Temporal patterns of diversification in Andean Eois, a species-rich clade of moths (Lepidoptera, Geometridae)


Patrick Strutzenberger, Department of Animal Biodiversity, University of Vienna, Rennweg 14, 1030 Wien, Austria. Tel.: +43 1 4277 57411; fax: +43 1 4277 9574; e-mail: patrick.strutzenberger@univie.ac.at


The timing of the origin of present day Neotropical animal diversity is still a matter of debate. For a long time, a preponderance of glacial (i.e. Pleistocene) radiations has been proposed. However, recent data from molecular clock studies indicate a preglacial origin for most of the examined taxa. We performed a fossil-calibrated molecular dating analysis of the genus Eois, which is a major component of one of the world’s most diverse assemblages of herbivorous insects. We found that diversification of Eois took place in the Miocene following a pattern best explained by density-dependent diversification. A strong slowdown of diversification towards the present was detected. Diversification of Eois does overlap with increased Andean uplift and diversification of the most commonly used host plant genus Piper. These findings match the patterns found for the majority of Neotropical tetrapods and for three other unrelated, ecologically different lepidopteran genera.


The biodiversity found in the Neotropics has fascinated naturalists since the very beginning of biological research. The Neotropics are the most diverse biogeographical region on species level (Myers et al., 2000) yet the origin of this astounding richness is still a matter of debate. Especially, the northern Andes qualify as one of the globally ‘hottest’ biodiversity hotspots (e.g. Brehm et al., 2005). The first explicitly formulated hypothesis trying to explain the origin of Neotropical species richness was the forest refugia hypothesis (Haffer, 1969; Hooghiemstra & van der Hammen, 1998) proposing that the origin of Neotropical biodiversity is rather recent. Repeated episodes of glaciation during the Pleistocene period (2.588–0.0117 Ma before present) led to forest fragmentation in the Amazonian lowlands and to oscillations of forest belts in Andean regions (Brunschön & Behling, 2010; Valencia et al., 2010). Under this model, speciation of forest-dwelling organisms would have occurred in vicariance: populations became isolated in remaining ‘forest islands’ as would be formed in valleys when the treeline drops in elevation during a glacial period and subsequently evolved into distinct species through drift and selection. Prior to the advent of ‘molecular clock’ analyses, it was virtually impossible to gather evidence to test any hypotheses on the origin of Neotropical animal diversity because of the paucity of fossils, especially for invertebrates. In recent years, an ever-increasing number of molecular dating studies has accumulated. Antonelli et al. (2010) reviewed the origin of Neotropical tetrapods and found that these mostly had diversified in the Neogene. According to a meta-analysis by Rull (2008), which incorporated all suitable studies on plants and animals, Neotropical taxa arose at a constant rate since the late Eocene. Insects were among the few taxa examined by Rull (2008) for which the majority of molecular dating studies indicated a predominantly glacial origin. Representing only 42 of 1115 evolutionary significant units examined, however, it is evident that insects are vastly under-sampled in regard to molecular dating studies. There are only two recent studies dealing with Neotropical Lepidoptera: Elias et al. (2009) found that two genera of ithomiine butterflies diversified in the late Miocene, whereas Casner & Pyrcz (2010) observed that diversification of the high Andean butterfly genus Lymanopoda took place in the late Miocene and throughout the Pliocene with very little diversification in the Quaternary. The main diversification of bees of the genus Euglossa (Ramirez et al., 2010) also took place in the Pliocene. Overall, there is a profound lack of studies dealing with radiation patterns in species-rich clades of Neotropical arthropods.

For quite a long time, it is known that the uplift of the Andes had a strong influence on Amazonian lowland climate, especially the spatial and temporal distribution of rainfall (Sepulchre et al., 2010). Fjeldsa & Lovett (1997) proposed that especially the northern Andes might also have acted as a ‘species pump’ for the Amazonian lowland. A number of recent molecular clock studies supported that at least to some degree (Wahlberg & Freitas, 2007; Santos et al., 2009; Sedano & Burns, 2010). Most Neotropical tetrapod clades diversified during the period of Andean uplift (Antonelli et al., 2010). The investigation of Andean clades is therefore essential to understand the origin of a substantial fraction of Amazonian lowland and Central American biodiversity.

With the advent of sophisticated dating methods and the ever-increasing amount of available sequence data, it became possible to study temporal patterns of diversification in greater detail. A commonly used measure is the γ statistic (Pybus & Harvey, 2000) to test whether the data fit a rate constant model or whether diversification rates have increased or decreased towards the present. Many studies detected a slowdown of diversification towards the present; though, this may partly be attributable to methodological artefacts (Cusimano & Renner, 2010). Soon after the first molecular dating studies had been published, it was recognized that ages inferred by molecular clocks tend to be inflated (Rodriguez-Trelles et al., 2002). This inflation increases in magnitude when going back in time, leading to implausibly old age estimates for deep splits in metazoan phylogeny (Wray et al., 1996). The use of more realistic relaxed clock models likely reduces age inflation. Inflated ages lead to an erroneously detected slowdown of diversification rates because of ages being pushed further back into the past. Alternatively, an apparent slowdown may be an artefact caused by phylogenetically informed taxon sampling (e.g. preferential inclusion of basal taxa; Cusimano & Renner, 2010). Furthermore, a slowdown may be erroneously detected because of incomplete species sampling (Pybus & Harvey, 2000). Inclusion of further species may obviously increase the number of young divergences that are missed when species sampling is incomplete. Pybus & Harvey (2000) developed a Monte Carlo simulation to correct for missing species.

In this study, we investigate the genus Eois, a speciose clade of small-sized tropical geometrid moths assigned to the subfamily Larentiinae. Scoble (1999) recognized 250 valid described species. The large majority of described species (211) occur in the Neotropical region where Eois comprises an important part of megadiverse geometrid moth assemblages in Andean mountain forests. Brehm et al. (2005) recorded at least 102 Eois morphospecies in one single small area of Ecuador, representing 8.1% of geometrid species and 10.2% of all geometrid individuals. More recent DNA barcoding data and additional field sampling suggest that the true number of Eois species in this area alone may easily exceed 160 species (Strutzenberger et al., 2010b). Hence, the overall species count of Eois is likely to be considerably higher than 250, and a rough estimate yields total species numbers for Neotropical Eois somewhere from 733 to 1710 species (G. Brehm, F. Bodner, P. Strutzenberger, F. Hünefeld & K. Fiedler, submitted). Species delimitations and therefore species numbers for Eois have to be considered preliminary. The best approximation is currently based on an integrative taxonomy approach (Strutzenberger et al., 2010b) where species were delimited by a combination of mitochondrial COI sequences (‘barcodes’) and external morphology.

The majority of Neotropical Eois feed on Piperaceae with a number of records from other plant families (Strutzenberger et al., 2010a). The piperacean genus Piper is by far the most commonly used plant genus. Eois species were so far found to be rather specific feeders, and one moth species does usually feed on only one or in some cases two or three species of Piper (Dyer et al., 2004; Connahs et al., 2009; Bodner et al., 2010). Strutzenberger et al. (2010a) confirmed that feeding on Piper is a common trait found in all major clades of Neotropical Eois whereas affiliations with other plant families are secondary occurrences. Feeding on Piper is also the most likely ancestral state for Neotropical Eois. These tight host plant connections stimulate the hypothesis that the availability of Piper plants had major influence on the evolution and radiation of Eois. Because of its abundance and species richness, inferences made on the timing of divergences within Eois as an exemplar taxon for the study of plant herbivore interactions hold great promise to foster our understanding of the historical dynamics of Neotropical insect herbivore diversity. We here analyse the temporal patterns of diversification within Neotropical Eois species. No other study has so far investigated such patterns in a species-rich clade of tropical moths down to species level. Hence, Eois may serve as a template for future studies on diversification in other species-rich clades of putatively co-evolved specialist herbivores.

Materials and methods

Taxon and sequence sampling is identical to a previous study on the phylogeny of Eois (Strutzenberger et al., 2010a). Taxa are listed with their Genbank accessions in Table S1. The vast majority of Eois taxa were collected at elevations from 1020–2700 m in the Reserva Biológica San Francisco, southern Ecuador (Fiedler et al., 2008). We included all available Eois taxa that may represent separate species delimited by a 2% COI sequence divergence. This is the most effective threshold for species delimitation in Eois as 106 of 118 species segregated at this threshold were confirmed by morphological re-examination, and all species were recovered as well-supported monophyletic clades in a maximum likelihood analysis (Strutzenberger et al., 2010b). Most species are therefore valid under the morphospecies concept, and all are valid under the phylogenetic species concept. Neotropical Eois form a monophyletic unit (Strutzenberger et al., 2010a); the closest known sister clade to Neotropical Eois are Old World Eois. The clade examined here is therefore a natural, biogeographically defined clade. No records of Eois are known in the Americas from outside the commonly used definition of the Neotropics, which ranges from southern Mexico to Tierra del Fuego.

Selection of taxa was random in respect to their phylogenetic position as taxon selection was performed prior to any knowledge on phylogenetic relationships. Effects caused by oversampling of deep nodes as described by Cusimano & Renner (2010) can therefore be excluded. Deep nodes may rather be under-sampled because of our limited geographical sampling. With respect to the entire genus, our taxon sampling of Neotropical Eois is not exhaustive but representative. All but two morphotypes known to occur among Neotropical Eois are represented. Morphological comparison of Eois from other Neotropical regions reveals that these do not represent separate complementary clades within Eois that are missing in our Ecuadorean taxon sample, but rather belong to the same species complexes as represented here (Strutzenberger et al., 2010a).

Lepidopteran fossils in general, and geometrid moth fossils in particular, are exceptionally rare (Grimaldi & Engel, 2005). Fossils suitable for calibration are even scarcer as these must fulfil two important characteristics: (i) be assignable to an extant clade with reasonable confidence and (ii) the fossil must be dated with high precision. There are only two fossils described in the literature suitable to calibrate our phylogeny: Geometridites larentiiformes from the Bembridge Limestone, Isle of Wight, UK (Jarzembowski, 1980), and Hydriomena protrita from the Florissant Formation Colorado, USA (Cockerell, 1922). Both fossils are of almost identical age. Because a high-resolution 40Ar/39Ar radiometric dating (Evanoff et al., 2001) and magnetic stratigraphy (Prothero & Sanchez, 2004) is available for the Florissant Formation, we decided to use H. protrita to calibrate the age of the Larentiinae. We were unable to determine the exact layer where the fossil was found; therefore, we used the youngest age obtained for fossil bearing rocks of the Florissant formation (34.07 ± 0.08 Ma). In addition, we used the estimated ages of the Geometridae and the clade (Larentiinae + Sterrhinae) obtained by Yamamoto & Sota (2007) in a larger scale dating study. Accordingly, we calibrated the root height with a normal prior (μ = 54.4 Ma, σ = 5 Ma); the clade (Larentiinae + Sterrhinae) with a normal prior (μ = 51.5 Ma, σ = 5 Ma). The clade Larentiinae was calibrated with a uniform prior (lowerbound = 34.06 Ma, upperbound = 54.4 Ma). The lower boundary was chosen to reflect the minimum age of the Florissant fossil and the upper boundary to reflect the mean for the best available estimate for the age of Geometridae per Yamamoto & Sota (2007). Definitions of geological ages are used in accordance with the IUGS 2009 timescale and are as follows: Miocene: 23.03–5.332 Mya; Pliocene: 5.332–2.588 Mya; Pleistocene: 2.588–0.0117 Mya and Holocene: 11.7 kya to present.

We employed beast (Drummond & Rambaut, 2007) for divergence time estimation. To generate valid starting conditions for beast runs, we used multidivtime (Thorne & Kishino, 2002) to obtain node ages for the Bayesian tree topology from Strutzenberger et al. (2010a). Likelihood parameters were estimated using the baseml function implemented in the paml4 package (Yang, 1997). Branch lengths were calculated with the Estbranches software (Kishino et al., 2001; Thorne & Kishino, 2002). Multidivtime was used to estimate divergence times; parameters were 5 × 106 generations, sampling frequency of 100 and a burn-in of 500 000 generations. Calibration was identical to the one used for subsequent beast runs.

beast input files were generated using BEAUti v1.5 (Drummond & Rambaut, 2007) and manual editing of xml files generated with BEAUti. The Bayesian tree topology from Strutzenberger et al. (2010a) with ages estimated by multidivtime was used as starting tree in beast analyses. We used separate substitution models for each codon position of the two genes. Substitution models were selected according to the recommendations by Modeltest under the Akaike Information Criterion, except for the second codon position of Ef1α where a HKY+G+I model was used instead of the recommended F81+G+I. Two runs of beast with 3 × 107 generations each and a sample frequency of 5000 were performed. Tree topology was constrained to the topology of the Bayesian combined analysis tree from Strutzenberger et al. (2010a). beast log files were combined and analysed using Tracer version 1.5.1 (Rambaut & Drummond, 2007). Tree files were combined using a text editor, and the maximum clade credibility tree was generated with TreeAnnotator version 1.5.1 (Drummond & Rambaut, 2007). Node heights for 10 000 trees were extracted from the posterior sample using Treestat v1.2.1 (Drummond & Rambaut, 2007).

Divergence times of Neotropical Eois were plotted against the number of lineages in a lineage through time (LTT) plot. The ‘R’ package laser (Rabosky, 2006) was used to examine the temporal patterns of diversification. The fitdAICrc command was used to test the fit of several rate constant and rate variable models on the observed pattern of diversification rates. Models tested for fit on our data were birth–death, pure-birth, density-dependent logistic (DDL), density-dependent exponential (DDX), yule2rate and yule3rate. Ints was set to 300 to provide a resolution of approximately 100 000 years. The gamStat function was used to calculate the diversification rate γ; mccrTest (Pybus & Harvey, 2000) was used to obtain γ values corrected for the number of unsampled species. Estimation of total Neotropical Eois species count is problematic because of lack of data. Only 211 Neotropical species are described to date. The number of known species calculated as the number of described species plus known but undescribed species is 393. Both of these numbers are clearly inadequate representations of the total species count. For our calculations, we used a rough estimate calculating the total number of species as 733–1710 species (G. Brehm, F. Bodner, P. Strutzenberger, F. Hünefeld, & K. Fiedler, submitted). mccrTest runs were performed with 5000 replicates.

Results and discussion

Examination of beast log files with tracer showed that both runs were converging, indicating a stable and reliable estimation of clade ages. The origin of Neotropical Eois was inferred at 30.96 Mya (+4.94 Ma; −5.17 Ma). A LTT plot of divergences of Neotropical Eois is presented in Fig. 1. The plot shows that diversification started at a high rate and was constantly declining towards the present. A sharp drop in diversification rates is observed at the Miocene–Pliocene boundary. Only three nodes were dated as being of Pleistocene age, one of them within a cryptic species complex and the two others between taxa where the species status is uncertain. The phylogenetic tree of Neotropical Eois with node heights and 95% credibility intervals indicated is presented in Fig. S1.

Figure 1.

 Lineage through time plot of Neotropical Eois. Solid lines represent 95% credibility intervals. Upright dashed lines indicate boundaries of geologic ages. Upright solid line indicates onset of increased Andean uplift. Shaded area indicates species-level diversification of Piper (according to Smith et al., 2008).

A constant rate of diversification could be rejected for Eois [γ = −7.37, P = 8.53 × 10−14 (one-tailed test)]. However, this may be an artefact caused by insufficient species sampling and/or a bias because of failure to distinguish young species. We consider the latter to be unlikely as species delimitation is based on an integrative taxonomy approach using a 2% sequence divergence threshold in combination with wing pattern morphology (Strutzenberger et al., 2010b) and host plant information wherever available. To exclude any potential effects of over- or under-splitting of recently diverged species, we calculated γ with divergences after the Miocene excluded, corresponding to the observed plateau in the LTT plot (Fig. 1). All nine splits in this time period exclusively occur within species complexes where morphological differentiation is minimal or absent. The extent of slowdown is even stronger in this case (γ = −8.34, P = 3.56 × 10−17). Unsampled species may pose a bigger problem. Monte Carlo simulations to correct for unsampled species do at least partially contest to the high robustness of the observed slowdown of diversification. When correcting for the lower estimate for the total number of species (733), a constant rate can still be rejected at P = 0.002. When using the upper estimate (1710 species) a constant diversification rate can however no longer be rejected (P = 0.35). Failure to reject a constant rate of diversification in this simulation does not imply that a constant rate is more likely than a decreasing or increasing rate. When using the truncated set of branching times, both the lower and upper estimate still yield a significant slowdown (P = 0.0002 and P = 0.039, respectively). Apart from incomplete species sampling, the observed slowdown may be an artefact caused by inflation of inferred ages. It is, however, not possible to determine whether and to what extent that phenomenon is present in our data.

Fitting of temporal models for rate changes showed that a DDL model is best suited to explain the observed pattern (ΔAICDDL, yule3rate = 7.4; ω = 0.97). The DDL model assumes that the speciation rate is inversely proportional to the number of species present at any given time. Density-dependent speciation points to processes of niche filling and saturation (Ricklefs, 2009). In the case of Eois, it is obvious that such niches could have arisen during the diversification of Piper, the major group of Eois host plants. Molecular dating of divergences within Piper and Peperomia (Smith et al., 2008) revealed that species-level diversification of Piper took place from approximately 21.5 to 5 Mya. These dates show broad temporal overlap with the diversification of Eois as deduced in our study (Fig. 1). This would be the required prerequisite for a niche-filling scenario where the diversification of Eois was driven by the growing number of available niches, i.e. species of Piper.

Apart from the diversification of Piper, the diversification of Eois coincided with an increased rate of Andean uplift. Twenty million years ago, the Andes were only 25–50% of their present day elevation (Gregory-Wodzicki, 2000). Central Andean uplift started to gain pace at a constantly high rate approximately 12 Ma ago and is continuing today (Gregory-Wodzicki, 2000). North Andean uplift took place at a rapid rate during the Pliocene epoch. The paleodistribution of Eois is unknown, and it is therefore not possible to determine whether central or north Andean uplift would have had greater impact. Reconstruction of paleodistribution with parsimony or likelihood methods is not possible because of the fragmentary knowledge of present day distributions of Eois species and the enormous degree of habitat distortion owing to recent human activity (e.g. deforestation as a major driver of habitat loss for forest-bound species: Hansen et al., 2008). Because the climate in the mid-Miocene was considerably warmer than today (Zachos et al., 2001), the potential geographical range of Eois and its host plants may well have been larger than today. It is evident that microhabitat structure, which is constantly generated and reshaped during an active orogeny, is a major factor influencing the composition of local moth communities. This is exemplified by the different moth communities observed in ravine vs. ridge forests. Ravine forests hold a moth fauna distinct from ridge forests (Günter et al., 2008), and this is also reflected in Eois communities at a spatial scale of a few 100 m (K. Fiedler, F. Bodner & G. Brehm, unpublished). The same applies to the flora with distinct vegetation assemblages occurring in ravines and on ridges (Homeier et al., 2010). This floristic distinction can be safely assumed to be a major factor influencing the small-scale distribution of host-specific herbivorous insects and can, on ecological time scales, serve as a powerful agent for vicariant speciation.

On a global scale, the most dramatic event taking place during the inferred diversification of Eois was global cooling (Zachos et al., 2001). Global cooling may well have influenced the evolution of Eois either directly, i.e. by restricting the suitable geographical and elevational range or indirectly through changes in the Neotropical flora. It is presently not possible to separate such effects from effects caused by Andean uplift.

The results of Elias et al. (2009) and Casner & Pyrcz (2010) on butterflies and this present study on nocturnally active moths are not compatible with a glacial origin of diversity in the examined groups, but are rather in accordance with the patterns observed in tetrapods (Antonelli et al., 2010). This is striking because these groups of organisms have widely divergent habitat requirements and life histories. Even the taxa examined in the three lepidopteran case studies have vastly different niche requirements. Ithomiines are host specific to Solanaceae plants, Eois to Piperaceae, and Lymanopoda to bamboos. Hence, these case studies support the notion that Paleogene/Neogene radiations have played a far more important role in generating the high Neotropical species diversity than Pleistocene processes. These concordant patterns of diversification in ecologically very different taxa provide additional support for the importance of the Andean uplift and the changing climate throughout the Neogene. Yet, given the lack of data regarding the temporal patterns of diversification in Neotropical insects, further such studies are required to gain a more robust picture.


In order of importance of their contribution we thank Manuela Zimmermann, Gunnar Brehm, Florian Bodner, Christine Truxa, Brigitte Gottsberger, Christian Schulze, Hermann Staude and Terence Whitaker. Special thanks go to Axel Hausmann for detailed information on Geometridites larentiiformis. Furthermore, we thank Michel Laurin and Gareth Dyke for thoughtful comments on the manuscript. This study was financially supported by grants from the Deutsche Forschungsgemeinschaft (FOR 402, Fi 547/6-3; FOR 816, Fi 547/10-1). The foundation Nature and Culture International (Loja/Ecuador, Del Mar/USA) provided access to their property for field work. The Ministerio del Ambiente (Ecuador) kindly issued the necessary research permit (002-PNP-DBAP-RLZCH/MA).