Reproductive capacity evolves in response to ecology through common changes in cell number in Hawaiian Drosophila

Summary

Lifetime reproductive capacity is a critical fitness component. In insects, female reproductive capacity is largely determined by the number of ovarioles, the egg-producing subunits of the ovary [e.g. ¹]. Recent work has provided insights into ovariole number regulation in Drosophila melanogaster. However, whether mechanisms discovered under laboratory conditions explain evolutionary variation in natural populations is an outstanding question. We investigated potential effects of ecology on the developmental processes underlying ovariole number evolution among Hawaiian Drosophila, a large adaptive radiation wherein the highest and lowest ovariole numbers of the family have evolved within 25 million years. Previous studies proposed that ovariole number correlated with oviposition substrate [2–4], but sampled largely one clade of these flies, and were limited by a provisional phylogeny and the available comparative methods. We test this hypothesis by applying phylogenetic modelling to an expanded sampling of ovariole numbers and substrate types, and show for these predictions across all major groups of Hawaiian Drosophila, wherein ovariole number variation is best explained by adaptation to specific substrates. Furthermore, we show that oviposition substrate evolution is linked to changes in the allometric relationship between body size and ovariole number. Finally, we provide evidence that the major changes in ovarian cell number that regulate D. melanogaster ovariole number also regulate ovariole number in Hawaiian drosophilids. Thus, we provide evidence that this remarkable adaptive radiation is linked to evolutionary changes in a key reproductive trait regulated at least partly by variation in the same developmental parameters operating in the model species D. melanogaster.

eTOC Blurb

Organisms leaving more offspring likely have higher fitness. Sarikara et al. use the adaptive radiation of Hawaiian Drosophila to investigate the evolution of fecundity. They find that habitat shifts played a strong role, and identify a developmental process that underlies evolutionary change in ovarian development and impacts egg laying capacity.

Results and Discussion

Adult reproductive traits of Hawaiian Drosophila

We measured three adult traits relevant to reproductive capacity (body size, ovariole number and egg volume), from field-collected females, lab-reared F1 offspring of field-collected females, and females from laboratory strains (Figure 1; Table S1). Species identities of field-collected females were assigned based on morphological keys or DNA barcoding (Tables S2, S3). All traits ranged over an order of magnitude within Hawaiian Drosophila: body size ranged from 0.71mm for Scaptomyza devexa to 3.12mm for D. melanocephala, ovariole number per female ranged from two for S. caliginosa to 88.5 for D. melanocephala, and egg volume ranged from 0.01um³ for Scaptomyza (Bunostoma) spp. group (S. palmae/ S. anomala) to 0.2um³ for D. adunca, highlighting the diversity of life history traits in Hawaiian Drosophila.

Figure 1. Reproductive and ecological traits of Hawaiian Drosophila in phylogenetic context.

Compiled adult life history traits (greyscale gradients) collected herein and by Kambysellis and Heed [2] mapped onto a phylogeny of Hawaiian Drosophila constructed from available mitochondrial and nuclear genes. Egg-laying substrate of each species is indicated by colored boxes: bark (brown), generalist (black), sap flux (yellow), leaf (green), fungus (purple), fruit (red), spider eggs (blue), flowers (pink), and unknown (gray). Boxes with solid outlines denote data collected in the present study; boxes with four notches denote data represented in our data and those of Kambysellis and Heed [2]; boxes with dotted outline denote data represented only in Kambysellis and Heed [2]. Missing boxes indicate data points that were either not previously reported [2] or that we were unable to obtain from field-caught samples. Black lines at right delineate the five major groups of Hawaiian Drosophila as follows: SCAP = Scaptomyza; PW = picture wing; MM = modified mouthparts; H = haleakalae; AMC = antopocerus-modified tarsus-ciliated tarsus. See also Tables S1, S2, and S3, and Figure S1.

Within the melanogaster subgroup species, species-specific differences in ovariole number are largely heritable [e.g. ⁵]. To test whether this is also the case in Hawaiian Drosophila, we compared ovariole number of wild-caught females and their lab-reared F1 offspring, across five species with different egg-laying substrates. We observed no significant differences between the ovariole numbers of these two generations regardless of natural substrate (Figure S1), indicating that species-specific differences in ovariole number are also strongly genetically determined in Hawaiian Drosophila.

Larval ecology influences ovariole number evolution

Major shifts in ovariole number have often been attributed to changes in life history strategies. Ovoviviparity is often correlated with reduced ovariole number in Diptera [6], suggesting that increased parental investment could be linked to reduced fecundity in these flies, as observed in other animals [7]. In the Drosophila melanogaster subgroup, previous studies have suggested that reproductive strategies and ovariole number evolve in response to oviposition or larval nutrition substrate [8, 9]. Most melanogaster subgroup species are generalists that oviposit on a variety of decaying fruits, and their mean ovariole number ranges from 18 to 43 per female [10]. Lachaise [9] proposed that the high ovariole number observed in the generalists D. melanogaster and D. simulans may be driven by the frequent oviposition opportunities available to generalist species compared to specialists. In contrast, D. erecta and D. sechellia are specialists on Pandanus fruit and the toxic Morinda fruit, respectively [11, 12], and D. sechellia has the lowest reported ovariole number of the group [13]. This reduction in ovariole number has been hypothesized to be the result of increased egg size as an adaptation to feeding on the toxic Morinda [14], or to be due to lower insulin signalling levels evolved in response to the relatively constant nutritional input provided by substrate specialization [15]. However, the melanogaster subgroup is not well-suited for a broader understanding of ovariole number evolution, as most species share similar oviposition substrates (i.e. rotting fruit) and there are few independent instances of evolution of specialists.

In contrast, Hawaiian Drosophila have evolved to specialize on a variety of oviposition substrates, including decaying flowers, leaves, fungi, sap fluxes, and bark of native plants, and eggs of native spiders [16]. Moreover, these flies exhibit the most extreme interspecies range of ovariole number reported in the genus, ranging from two to 101 per female [2]. Hawaiian Drosophila have undergone rapid island radiation from a common ancestor in the last 25 million years, leading to approximately 1000 extant species [17, 18]. Most of the species diversity of Hawaiian Drosophila is spread across five monophyletic species groups that share genetic, morphological and ecological similarities, and rely on different oviposition substrates [18–21], as follows (Figure 1): Scaptomyza are small species that lay eggs on leaves, flowers and fruits, and only approximately one third of Scaptomyza species are reported to be generalists. Picture wing (PW) species are larger species with striking pigment patterns on their wings . PW species primarily lay eggs on decaying bark or branches of native trees, though some specialize on sap fluxes [16]. Modified mouthpart (MM) species, which have male-specific modifications on mouthparts used during mating [22], have the largest range of egg-laying substrates, including bark, leaf, fruit and sap-fluxes. However, the predominant egg laying substrate for MM species is bark, followed by generalists and leaf specialists [16]. Haleakalae species are darkly pigmented flies that only lay eggs on fungi. Lastly, most antopocerus-modified tarsus-ciliated tarsus (AMC) species are leaf breeders, though there are a few exceptions that have evolved bark-breeding [18].

Within the five major clades of Hawaiian Drosophila, ovariole number is highest in the PW species (up to 88 per female), and lowest in Scaptomyza and AMC species (as few as two per female) [2]. Dramatic differences in ovariole number between species were historically suggested to be associated with evolutionary shifts between oviposition substrates [2]. Subsequent studies [3, 4] found significant associations between ovariole numbers and some substrate types, in support of the earlier predictions. However, these studies either lacked a phylogenetic framework [2], or incorporated phylogenies including largely PW species [3, 4] that have since been improved upon with expanded taxon and locus sampling [18, 19, 21, 23].

Using an updated phylogenetic framework (see Supplementary Information) and expanding taxon sampling across all major groups of Hawaiian Drosophila, we tested this hypothesis by comparing the fit of evolutionary models of ovariole number that accounted for ecologically driven evolution, to those that did not. We combined original observations reported in this current study with data previously published by Kambysellis and Heed [2] (see Supplementary Information). The combined dataset nearly doubles the number of species previously studied, includes both specialist species that oviposit on one of bark, sap flux, leaf, fungus, fruit, flower or spider-eggs, as well as generalist species that oviposit on multiple decaying substrates, and adds new substrate types (spider eggs and flowers). We compared the fit of five models to our data, two of which ((i) Brownian Motion, BM, and (ii) an Ornstein Uhlenbeck model with a shared optimum for all species, OU1) do not take into account the oviposition substrate, and three of which were nested ecological models based on alternative methods of substrate classification: (iii) OU2 assumed two states, bark breeders and all other species, to test previous suggestions that bark-breeding may drive evolution of ovariole number [3, 4]; (iv) OU3 assumed three states, Scaptomyza specialists on spider eggs and flowers, bark-breeders, and species using any other substrate, to test hypotheses that substrates influence ovariole number evolution because of their differences in carrying capacity and field predictability [2, 8]; and (v) OU8 categorized each oviposition substrate separately. These five models were fit over 100 trees sampled from the posterior distribution of a Bayesian phylogenetic analysis to account for phylogenetic uncertainty.

We found that models accounting for larval ecology explained the ovariole number diversification in Hawaiian Drosophila (Figure 2A) better than those that did not. Comparing the three ecological models, we found that the three-state model (OU3), which accounted for both bark breeders and Scaptomyza specialists, was supported as the best-fit model across a majority of trees for ovariole number (∆AICc > 2 as compared to OU2 and OU8 models; Data S1). Estimated theta values for the OU3 model showed that bark breeders have more ovarioles than species that oviposit on other substrates, suggesting that evolution of higher ovariole numbers accompanied the transition to bark breeding from likely non-bark breeding ancestors (Figure 2B,C; Data S1), consistent with earlier hypotheses [3, 4]. In contrast, Scaptomyza species may have experienced a dramatic decrease in ovariole number as they independently specialized on spider eggs and flowers (Figure 2C). Taken together, these results confirm and extend previous work in two important ways. First, they support the suggestions of Kambysellis and colleagues [2–4] that shifts in oviposition substrate may have contributed to the evolution of diverse ovariole numbers. Further, these results account for phylogenetic history using robust comparative methods, and expand the previous taxon sampling to show that this trend applies not only to the PW flies that were most heavily studied previously [3, 4], but also across all major groups of the adaptive radiation of Hawaiian Drosophila.

Figure 2. Different ecological states tested for OU analysis.

(A) Comparison of AICc and weighted AICc values for models testing the relationship between oviposition substrate and ovariole number. Values are for model fit of Brownian motion (BM) and Ornstein-Uhlenbeck with one optimum (OU1) or with multiple optima (OUM) with different combinations of oviposition substrate categories, calculated with the R package OUwie v.1.48 [38]. Oviposition substrates were categorized as follows: OU2 categorizes species that lay eggs on bark and non-bark; OU3 categorizes species into bark-breeder, spider egg/flower breeder, and other; and OU8 categorizes each species according to the eight oviposition substrates represented (bark, flower, spider egg, fruit, leaf, generalist, fungus, sap flux). Models were tested over 100 posterior distribution BEAST trees using nuclear and mitochondrial gene sequences. Bold indicates the best supported model. (B) A two-state model (OU2) of bark-breeders (brown) and non-bark breeders (white). (C) Three-state model (OU3) that codes bark-breeders (brown), spider egg and flower breeders (blue), and other oviposition substrates (white). (D) Eight-state model (OU8) that codes each egg-laying substrate separately, color coded as in Figure 1. Pie charts show the maximum likelihood ancestral state estimates at each node, calculated with the rayDISC function in the R package corHMM,v.1.18 [39]. See also Data S1.

Evolution of specialist habitats changes allometry of reproductive traits

Across animals, potential fecundity sometimes correlates positively with body size [e.g. ^{24, 25}]. The range of Hawaiian Drosophila body sizes is greater than that of other members of the genus, spanning an order of magnitude (Table S1). To determine whether changes in allometric growth might underlie reproductive trait evolution, we analyzed the allometric ratio of such traits using a phylogenetic least squares (PGLS) analysis and thorax volume (thorax length³) as a proxy for body size. We found that across all Hawaiian Drosophila, thorax volume was significantly positively correlated with both ovariole number (Figure 3A; Table S4) and egg volume (Figure 3B; Table S4). However, individual species groups show differences in trends for allometric ratios of reproductive traits. In PW and MM species, body size is correlated positively with ovariole number (Figure 3A1, A2), but not with egg volume (Figure 3B1, B2). In contrast, AMC and Scaptomyza species have a positive correlation with body size and egg volume (Figure 3B3, B4), but not ovariole number (Figure 3A3, A4). For PW, MM, and AMC, there is a negative correlation between ovariole number and proportional egg size (Table S2; Table S4), and there is a negative correlation between ovariole number and egg volume in AMC and Scaptomyza (Table S4).

Figure 3. Allometric relationship between life history traits in Hawaiian Drosophila.

Scatter plots of log transformed adult measurements with phylogenetically transformed trend lines generated by averaging runs from PGLS analysis across 100 posterior distribution BEAST trees, performed with the R package nlme v.3.1–121 [40]. Trend line of the consensus tree is denoted in red when there was a significant relationship between the two traits, and black when PGLS analysis did not support a significant relationship (Data S1). (A, A1-A4) Ovariole number plotted against thorax volume (mm³) in (A) all specimens, (A1) PW, (A2) MM, (A3) AMC, and (A4) Scaptomyza. (B, B1-B4) Egg volume (mm³) plotted against thorax volume (mm³) in (B) all specimens, (B1) PW, (B2) MM, (B3) AMC, and (B4) Scaptomyza. See also Table S4.

We note that these trends are associated with differences in life history strategies between groups. In PW and MM group species, ovariole number increases with increasing body size (Figure 3A1, A2): PW are primarily bark breeders that oviposit eggs in clutches of up to 100 eggs [2], and about half of MM species are bark-breeders [16]. Our analysis suggests that bark-breeding was ancestral to both species groups (Figure 2C). In contrast, the AMC and Scaptomyza species groups, in which ovariole number and body size are decoupled (Figure 3A3, A4), contain very few bark breeding species, and instead have evolved to use a variety of different substrates. AMC group species are primarily leaf breeders. Scaptomyza species include specialists on leaf, flower, fruits and spider-eggs, as well as host plant specialist species that oviposit on all parts of the plant, and fewer than 5% of Scaptomyza are bark specialists [16]. In sum, while a positive correlation between body size and fecundity is commonly posited in egg-laying animals [e.g. ^{24, 25}], we did not find universal support for this trend across Hawaiian Drosophila. This is consistent with previous studies on Diptera, wherein trends toward higher fecundity or ovariole number in larger animals observed within species [24] contrast with between-species differences in ovariole number that do not always correlate with body size [9, 26].

Larval ovary somatic cell number determines ovariole number

Ovariole number is determined during larval development, when a specific group of cells called terminal filament cells (TFCs) form stacks called terminal filaments (TFs) that serve as the beginning point of each ovariole [27]. While in at least some insects, TF destruction during pupal stages can also contribute to final ovariole number [28], TF formation appears to be a prevalent mechanism determining ovariole number between and within Drosophila species [29–31]. This then begs the question of what developmental processes determine how many TFs will form. We previously identified two cell number and cell behaviour parameters that can alter TF number, and thus ovariole number: changes in TFC number per TF, and change in total TFC number [29]. To determine whether the same developmental processes that regulate ovariole number in laboratory populations, also underlie the evolution of ovariole number in natural populations, we measured TF and TFC numbers in the developing larval ovaries of Hawaiian Drosophila. Our analysis of 12 species representing four of the major Hawaiian Drosophila species groups showed that even over a range of ovariole numbers spanning an order of magnitude (Figure 4; Table S5), larval TF number per ovary essentially corresponded to adult ovariole number per ovary (Table S5). Although TFC number per TF varied somewhat between species (Figure 4A; Table S5), PGLS analysis showed no correlation between TFC number per TF and total TF number (Table S6). In contrast, average total TFC number was strongly positively correlated with TF number (Table S5; Figure 4B; Table S6), suggesting that, as in laboratory populations of D. melanogaster, changes in TFC number underlie ovariole number evolution in Hawaiian Drosophila.

Figure 4. Terminal filament cell (TFC) number predicts terminal filament (TF) number in Hawaiian Drosophilids.

(A-C) Bar graphs for (A) TFC number per TF, (B) total TFC number per larval ovary, and (C) TF number per larval ovary representing the mean and standard deviation, as well as the phylogenetic relationship between the species shown (bottom). Grey panel below (B) indicates sample sizes used to determine total TFC number (n = number of ovaries from lab-born larvae of wild-caught females; Table S5). Grey panel below (C) indicates sample sizes used to determine mean TF number (n = number of ovaries from lab-born larvae of wild-caught females; Table S5). Grey-outlined Box below (C) indicates mean ovariole number (ON) per adult ovary (Table S1), and sample sizes used to determine mean ovariole number (n = number of wild-caught adult females; Table S1) for each species. (D-F) Late third instar larval ovaries stained for nuclei (purple) and F-actin (green) for (D) S. caliginosa (flower breeder), (E) D. silvestris (bark breeder), (G) D. mitchelli (egg-laying substrate unknown), and (F) D. tanythrix (leaf breeder). White arrowheads indicate TF structures in the ovary. See also Tables S5 and S6.

The developmental processes underlying ovariole number evolution is particularly interesting in light of the allometric changes in Hawaiian Drosophila species groups. There has been some debate as to whether allometry constrains or facilitates adaptive evolution [e.g. ³²]. In Hawaiian Drosophila, the allometric relationship between two important female reproductive traits, ovariole number and egg size, was coupled to body size in different groups in different ways: when ovariole number was coupled with body size, egg size was not, and vice versa (Figure 3). These trends were associated with bark breeding in the PW and MM groups, where ovariole number was coupled with body size (Figures 1, 2). While the phenotypic integration of ovariole number and egg volume appears tightly regulated across insects [33], the coupling of ovariole number to body size appears more flexible in Hawaiian Drosophila, suggesting that in this context, heritable changes in allometry may contribute to adaptive evolution.

Ovariole number is highly polygenic [e.g. ^{5, 34}], and is regulated by both intrinsic and extrinsic growth factors. Many of these genes, including Hippo signalling, ecdysone and insulin-like peptides, are pleiotropic and can also regulate body size [e.g. ^{15, 35, 36, 37}]. Thus, we propose that the mechanistic basis for evolutionary change of ovariole number on different substrates, may be changes in the relative influence of nutritionally regulated circulating growth factors on the one hand, and cell-autonomous growth on the other hand, on ovarian development during larval and pupal stages. For example, we speculate that on certain substrates, the larval ovary may become less sensitive to nutritionally-mediated growth factors by evolving lower expression levels of growth factor receptors, and relying more on tissue-specific growth factors, which could include local insulin release or cell proliferation pathways such as Hippo signalling.

In summary, by combining phylogenetic comparative methods with comparative developmental analyses of both wild-caught flies and laboratory strains, we have identified potential mechanisms of evolutionary change in ovariole number operating at three levels of biological organization. First, evolutionary shifts in ecological niche can predict the dramatic differences in ovariole number in Hawaiian Drosophila. Second, whether adult body size is coupled with ovariole number or egg volume differs between species groups with different oviposition substrates, suggesting that the allometric growth relationships between these traits evolves dynamically. Finally, changes in ovariole number from two to 60 per individual can be explained by changes in total TFC number, suggesting that ovariole number diversity evolves through the same developmental processes, regardless of the specific ecological constraints or selective pressures. Thus, by integrating ecology, organismal growth, and cell behavior during development to understand the evolution of ovariole number, this work connects the ultimate and proximate mechanisms of evolutionary change in reproductive capacity.

STAR METHODS

Contact for Reagent and Resource Sharing

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Cassandra G. Extavour (extavour@oeb.harvard.edu).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Field collections of Hawaiian Drosophila

Field collections of Hawaiian Drosophila were conducted under the Department of Land and Natural Resources of Hawai’i native invertebrate scientific collection permits FHM14–305, FHM14–353, and HAVO-2013-SCI-0002. Collections were made at the Koke’e State Park and Kui’a Natural area reserve (NAR) on Kauai, West Maui Forest Reserve, Makawao Forest Reserve, and the Nature Conservancy’s Waikamoi Preserve on Maui, and Hawaii Volcanoes National Park and Upper Waiakea Forest Reserve on Hawai’i island. Flies were collected by aspirating flies from traps or sponges containing fermenting fruit and fungi, or by sweeping and sorting leaf litter in forests.

Husbandry of Hawaiian Drosophila

Field-caught females were maintained on yeast-less Wheeler-Clayton medium or Drosophila standard laboratory medium at 18°C at 80% humidity. Each vial contained a piece of tissue paper (Kleenex) moistened with distilled water that was steeped with bark of Clermontia spp. at ambient temperature for several days to stimulate oviposition [41]. In addition, AMC species were supplemented with pieces of decaying Cheirodendron leaves, and S. caliginosa was supplemented with morning glory flowers (Ipomea acuminata), both of which were collected from the field and frozen overnight to eliminate mites.

Larvae of picture wing subgroup and antopocerus species of the AMC subgroup have longer development times than the other species studied herein, and larvae of these species were fed additional food, which was made as follows: 6g of Agar and 225mL distilled water were mixed in a 1L beaker and microwaved for two minutes. 60g cornmeal, 6.6g roasted soybean meal and 7.5g brewer’s yeast were mixed, blended and added to the beaker along with an additional 300mL distilled water, and mixed with a spoon. Lastly, three tablespoons of Karo light corn syrup and one tablespoon of unsulfured molasses was added to the mix, and the mixture was microwaved for three minutes. Food was mixed every minute during microwaving until the mixture was close to boiling point and started to rise up within the beaker. The beaker containing hot food was placed at room temperature until the mixture was warm enough to touch. 3mL of propionic acid and 3mL of 99% ethanol were added, and the solidified food was stored at 4 °C. The solidified food was mixed with a small quantity of water to soften the consistency before being used to feed larvae.

Non-picture wing Hawaiian Drosophila species pupariated on the side of the glass vials, and hatched F1 offspring were transferred into new vials. Larvae of the picture wing subgroup species pupate in the soil. To accommodate this behavior, food vials with wandering picture wing larvae were placed in a large jar containing 1–2cm of moist sand at the bottom. A piece of cloth or paper towel held in place using a rubber band was used close the opening of each jar. Larvae migrated from the vials to the sand to pupariate, and thus adults emerged from the sand, and were aspirated out of the jar into a fresh adult food vial.

The effect of larval substrate on ovariole number in lab-reared and wild Hawaiian Drosophila

Evolutionary modelling analysis showed that ovariole number in Hawaiian Drosophila is best explained by evolutionary forces related to egg-laying substrate. Specifically, the three-state model that we tested (OU3), which distinguishes between bark breeding, the specialist substrates of Scaptomyza specialists, and other substrates, was the best fit for our ovariole number data across a majority of trees for ovariole number (∆AICc > 2 as compared to OU2 and OU8 models). The Brownian motion (BM), one-state model (OU1) and two-state (OU2) model lacked support as compared to the OU3 model (∆AICc > 2). Along with strong support for the OU3 model, we obtained occasional support for the full eight state model (OU8) (Data S1). It is possible that the limited sample size for some substrate categories in this model contributed to the poor fit of OU8 to our data. Further studies with deeper sampling to obtain increased representation of non-bark oviposition substrates will be needed clarify the extent to which finer distinctions between specific specialist substrates may contribute to adaptive changes in ovariole number.

These results were largely unchanged when comparing models using an alternative topology generated from mtDNA sequences (Data S1), with all analyses supporting a role for ecology in driving trait evolution. Since five out of the 66 species represented in the analysis were categorized into species groups rather than a single species based on the DNA barcoding, we assigned two different species IDs and ran the OU analysis to address whether group assignments had an impact on our analysis. Assigning IDs to closely related species within the group did not alter the results (Data S1). Similarly, results were unchanged when only the data collected for this study were considered, excluding data previously reported by Kambysellis and Heed [2] (Data S1).

Here we note some of the difficulties faced by researchers wishing to rigorously and thoroughly account for oviposition substrate in these analyses. Species keys are not available for females of most non-PW Hawaiian Drosophila species, and DNA barcode data that were used to identify the species were not available for many samples that were collected. Since Haleakala species were difficult to key, most samples from this group were excluded from the analysis. Further, presence of specific Hawaiian Drosophila species in the field can be unpredictable, and difficulty in encountering them during field work is increasingly compounded by the declining numbers of endemic species in Hawai’i. For example, sap flux specialists have been documented to exist in the PW, MM, and AMC groups [16], but we were only able to collect data from two PW sap flux breeders, one from the field (D. picticornis) and another from a laboratory line (D. hawaiiensis). Lastly, certain egg-laying substrates are observed in very few species. For example, Titanocheta Scaptomyza species are spider-egg breeders, and this trait appears to have evolved only once [19, 42]. Despite these challenges to taxon sampling, however, our analysis rejected the null model of Brownian evolution, and supported the hypothesis that ovariole number evolves in response to changes in egg-laying substrate across Hawaiian Drosophila.

The effect of larval substrate on fecundity in lab-reared and wild Hawaiian Drosophila

While species-specific ovariole numbers clearly have a strong heritable component even in the absence of the native substrate (Figure S1), the substrate does appear to provide chemical cues that are important to drive egg-laying behaviour. In other words, substrate components may contribute to reproductive output by inducing or facilitating oviposition, but not by determining ovariole number. We therefore speculate that evolution of host specialization may have resulted from or in changes in heritable mechanisms that also, perhaps due to pleiotropy, determine ovariole number.

We note that other species of flies have also evolved differences in ovariole number when they shift between host substrates. For example, in African drosophilids and tephritid Dacus flies, generalist species that oviposit on a variety of egg-laying substrates have higher fecundity than specialists [9, 13, 26]. Moreover, specialist species of African and Central American Drosophila species are more fit in the presence of host-specific compounds [14, 43–45], some of which are toxic to other species of Drosophila. For example, D. sechellia is best reared on lab media supplemented with Morinda fruit [14], while D. pachea cannot be reared in laboratory conditions without supplementing media with sterols from its host cactus [46]. Egg-laying substrates for Hawaiian Drosophila have divergent chemical cues and fungal populations [47]. Laboratory populations of Hawaiian Drosophila often require extracts or pieces of egg laying substrates to stimulate oviposition, but can undergo development completely on laboratory food to give rise to adults with similar ovariole numbers as wild-caught flies (see Supplementary Information; Figure S1). We therefore speculate that specific substrate components may not only allow females to distinguish between hosts for oviposition, but also may contribute to species- and substrate-specific egg laying behavior in Hawaiian Drosophila.

The effect of larval substrate on body and egg size

In addition to ovariole number, we tested whether shifts in larval ecology influenced the evolution of body size and egg index (calculated as the phylogenetic residual of egg volume to thorax volume), as these traits are often correlated with ovariole number and have been predicted to evolve in response to changes in ecology and reproductive strategy. For body size, we found that models that accounted for ecological evolution did not fit the data better than a Brownian Motion model (BM, ∆AICc > 2). For egg index, models that accounted for ecological evolution fit the data better than BM and OU1 (∆AICc > 2), but we were unable to distinguish within between models (∆AICc < 2) (Table S6). These results suggest that the evolution of ovariole number, but not overall body size, has been linked to changes in larval ecology within the Hawaiian radiation of Drosophila.

The relationship between ovariole number and egg size

One of the life history characteristics commonly observed in animals is the inverse relationship between high reproductive capacity and investment into offspring [7]. Egg size is often considered a proxy for maternal investment in insects, and life history theory predicts a trade-off between maternal investment egg size and reproductive output, thereby predicting a negative correlation between ovariole number and egg size [48]. Previous empirical studies have found evidence for this predicted inverse correlation between egg size and ovariole number in some insects [14, 33].

We compared the evolutionary relationship between ovariole number and egg size across Hawaiian Drosophila, accounting for the relationship to body size in each variable using phylogenetic residuals [49]. We observed a significant negative correlation that explained most of the variation in relative egg size (Figure 3C-D; Data S1). Specifically, we found that when controlling for body size, species with more ovarioles have proportionally smaller eggs (Figure 3C-D; Data S1). This result suggests that the allometric relationship between ovariole number and egg size is complex, and implies that there are constraints preventing the evolution of both large eggs and high ovariole number.

METHOD DETAILS

Measurement of adult phenotypes

Adult ovaries were dissected in 1X PBS, and placed in 2% paraformaldehyde in 1X PBS overnight at 4°C. Ovaries were then stained with the nuclear dye Hoechst 33342 (Sigma, 1:500 of 10mg/ml stock solution) in 1X PBS for two hours at room temperature, then washed with 1X PBS for a total of one hour. Ovaries were mounted on glass slides in Vectashield mounting medium (Vector Labs), and ovarioles were spread apart using tungsten needles for species with high ovariole number. Adult ovariole number was counted from both ovaries to obtain ovariole number per female under fluorescent and white light microscopy using a Zeiss AxioImager microscope. Images of eggs were taken from these slides using DIC white light settings. Egg volume was calculated when mature eggs were laid by captive adult flies or when present in dissected ovaries. In these cases, egg volume was estimated by measuring the straight lines across the longest and widest points of the egg, and assuming a prolate spheroid shape following a previously published protocol [50] using ImageJ.

Adult bodies were placed in 99% ethanol after dissection for DNA extraction and adult size analysis. Lateral view images of the thorax were captured using a Zeiss Lumar Stereomicroscope. The highest point of the anterior tip of the thorax and the posterior-most point of the scutellum in the same image plane were used to measure thorax length. A straight line was drawn between these two points in these images using ImageJ’s measure function. Thorax volume was calculated as thorax length (mm)³ as a proxy for body size, and proportional egg volume was calculated by dividing egg volume by thorax volume. Raw data measurements for adult traits are publicly available at https://github.com/shchurch/hawaiian_drosophila_ovaries_2018.

Measurement of larval phenotypes

Wandering larvae or early pupal stage individuals were dissected in 1X PBS + 0.1% Triton-X and fixed in 4% Paraformaldehyde in 1X PBS for 20 minutes at room temperature. Larval ovaries were stained as previously described [29] using mouse anti-Engrailed/Invected (4D9, Developmental Studies Hybridoma Bank, 1:50), FITC-conjugated Phalloidin (Sigma, 1:120), and Hoechst 33342 (Sigma, 1:500 of 10mg/ml stock solution). Samples were post-fixed in 4% paraformaldehyde in 1X PBS for 15 minutes at room temperature and mounted in Vectashield mounting medium (Vector Labs) for imaging using a Zeiss LSM780 Confocal Microscope at the Harvard Biological Imaging Center. Quantification of TFCs and TFs was conducted as previously described [29]. Larval ovary TF and TFC number measurements were obtained per ovary, unlike adult measurements, which were collected from both ovaries for each female.

For some samples, fewer collected specimens could be used for measuring total TF number than for others, as total TF number can only be counted in larval ovaries where TF morphogenesis has completed, which is usually near the end of larval development (2). At the time of larval ovary dissection, some ovaries contained completed TFs, while others were still undergoing morphogenesis and could not be used to gather data on TF number. In the latter cases, TFC number per TF was measured for those TFs that had completed morphogenesis, and total TFC number for that species was assigned based on the average TF number from other specimens from the same species. Larval trait raw data measurements are publicly available at https://github.com/shchurch/hawaiian_drosophila_ovaries_2018.

Notes on analysis of larval ovarian development in Hawaiian Drosophila

Overall larval ovary morphology of Hawaiian Drosophila was similar to that of the melanogaster subgroup species (Figure 4D-F), with characteristic TFC stacks forming toward the end of larval development (white arrowhead, Figure 4D-F). To determine whether, as in D. melanogaster [29, 30, 35], ovariole number is established by the end of larval development and does not change during the pupal phase, we compared total TF number in Hawaiian Drosophila ovaries that had completed TF morphogenesis to adult ovariole number per ovary. We found a close to 1:1 correlation between TF number per ovary and ovariole number per ovary (Table S5).

One notable difference was in S. caliginosa, which had one ovariole per ovary in our adult samples, and two TFs per ovary in the developmental analysis. Given that Kambysellis and Heed [2] previously reported S. caliginosa females with more than two ovarioles, our result may be due to the small sample size of adults of S. caliginosa in our study (n = 5 versus n = 24 in the previous study), or to the difficulty of counting ovarioles in this species. However, we note that honeybees can destroy ovarioles that are formed during larval stages through programmed cell death during pupal development [51]. Thus, we cannot exclude the possibility that the difference between larval TF number (two) and adult ovariole number (one) observed in S. caliginosa may be a result of a similar developmental process as that reported in honeybees.

PCR amplification of mitochondrial genes for species identification

While there are detailed dichotomous keys for species identification of Hawaiian Drosophila [52], these keys focus on male-specific traits including male genitalia and other sexually dimorphic characters. Therefore, we identified female flies using a combination of morphological features [22, 53], collection site information, and DNA barcode-based methods as previously described [54]. Following ovary dissection, abdominal at tissue was used for DNA extraction using the Qiagen Blood and Tissue kit. PCRs were conducted using primer sets (from 5’ to 3’) as previously published and listed in the KRT [18].

PCR was conducted using Dynazyme DNA polymerase (Thermo Scientific) as follows: 95 °C 5 minutes, (95 °C 30 seconds, 50 °C (COI and COII) or 54 °C (16S) 30 seconds, 72 °C 30 seconds) x 30, 72 °C 5 minutes. PCR products were cleaned using ExoSAP-IT (Affymetrix) and sequenced by Genewiz (Cambridge, MA). Sequences were analyzed using 4Peaks (Nucleobytes).

QUANTIFICATION AND STATISTICAL ANALYSIS

Analysis of mitochondrial sequences for species delineation

All sequences were first analyzed by BLASTn alignment against the NCBI Nr/Nt collection, and the accession numbers were noted for those where there was one clear hit with 98–100% sequence identity to one species (Table S2). Species that returned multiple 99–100% BLASTn hits are summarized in Table S3. We note that in many cases, multiple hits are within a closely related species subgroup (Table S3), as reported in previous phylogenetic studies of Hawaiian Drosophila [18–21, 23, 55]. When BLAST results of all three barcodes were consistent, the sample was assigned a species identity. In cases where the BLAST analysis did not provide a clear identity, we tested whether the sample was sister to a single reference species in gene trees constructed using RAxML, as described below in ‘Phylogenetic Inference’. Samples with unambiguous BLAST and/or tree-based support were assigned a species identity. Some samples resulted in BLAST and tree-based support for a closely related species group; in these cases, a species group identity was assigned instead of a specific species. Samples that did not have clear support for any species group hit were discarded from the dataset. The published and original sequence data, resultant gene trees, and custom scripts used for this analysis are available at https://github.com/shchurch/hawaiian_drosophila_ovaries_2018, commit 5c70803, directory ‘sequence_data_methods’.

Taxon sampling in this and previous studies

All measurements were taken following methods from [2], and 15 species from that study were also included in our study. The majority of the overlapping data were within two standard deviations of the previous study (Supplemental Table 1; >3 standard deviations highlighted in yellow), suggesting that these traits have remained stable over the last 40 years, such that our measurement methods are comparable to those of the previous study. We therefore included the data from Kambysellis and Heed [2] for subsequent analyses. Our final dataset for analysis contains 35 newly characterized species, 15 species included in both our field-caught dataset and Kambysellis and Heed [2], and 16 additional species discussed by Kambysellis and Heed [2] but not found by us in the field, yielding a total of 66 Hawaiian Drosophila species across all major species groups that were used in the analysis herein (Figure 1).

Our taxon sampling for phylogenetic inference combines the efforts of four previous studies [18, 19, 21, 23] and additional newly identified mitochondrial sequences (GenBank accession numbers MK276992 - MK277193). This sampling includes members of all major lineages of Hawaiian drosophilids. Nucleotide sequences from each of these four studies were downloaded from GenBank, totalling 18 genes. The sequence IDs were parsed using the program phyutility v2.2.6 [56], and the 18 genes were aligned individually using MAFFT v7.130b [57] with the “auto” option selected, and trimmed with Gblocks v0.91b [58] with the “with half” option selection. Trimmed sequences were concatenated using phyutility into two alignments, one including all 18 available genes and one including only the four mitochondrial genes. This second alignment reflects the analysis performed by O’Grady and colleagues [18]. PartitionFinder v1.1.1_Mac [59]; options ‘raxml’) was used to find the best fitting model for each partition; GTR + Γ +I was found for nearly all partitions. For species delineations, sequences of the three targeted genes generated in this study were combined with homologous sequences from the four previous studies and aligned and trimmed using the same procedure as above. Gene trees were generated in RAxML v8.2.3 [60] using a GTR + Γ +I model of sequence evolution).

Phylogenetic inference

Phylogenetic relationships and divergence time estimates were inferred simultaneously using both the mitochondrial and mitochondrial+nuclear alignments in a Bayesian framework in BEAST v. 2.3.2 [61, 62]. A single calibration at the root (i.e., at the base of Hawaiian Drosophila + Scaptomyza) was used to infer divergence times; this was assigned a uniform prior from 23.9–37.1 Ma, following [23]. Rate-smoothing was performed using a relaxed lognormal clock model [63]. The BEAST analyses followed the same partitioning scheme as RAxML, as determined by PartitionFinder [59] and utilized a birth-death tree prior. Four separate chains were allowed to run for 100 million generations (sampling every 10,000) using the CIPRES supercomputer cluster (https://www.phylo.org). Convergence was assessed using the AWTY web interface [64] and effective sample size (ESS) values of the runs (using values >200 as a cutoff) in Tracer [65]. After convergence was reached, the individual runs were combined and the maximum clade credibility tree, including credibility intervals (CI) for ages and posterior probabilities (PP) for node support, was assembled in TreeAnnotator [66]. Upstream phylogenetic comparative analyses used either the maximum clade credibility (MCC) tree or a subset of 500 trees from the posterior distribution of trees, as appropriate; analyses were repeated for each of the two BEAST analyses. The final alignments, mcc trees, and posterior distributions are available at https://github.com/shchurch/hawaiian_drosophila_ovaries_2018, commit 5c70803, directory ‘sequence_data_methods’.

Phylogenetic relationships between Hawaiian Drosophila in this study

Our study focuses on the Hawaiian clade Drosophilidae, which comprises an estimated 1000 species in two genera, Drosophila and Scaptomyza. The phylogeny of this clade and its substituent subclades has been the focus of many recent studies [18–21, 23, 55]. The monophyly of Hawaiian drosophilids is well-supported, as is the monophyly of Scaptomyza and Hawaiian Drosophila within them. The relationships between and within the subgroups of Hawaiian Drosophila (Scaptomyza, PW, MM, Haleakala, and AMC species groups) remain less resolved. Specifically, the monophyly of each of these individual groups is generally well-supported (but see the moderate support values for the MM group reported in [21] and [18]), but the relationships between these large groups are in conflict across studies.

To generate a phylogenetic tree for evolutionary modelling analysis, we compiled data from published phylogenetic studies and captured the two topologies of Hawaiian Drosophila species groups that have been published to date. The consensus tree topology that we recovered matched the topology recovered in recently published phylogenetic analyses [18, 21], and the second most common basal topology matched the alternative topology presented in [21]. Our phylogenetic analysis also captured monophyletic species subgroups in AMC, PW and Scaptomyza [19, 21, 23]. Given that species relationships at the species group and subgroup level were recovered in the phylogenies used for phylogenetic comparative methods analysis, we believe our analysis represents an accurate estimate of our current knowledge of Hawaiian Drosophila phylogeny.

Phylogenetic Generalized Least Squares Analyses

All phylogenetic comparative analyses and corresponding figures were computed in R version 3.2.0 [67]. The evolutionary relationship between ovariole number, egg volume, egg volume over thorax volume, and thorax volume was analyzed in pairs using phylogenetic generalized least squares in nlme v.3.1–121 [40], using the phylogenetic correlation matrix generated using the corMartins function in ape v.3.3 [68, 69] with a small starting alpha value. All traits were natural log transformed prior to analysis. Pairwise comparisons were performed over 100 trees randomly drawn from the posterior distribution generated in BEAST. The range and average of both the p-value and the slope from the PGLS models across the 100 trees were calculated, and a cutoff threshold of 0.05 was used to determine significance of the p-values.

Analysis of Evolutionary Regimes

We used all reported ecological information about Hawaiian Drosophila as summarized by Magnacca and colleagues [16] to code oviposition site for the species in our dataset. Three different coding schemes were compared: (1) OU8, which considered eight ecological substrates (bark, flower, fruit, fungus, generalist, leaf, sap and spider egg breeders); (2) OU3, which considered three states (bark, flower & spider egg, and ‘other substrate’ breeders); and (3) a final one with two states (OU2: bark and non-bark breeders). We categorized species as bark-breeders if they utilized the tree stem or trunk, though previous studies distinguished between the two [3, 4]. Ancestral states for each of these character codings were calculated over 100 trees randomly drawn from the posterior distribution of trees generated with BEAST. The most likely ecological state was mapped at each node using the rayDISC function in the R package corHMM, v.1.18 [39], and the resulting tree was pruned to include only tips with ovariole number data.

The fit of three models of trait evolution were assessed on pruned trees using the R package OUwie v.1.48 [39]. The three models tested were Brownian Motion (BM1), Ornstein-Uhlenbeck with a single optimum for all species (OU1), and Ornstein-Uhlenbeck with optima for each ecological state (OUM; OUM was fit for eight, three and two state models for each of the distinct character codings, respectively OU8, OU3, and OU2 as described above). Corrected Akaike Information Criterion (AICc) values were compared for each of these models for each of the trait coding schemes, where the best-fit model (i.e. the model with the lowest delta AICc score) was moderately supported when other models had delta AICc of 2–10, and strongly supported when delta AICc>10. This analysis was repeated over each of the 100 trees, and the frequency of each best-fitting model was recorded. Optimized theta values for each of the three OUM analyses were untransformed and recorded (Data S1).

DATA AND SOFTWARE AVAILABILITY

All of the commands and data used to perform phylogenetic comparative analyses, as well as the corresponding commands to generate the figures, are available in the public repository https://github.com/shchurch/hawaiian_drosophila_ovaries_2018, commit 5c70803.

Supplementary Material

1

Data S1. Summary of OU analyses, Related to Figure 2.

(A)Summary of OU analyses. Values are for model fit of Brownian motion (BM) and Ornstein-Uhlenbeck with one optimum (OU1) or with multiple optima (OUM) with different combinations of oviposition substrate categories, estimated with the R package OUwie v.1.48 [S3]. Oviposition substrate was categorized as follows: OU2 categorizes species that lay eggs on bark and non-bark, OU3 categorizes species into bark-breeder, spider egg and flower breeder, and other, and OU8 categorizes each species according to the eight oviposition substrates occupied by the studies considered here (bark, flower, fruit, spider egg, leaf, generalist, fungus, sap flux). Models were tested on phylogeny containing nuclear and mitochondrial gene sequences. Analyses were conducted across 100 trees from a BEAST posterior distribution using nuclear and mitochondrial genes, or mitochondrial genes only. “Combined data” includes data from our study and from that of Kambysellis and Heed [S2]. “Present study” indicates analyses conducted using data collected from this study alone.

(B) Estimated OU theta values from OUwie analyses. Values are the estimated OU theta values for each ecological regime, under different combinations of oviposition substrate categories, calculated using the R package OUwie v.1.48 [S3]. Oviposition substrate was categorized as follows: OU2 categorizes species that lay eggs on bark and non-bark, OU3 categorizes species into bark-breeder, spider egg and flower breeder, and other, and OU8 categorizes each species according to the eight oviposition substrates occupied by the species in our study (bark, flower, fruit, spider egg, leaf, generalist, fungus, sap flux). Table areas shaded in dark grey denote oviposition substrates that are not relevant for the particular model. Values were estimated on a phylogeny based on both nuclear and mitochondrial gene sequences. Analyses were performed across 100 trees from a BEAST posterior distribution using nuclear and mitochondrial genes. The values represent the average of the theta values and standard error over 100 trees.

(C) Comparison of AICc and weighted AICc values for models testing the relationship between oviposition substrate and egg volume, residual egg volume and thorax volume. Values are for model fit of Brownian motion (BM) and Ornstein-Uhlenbeck with one optima (OU1) or with multiple optimum (OUM) with different combination of oviposition substrate categories, calculated with the R package OUwie v.1.48 [S3]. Oviposition substrates were categorized as follows: OU2 categorizes species that lay eggs on bark and non-bark; OU3 categorizes species into bark-breeder, spider egg/flower breeder, and other; and OU8 categorizes each species according to the eight oviposition substrates represented (bark, flower, spider egg, fruit, leaf, generalist, fungus, sap flux). Models were tested over 100 posterior distribution BEAST trees using nuclear and mitochondrial gene sequences.

4

Table S1. Summary of ovariole number, body size, and egg size from Hawaiian Drosophila species documented in this study, compared to a previous study, Related to Figure 1. Species name, species subgroup, and egg-laying substrate of each species is listed along with measurements of adult life history traits, along with the standard deviation and number of specimens. Ovariole numbers are per adult female. Left columns represent data collected in the present study, and right-hand columns represent data from Kambysellis and Heed [S2]. Species with overlapping data points that were not within two standard deviations are denoted in yellow. For egg volume, the length and width of the eggs were compared between the two studies, as Kambysellis and Heed [S2] did not include egg volume calculations and their standard deviations. Thorax volume, calculated as the cube of thorax length, is used as a proxy for body size.

5

Table S2. Results of best BLAST hit and parsimony analysis using mitochondrial gene sequences from field-caught females, Related to STAR Methods. For each sample, which is denoted by the sample ID, the island of origin is listed. A BLASTn best hit was assigned when there was a hit to a single species that was >98%, and the accession number and species ID are listed. “m.h.” denotes cases where there were multiple 98–100% top hits, or there were multiple 92–97% hits. Table S3 contains an expanded list of the m.h. accession. For samples with multiple hits that fell into one specific species group, we noted the name of the species group in the column for BLASTn best hit species. “n.s.” indicates samples where sequence data are not available. Parsimony tree results indicate the species group or sister species that the sample was included in by a phylogenetic analysis, for details of the analysis see methods. Final species IDs were assigned to a specific species when BLAST and phylogenetic analysis showed matching results, and were assigned to a species group when BLAST and phylogenetic analysis indicated close similarity to a group of species. For species with morphological keys, the final species ID indicates the species ID based on morphology. Species names with grey background indicate samples where the barcoding and the morphological key did not provide consistent results. See also Figure 1.

6

Table S3. Multiple BLAST hit results from mitochondrial gene sequences of field-caught females, Related to STAR Methods. The accession number and species ID for mitochondrial sequences obtained from field-caught females are listed along with percent sequence similarity to the published sequences. Note that there were several instances where the 16S sequences had 99% sequence similarity to more than 20 Hawaiian Drosophila species of various groups; these are indicated as “Over 20 HI Dros spp.”. See also Figure 1.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse anti-Engrailed/Invected	Developmental Studies Hybridoma Bank	4D9
Biological Samples
DNA from field collected Hawaiian Drosophila	This study	Supplemental Table 2
Chemicals, Peptides, and Recombinant Proteins
FITC-conjugated Phalloidin	Sigma	Cat# P5282
Hoechst 33342	Sigma	Cat# B2261
Vectashield Anti-fade mounting medium	Vector Labs	Cat# H-1000
Dynazyme DNA Polymerase	Thermo Scientific	Cat#: F501
ExoSAP-IT	Thermo Scientific	Cat#: 78201
Critical Commercial Assays
DNeasy Blood and Tissue DNA extraction kit	Qiagen	Cat#: 69506
Deposited Data
COI sequences from picture wing species identified using morphological keys	This study	GenBank: MK276992–277193
Mitochondrial sequences from field caught Hawaiian Drosophila for DNA barcoding	This study	https://github.com/shchurch/hawaiian_drosophila_ovaries_2018/tree/master/sequence_data_methods/sarikaya2018_original_sequence_data
RAxML phylogenetic trees for species delineation using DNA barcoding	This study	https://github.com/shchurch/hawaiian_drosophila_ovaries_2018/tree/master/sequence_data_methods/species_delineation_gene_trees
Adult and larval phenotypic data	This study	https://github.com/shchurch/hawaiian_drosophila_ovaries_2018/blob/master/data.txt
Hawaiian Drosophila phylogeny	This study using sequences from [18, 19, 21, 23]	https://github.com/shchurch/hawaiian_drosophila_ovaries_2018
Oligonucleotides
COI F: ATT CAA CCA ATC ATA AAG ATA TTG G	[18]	NA
COI R: TAA ACT TCT GGA TGT CCA AAA AAT CA	[18]	NA
COII F: ATG GCA GAT TAG TGC AAT GG	[18]	NA
COII R: GTT TAA GAG ACC AGT ACT TG	[18]	NA
ND2 F: AGCTATTGGGTTCAGACCCC	[18]	NA
ND2 R: GAAGTTTGGTTTAAACCTCC	[18]	NA
Software and Algorithms
ImageJ	NIH	https://imagej.nih.gov/ij/
4Peaks	Nucleobytes	https://nucleobytes.com/4peaks/index.html
Phyutility v2.2.6	[56]	https://github.com/blackrim/phyutility
MAFFT v7.130b	[57]	https://mafft.cbrc.jp/alignment/software/
Gblocks v091b	[58]	http://molevol.cmima.csic.es/castresana/Gblocks/Gblocks_documentation.html
PartitionFinder v1.1.1_Mac	[59]	http://www.robertlanfear.com/partitionfinder/
RAxML v8.2.3	[60]	https://cme.h-its.org/exelixis/web/software/raxml/index.html
BEAST v2.3.2	[61, 62]	http://beast.community/
CIPRES cluster	Cyberinfrastructure for Phylogenetic Research	http://www.phylo.org/
AWTY	[64]	http://king2.scs.fsu.edu/CEBProjects/awty/awty_start.php
Tracer	[65]	http://www.beast2.org/tracer-2/
TreeAnnotator	[66]	http://beast.community/treeannotator
R v3.2.0	[67]	https://www.r-project.org/
Nlme v3.1–121	[40]	https://cran.r-project.org/web/packages/nlme/
Ape v3.3	[68,69]	https://cran.r-project.org/web/packages/ape/
corHMM v1.18	[39]	https://cran.r-project.org/web/packages/corHMM/
OUwie v1.18	[38]	https://cran.r-project.org/web/packages/OUwie/

Highlights.

• Ecology and developmental predict fecundity evolution in Hawaiian Drosophila.

• Where Hawaiian flies lay their eggs influences evolution of reproductive capacity.

• Allometric relationship between body and ovary size differs by habitat type.

• Changes in somatic gonad cell number explain convergent ovariole number evolution.