We used mitochondrial gene COI, Cytb, and nuclear gene EF-1α to reconstruct the phylogenetic relationships of Mollitrichosiphum. All sequences were submitted to GenBank (see Table 2 for accession numbers). The detailed characters of the individual sequences and the combined dataset are listed in Table 3. The monophyly of the sampled Mollitrichosiphum species was supported by all phylogenetic analyses. The results based on the combined dataset of COI and Cytb were similar to that of a previous study [30] (data not shown here). Phylogenetic trees based on EF-1α sequences and the combined dataset of the EF-1α and mitochondrial genes are shown in Figure 1. In the phylogenetic analyses based on the EF-1α gene, four representatives of Thelaxinae, Cervaphidini and Greenideini were used to root the trees. Maximum parsimony (MP), maximum likelihood (ML) and Bayesian (BI) analyses yielded congruent trees, and the tree obtained from the ML analysis is shown (Figure 1A). The monophyly of the genus Mollitichosiphum was highly supported (MP/ML/BI: 82/93/1.00). Three samples of Greenideini, Cervaphidini and Thelaxinae were used as outgroups in the Bayesian and ML analyses of the combined dataset. The combined dataset also supported the monophyly of this genus (ML/BI: 80/1.00) and of most species (Figure 1B). All analyses supported the sister relationship between M. nandii and M. montanum. Interestingly, the samples of M. luchuanum and M. rhusae clustered together and formed a paraphyletic clade in the phylogenetic tree based on the EF-1α, while the two species could be distinguished in the trees based on combined datasets of the two mitochondrial genes as well as the total three genes. Moreover, the position of M. nigrum as a sister clade of M. luchuanum and M. rhusae is well-supported by all analyses. The basal position of Mollitrichosiphum sp. in the subgenus Metatrichosiphon is well-supported in the combined dataset, although the supports were lower than 50% with the EF-1α gene. The phylogenetic relationships of M. nigrofasciatum with other species are uncertain. In the EF-1α tree, M. nigrofasciatum grouped together with M. nigrum, M. rhusae and M. luchuanum (bootstrap values and posterior probabilities were all lower than 50%), and based on the combined dataset, this species forms a sister clade with M. nigrum, M. rhusae, M. luchuanum, M. montanum and M. nandii (ML/BI: 76/0.86). In the phylogenetic analyses based on the combined dataset, the samples of M. tenuicorpus were subdivided into three major clades (Figure 1B), which was also supported by the results obtained with EF-1α to some extent (Figure 1A).

The divergence times were estimated based on the mitochondrial and nuclear data separately, and the mean age estimates of most nodes were largely consistent. The molecular dating results indicate that clade A within M. tenuicorpus diverged at approximately 11.0 mya (95% Highest Posterior Density, HPD: 5.5–16.4 mya) (Figure 2), while clades B and C diverged at 8.2 mya (HPD: 3.7–13.1 mya). The ancestors of Mollitrichosiphum sp. and the other species of subgenus Metatrichosiphon diverged at approximately 15.8 mya (HPD: 10.8–18.6 mya). The most recent common ancestors (MRCAs) of the other six species were 13.6 mya (HPD: 8.4–17.6 mya). Divergence among the individuals of six species (M. rhusae, M. nigrum, M. nandii, M. luchuanum, M. nigrofasciatum and M. montanum) were between 10.6 mya (HPD: 6.2–15.1 mya) and 3.9 mya (HPD: 0.7–8.2 mya), with M. nigrofasciatum exhibiting the earliest within species divergence (5.3 mya), and MRCAs of M. montanum and M. nandii was estimated at 7.0 mya (HPD: 3.1–12.0 mya). M. tenuicorpus widespread from northeastern India, southern China, Thailand, Indonesia, Japan and Korea (Table S1); ranges of other seven species appear to be constrained. M. montanum and M. nandii are distributed on the south edge of the QTP and Hengduan Moutians; M. luchuanum, M. nigrum, M. nigrofasciatum, M. rhusae and Mollitrichosiphum sp. are limited to southeastern China. Among them, M. rhusae and Mollitrichosiphum sp. are distributed in the Hainan and Taiwan islands, respectively. Distributions of these species were mapped roughly in Figure 2.

According to collection records of specimens and published reliable literature [25,27,31], we summarized host plants of other 18 species in Mollitrichosiphum. Except for the host plant of M. kazirangi still unknown, the hosts of all other species related to 14 families were listed in Table 1. Host plants of Metatrichosiphon have high diversity, referring to all 14 families mentioned above. While in the subgenus Mollitrichosiphum, three species are all restricted on Fagaceae and just a few samples of widespread species, M. tenuicorpus feed on plants of Betulaceae and Sabiaceae.

The results of ancestral state reconstruction suggest that Fagaceae was highly favored among all host plants. Node 61 and 77 (Figure 3), which are the MRCAs of the subgenus Metatrichosiphon and Mollitrichosiphum, showed high proportion of Fagaceae, 0.99 and 0.97, respectively; thereby the common ancestor of Mollitrichosiphum (node 78, 0.99) feeds on Fagaceae received strong support.

The monophyly of the genus Mollitrichosiphum was supported by phylogenetic analyses based on different approaches and genes (Figure 1). The phylogenetic trees of the combined dataset supported the division of the two subgenera (Mollitrichosiphum and Metatrichosiphon) based on morphological characters [24,25]. The molecular phylogenetic relationships between the species were also supported by morphological characters, such as the ultimate rostral segments of M. luchuanum, M. rhusae and M. nigrum, which are at least twice the length of the second hind tarsal segment, while the ultimate rostral segment of M. nandii and M. montanum are not more than twice the length of the second hind tarsal segment.

The phylogenetic placement of M. nigrofasciatum with respect to the other species remains undetermined, as the results based on different datasets were inconclusive. A possible explanation for the incongruent patterns of M. nigrofasciatum is incomplete lineage sorting during successive speciation events. However, more data and further study are needed to verify the real reasons that contribute to these results. Thus, additional markers are required for future studies to reveal the position of M. nigrofasciatum. Analysis of the EF-1α gene yielded a paraphyletic clade that included M. luchuanum and M. rhusae, while the combined datasets resulted in independent clades of the two species. This finding may be explained by the differential effectiveness of the genes. The nuclear EF-1α gene is usually used as a marker to investigate the phylogenetic relationship of aphids above the genus level, due to its slow rate of evolution, whereas the mitochondrial genes are more suitable for revealing the relationships at intra- and interspecies levels [32]. M. luchuanum and M. rhusae may have diverged recently, therefore, only mitochondrial genes may provide phylogenetic signals.

Southern China, which appears to be a center of diversity for Mollitrichosiphum, has experienced great topographical and climatic changes during the past million years. Estimated divergence times of this genus provide a reasonable temporal framework to reconstruct historical patterns of species divergence and distribution. Despite a degree of controversy regarding the exact timing of the uplift of the Tibetan Plateau [33], geological estimates date the period of rapid uplift unequivocally within the Late Tertiary/mid-Miocene [34], and episodes of uplift probably continued throughout the Late Pliocene (c. 3 mya) and into the Quaternary [5,35]. Clade A within M. tenuicorpus (Figure 2) is only currently distributed throughout the QTP, for which the divergence time was at approximately 11.0 mya, thereby falling within the time frame of the mid-Miocene uplift of the QTP. The clade including M. nandii and M. montanum was restricted to the Hengduan Mountains, which are located in the southeastern region of the QTP, and diverged from other species at approximately 10.6 mya (HPD: 6.2–15.1 mya). This result indicates that the distribution of these two species was probably also influenced by the uplift of the QTP.

The divergence of clades B and C within M. tenuicorpus was estimated at approximately 8.2 mya (Figure 2). These two clades are distributed east or west side of the Lancang River [30], which concurs with the long-held notion that the “Mekong-Salween Divide” (MSD) acted as a significant historical and geographical barrier to species dispersal [18,36,37]. The rivers located at the southeast margin of the Tibetan Plateau, the Dadu River, Mekong (Lancang River), Salween (Nu River), and Tsangpo-Brahmaputra (Yalu-Tsangpo River), were once tributaries of a single southward flowing river system, the Paleo-Honghe (Red River). The uplift of the Tibetan Plateau dramatically changed the characteristics of the major river drainages [38]. Time scales of formation of the MSD have rarely been estimated [18], thereby providing little insight into whether there is a direct link between the divergence of the B and C clades with the formation of the MSD. However, these historical events may have isolated the populations of M. tenuicorpus and stimulated differentiation among the populations.

M. rhusae and Mollitrichosiphum sp. are restrictedly distributed in the Hainan Island and Taiwan Island, respectively. The MRCAs of these two species are estimated at approximately 3.9 mya and 15.8 mya, respectively. Based on the results of geological studies, Hainan Island was first separated from the mainland of China during the Early Pleistocene (c. 2 mya) [7], and Taiwan Island was separated approximately 5 mya [39]. The divergence times show that the MRCAs of M. rhusae and Mollitrichosiphum sp. originated much earlier than the separation of the two islands. These results indicate that the geographical isolation of the two islands was probably not the primary factor triggering the differentiation of the two species. However, the isolation may have limited the subsequent species dispersal between the islands and mainland and thereby promoted the present distribution pattern. Although both islands were connected to the mainland when the sea level retreated during the Pleistocene, it has been suggested that the flora covering the emerged areas during the glaciations were poorly developed [40] and not suitable for aphid species dependent on woody host plants.

Fossils of Mollitrichosiphum have been found exclusively in Europe, while all extant species of this genus are currently restricted to Southeast Asia [25,41]. Peñalver et al. [42] demonstrated that the geographic ranges of Greenideinae were markedly wider during the Miocene and comprised the northern coasts of the Tethys Sea. During the Lower and Middle Miocene, the southern region of Europe was characterized by a hot subtropical climate. The terrestrial plant communities of Europe were dominated by needle-leaved forests, Cupressaceae species, and some broadleaf forest species [41]. Later geological events, such as the uplift of the Tibetan Plateau, affected both the climatic conditions of southern Europe and the composition of the flora [43]. During this period, a suitable climate and host plants for Mollitrichosiphum disappeared [44,45]. However, the climate in Southeast Asia remained unchanged. This may have determined the current distribution pattern of Mollitrichosiphum in southeastern Asia, including southern China.

Speciation processes are generally driven by the advent of barriers to gene flow between interbreeding populations [46,47]. The life history of many phytophagous insects often depends on specific host plants. Differences in host plant preference can result in reproductive isolation [48]. For example, studies of the pea aphid Acyrthosiphon pisum demonstrated that species differentiation occurred among different populations that fed on pea, clover and alfalfa [49,50].

Phylogenetic analyses show that Mollitrichosiphum sp. and M. nigrofasciatum are two basal clades in the subgenus Metatrichosiphon; together with M. tenuicorpus, these three species mainly feed on Fagaceae. The divergence times of these species were 15.8 mya, 10.6 mya and 18.4 mya, respectively (Figure 2), which are much earlier than the other species. In contrast, the host plants of a younger clade (with an age of 7.7 mya) including M. nigrum, M. rhusae and M. luchuanum are distinct (Table 1), and each of the three aphid species has hosts belonging to at least two plant families. Similarly, the sister species M. nandii and M. montanum with divergence times of 3.5 mya and 1.5 mya, are associated with host plants belonging to three and four families, respectively. It seems that these species have acquired a wider range of host plants during their evolutionary history.

The phylogenetic pattern of host association shows that basal species much more exclusively feed on Fagaceae, which is a common host plant for most species. Results of ancestral state reconstruction also supported the fact that Fagaceae may be the ancestral host plant of Mollitrichosiphum. Notably for the subgenus Metatrichosiphon, the acquisition of new hosts and the expansion of the host range might be the main factor driving species differentiation. Of the species in the subgenus Mollitrichosiphum, M. tenuicorpus is widely distributed in Southeast Asia and southern China, while the other three species are restrictedly distributed in Nepal, the northeastern region of India (Sikkim), and the southern edge of the QTP [25,27,31]. The divergence of the three clades within M. tenuicorpus (Figure 2) may be due to the combined effects of geographical isolation (as discussed above) and the differentiation of host plants (clade A on Betulaceae, clade B on Sabiaceae and Fagaceae, and clade C on Fagaceae). However, it seems that geographical isolation probably represents the key factor promoting the species divergence of the subgenus Mollitrichosiphum.