Ecological dimensions of speciation
This study is to our knowledge the first one explicitly using ecological niche models and phylogeny to understand speciation in a clade of African rain forest trees. Our results indicate little geographical overlap between sister species (7/11 comparison with less than 30% overlap, Table 1). Thus, most species pairs occur in allopatry, a result consistent with the geographical isolation hypothesis, although this would also be expected under the ecological gradient hypothesis. In one case the overlap was relatively large, for example between M. junodii and M. hastipelata (~ 90%). However, this might be an artifact in the way the overlap statistic was calculated (see Methods) as both these species never grow in complete sympatry (i.e. next to each other, pers. obs.). Monodora hastipelata is a local endemic to a small area in the Matumbi Hills in Tanzania  while M. junodii has an extensive distribution across East Africa .
Most species pairs show visual overlap in ecological space (Figures 4 and 5) and are never completely separated by PCA analysis (expect for the M. carolinae and M. stenopetala pair, see below). In contrast, the niche similarity tests  based on the ecological niche models demonstrated that all species pair had highly significantly different niches (Table 1). This test, however, is known to be very strict, and rejection of similarity can occur based on very small niche differences, especially for allopatric species . At the scale studied here (distribution of species at the continental level) the probability of two species having completely identical niches is very low  and these tests might be too stringent, something already noted in other studies (e.g. ). The background tests appear better suited for this type of analyses and is generally used when species occur in allopatry . Just under half of the tests undertaken (48%, Table 1) supported the hypothesis of niche similarity between sister species whereas 13% (6/44) of the tests supported a significant difference. However, four of these latter tests were found for the species pair Monodora junodii/M. hastipetala, and could be linked to the few data points associated with M. hastipetala (see Methods). In several cases (17/44) a non-significant (NS) result was found, which is suggestive of a lack of power to detect niche differentiation/similarity, either linked to a low sample size or to the distribution of the habitat . NS results were found either between widely distributed species with a fair number of collections (e.g. I. hexaloba and M. myristica), suggesting habitat heterogeneity as a source of lack of power, or between species with a small number of collections (M. stenopetala /M. carolinae; I. thonneri /I. dewevrei).
Finally, we also tested each bioclim variable for phylogenetic signal, i.e. the statistical nonindependence among species of the variables given their phylogenetic relationships. Direct interpretation in terms of evolution of significant phylogenetic signal of traits is difficult and should be done with caution [44, 45]. These tests yielded different results depending on the level of the analysis (genus versus clade). Nevertheless, both statistics identified several bioclim variables within each genus and for the clade as containing significant phylogenetic information (Table 2). Such a result was already suggested for Monodora  but not for Isolona or the clade as a whole. However, it should be noted that for the genus tests the power of the randomization approach to detect significance of the K statistic might be slightly low for 14-16 tip phylogenies . Here, we do not attempt to draw strong conclusions about the rate of evolutionary change or the pattern of the evolutionary process linked to these variables. We simply underline that many variables are not randomly placed on the phylogeny and that they deviate from a random evolutionary process as generated under the Brownian motion model (as interpreted by the K statistic). This is what would be expected under niche conservatism and the geographical speciation model, although exactly how they have influenced its diversification would require more in depth model fitting analyses [44, 46]. In addition, we failed to identify any phylogenetic signal of ecological niche overlap (Figure 7) . Absence of phylogenetic signal can be the result of a mixture in the speciation pattern , or it may be directly related to the quality of the data. In Monodora, the latter would appear to be the case as phylogenetic resolution within the West species is low, and specimen locality data from two sister species pair in East Africa are few, both being sources of error.
The analyses at several levels using different approaches provide a mixed signal on the role of ecology in speciation. Although sister species within this clade do not have identical niches, which is to be expected , they are in several cases significantly more similar between each other than by chance alone (Table 1). Overall however, based on PCA, ecological niche modeling and phylogenetic signal analyses, our results do provide some support to the idea that in terms of diversification, ecological speciation as viewed through climate has not played a major role in the evolution of Isolona and Monodora species. This contrasts with several recent publications where significant ecological divergence was generally demonstrated for northern hemisphere plant sister species such as in Lonicera , Cyclamen , five Andean Solanum species  and in most clades of the South American genus Hordeum . They all concluded that ecological speciation was an important factor of diversification within these genera. Moreover, in the mainly African distributed cucurbit genus Coccinia, frequent biome shifts were inferred during a period of 6 Ma between forest, woodland and semi arid habitats, implying an important role of ecological diversification . Finally, the spread of arid environments in Africa during the Neogene was suggested to be an important driver of diversification of some partially TRF genera such as Acridocarpus . Adaptation to alternative (more arid) environments appears extremely limited within Isolona and Monodora mainly because of the strong evolutionary constraint applied by the precipitation variables (Table 2). A strong correlation between rainfall and species distribution was also found within Neotropical Annonaceae in general .
The only species pair for which speciation might have relied on ecology is found in the East African species M. carolinae et M. stenopetala (Figure 5), the former occurring in moist semi-deciduous coastal forests of southern Tanzania and Mozambique while the latter is found in dense thickets and woodlands of Malawi (Figure 7). However, the background test indicated that M. stenopetala had a significantly more similar niche when compared to the environmental background of M. carolinae (in the other direction, i.e. the similarity of M. carolinae using the background of M. stenopetala, was non-significant, and thus inconclusive). This result is quite intriguing given the important ecological differences identified using PCA between these two species, and could be an artifact due to the low sample size of both species (7 for M. carolinae; 6 for M. stenopetala). It remains unclear exactly how many samples are necessary to produce a robust ecological niche model, although some authors have suggested more than 10 [23, 27] or even over 100 [see ]. Also, these results could be due to the established background that was used for the tests, here defined as all grid cells within 20 km of known occurrence localities. The background tests are known to be sensitive to the definition of species 'background' and therefore a finer or coarser background region could yield different results .
Temporal dimensions of speciation
The estimated mean ages of the origin of species in both genera inferred by Couvreur et al.  (Figure 1, Table 1) are dated to before or at the beginning of the Pleistocene (9/11 speciation events are older than 2.4 Ma, Table 1). This would suggest that the Pleistocene climatic fluctuations had little effect on species diversity within these genera. This result is intriguing because the geographical isolation hypothesis was thought to be especially important during the Pleistocene because of the numerous successive rain forest expansion and fragmentation [[52–54]]. Thus, although ecology plays a little role in the speciation processes of these genera in Africa, the timing of these events pre dates a period of intensive climatic variation and hence potential isolation events. Rather, most speciation events occurred in the Late Miocene and during the Pliocene (between 6-2.4 Ma). It is important to note that these conclusions are based on the mean age for each node and that the 95% confidence intervals largely overlap with the Pleistocene (see Table 1), and thus should be treated with caution. However, these estimates are in line with numerous other studies that have identified pre-Pleistocene diversification in African TRFs, such as in Afromomum , an estimated 60% of Begonia species  as well as in several animal clades such as African birds  and African clawed frogs species . This was also recorded for the genus Coccinia , although this genus is not restricted to TRF. The Late Pliocene corresponds to a period of renewed rain forest re-expansion which was preceded by a fairly long period of aridification and savanna expansion during the Late Miocene . In contrast, a recent temporal analysis of the herbaceous and mainly montane genus Impatiens suggested an important role of Pleistocene refuges on the diversification of the genus . Most of the studies that have detected (some) Pleistocene diversification focused on herbaceous plant clades (e.g. Impatiens and Begonia [[9, 56, 59]]) which are known to have faster rates of molecular evolution when compared to woody taxa (e.g. trees)  and or dispersal abilities. Intraspecific analyses of genetic diversity (phylogeography) of widespread African tree (woody) species (including an African Annonaceae species Greenwayodendron suaveolens ) suggest that Pleistocene refuges did have some effect but mainly at the infra specific genetic structure level only [61, 62]. The small role of the Pleistocene in generating species was also suggested to be the case in Afromomum . Even though Afromomum species are herbs, individuals can live up to 10 years which could imply lower molecular evolution  when compared to other herbaceous taxa. The climatic variations of the past 2.5 Ma might have been too quick to allow allopatric speciation with little ecological divergence to operate, especially on organisms with lower rates of molecular evolution such as trees, something that is supported by our data as well as in other studies . Thus, the late Miocene and Pliocene epochs appear to have played a significant role in diversification of African TRF trees possibly related to the longer phases of unfavorable climatic conditions and isolation of populations allowing proper genetic isolation between them.