (presence/absence, cultural prescription, universality, age of occurrence). For the remaining groups (which had relatively low population numbers), we assumed practices averaged over geographically and culturally (as ascertained by linguistic similarity [54]) very close ethnic groups (see details in S2 Dataset).

Interpolating male circumcision frequency per ethnic group

We aimed to estimate MC frequency over time, c(t), between the late 19^th century and the present, by performing interpolations between the early, ethnography-based estimate, and the current, DHS survey-based estimate. The ethnographic estimates can be considered reliable when they state that MC was either (almost) totally absent or (almost) universal. In intermediate cases, we had to estimate a plausible frequency range based on the verbal account. In our default model, interpolation was based on logistic growth, but we also used other models, as explained below.

For all ethnic groups, we estimated MC frequency, c(t), for the whole period between the years t_min = 1851 and t_max = 2016. For all groups, there was a survey-based estimate of MC frequency, c_d, at time t_d. For many groups, ethnographic sources state that MC was culturally obligatory, universal, and performed before adulthood in early 20^th century. These always corresponded to survey-based MC, c_d>90%. We thus labelled these groups "FIXEDUNIV" and set MC frequency c(t) = c_d for the whole t_min–t_max period.

For the remaining, non-FIXEDUNIV groups, MC was either not practiced or not universally practiced in early 20^th century. To avoid working with many assumptions, we developed simplified models for MC interpolation, working mostly with parameters and assumptions common to all ethnic groups, and keeping the ethnic group-specific parameters at a minimum.

Hence, for the non-FIXEDUNIV ethnic groups, we defined our default model as follows. We assumed that MC frequency was initially low, and started to increase at time t₁. For most groups, the parameter t₁ was set equal to t_1-colrst, a general parameter sampled from the period 1895–1905. The rationale for this assumption is that this was the time of establishment of colonial rule in all colonies, which prompted rapid and partly enforced mixing between ethnic groups [57,77,78,80,81]. Since several sources suggest that many previously non-circumcising groups were borrowing MC practice from other groups by that time [57,77,78], we assume that the wave of MC adoptions inferred by our preliminary survey (see Introduction) started or greatly accelerated at the start of colonial rule. We call this default scenario the Colonial Rule Start (COLRST) model. For a minority of ethnic groups, for which the ethnographic accounts justified different starting time for MC frequency increase, we used an ethnic group-specific t₁, rather than the general t_1-colrst. At time t₁, we assumed the MC frequency c₁ to be in a (c_1-low–c_1-high) plausibility range. If the ethnographic accounts stated that MC was unknown or not practiced, we assumed this range to be 0–5%, and if they stated that MC existed but was not universal, or practiced late in life (after age 25), we assumed other plausibility ranges based on the ethnographic statements.

Between t_min and t₁, MC frequency was constant at c(t) = c₁, and then, either: i) increased following a logistic curve saturated at 1, and passing through the point (t_d, c_d) (COLRST-LOGIS, our default model); or ii) increased linearly between the points (t₁, c₁) and (t_d, c_d), and then remained constant until t_max (COLRST-LIN model). In the logistic model, we reasoned that any cultural pressures causing adoption of MC will tend to produce universal MC in the long term, and so we used logistic curves saturated at 1 in the distant future, and obtained their parameters analytically as follows.

The logistic progression of MC over time, c(t) is given by:

For a period t₀–t₁, we have estimates of MC frequency at its extremes, c₀ and c₁ (either based on ethnographic data or in DHS). Assuming that the logistic growth curve passes by the two points, (t₀, c₀) and (t₁, c₁), we can write:

The logistic parameters, k and t_m, can then be obtained by solving the above two equation system. The result is: Where and

To model a likely acceleration of MC adoption during the urbanization boom (mainly in the period 1950–60), we defined a further Colonial Rule Start and Urbanization (COLRSTURB) model as follows. We used a t_1-colrst and a c₁ as in the previous two models, and a t_2-urb (sampled from the 1950–60 range). We calculated the overall implicit exponential growth rate of MC between t_1-colrst and t_d; we assumed that MC grew exponentially by half that rate in the period t_1-colrst–t_2-urb, and assumed logistic growth (saturated at 1) in the period t_2-urb–t_max.

S1 Fig illustrates the interpolation of c(t) for three ethnic groups, under the three models.

We ran 1,000 different calculations for each of the 3 models described above. For each group, we sampled c₁ from its plausibility range (c_1-low–c_1-high) using Latin Hypercube Sampling (LHS). We performed all our calculations with the R programming environment (http://www.r-project.org/ [102]). All the ethnic group-specific data, estimates, and underlying references, are given in S2 Dataset. The source code is supplied in S1 Computer Code.

Estimating male circumcision frequency per city

For all the cities included in the study we searched for tables of ethnic partitioning of the population (TEPP) in the demographic (including censuses), historical, or other social science literature, at different moments of the 20^th century. We found 24 TEPPs based on census data read from either printed censuses or from articles and books referring them. We found three TEPPs based on published surveys of particular cities, and one based on expert estimates. For several cities no direct city-level TEPP data or estimates were available; in these cases we relied on four tables based on census data of the whole urban population of the administrative division to which the city belonged (Korhogo, Bondoukou, Man, and Dabou); and on two tables based on published estimates for the whole administrative division (Bobo-Dioulasso and Banfora). For Buchanan, Saint Louis, and Nouakchott no tables were available. We assumed for Buchanan a TEPP similar to nearby Monrovia; for Saint Louis, we assumed a TEPP similar to Senegal as a whole, without the Casamance region; for Nouakchott we assumed a TEPP similar to Mauritania as a whole. This information is summarized in S3 Dataset, which lists all relevant TEPPs and HIV-2 prevalence rates. In S4 Dataset this information is detailed per ethnic or regional groups. S2 Text lists the references that support these data. S1 Protocol contains permissions to cite personal communications given by their authors.

For many TEPPs, sources often mentioned broad ethnic or regional categories encompassing several ethnic groups; in such cases we subdivided them into the component ethnic groups following their proportions in the country population. Entries labeled as "Others" were subdivided into all groups belonging to the country and with no direct entry in the TEPP. To accommodate this we built a table of functional regions (corresponding to either entire countries or parts of countries), with aggregates of ethnic groups weighted by their population numbers (S5 Dataset). Entries referring to a foreign nationality were subdivided in all ethnic groups of the referred country proportionally to their numbers. Ethnic proportions in countries were based on the Ethnologue database [55]. For several ethnic groups known to be present in a city at a given time, often no data was given; in such cases we estimated them based on their proportion in the same city at another time or, when using another method of estimation, we justify it explicitly in comments in S4 Dataset.

We estimated MC frequency for the cities at several time points: 1930, 1950, 1970, 1980, and 1988. For each estimation, we used the TEPP nearest in time available for the city, and computed its MC frequency as the average of the MC frequencies of its ethnic groups, weighted by their numbers in the city. We performed 1,000 calculations for each city, time, and calculation mode, based on the ethnic groups' calculations. For each calculation mode and time, we obtained 1,000 calculated values of MC frequency per city.

Correlation between MC frequency and HIV-2 prevalence

Correlations (Spearman’s rho) were calculated on logit-transformed estimates of MC frequency and HIV-2 prevalence. In a sensitivity analysis, we repeated the correlation tests on several subsets of the cities. As a result of LHS sampling of the parameter values, 1,000 measures of the correlations were obtained for each studied year and interpolation model; we present interquartile ranges (IQR) of the correlation coefficients.

Ethics statement

The study involved a re-analysis of published HIV prevalence data; individual patient data were not included in the analysis. This research was approved by the Commissie Medische Ethiek of the Katholieke Universiteit Leuven, approval number S59252.