Secondary Logo

Journal Logo


Spatial accessibility and the spread of HIV-1 subtypes and recombinants

Tatem, Andrew J.a,b,c; Hemelaar, Jorisd; Gray, Rebecca R.e; Salemi, Marcob,f

Author Information
doi: 10.1097/QAD.0b013e328359a904



HIV-1 originated in west-central Africa via cross-species transmission from chimpanzees around the beginning of the twentieth century, and has since then diversified in the human population [1–3]. Today, the most prevalent group of HIV-1 is the main (M) group, which has been divided into subtypes (A–D, F–H, J–K), some of which are further divided into sub-subtypes (A1–A4, F1–F2). Recombination between strains has resulted in the generation of 52 circulating recombinant forms (CRFs) (, which together are estimated to comprise up to 20% of all global HIV infections [4]. In the second half of the twentieth century, the global spread of HIV-1 group M resulted in a complex global distribution of HIV-1 subtypes and recombinants [4,5]. HIV diversity impacts HIV diagnosis and viral load measurements and may affect the response to antiretroviral treatment and the emergence of drug resistance [6]. Subtypes may differ in the rate of disease progression and some evidence suggests that subtypes are transmitted at different rates [7,8]. Moreover, HIV diversity, in populations and in individuals, is one of the major challenges in HIV vaccine development [9]. It is, therefore, essential that the HIV subtype distribution is monitored and that we understand the factors that drive the global spread of HIV variants [10].

The greatest HIV-1 diversity is found in Africa where all known subtypes, many unique sequences and many CRFs and unique recombinant forms (URFs) circulate and form distinct spatial patterns. Explanations for the distribution patterns observed are probably multifactorial and include founder effects, population growth, urbanization, transport links and migration. It is uncertain at present whether biological properties of different subtypes and recombinants play a role in their differential spread. However, transportation links and related human migrations and movements have been consistently cited as a strong driver of the spread of HIV, with recent studies underlining their major role [11].

Human mobility is a key driver of disease spread across multiple spatial and temporal scales [12–14]. As air, sea and land transport networks continue to expand in reach, speed of travel and volume of passengers carried, pathogens are now able to move further, faster and in greater numbers than ever before. The spread of the HIV/AIDS pandemic worldwide is a travel story whose episodes can be traced by molecular tools and epidemiology [15–18], and studies have shown that certain groups of mobile and sexually active individuals including immigrants, IDUs, tourists, truck drivers and military troops are important in seeding local epidemics [19,20]. The ‘accessibility’ (or ease by which these groups can move around a region [21,22]) through dense and efficient transport networks often result in higher infection rates near main roads and in trading centers than in isolated villages [23,24]. However, the role of such accessibility on the spread of HIV-1 subtypes and recombinants over large scales has never been quantified.

Here we present a spatial analysis of the mainland sub-Saharan Africa-wide distribution of HIV-1 subtypes and recombinants for the period 1998–2008 using cross-sectional HIV-1 molecular epidemiology data [4], and examine the role of transportation networks and the landscape in explaining the observed patterns.


HIV-1 subtype and recombinant distribution data

Data on the distribution of HIV-1 subtypes and recombinants (CRFs and URFs) in sub-Saharan countries were obtained from researchers in the field and from a comprehensive literature review as previously described [4]. Briefly, research laboratories across the globe specializing in subtyping of HIV-1 samples were solicited for cross-sectional HIV-1 subtyping data of samples collected between 1998 and 2008. The MEDLINE literature database was searched for HIV-1 subtyping data for each country using the terms ‘HIV’, ‘subtype’ and the relevant country names. All potentially relevant articles in English were retrieved and HIV-1 subtyping data from samples collected between 1998 and 2008 were included in our study. Data submitted to us and derived from the literature were combined to determine the overall proportions of HIV-1 subtypes and recombinants at each sampling site in the period 1998–2008. All CRFs, apart from CRF01 and CRF02, were pooled together; thus, any additional variant diversity within the CRF and URF categories was not included in the analyses presented here. There was insufficient data to allow temporal analysis of subtype distributions at each site. Where possible, each sample was georeferenced through assignment of latitude and longitude coordinates. Only samples that could be assigned to a single location were used in the analysis and aggregated samples that were obtained from multiple surveys across wide areas were removed. The remaining samples are mapped in Fig. 1, features of them are documented in Table 1[4] and source references are provided in Supplemental Digital Content ( In total, samples from 77 individual locations and 6873 individuals across 29 of the 42 countries in mainland sub-Saharan Africa were used.

Fig. 1:
The spatial distribution of HIV-1 subtype and recombinant samples in mainland sub-Saharan Africa overlaid on a map showing estimated travel times of less than 5 h to the nearest city of more than 100 000 population.CRF, circulating recombinant form; URF, unique recombinant form.
Table 1:
Summary of the data used in the study.

Transport infrastructure data

Spatial datasets on Africa-wide road networks [GRoads,; VectorMap0,; and National Transportation Network Geographical Information System (GIS) datasets from Kenya, Namibia, Tanzania, Swaziland, Rwanda, Niger, Zambia, Angola, Somalia and Djibouti, see Supplemental Digital Content (], land cover [25], settlement locations ( [22], inland water bodies [26] and topography [27] were obtained and assembled within a GIS. The datasets were the most detailed and complete available, and all were constructed within the last 20 years to represent conditions during this period. Although these, therefore, do not necessarily match the periods of greatest HIV-1 spread that may have occurred earlier, the vast majority of the travel routes across sub-Saharan Africa in place today, and mapped by these datasets, were developed or initiated during colonial times in the first half of the twentieth century [28]. These datasets formed the basis for calculating road distances between samples, and constructing the ‘friction’ surface used to calculate ‘access’ distances (see next section).


The similarity between sample HIV-1 subtype distributions across the continent was quantified through calculation of root mean differences. For samples A and B with subtype proportions of AA, BA, …, AURF, BURF (see Fig. 1 for full list of subtypes), the root mean subtype difference (RMSD) is as follows:

RMSD is 0 if the subtype proportion distribution is identical between samples A and B, and is more than 0.35 if the distributions are completely different. RMSD was calculated between each sample and every other sample to create a RMSD matrix. The RMSD measure provided a metric to compare against differing candidate measures of distance between samples to examine the extent to which regional distance and accessibility could explain the patterns in subtype distributions seen.

Candidate measures of distance calculated between samples were great circle distance; road distance; and ‘access’ distance. First, great circle distance (straight line or linear distance between samples that takes into account the curvature of the Earth) between sample locations were calculated using ArcGIS 10.1 (ESRI, Redlands, California, USA) to create a matrix documenting distance between each sample and every other sample. Second, distances on roads classified as primary or secondary between sample locations were calculated. The networks of these routes are shown in Supplemental Digital Content (, and shortest distances between samples along them were calculated using ArcGIS 10.1 again. Following travel speed assumptions of previous approaches [29,30], the secondary road distances were increased by a factor of one-third, based on the assumption of increased ease and capacity of travel on primary roads compared to secondary roads.

Third, an ‘access’ distance matrix was calculated based on a combination of GIS datasets. ‘Accessibility distance’ is a measure of friction between one location and another that takes into account land cover types, transport network and gradient. It is generally thought to be a more representative measure of ease of human travel across a landscape than simple linear distance, due to compensating for impedances to travel. Following previously developed methodologies [29,30], land cover, water bodies, slope and road network datasets were combined in a 1-km spatial resolution grid and empirically derived travel speeds [29,30] were assigned to each land use type and modified based on topography to create a ‘friction surface’ for Africa (shown in the Supplemental Digital Content, Country borders were not included as a travel impediment in this case, as national borders in sub-Saharan Africa are often porous, and present few obstacles to crossing. The friction surface can be converted to a map of estimated travel times to features of interest, and Fig. 1 shows travel times of less than 5 h to the nearest settlement of more than 100 000 people mapped out.

To calculate ‘access’ distances between sample locations across the friction surface, electrical circuit theory was used. Previously, least-cost path modeling has been used to derive estimates of travel time across friction surfaces [31–33], but such models assume that a single, optimal pathway is always taken, when in reality multiple pathways are often taken due to personal preferences, transport availability, changing route conditions and costs. The isolation by resistance model [34,35] uses a graph-theoretic distance metric based on circuit theory to simultaneously consider all possible pathways connecting sample location pairs, thus overcoming the limitations of least cost path modeling. Here, the isolation by resistance model was implemented using the Circuitscape tool (, [35]) to create a matrix of access distances across the friction surface from each sample location to every other sample location.

To quantify the relationships between differences in subtype distribution patterns and the different measures of distance, generalized linear models for each country were constructed, considering the RMSD to be the mean of a Gaussian process. Each country model compared the RMSD values between samples from within its borders and from neighboring countries with the corresponding distance measures. This meant that only regional distances and accessibility were examined and, thus, relationships between samples at long distances from each other that realistically should show little spatial interaction with one another were removed from the analyses. The deviance from a null model was used to examine the additional explanatory power that each type of distance metric had in describing the subtype distribution patterns observed.

Finally, the relationship between subtype diversity at each location and its relative accessibility to surrounding areas was assessed through first calculating the inverse Simpson diversity index [36], S, for each sample location (i). This was calculated as follows:

in which p is the proportion on a 0–1 scale of subtype j. The resulting index values for each location are mapped in the Supplemental Digital Content ( These values were compared against the mean friction values within a 500-km radius of each sample location, which provided a relative measure of accessibility to and from the sample location from the surrounding region.


Spatial autocorrelation in HIV-1 subtype and recombinant distribution patterns is clearly evident (Fig. 1). Overlaying these onto a map of estimated travel times to large settlements shows how clusters of similar subtype distributions are well connected and easily accessible from one another, whereas regions of low accessibility separate groupings of similar subtype distributions. Particularly, we find high levels of subtype similarity and connectivity in western Africa where CRF02_AG predominates in most sites and the travel ‘friction’ is generally low (Fig. 2a, the full friction surface dataset can be seen in Supplemental Digital Content,, and also in southern Africa and Ethiopia where subtype C near-exclusivity occurs. This can be seen, for example, for Ethiopia in Fig. 1, where the further away samples are from those in the country, the more different the subtypes and proportions are, whereas the closely clustered samples within the country are very similar. Moreover, a strong transport network and a majority of subtype A is seen in eastern Africa. In central Africa, substantial HIV diversity corresponds to relatively poor transport connectivity, poor standards of roads (Supplemental Digital Content, and much higher landscape friction (Fig. 2b, Supplemental Digital Content, Such patterns suggest that the transport networks across sub-Saharan Africa, and the ease of traversing them, are linked to the subtype distributions seen, and statistical analyses provide further evidence for this link.

Fig. 2:
Friction surface maps.(a) a region of coastal west Africa, covering Cote d’Ivoire in the west, to Nigeria in the east; (b) a region of central Africa, covering Gabon in the west to the Democratic Republic of Congo in the east. In both maps, the lowest friction (least difficult terrain to cross, e.g. roads) is shown in dark blue, and the highest friction (most difficult terrain to cross, e.g. rainforest) in red, with values on a unit-less relative scale representing the time taken to cross each 1-km grid square.

For the vast majority of countries, the incorporation of straight line (linear) distance as an explanatory variable for the subtype distribution differences seen produces substantial reductions in deviance compared with a null model (Fig. 3). Moreover, accounting for transport infrastructure in terms of road distance explains more of the variation seen, and additionally accounting for road quality, topography and land cover in the form of ease of access between samples [quantified using landscape friction (Fig. 2, Supplemental Digital Content,] consistently improves the power of the models further, giving deviance reductions of up to 80%. An illustrative example of this is Kenya, where the pattern of subtype similarity changing with distance to surrounding country samples can be seen (Fig. 1), but the relationship is nonlinear, and the change in subtype proportions by distance is actually lower between samples connected by major roads than those that are not, highlighting the importance of accounting for transport routes in describing the patterns seen. Significant relationships at P value less than 0.05 level were found using access distance for 22 of the 29 countries examined, with many of these relationships significant at P value less than 0.01 level and the remaining countries generally showing relatively small sample sizes.

Fig. 3:
Percentage reduction in deviance from a null model through the addition of different distance variables in explaining subtype and recombinant distribution patterns seen for each country with sufficient samples within and surrounding them.The asterisks indicate the significance of the relationships, with *P < 0.05 and **P < 0.01.

It appears clear that the construction of a more realistic landscape and use of a modeling approach that can account for multiple movement routes likely approximates movement patterns relevant to HIV-1 spread more accurately than simple distance. This is evidenced by the size of deviance reductions between the linear distance model and access distance model, which show substantial variations by country. This is related to the road connectivity, density and quality within and between countries that determine their overall accessibility [22] (Fig. 2, Supplemental Digital Content, For instance, the well connected and densely populated coastal west African countries show only small improvements over linear distance when using access distance, compared with other more sparsely populated countries such as Tanzania.

Spatial accessibility also appears to play a role in determining subtype diversity in an area, with the difficulty in traversing landscapes surrounding them being significantly higher (Mann–Whitney U-test, W = 331, P = 0.0042) for locations with relatively high subtype diversity (Simpson's inverse diversity index >2.5) than those with lower diversity (Supplemental Digital Content, Although the relationship between diversity and accessibility is not especially strong (R2 = 0.1129), it is significant (F = 8.8, P = 0.00417) (Fig. 4, Fig. S2 shows the friction map). This relationship can be illustrated through comparing South Africa with the Democratic Republic of Congo (DRC) in Fig. 1. The road network in South Africa is relatively dense and efficient, and has remained so for many decades, promoting substantial population movements and, thus, also the ease of HIV-1 subtype spread, meaning that subtype C likely spread rapidly upon introduction. In contrast, travel around the DRC has always been challenging due to poor infrastructure (Fig. 1), likely restricting the spread of HIV-1 and maintaining subtype diversity (Fig. 4).

Fig. 4:
Scatterplot showing the relationship between the Simpson index of diversity for each sample location and the mean friction value within a 500-km radius of each sample location.The linear regression fit is shown: R 2 = 0.1129, F = 8.8 and P = 0.00417.


Previous studies on local scales have shown that transport infrastructure and human mobility are strongly related to HIV spread in Africa, and are likely major driving forces [20,23,24]. Moreover, previous phylogenetic studies [11] have also postulated on the importance of travel accessibility in driving the spread of HIV subtypes in eastern Africa. Here, for the first time, we have examined the role of transport infrastructure and accessibility at regional scales, and confirmed its importance across sub-Saharan Africa. We find clustering of certain subtype distributions in well connected regions of sub-Saharan Africa (western, eastern and southern Africa, and Ethiopia), which are separated by areas of limited connectivity, and quantify the importance of spatial accessibility in explaining these patterns.

Zoonotic transmission of HIV-1 group M to humans is thought to have occurred at the beginning of the twentieth century in south-eastern Cameroon [2,37]. From here infected individuals travelled to Kinshasa, DRC and it is thought that the rise of cities, and consequently large congregations of people, in the first part of the twentieth century facilitated the initial establishment and early spread of HIV-1 [2]. The relatively poor connectivity in central Africa likely contributed to the slow initial growth of the epidemic in the first half of the twentieth century as well as the relatively low HIV prevalence rates (around 5%) observed in central Africa throughout the course of the HIV pandemic, even when HIV prevalence rates soared in southern Africa [11,38]. The diversification of HIV-1 group M into different subtypes took place in central Africa in the first half of the twentieth century [2,39]. The global spread of HIV subsequently took place in the second half of the century, leading to the differential global and regional spread of HIV variants. Phylogenetic studies have shed some light on the history and spread of individual subtypes in Africa. For instance, subtypes A and D arrived in eastern Africa in approximately 1950s and 1960s, respectively [11], whereas subtype C started the epidemic in southern Africa around 1970s [40], and from here travelled to Ethiopia around 1982 [41]. The relatively poor accessibility of eastern and southern Africa to the epicenter of the epidemic in central Africa might explain why it took some time for HIV to reach these areas, whereas the good connectivity within eastern and southern Africa helps explain the explosive growth of the epidemics seen there [11,19].

Large variations in the size of deviance reductions between countries can be seen (Fig. 2) and, in some locations, big differences in subtype distributions between samples separated by small distances (Fig. 1). In some cases, the subtype distributions for a country and its surrounding region appear to be unrelated to transport and distance factors. This likely occurs when, first, a country is especially poorly connected, both domestically and to surrounding countries, for example in the case of the DRC, making the influence of distance beyond a certain local-scale threshold irrelevant and, second, where additional factors, such as ethnic differences or seasonal migration for which detailed data Africa-wide do not exist, likely have a larger impact than transport connectivity, for example in the cases of Nigeria, Niger and their surrounding countries [42,43]. Moreover, although the spread of HIV subtypes has occurred over a period of decades, spatial datasets representing single time points have been used here due to data availability. Thus, it is likely that some of the unexplained variance in the distance model fits could be due to temporal mismatches between the timings of HIV spread and the years represented by the transport network data used here. Temporal changes, including the development of new roads or decline in old transport modes, such as the deterioration of the efficient river and rail-based systems used in colonial times in the DRC [44], are not represented in the spatial datasets.

Variations in deviance reductions between countries are also likely the result of factors relating to the underlying subtype data. It is important to recognize that the proportions and distributions of subtypes and recombinants in many African countries remain unclear as a result of the small numbers of HIV-1 samples that have been subtyped and the potential bias in the way samples have been collected [4]. Some countries and regions had small sample sizes, particularly those which harbor the largest number of infections and have the highest subtype diversity (Table 1). Sampling biases may have occurred due to patient selection (risk groups, treatment failure and disease progression), limited geographical coverage and consistency within countries and unknown dates and places of infection. In addition, subtyping methods used and the type and number of genome segments analyzed may have affected results. HIV-1 subtyping may have been based on one or a few genes, and recombinant strains may have been missed [45]. Moreover, the pooling of CRFs and URFs undertaken here likely means that the diversity measured is an underestimate for many samples. Publication bias may also have occurred in the data derived from the literature. Finally, changes in the subtype proportions across the period of the dataset assembled here may have occurred, although a recent analysis [4] found that the global distribution of HIV-1 subtypes was broadly stable over the 2000–2007 period. Despite this set of uncertainties, the coherent patterns of subtype distributions (Fig. 1) and resemblances to other analyses of sub-Saharan African human and pathogen movement [43,46] suggest that the data used here is a representative sample of subtype distributions across the continent.

Ideally, these analyses should be extended to undertake more complete multivariate tests of the drivers of the subtype distribution patterns, with the inclusion of additional variables on such factors as prevalence, urbanization, wealth, ethnicity, religion and migration, as well as factors influencing the transmissibility of HIV, such as male circumcision, sexual networks and concomitant infections. However, at present, such variables do not exist at the spatial scales and levels of coverage to match the HIV-1 subtype sample data assembled, and their inclusion would require the summarization of sample data to national-level analyses, masking much of the subnational variation seen in Fig. 1. Moreover, to undertake a more complete analysis here, ideally data on actual volumes and flows of human travel between sample locations should have been included as a model variable. Data on human movements, especially in resource-poor settings, remain sparse. Novel approaches and datasets are however being developed to enable the derivation of quantitative movement estimates at a range of spatial and temporal scales. These include mobile phone call data records [47,48], Global Positioning System technologies [13,49] and satellite nightlights [50], with accompanying modeling frameworks under development ( Future work will build on the methodologies and findings here, and include the incorporation of modeled population movement estimates; the development of gravity-type or radiation [51] spatial interaction models parameterized by these movement data; and the extension of these analyses beyond Africa and to other pathogens. One possible avenue for extension of this work involves the development of simple predictive models of subtype and recombinant spread, given planned transport network developments. Although substantial uncertainty would remain in predictions made, for specific areas, the likely directions in terms of subtype composition similarity change toward nearby regions that new roads are connecting them to could be modeled to aid strategic planning for the health impacts of transport infrastructure projects. The deterministic-type models presented here should ideally be extended to include stochastic factors though that takes into account the more ’random’ long distance movements [52] that can bring new subtypes to an area.

The global and regional diversity of HIV continues to generate new recombinants in regions where multiple variants co-circulate [9]. The increased travel and mobility of people may lead to the accelerated spread of new variants and the further diversification of the global HIV epidemic. HIV genomic variability continues to pose challenges for diagnosis, viral load measurement, treatment and resistance development and vaccine development. Continued and increased efforts to monitor HIV molecular epidemiology are, therefore, required to inform prevention and treatment programs. As much of Africa undergoes rapid population growth and economic development, with many new transportation projects underway or planned [53], the spatial accessibility landscape of the continent will continue to change. The results here suggest that a consequence of this will be future substantial changes in the spatial epidemiology of HIV-1. Further, the results point toward the wider implications of disease spread and circulation in Africa. Recent analyses of the distribution of malaria drug resistance markers show a similar spatial pattern to those in Fig. 1[46], highlighting that accessibility and mobility likely drives the spread of other pathogens and the spread of resistance to measures to control them [54]. A comprehensive understanding and evidence base on accessibility, travel and mobility in resource-poor settings would, therefore, provide a valuable resource for the strategic planning of disease control.


A.J.T. designed the study, undertook data analyses and wrote the manuscript. J.H. undertook data collation, provided input to the study design and helped write the manuscript. R.R.G. and M.S. provided input to the study design and helped write the manuscript. All authors contributed to writing the manuscript.

The authors thank Peter Ghys of the Joint United Nations Programme on HIV/AIDS for critical input on the manuscript.

This paper forms part of the output of the Human Mobility Mapping Project (

A.J.T. acknowledges funding support from the RAPIDD program of the Science and Technology Directorate, Department of Homeland Security, and the Fogarty International Center, National Institutes of Health, and is also supported by grants from the Bill and Melinda Gates Foundation (#49446 and #1032350).

J.H. is a recipient of an Academic Clinical Fellowship from the National Institute of Health Research, UK. R.R.G. is supported by a UK Medical Research Council Fellowship Award.

Conflicts of interest

There are no conflicts of interest.


1. Korber B, Muldoon M, Theiler J, Gao F, Gupta R, Lapedes A, et al. Timing the ancestor of the HIV-1 pandemic strains. Science 2000; 288:1789–1796.
2. Worobey M, Gemmel M, Teuwen DE, Haselkorn T, Kunstman K, Bunce M, et al. Direct evidence of extensive diversity of HIV-1 in Kinshasa by 1960. Nature 2008; 455:661–664.
3. Salemi M, Strimmer K, Hall WW, Duffy M, Delaporte E, Mboup S, et al. Dating the common ancestor of SIVcpz and HIV-1 group M and the origin of HIV-1 subtypes using a new method to uncover clock-like molecular evolution. FASEB J 2001; 15:276–278.
4. Hemelaar J, Gouws E, Ghys PD, Osmanov S. Global trends in molecular epidemiology of HIV-1 during 2000-2007. AIDS 2011; 25:679–689.
5. Tebit DM, Arts EJ. Tracking a century of global expansion and evolution of HIV to drive understanding and to combat disease. Lancet Infect Dis 2011; 11:45–56.
6. Peeters M, Aghokeng AF, Delaporte E. Genetic diversity among human immunodeficiency virus-1 non-B subtypes in viral load and drug resistance assays. Clin Microbiol Infect 2010; 16:1525–1531.
7. Kiwanuka N, Laeyendecker O, Quinn TC, Wawer MJ, Shepherd J, Robb M, et al. HIV-1 subtypes and differences in heterosexual HIV transmission among HIV-discordant couples in Rakai, Uganda. AIDS 2009; 23:2479–2484.
8. Kiwanuka N, Laeyendecker O, Robb M, Kigozi G, Arroyo M, McCutchan F, et al. Effect of human immunodeficiency virus type 1 (HIV-1) subtype on disease progression in persons from Rakai, Uganda, with incident HIV-1 infection. J Infect Dis 2008; 197:707–713.
9. Hemelaar J. The origin and diversity of the HIV-1 pandemic. Trends Mol Med 2012; 18:182–192.
10. Salemi M. Toward a robust monitoring of HIV subtypes distribution worldwide. AIDS 2011; 25:713–714.
11. Gray RR, Tatem AJ, Lamers S, Hou W, Laeyendecker O, Serwadda D, et al. Spatial phylodynamics of HIV-1 epidemic emergence in east Africa. AIDS 2009; 23:F9–F17.
12. Prothero RM. Disease mobility: a neglected factor in epidemiology. Int J Epidemiol 1977; 6:259–267.
13. Stoddard ST, Morrison AC, Vazquez-Prokopec GM, Paz Soldan V, Kochel TJ, Kitron U, et al. The role of human movement in the transmission of vector-borne pathogens. PLoS Negl Trop Dis 2009; 3:e481.
14. Tatem AJ, Rogers DJ, Hay SI. Global transport networks and infectious disease spread. Adv Parasitol 2006; 62:293–343.
15. Lemey P, Pybus OG, Wang B, Saksena NK, Salemi M, Vandamme AM. Tracing the origin and history of the HIV-2 epidemic. Proc Natl Acad Sci U S A 2003; 100:6588–6592.
16. Perrin L, Kaiser L, Yerly S. Travel and the spread of HIV-1 genetic variants. Lancet Infect Dis 2003; 3:22–27.
17. Salemi M, Goodenow MM, Montieri S, de Oliveira T, Santoro MM, Beshkov D, et al. The HIV type 1 epidemic in Bulgaria involves multiple subtypes and is sustained by continuous viral inflow from west and east European countries. AIDS Res Hum Retroviruses 2008; 24:771–779.
18. Faria NR, Suchard MA, Abecasis A, Sousa JD, Ndembi N, Bonfim I, et al. Phylodynamics of the HIV-1 CRF02_AG clade in Cameroon. Infect Genet Evol 2012; 12:453–460.
19. Lurie MN, Williams BG, Zuma K, Mkaya-Mwamburi D, Garnett GP, Sweat MD, et al. Who infects whom? HIV-1 concordance and discordance among migrant and nonmigrant couples in South Africa. AIDS 2003; 17:2245–2252.
20. Salit IE, Sano M, Boggild AK, Kain KC. Travel patterns and risk behaviour of HIV-positive people travelling internationally. CMAJ 2005; 172:884–888.
21. Yoshida N, Deichmann U. Measurement of accessibility and its applications. J Infrastructure Devel 2009; 1:1–16.
22. Linard C, Gilbert M, Snow RW, Noor AM, Tatem AJ. Population distribution, settlement patterns and accessibility across Africa in 2010. PLoS One 2012; 7:e31743.
23. Boerma JT, Urassa M, Senkoro K, Klokke A, Ngweshemi JZ. Spread of HIV infection in a rural area of Tanzania. AIDS 1999; 13:1233–1240.
24. Lagarde E, Schim van der Loeff M, Enel C, Holmgren B, Dray-Spira R, Pison G, et al. Mobility and the spread of human immunodeficiency virus into rural areas of west Africa. Int J Epidemiol 2003; 32:744–752.
25. Arino O, Bicheron P, Achard F, Latham J, Witt R, Weber JL. GLOBCOVER: the most detailed portrait of Earth. ESA Bull 2008; 136:24–31.
26. Lehner B, Doll P. Development and validation of a global database of lakes, reservoirs and wetlands. J Hydrol 2004; 296:1–22.
27. Rabus B, Eineder M, Roth A, Bamler R. The shuttle radar topography mission: a new class of digital elevation models acquired by spaceborne radar. Photogram Eng Rem Sens 2003; 57:241–262.
28. Martinez AJT. Road maintenance policies in sub-Saharan Africa: unsolved problems and acting strategies. Transport Policy 2001; 8:257–265.
29. Nelson A. Estimated travel time to the nearest city of 50 000 or more people in year 2000. Ispra, Italy: Global Environment Monitoring Unit, Joint Research Centre; 2008.
30. Uchida H, Nelson A. Agglomeration index: towards a new measure of urban concentration. Washington DC: World Bank; 2009.
31. Adriaensen F, Chardon JP, De Blust G, Swinnen D, Villalba D, Gulinck H, Matthysen E. The application of ’least-cost’ modelling as a functional landscape model. Landscape Urb Plan 2003; 64:233–247.
32. Alegana VA, Wright JA, Pentrina U, Noor AM, Snow RW, Atkinson PM. Spatial modelling of healthcare utilisation for treatment of fever in Namibia. Int J Health Geogr 2012; 11:6.
33. Tanser F, Gijsbertsen B, Herbst K. Modelling and understanding primary healthcare accessibility and utilization in rural South Africa: an exploration using a geographical information system. Soc Sci Med 2006; 63:691–705.
34. Klein DJ, Randic M. Resistance distance. J Math Chem 1993; 12:81–95.
35. McRae BH, Beier P. Circuit theory predicts gene flow in plant and animal populations. Proc Natl Acad Sci U S A 2007; 104:19885–19890.
36. Simpson EH. Measurement of diversity. Nature 1949; 163:688.
37. Keele BF, Van Heuverswyn F, Li Y, Bailes E, Takehisa J, Santiago ML, et al. Chimpanzee reservoirs of pandemic and nonpandemic HIV-1. Science 2006; 313:523–526.
38. Mulanga-Kabeya C, Nzilambi N, Edidi B, Minlangu M, Tshimpaka T, Kambembo L, et al. Evidence of stable HIV seroprevalences in selected populations in the Democratic Republic of the Congo. AIDS 1998; 12:905–910.
39. Zhu T, Korber BT, Nahmias AJ, Hooper E, Sharp PM, Ho DD. An African HIV-1 sequence from 1959 and implications for the origin of the epidemic. Nature 1998; 391:594–597.
40. Dalai SC, de Oliveira T, Harkins GW, Kassaye SG, Lint J, Manasa J, et al. Evolution and molecular epidemiology of subtype C HIV-1 in Zimbabwe. AIDS 2009; 23:2523–2532.
41. Abebe A, Lukashov VV, Pollakis G, Kliphuis A, Fontanet AL, Goudsmit J, de Wit TF. Timing of the HIV-1 subtype C epidemic in Ethiopia based on early virus strains and subsequent virus diversification. AIDS 2001; 15:1555–1561.
42. Bharti N, Djibo A, Ferrari MJ, Grais RF, Tatem AJ, McCabe CA, et al. Measles hotspots and epidemiological connectivity. Epidemiol Infect 2010; 138:1308–1316.
43. Tatem AJ, Smith DL. International population movements and regional Plasmodium falciparum malaria elimination strategies. Proc Natl Acad Sci U S A 2010; 107:12222–12227.
44. Foster V, Benitez DA. The Democratic Republic of Congo's infrastructure: a continental perspective. Washington DC: the World Bank; 2011.
45. Papathanasopoulos MA, Hunt GM, Tiemessen CT. Evolution and diversity of HIV-1 in Africa: a review. Virus Genes 2003; 26:151–163.
46. Pearce RJ, Pota H, Evehe MS, Ba el H, Mombo-Ngoma G, Malisa AL, et al. Multiple origins and regional dispersal of resistant dhps in African Plasmodium falciparum malaria. PLoS Med 2009; 6:e1000055.
47. Tatem AJ, Qiu Y, Smith DL, Sabot O, Ali AS, Moonen B. The use of mobile phone data for the estimation of the travel patterns and imported Plasmodium falciparum rates among Zanzibar residents. Malar J 2009; 8:287.
48. Gonzalez MC, Hidalgo CA, Barabasi AL. Understanding individual human mobility patterns. Nature 2008; 453:779–782.
49. Vazquez-Prokopec GM, Stoddard ST, Paz-Soldan V, Morrison AC, Elder JP, Kochel TJ, et al. Usefulness of commercially available GPS data-loggers for tracking human movement and exposure to dengue virus. Int J Health Geogr 2009; 8:68.
50. Bharti N, Tatem AJ, Ferrari MJ, Grais RF, Djibo A, Grenfell BT. Explaining seasonal fluctuations of measles in Niger using nighttime lights imagery. Science 2011; 334:1424–1427.
51. Simini F, Gonzalez MC, Maritan A, Barabasi AL. A universal model for mobility and migration patterns. Nature 2012; 484:96–100.
52. Brockmann D, Hufnagel L, Geisel T. The scaling laws of human travel. Nature 2006; 439:462–465.
53. Dobson AP, Borner M, Sinclair AR, Hudson PJ, Anderson TM, Bigurube G, et al. Road will ruin Serengeti. Nature 2010; 467:272–273.
54. Lynch C, Roper C. The transit phase of migration: circulation of malaria and its multidrug-resistant forms in Africa. PLoS Med 2011; 8:e1001040.

accessibility; HIV; mapping; migration; recombinants; subtypes; transport network

Supplemental Digital Content

© 2012 Lippincott Williams & Wilkins, Inc.