In addition to using the MPXV isolates used in the phylogenetic analysis by Likos et al. (7), we included 6 more isolates: 2 of these isolates correspond to strains that cause monkeypox outbreaks in laboratory animals (Copenhagen and Walter Reed) (24), 1 from Sierra Leone (24), 1 from Yandongi in the Democratic Republic of Congo (DRC), and 2 from the Sudan outbreak during 2005. The latter 2 isolates were found to be identical after alignment and were regarded as 1 genome, MPXV_Nuria_Sudan_2005, during our analysis (Table 1, Figure 1). In total, 11 isolates were used in the phylogenetic analysis.

A maximum clade credibility tree was generated from the sequence matrix by using MrBayes (30,31) under the general time reversible + proportion invariant + Γ model with the following settings: lset nst = 6, rates = invgamma, mcmc ngen 5,000,000, sample freq = 1,000, nchains = 4, startingtree = random, burnin = 500. The model was chosen because it allows for variable base frequencies, rate variation among sites, and a proportion of invariable sites within the matrix. No priors were specified because the default priors are expected to work well for most analyses and parameters are then estimated from the data assuming no prior knowledge of their values. The final average standard deviation of split frequencies for the 2 runs was 0.00000, demonstrating convergence.

We used 2 algorithms, Genetic Algorithm for Rule-Set Production (GARP [http://openmodeller.sourceforge.net/index.php?option=com_content&task=view&id=8&itemid=4]) and Maxent (www.cs.princeton.edu/~schapire/maxent/), to create ecological niche models of monkeypox transmission by using localities where the 2 recognized clades (Congo Basin and West African) were identified in human samples. GARP is based on a genetic algorithm and is used to find a set of rules describing nonrandom associations between localities where disease transmission has been reported and environmental conditions in those localities (32,33). These rules are built and selected through an iterative process of creation, evaluation, modification, and inclusion or exclusion of rules that follow 4 basic forms (bioclimatic, atomic, negated, and logistic regression); this process stops when a maximum number of iterations (1,000) is met or an optimization parameter changes by <1% from 1 generation to the next. We tested 500 models by using Desktop GARP (www.nhm.ku.edu/desktopgarp/index.html). We used 50% of points for training the model and 50% to test it. We then selected 50 models by using the best subset option (20% of soft omission threshold distribution and 50% of commission threshold distribution); all other parameters were left as default values. The 50 selected models were then combined. The predicted area was defined as a combination of those areas with higher or equal model coincidence value than the lowest model coincidence value for the human case-patient localities.

Maxent is used to estimate a probability distribution by comparing the environmental conditions at localities in which disease transmission was recorded with environmental conditions across the landscape (34,35). We used the maximum entropy principle calculation of the program to find this estimated distribution; thus, the result was the closest to a uniform distribution with a mean that was closest to the observed mean value from known occurrences, achieved by a regularization parameter (β). We used the default values for all parameters; the predicted area was then selected by using a lowest probability threshold value to differentiate between suitable and nonsuitable areas (36).

By using the 7 environmental variables described, we performed a principal component analysis to describe the environmental variability in sub-Saharan Africa and identify those conditions in which monkeypox cases in humans have been recorded and those conditions identified as suitable by ENM analysis. We used the principal components tool in ArcGIS 10 (Esri, Redlands, CA, USA) to calculate the principal components and used the rgl package 0.92.798 (http://ftp.osuosl.org/pub/cran/) for R 2.13.1 (37) to visualize the environmental conditions at the localities where monkeypox had been reported, in the areas predicted by ENM algorithms to be suitable for monkeypox transmission, and in the localities of the Sudan outbreak in 2005 in association with the conditions in sub-Saharan Africa. Additionally, we calculated basic statistical parameters (mean, maximum, minimum, and SD) for each variable within the area predicted by the ENM algorithms, which, in turn, were compared with the environmental conditions found at the locations proposed as transmission sites in Sudan.

We used a maximum clade credibility tree, which displayed high support at all nodes (Figure 1). Six isolates were grouped within the clade identified in western Africa: Liberia, Sierra Leone, 2 isolates from Ghana, and 2 isolates acquired from primate colony outbreaks and subsequently extensively passaged in cell culture (Copenhagen and Walter Reed). The remaining 5 isolates were grouped within the Congo Basin clade, including the 2 isolates obtained during the Sudan outbreak in 2005 that were processed and combined for analysis. The isolate from Sudan was most closely related to isolates from Yandongi and Mindembo, located in north central DRC. Of isolates examined, these were from locations nearest to Sudan (Figure 2).

Figure 3 shows a genotypic map of a unique indel region in the right side of the MPXV_Nuria_Sudan_2005 genome that represents a large inverted duplication originating from the left end. This unusual (10.8 kbp) duplication found in Sudan isolates 1 and 2 is composed of open reading frames of several host immune modulator genes (MPXV_Zaire_1979–005 open reading frames 5–16) and some fragments of the inverted terminal repeats. This duplication was partially lost in 1 of the 2 isolates after the second BSC40 tissue culture passage. Additional sequence variations between Sudan isolates 1 and 2 occurred at 4 locations (these variations were also excluded from the phylogenetic analysis because they created gaps that were removed during the alignment): at nt position 10838, Sudan isolate 2 has additional copies of a repeat TTAGA (this variation is in the inverted terminal repeat and so is also reflected at the right end of the genome); at nt 20935, Sudan isolate1 has an additional T in a homopolymer string; at nt 138827, Sudan isolate 1 has an additional ATC repeat; at nt 179133, Sudan isolate 1 has a repeat of 23[ATATACATT] not present in Sudan 2. One of these 4 variations occurs in a coding region. The additional ATC repeat is found in the P4c precursor gene where it codes for an additional aspartic acid residue. The complete genome sequence, absent the full inverted terminal repeat regions and hairpin ends, of the Sudan 1 isolate was 206,346 nt.

A comparison of the 189,830-nt alignment data (Table 2) revealed no nt differences between Sudan isolates 1 and 2, 21 nt differences between MPXV_Nuria_Sudan 2005 and MPXV_Yandongi_DRC_1986, and 34 nt differences between Sudan and Mindembo. Notably, isolates from Yandongi and Mindembo were collected in 1986 and 1979, respectively (19 and 26 years before the monkeypox outbreak in Sudan), and there are 34 nt differences between these 2 isolates. We compared data from the monkeypox outbreaks in Yandongi, Sankuru, and Impfondo and found 55 nt differences between isolates from Yandongi and Sankuru and 44 between those from Yandongi and Impfondo; locations in these pairs are separated by a geographic distance less than that between Sudan and Yandongi. Consequently, the genetic distance between Sudan and Yandongi isolates is low in comparison with other pairs, especially considering the greater number of years between collection dates of the 2 isolates and greater geographic distance.

The areas with suitable environmental conditions obtained from the GARP and Maxent models were similar; thus, we present only the results from Maxent (Figure 2). We used Maxent probability >0.0903047 and GARP model coincidence ≥34 to determine the areas predicted by the model because these are the lowest probability and coincidence values at the localities used to train the models (i.e., omission error = 0). Reported cases from southern Sudan do not fall within the suitable areas predicted by the ENM algorithms (Figure 2).

Table 3 summarizes the environmental conditions found in the areas predicted as suitable by the ecological niche models for all 7 variables. In general, monkeypox cases in humans in Sudan were reported from areas with higher mean temperatures, lower annual precipitation, and higher temperature ranges than those from areas with indigenous monkeypox occurence. Annual precipitation values for the localities in Sudan are lower than suitable values predicted by Maxent and are at the drier end of values predicted by GARP; furthermore, values for precipitation of the driest month for Sudan localities reach 0 mm, although neither algorithm predicts such conditions to be suitable for MPXV transmission. Annual mean temperatures of Sudan localities are higher than mean values from ENM algorithm predictions but slightly lower than the maximum values from the models. Maximum temperature of the warmest month is consistently higher in Sudan than in either ENM prediction.

The first 3 components describe >99% of the environmental variability in sub-Saharan Africa on the basis of the 7 selected variables (principle component [PC]1 = 97.47%, PC2 = 1.35% and PC3 = 0.84%). Figure 2 shows the distribution within sub-Saharan environments, historic MPXV case localities, and localities corresponding to the 2005 Sudan monkeypox outbreak. The latter localities fall outside the suitable environmental conditions for MPXV transmission predicted by ENMs.