Table 1 summarizes the MPXV isolates used in our phylogenetic analysis and provides references for the original description of the cases; they include (a) four isolates that correspond to MPXV obtained from outbreaks recorded in laboratories; (b) the five isolates included in the study by Likos et al.; (c) twelve isolates available from human case reports in Sub-Saharan Africa between 1970 and 2010; (d) one isolate obtained from a squirrel from Yambuku, DRC, which is the only MPXV isolate derived directly from wildlife included in this study; and (e) 23 isolates from human cases reported in Sankuru District, DRC between 2006 and 2007 [31]. In all, the 45 analyzed isolates cover the known range for MPX in Central and West Africa (Figure 1). Isolates form Nigeria, Cameroon, and Gabon are of particular interest for the present analysis since they are closer to one of the proposed biogeographic barriers (Cameroon Highlands) than the rest of the isolates from Western Africa or Congo Basin.

Genomes were sequenced using Sanger sequencing or Illumina^® (Illumina Inc., San Diego, CA USA) paired end sequencing. Whole genomes of MPXV isolates were aligned using MAFFT v7.017 [32]. Cowpox virus (Grisham 1990, X94355) and horsepox virus (Mongolia 1976, DQ792504) isolates were included as outgroup taxa. The original alignment was 241,258 bp in length. The first 25 kb and last 26 kb are highly variable between CPXV and other Orthopoxviruses since they contain a large number of indels; thus, they were trimmed. Subsequently, all columns containing indels were removed, resulting in an alignment of 173,804 bases from the central conserved region of the genome. A majority-rules consensus tree was estimated from the alignment of all genomes using MrBayes v3.2.2 [33,34]. Settings included a general time reversible model (lset nst = 6) with estimated stationary state frequencies and substitution rates, and a model of gamma-shaped rate variation across sites (rates=gamma). The tree search was carried out over five million generations.

A patristic distances matrix was obtained from the consensus tree and separated into the groups representing the two recognized MPXV clades. We tested the distance values within each group for normality via the Shapiro-Wilk test, compared their variances via the F-test and compared the two groups using a Student’s t-test; all statistical test were performed in R 3.1.1 [35]. MEGA v6.06 [36] was used to calculate within group uncorrelated p-distances for a subset of samples from the Lomela Health Zone in DRC.

Environmental data: we used a set of seven diverse and relatively uncorrelated variables [37] from the bioclimatic dataset of WorldClim [38] used in a previous monkeypox ENM study [39]: annual mean temperature, mean diurnal range, maximum temperature of the warmest month, minimum temperature of coldest month, annual precipitation, precipitation of the wettest month, and precipitation of the driest month. These variables were obtained from the WorldClim website (http://www.worldclim.org/) in their digital format at a spatial resolution of ~4 km (2.5 min) for the Last Glacial Maximum (LGM, ca. 21,000 years ago) and Mid-Holocene (MidHol, ca. 6000 years ago) periods; and ~1 km spatial resolution (30 second) for the Last Interglacial period (LIG, ca. 135,000 years ago) [38,40]. A buffer of 5 degrees (~555 km) of radius was created from all monkeypox case localities (see below) and was used to delineate the area in which ENMs is calibrated.

Human case data: we used reports of human MPX from the World Health Organization (WHO) surveillance efforts in DRC between 1970 and 1986. Case localities were georeferenced matching the patient’s village of residence to digital versions of 1:250,000 Joint Operational Graphic (JOG) topographic maps (GNS; http://earth-info.nga.mil/gns.html/) following the georeferencing procedures from MaNIS [41]; details of these procedures are included in a separate publication [39]. The aim of the present study is to identify the environmental conditions required for the disease to be transmitted from its wildlife reservoir to humans; therefore, cases suspected to have resulted from secondary transmission (i.e., the disease acquired by contact with a sick person) were not included in the analysis. In all, 90 unique localities representing the entire geographic distribution of human MPX occurrence were used (white circles in Figure 3A,B). Given that most of the localities were obtained from the active surveillance efforts at the end of the smallpox eradication, this data is biased with better representation of localities in DRC. To account for this bias, we produced 25 subsets of the original dataset by reducing the density of the reported localities in DRC; specifically, we selected subsets of the original dataset in which localities were separated by a distance of 30 min (Approx. 50 km); this process produced datasets containing 41–43 localities.

Ecological niche algorithms: two of the most widely used ecological niche modeling algorithms (GARP and Maxent) have also been used for the study of disease transmission [22,23,26,42,43,44,45,46,47]. Both algorithms search for non-random associations between environmental conditions and the localities where the disease has been recorded; however, models may differ from one algorithm to another in their ability to estimate distributions [48] and to transfer such models into other environmental conditions or geographic areas [49], therefore, independent models were generated with each algorithm.

GARP v1.1.6 produces a model consisting of a set of rules that describe associations between environmental conditions and disease transmission events [50,51]. These models are built via an iterative process of creation, evaluation, modification, and inclusion/exclusion of rules of four basic forms (bioclimatic, atomic, negated, and logistic regression rules). This process stops when the optimization parameter changes by less than 1% from one iteration to the next, or when the maximum number of iterations is reached (1000). For each of the 25 subsets of localities, we ran 100 replicates using 50% of occurrence points to train models on environmental conditions in the present and selected the replicates with the highest performance via the “best subset” consensus approach: retaining the 20% with the lowest omission error and, from this subset, the central 50% in terms of commission error [52]. The models selected by this process (n = 10) were summed and evaluated using the modified ROC (E = 10%, 1000 bootstrap iterations and 50% of points), in which models performing better than random have AUC ratio values higher than 1 [53]. For a particular subset of localities, we selected the areas containing 90% of the training localities to represent the distribution of suitable environments for MPX transmission and created a binary map where 1 = suitable and 0 = unsuitable. Finally, the areas selected from each subset of localities were combined to find areas of agreement by summing the binary maps into one map with values ranging from 0 to 25.

The Maxent algorithm is based on the idea that the best solution for an unknown phenomenon maximizes entropy; thus, it calculates a probability distribution that is closest to a uniform distribution, constrained by the parameters calculated for the locality cases and a regularization parameter (β) that prevents the estimated mean values of the distribution from deviating from the observed mean value [54,55]. We used default settings in Maxent v3.2.1 (i.e., regularization multiplier = 1.0, 1500 maximum iterations, 10,000 background points, convergence limit = 10⁻⁵) to create models for each locality subset. We evaluated the models using the modified ROC as described above and selected the areas containing all localities used for training the model for each of the 25 subsets of localities to create a binary map; these maps were combined to find areas of agreement as described above.

ENMs from both algorithms were projected into the environmental conditions during the Mid-Holocene, Last Interglacial and Last Glacial Maximum periods using the environmental datasets described in the previous section. Areas with suitable environments were selected using the methodology described above for each algorithm. The final maps for each period are the combination of the models produced by all 25 locality subsets.

Results of the phylogenetic analysis are shown in Figure 1. The average standard deviation of split frequencies was 0.001,718. Branch lengths between MPXV isolates are shown to scale. The Bayesian analysis grouped 11 isolates in the WA clade and 34 in the CB clade. Geographically, the WA clade isolates correspond to those cases reported from Nigeria, Ghana, Liberia, Cote d’Ivoire, and Sierra Leone, including the four isolates from outbreaks in captive animals; while the CB clade includes all isolates from DRC, Republic of the Congo, Cameroon, and Gabon. Although four polytomies resulted from the phylogenetic analysis, node support was very high (0.99–1.0) throughout most of the tree except for the polytomy that includes groups I-IV, which has support of 0.868. Eight isolates from Lomela Health Zone are included in a polytomy within group III (JX878413, JX878414, JX878415, JX878416, JX878421, JX878422, JX878427, and JX878428), with an average distance between them of 0.002%. Comparison of within clade patristic distances showed that branches in the CB clade are significantly shorter than branches within the WA clade (Shapiro-Wilk p < 0.001 in both clades, F-test p = 0.001, and t-test [unequal variance] p < 0.001). The two Nigerian isolates group together in spite of having high genetic differentiation between them and are from a group that is sister to the one formed by the remaining isolates of the clade; in addition, all isolates obtained from captive animals form a single monophyletic group.

Within the CB clade, five monophyletic groups were formed after the phylogenetic analysis: group I includes the two western most isolates from this clade: Cameroon 1989 and Gabon 1987; group II contains 11 isolates, five from Sankuru District (JX878407, JX878423-25, and JX878429), one from South Sudan (Sudan 2005), one from the Republic of the Congo (Impfondo 2003), and four from other parts of DRC (Boende 2008, Bokenda 1970, Yandongi 1986, and Mindembo 1979); group III is the closest sister group to group II, twelve of the thirteen isolates in this group are from Sankuru District (JX878409-16, JX878421-22, JX878427-28), the remaining one is from Yambuku; six of the seven isolates in group IV are from Sankuru District (JX878408, JX878426, JX878417-20, and Sankuru 1996), the remaining one is Ikubi; group V only has one isolate from Sankuru District (JX878417). The groups are indicated in Figure 2 and Table 1.

ENM projections into present day environmental conditions are similar to models presented in previous works [25,39] and identify areas in Central and Western Africa in which MPX is known to occur. Models based on each of the 25 locality subsets performed better than random expectations with modified ROC AUC values between 1.31 and 1.42 for GARP (average = 1.37, standard deviation = 0.03); and between 1.33 and 1.48 for Maxent (average = 1.42, standard deviation 0.04). These models are able to capture those environmental conditions that are suitable for MPX transmission with high confidence. ENM maps presented some differences: (a) GARP models predicted a larger area of suitable environmental conditions in Cameroon and Gabon than the models produced by Maxent; (b) Maxent models show a more uniform distribution of suitable conditions in West Africa, while GARP shows fragmentation; and (c) GARP models show an area of low model agreement in the middle of the Congo River Basin (Figure 3A,B).

GARP model projections onto climatic conditions during the Mid-Holocene show more connection between areas with suitable environments for MPXV in West Africa and also between those suitable areas in Nigeria and Central Africa; the most evident discontinuity of suitable areas separates such areas in costal Nigeria from the rest of West Africa and is located in Benin (Figure 3C). Maxent, however, shows a noticeable reduction of suitable areas in West Africa concentrated in the coasts of Nigeria, Togo and Benin with some smaller suitable areas in Sierra Leone, Liberia and Ghana (Figure 3D). ENM projections onto environmental conditions during the LGM with both algorithms identified smaller areas with suitable environments for MPX transmission. The area with highest model agreement for Maxent is located in western DRC with a few other areas with lower model agreement within this country; however, only few small areas in West Africa were identified as suitable by a few models (Figure 3F). Areas of high model agreement for GARP are also restricted to small areas in northern and western DRC, as well as along the coast throughout West Africa (Figure 3E).

Finally, ENM projections onto environmental conditions during the LIG identified a larger and more continuous area of model agreement for both algorithms than that for the LGM located in central Africa with a few small patches along the coast in West Africa for GARP and Gabon for Maxent (Figure 3G,H).