Behavioral evolution accompanying host shifts in cactophilic Drosophila larvae

Abstract For plant utilizing insects, the shift to a novel host is generally accompanied by a complex set of phenotypic adaptations. Many such adaptations arise in response to differences in plant chemistry, competitive environment, or abiotic conditions. One less well‐understood factor in the evolution of phytophagous insects is the selective environment provided by plant shape and volume. Does the physical structure of a new plant host favor certain phenotypes? Here, we use cactophilic Drosophila, which have colonized the necrotic tissues of cacti with dramatically different shapes and volumes, to examine this question. Specifically, we analyzed two behavioral traits in larvae, pupation height, and activity that we predicted might be related to the ability to utilize variably shaped hosts. We found that populations of D. mojavensis living on lengthy columnar or barrel cactus hosts have greater activity and pupate higher in a laboratory environment than populations living on small and flat prickly pear cactus cladodes. Crosses between the most phenotypically extreme populations suggest that the genetic architectures of these behaviors are distinct. A comparison of activity in additional cactophilic species that are specialized on small and large cactus hosts shows a consistent trend. Thus, we suggest that greater motility and an associated tendency to pupate higher in the laboratory are potential larval adaptations for life on a large plant where space is more abundant and resources may be more sparsely distributed.

adaptations have largely focused around plant chemistry, and the ability of insects to survive in and utilize novel chemical environments (Becerra, 1997;Ehrlich & Raven, 1964;Futuyma & Agrawal, 2009).
One ecological variable that has received considerably less attention is the physical structure of the host plant, such as its shape, volume, and more importantly the usable resource distribution within the host. While adult insects are known to use aspects of plant or fruit shape as cues for oviposition on recently adapted hosts (Alonso-Pimentel, Korer, Nufio, & Papaj, 1998;Kanno & Harris, 2000;Prokopy, 1968), it is unclear how frequently insects have adapted specifically to maximize fitness on plants of different physical structure. Among the traits that potentially are influenced by the shape, volume, and size of the host plant are the foraging behavior of larvae. Adult foraging behavior has long been thought to be controlled by habitat structure (Moermond, 1979;Robinson & Holmes, 1982;Uetz, 1991). For insects, especially holometabolous insects, host structure is unlikely to define the foraging environment for adults, which is more likely related to the distribution of plants throughout the broader landscape. However, for larvae that can only crawl, the shape and volume of the host plant should present strict boundaries to the available foraging habitat. Therefore, we hypothesize that variation in the physical structure of the host plant, that is, its shape and volume, should influence larval insect behaviors related to foraging or motility.
Furthermore, additional closely related cactophilic species within the D. repleta species group have colonized variable cactus environments (Oliveira et al., 2012), providing the potential for deeper evolutionary comparisons.
If the shape and volume of the necrotic cactus resources define the boundaries of the foraging environment for larvae, we predicted that movement-related behaviors should differ between flies inhabiting cacti of different shapes, and specifically that the behavior of larvae native to larger and longer columnar cacti should reflect an ability to forage across greater distances, potentially allowing access to additional or preferable sources of nutrition. Conversely, individual larvae utilizing prickly pear cladodes will be restricted by the size of cladode itself, limiting the need to travel long distances to forage.
To test these predictions, we quantified pupation height and third-instar speed across the four D. mojavensis populations under common garden conditions. We found that flies from the Catalina Island population (prickly pear) were both slower and pupated closer to the food resource than flies from the other populations, especially the Sonoran population (columnar). Furthermore, the specialist species D. navojoa and D. nigrospiracula, which inhabit primarily small (prickly pear) and large cactus (saguaro and cardón), respectively, display consistent results for larval speed. The generalist D. arizonae displays intermediate phenotypes for both pupation height and speed. Lastly, F 1 crosses between the Catalina Island and Sonoran populations suggest that speed and pupation height are genetically independent phenotypes in D. mojavensis. We argue that both phenotypes are likely related to the shape, volume, and size of the host cacti.

| Experimental insects
We utilized isofemale lines of D. mojavensis originally collected from Drosophila nigrospiracula was maintained on potato flake media with a necrotic saguaro homogenate mixture (Castrezana, 1997).
We quantified both pupation height and larval activity in isofemale lines of each of the four D. mojavensis populations. We

| Larval activity assays
To assay larval activity, we placed 7-10-day old virgin flies in mixed sex vials with the appropriate media under a 14:10-hr light:dark cycle at 50% humidity and 25°C for 24 hr before removing all adults and allowing eggs to hatch and develop undisturbed.
Upon reaching the third instar, we removed larvae from vials in randomly selected groups of five and placed them on a 10 cm petri dish partially filled with 1% agar. Larvae in the third-instar stage were determined by body size and used irrespective of specific age, because species as well as the four D. mojavensis populations vary substantially in developmental time (Etges, 1990; J.M.C. pers. obs.). We then recorded each group of larvae for 5 min using a Point Grey video camera (FLIR Systems, Wilsonville, OR, USA), and retained images taken every 5 s. To ensure that we disregarded an initial period of low activity after transfer, we utilized only the 50 s before each larva reached the wall of the experimental chamber (whereupon estimating the position of the larva became imprecise) for analysis. We analyzed the mean speed of each larva during this 50 sec period using the TrackMate plugin of the ImageJ software package (https://imagej.nih.gov/ij/). All activity trials were performed in full light conditions, in the afternoon between 12:00 and 3:00 p.m.

| Pupation height assays
To measure pupation height, mated flies from stocks (see above) were maintained at low density in glass 8-dram vials with bananamolasses media and allowed to oviposit for 24 hr before being removed. Eggs were allowed another 24 hr to hatch into first instar larvae before being collected. Using a needle, 40 newly hatched larvae were placed in fresh 95 mm tall 8-dram glass vials containing approximately 10 ml of banana-molasses medium. Vials were capped using a packed cotton plug (Genesee Scientific), and then incubated at 25°C in 50% humidity on a 14:10 hr light:dark cycle. Larvae were then allowed to develop without disturbance. Once the larvae pupated, the distance between the surface of the food and the highest tip of the pupae was measured in millimeters using a digital calliper.

| Statistical analyses
We analyzed pupation height data using GLMs modeled with quasipoisson error structures to account for non-normality of the data, which contained a high number of zero values. We analyzed larval speed data using GLMs modeled with gaussian error structures.
We used Tukey's HSD from the multcomp package (Hothorn, Bretz, & Westfall, 2008) in R to perform all pairwise post hoc comparisons for each GLM. We calculated Pearson's coefficient to estimate correlations between mean pupation height and mean speed across isofemale lines within the D. mojavensis Catalina Island and Sonora populations. To assess the effects of genotype on each phenotype, we analyzed isofemale line as a nested effect within population using the lme4 package (Bates, Mächler, Bolker, & Walker, 2015) in R. We then used the ANOVA function to compare the performance of the mixed models to identical models with the nested term removed. All statistical analyses were performed in R 3.4.0 (https://www.R-project.org).

| Phenotypes of F 1 crosses
Third-instar speed of the Sonoran population was not significantly different from either F 1 cross, but was, as expected, greater than the Catalina Island ( Figure 3). Speed of the Catalina Island population was slower than the Catalina female by Sonora male cross but not the reciprocal. The F 1 crosses were also not different from each other (Appendix 5).
The Sonoran population had significantly higher pupation heights than Catalina Island flies or either F 1 cross (Figure 4). Both F 1 crosses and Catalina Island populations displayed no difference among them in pupation (Appendix 6).

| D ISCUSS I ON
The hosts used by phytophagous and saprophytic insects display dramatic phenotypic variation along several axes. Many of these traits, including chemical traits, have been found to incur selective pressures on their insect cohabitants, leading to local adaptation (Bernays & Chapman, 1994;Thompson, 1968;Throckmorton, 1975;Wiens et al., 2015). However, physical characteristics such as shape, volume, and size of host plants or usable resource within a host plant (e.g., necrotic section) have seldom been investigated for their role in creating novel selective environments for insects. For insect larvae, the shape and volume of the host on which eggs are oviposited should constrain their movement, defining the boundaries of their foraging environment. Thus, we considered larval behavioral phenotypes as strong candidates for local adaptation to plants with variable physical characteristics. To examine this prediction, we measured pupation height and third-instar activity in four populations of D. mojavensis, which has colonized multiple cactus host species throughout southwestern North America.
Larval speed was greater in D. mojavensis populations living on taller, larger cacti, and lowest in the Catalina Island population inhabiting necrotic prickly pear cladodes. Interspecific data also support this relationship, as D. nigrospiracula specializing on the saguaro and cardón cactus have very fast larvae, while D. navojoa inhabiting prickly pear are especially slow. Larval speed has generally been interpreted as a foraging related trait in Drosophila, exemplified by the rover/sitter polymorphism of D. melanogaster (Sokolowski, 1980). Furthermore, Sokolowski (1980) has suggested that increased speed may be an adaptation to widespread or discontinuous food sources.
This type of distribution is likely to be a characteristic of the larger columnar arms of organ pipe, agria, and saguaro, where necroses are known to occur in patches at the ends of arms (Nobel, 1980).
Organ pipe cactus has been previously observed to be associated with the lowest egg-to-adult viability when flies from Sonora and other populations are reared on it (Date, Crowley-Gall, Diefendorf, & Rollmann, 2017;Etges & Heed, 1987). This could partly explain the pattern of increased motility of Sonoran larvae, but not for all populations (e.g., agria rots are of better quality, but Baja California larvae are among the fastest). On the small cladodes of prickly pear, slow speed may be advantageous because of energetic costs (Roff, 2002) when the potential advantages of high motility and long foraging distances have been removed. Furthermore, increased larval activity on a smaller resource might prove detrimental given that a potential severe fitness cost would be imposed to those larvae that wander out of the necrotic host resource. However, it is unlikely that cactus shape is the lone selective pressure shaping larval speed differences. Other environmental factors potentially influencing speed between species and populations might include temperature, toxicity, nutritional composition, competition between con-or heterospecific larvae, predation or parasitism. High toxicity is expected to select for slower speeds (Borash, Teotonio, Rose, & Mueller, 2000;Mueller et al., 2005), and because there are differences in chemical composition of columnars and prickly pear cacti (Kircher, 1982;Stintzing & Carle, 2005), this represents another environmental factor which could affect speed differences.
Pupation height also differed between D. mojavensis populations; larvae from larger columnar or barrel cacti pupated higher than larvae from shorter cacti, specifically, those living on prickly pear. As with speed, this behavioral difference may be related to the ability to efficiently utilize space. Studies of pupation in a variety of Drosophila species have attributed pupation location behavior to a variety of environmental factors, including biotic conditions such as conspecific density (Beltrami, Medina-Muñoz, Arce, & Godoy-Herrera, 2010;Sokal, Ehrlich, Hunter, & Schlager, 1960) and abiotic factors such as temperature (Dillon, Wang, Garrity, & Huey, 2009), light (Manning & Markow, 1981), moisture (Sameoto & Miller, 1968), and chemical composition (Beltrami et al., 2010). Therefore, pupation behavior in a given necrotic cactus host may reflect the ability of a larvae to take advantage of increased opportunities to find optimal pupation sites and mitigate the many biotic and abiotic risks listed above. We suggest that it is the ability to pupate higher, rather than actual pupation height, that may be advantageous on larger cacti. This could help explain the broad distribution of pupation phenotypes in the columnar and barrel D. mojavensis populations, all of which contain many individuals which pupate at low heights.
Variation in speed or pupation height might also be related to experimental conditions. It has been predicted that inbreeding should result in slower larval speeds in Drosophila (Bauer & Sokolowski, 1984). Consistent with this explanation in our dataset is the fact that D. nigrospiracula, the most active species, is also by far the most recently introduced to the laboratory, and therefore the most outbred.
However, among the D. mojavensis populations, the slow Catalina Island population isofemale lines were actually the most recently established in the laboratory, though all have been maintained for well over the 20 generations of inbreeding required to purge nearly all genetic variation in Drosophila (Falconer & Mackay, 1996;Huang et al., 2014). Furthermore, the Catalina Island population has a significantly lower level of segregating variation and effective population size compared with the other three D. mojavensis populations (Machado, Matzkin, Reed, & Markow, 2007;Matzkin, 2004;Matzkin & Eanes, 2003;Reed, Nyboer, & Markow, 2007) and hence, the possible effect of inbreeding would be the least in this population.
Also, the overestimation of phenotypic differences between isofemale lines is only a problem when lines are maintained with small numbers of individuals or for a handful (<10) generations (Hoffmann & Parsons, 1988), neither of which apply to our data. Thus, though inbreeding may be influencing pupation height or larval activity broadly, it is not likely leading to the specific pattern of differences reported here.
An additional possibility regarding pupation is that larval speed and pupation height are not truly distinct behavioral phenotypes.
Might larvae pupate higher simply because their activity is greater, thus simply increasing the chances of being higher at the time of pupation? This would also be consistent with the broad distribution of pupation heights in the columnar/barrel D. mojavensis populations.
However, we present several lines of evidence suggesting that this is not the case. First, we crossed the Catalina Island and Sonoran populations, which had the most extreme phenotypes for both traits.
While crosses indicated a quantitative basis for both characters, both F 1 hybrids displayed pupation height phenotypes insignificantly different from the Catalina Island parental population. This suggests a dominance effect to the alleles governing pupation height that we did not observe for third-instar speed. Second, though sample sizes were small, we observed no evidence for a positive correlation of pupation height and speed within either the Catalina Island or Sonora populations. If pupation height was simply a consequence of increased speed, then lines within a population with higher speed should also exhibit higher pupation heights. Lastly, despite its remarkably high speed, previous work on D. nigrospiracula suggests that it does not pupate especially high compared to other cactophilic species (Fogleman & Markow, 1982). This also matches findings that pupation height and larval speed have distinct genetic bases in D. melanogaster (Bauer & Sokolowski, 1985;Sokolowski, 1985). Therefore, we suggest that speed and pupation height are truly separate traits potentially responding to environmental conditions on different cactus hosts.
We argue that host physical structure represents a strong candidate for a selective environment acting on both larval activity and pupation behavior. Given the importance of larval feeding on adult fitness, we expect host shape, volume, and size to impart similar selection pressures on other phytophagous and saprophytic insects and predict that activity and other traits related to resource utilization should consistently differ in larvae when insects have undergone host shifts to resources with novel physical characters.

ACK N OWLED G M ENTS
We thank C. Allan for technical assistance. J. Hunt and two anonymous reviewers provided helpful comments on the manuscript. This project was supported by a National Science Foundation grant (IOS-1557697 to LMM).