Discussion
Analysing the multifarious factors potentially generating population differentiation across six related Neotropical plant species differing in pollination strategy, we here detected consistently stronger isolating effects (particularly IBD) across localities of bee- than vertebrate-pollinated species (Table 1, Fig. 4). The two bee-pollinated species were the only ones where current climatic conditions contributed substantially to explaining population differentiation. These results are in line with the expectation that more mobile pollinators (i.e. flying vertebrates) may connect populations more effectively than less mobile (i.e. (small) bee) pollinators (Wessinger 2021). Even though large bees (pollinators of M. maxima ) are generally considered as relatively mobile (Gamba & Muchhala 2020), their flight activity is strongly reduced under adverse weather conditions in tropical mountains, likely limiting (large-distance) pollen dispersal (Dellinger et al. 2021). Within localities, we also detected IBD more frequently in bee- than vertebrate-pollinated localities (Table S14). Against expectations, however, we did not find consistent differences in nucleotide diversity, heterozygosity or disparity according to pollination strategy (Fig. 2). This suggests that, at least across small spatial scales (i.e. within localities, between adjacent localities), bees may be equally effective outcross pollinators as vertebrates (Opedal et al. 2017, Schmidt-Lebuhn et al. 2019). In the following, we discuss our findings in the context of other factors potentially influencing present-day population genetic structure such as mountain topography, habitat suitability and Pleistocene climatic fluctuations.
Our study species share the same macroevolutionary background (same tribe), but differ in distribution ranges and ecosystems colonized. Bee-pollinated Ad. adscendens has a wide distribution in lowland rainforests in tropical Latin America, while the other five species inhabit montane cloud forests with relatively continuous (M. phlomoides , M. tomentosa ) or patchy (A. costaricensis ,M. maxima , M. sanguinea ) distributions (Fig. S1). We may, hence, expect partly idiosyncratic responses to isolating barriers and (current and past) climatic habitat suitability. Following the expectation that mountain terrain generates strong physical barriers, we did indeed detect significant IBRTerrain in all species but A. costaricensis (Table 1). It was, however, never recovered as factor best explaining population differentiation (Tables 2). While rugged mountain terrain may act as effective barrier even across small spatial scales (i.e. populations three and six in M. phlomoides , Fig. 1), admixture was overall high among close (< 12km) localities in all species, suggesting considerable connectivity by both insect and vertebrate pollinators among localities (Fig. 3).
Ectothermic insect pollinators may be affected more strongly by harsh abiotic climatic conditions (i.e. low temperatures, high precipitation, strong winds) than vertebrate pollinators (Cruden 1972, Dellinger et al. 2021). Accordingly, we found evidence of isolating effects related to current climatic conditions (Table 2, Fig. 4) only in the two bee-pollinated species (IBE in Ad. adscendens and IBRHabitat in M. maxima ). In Ad. adscendens, the marked separation into two clusters (Fig. 1A, 3A) clearly reflects the combination of IBD, isolating effects related to mountain topography, and current climatic conditions. The south-western localities are cut off from the north-eastern localities by the central Costa Rican mountain range. These mountains feature moist and cool cloud forests, which are generally unhostile to smaller bees, the pollinators of Ad. adscendens (Dellinger et al. 2019). Along the western coast, occasional dry habitats (i.e. Nicoya peninsula), on the other hand, represent unsuitable habitats for the moisture-adapted plants (Pröhl et al. 2010). In accordance with this, our niche models estimated the ‘least-cost’ path connecting the southern and northern localities of Ad. adscendens through the south-eastern lowlands along the Caribbean side of the high Central American mountains (Fig. 1A, Patten & Smith-Patten 2008). In M. maxima , the marked differentiation among localities 3 and 4 may further indicate disproportionately strong effects of climatic conditions on ectothermic bee-pollinator activity. These two localities were significantly differentiated from each other genetically, albeit only 20 km apart. Our habitat-suitability models indicated that locality 4 lies in a climatically less suitable area (Fig. 1). Our own experimental work in montane Meriania species in the Andes has established a strong link between bee pollinator activity and current weather conditions, with high bee activity under sunny conditions, but almost no flower visitation under cool and rainy conditions (Dellinger et al. 2021). It is hence possible that the difference in habitat suitability strongly reduced the probability of bee flight among these localities, hence generating strong population differentiation. Interestingly, we also found individuals of M. maxima in locality 4 to differ morphologically (i.e. smaller flowers, non-revolute leaf basis, Dellinger, pers. obsv.). Whether these observed phenotypic differences are the result of random genetic drift, or a response to different selection pressures, remains to be investigated.
Among the vertebrate-pollinated species, we did not find isolating effects caused by habitat suitability (IBRHabitat) or environment (IBE; Fig. 1, Table 2). Indeed, cloud forests form a relatively continuous ecosystem particularly on the eastern slopes of the Andes and Central American mountains (Balslev 1988, Luebert & Weigend 2014), and possibly provide continuously suitable habitats for the cold-adapted vertebrate pollinators. The marked population differentiation observed between Northern and Southern Ecuadorian populations of M. sanguinea and M. tomentosa , on the other hand, follows the well-known biogeographic barrier of the dry Girón-Paute valley (Jørgensen & Ulloa Ulloa 1994, Escobar et al. 2020). This demarcation line, part of the Amotape-Huancabamba zone, has acted both as dispersal barrier for montane species as well as a corridor for lowland species (Weigend 2002, Trénel et al. 2008).
Understanding how tropical plants reacted to Pleistocene climatic fluctuations, i.e. whether they retracted into refugia (‘dry-refugia’ hypothesis) or underwent down- and upslope migrations (‘moist-forest’ hypothesis), remains a major conundrum (Ramírez-Barahona & Eguiarte 2013). If species retracted into small refugia, climatic instability through time should explain population genetic variation (Helmstetter et al. 2020). In our study, we found significant associations between genetic differentiation and climatic instability in five species (Table 1), but IBI was recovered as factor best explaining variation inFST only in M. tomentosa . Indeed, modelling past climatic habitat suitability indicated that the cloud forest species A. costaricensis, M. maxima, M. phlomoides andM. tomentosa retained relatively continuous suitable habitats along the Central and Northern South American mountain ranges throughout the Pleistocene (Fig. S3). These results support the ‘moist-forest’ hypothesis, with downslope and subsequent upslope migration during Pleistocene climatic fluctuations (Fig. S14), and even range expansion in M. phlomoides (Fig. S3c). Further, at the scale of our study localities, there was some evidence of habitat contraction during the LGM in A. costaricensis (continuously suitable around locality 4, Fig. S6) and M. maxima (continuously suitable around localities 1, 2 and 5, Fig. S7). This suggests the possibility for in-situ persistence of these cloud-forest species in part of the distribution range (without necessarily contraction into isolated refugia), a pattern documented for other montane Neotropical plant lineages (Ornelas et al. 2019). M. sanguinea , on the other hand, is the only species in our sample occurring in the high-elevation cloud-forest-Páramo-ecotone. This species may have undergone more prominent refugial retraction in Southern Ecuador and Peru (Fig. S3, S14). Indeed, our models indicate little connectivity among the southern Ecuadorian localities (south of the biogeographic barrier of the Amotape-Huancabamba zone) and markedly differentiated Northern Ecuadorian locality during the LGM (Fig. S7). Finally, in lowland bee-pollinated Ad. adscendens , suitable habitats were likely extensive in lowland Amazonia during the LGM, with mostly continuously suitable habitats along the (eastern) coast of Central America (Fig. S6, Pröhl et al. 2010).
Taken together, our results are highly valuable in adding to the limited data available on the diverse processes shaping population genetic differentiation of tropical plants, including differences in pollination strategy. While we highlight that a wider sampling across bee-pollinated Merianieae is required to firmly nail the role of pollinators in promoting population differentiation, our result on stronger isolation among bee- than vertebrate-pollinated populations suggests a critical role of pollinator mobility in shaping population-level processes. Extrapolating to a macroevolutionary perspective, pollinator shifts are often invoked as “key innovations” spurring diversification (van der Niet et al. 2014). Potentially, pollinator shifts may also alter a population’s susceptibility to isolation and, consequently, its potential for allopatric divergence. Interestingly, in various Neotropical/Andean plant groups, shifts from bee to vertebrate (particularly hummingbird) pollination go hand in hand with increases in diversification rates (e.g. Serrano-Serrano et al. 2017, Lagomarsino et al. 2017). This is somewhat counter intuitive, however, if vertebrates are expected to buffer isolating effects among populations (see Serrano-Serrano et al. 2017 for discussion on additional factors). Clearly, comparative studies, ideally focusing on small monophyletic plant complexes, and documenting both pollination ecology and population genetics of multiple populations across the landscape (e.g. Opedal et al. 2017), will be essential for resolving the relative contribution of pollinator shifts in spurring or limiting speciation through gene flow (Abrahamczyk et al. 2014). To date, we largely lack pollinator observations from multiple populations of the same plant species in the (Neo)tropics, and hence know little about the variability of pollinator composition across a species’ distribution range (pollinator mosaic, Gowda & Kress 2013). Within our own dataset, we have such information for a subset of all populations only (Table S1). While we documented the same functional groups acting as pollinators in most populations, we documented effective hummingbird and rodent pollinators in Southern Ecuadorian populations of M. sanguinea , and hummingbird and bat pollinators in Northern Ecuador. Obtaining such natural history information, in addition to population genomic data, will be key for adding a more realistic understanding to the processes governing speciation and hence contributing to the exceptional diversity observed in the (Neo)tropics today.