The ability of Drosophila hybrids to locate food declines with 1 parental divergence

16 17 Hybrids are generally less fit than their parental species, and the mechanisms underlying 18 their fitness reductions can manifest through different traits. For example, hybrids can 19 have physiological, behavioral, or ecological defects, and these defects can generate 20 reproductive isolation between their parental species. However, the rate that mechanisms 21 of postzygotic isolation other than hybrid sterility and inviability evolve has remained 22 largely uninvestigated, despite isolated studies showing that behavioral defects in hybrids 23 are not only possible but might be widespread. Here, we study a fundamental animal 24 behavior – the ability of individuals to find food – and test the rate at which it breaks 25 down in hybrids. We measured the ability of hybrids from 94 pairs of Drosophila species 26 to find food and show that this ability decreases with increasing genetic divergence 27 between the parental species and that male hybrids are more strongly (and negatively) 28 affected than females. Our findings quantify the rate that hybrid dysfunction evolves 29 across the diverse radiation of Drosophila and highlights the need for future 30 investigations of the genetic and neurological mechanisms that affect a hybrid’s ability to 31 find a suitable substrate on which to feed and breed. 32 33 INTRODUCTION 34 35 Identifying and understanding the evolution of traits underlying reproductive 36 isolation (RI) between species is central to our understanding of how biodiversity is 37 generated and maintained (Palumbi 1994; Turelli et al. 2001; Coyne and Orr 2004; 38 Widmer et al. 2009; Nosil 2012). Traits that generate RI between populations or species 39 (collectively referred to as ‘taxa’) can act at different stages in the lifecycle of an 40 organism. Traits affecting prezygotic isolation act before, and postzygotic traits act after, 41 the formation of a zygote (Via 1999; Schluter 2001; Turelli et al. 2001; McKinnon et al. 42 2004). Prezygotic traits therefore tend to affect mating or gametic compatibility between 43 taxa, while postzygotic traits typically involve developmental and behavioral dysgeneses 44 that reduce the fitness of hybrids relative to pure-species individuals (Turelli et al. 2001; 45 Coyne and Orr 2004). 46 A large diversity of traits affecting RI have now been identified (Coyne and Orr 47 2004; Widmer et al. 2009; Nosil 2012) and a number of studies have employed 48 phylogenetic comparative methods to measure the rates that their effects change with 49 increasing divergence. In their seminal paper Coyne and Orr (1989) analyzed how mate 50 discrimination (a prezygotic trait) and hybrid sterility and inviability (both postzygotic 51 traits) evolve between pairs of Drosophila species as the divergence between them 52 increases. Coyne and Orr (1989) showed that when two species were allopatric, mate 53 discrimination, hybrid sterility, and hybrid inviability evolved at similar rates; however, 54 when the species were sympatric, mate discrimination evolved more rapidly than both 55 sterility and inviability. The general finding that traits affecting RI tend to become 56 stronger as a monotonic function of the genetic distance between species has now been 57 documented in diverse taxa including fungi (Giraud and Gourbière 2012), fish 58 (Mendelson 2003; Bolnick and Near 2005), frogs (Sasa et al. 1998), lizards (Singhal and 59 Moritz 2013), butterflies (Presgraves 1998), and angiosperm plants (Moyle et al. 2004; 60 Owens and Rieseberg 2014). These studies provide evidence that mismatched or 61 dysfunctional phenotypes evolve faster in prezygotic than in postzygotic traits (Coyne 62 and Orr 1989; Mendelson 2003; Bolnick and Near 2005; but see Edmands 2002), 63 manifest more rapidly in hybrids of the heterogametic sex than in the homogametic sex 64 (Coyne and Orr 1989; Presgraves 1998), and have asymmetric patterns of penetrance 65 with respect to the maternal direction of a given cross (Bolnick and Near 2005). 66 Traits that affect the sexual compatibility between taxa or the fertility and survival 67 of their hybrid offspring frequently have phenotypes that are easily measured in the lab 68 and manifest independently from environmental cues. In addition to infertility and 69 increased levels of mortality, hybrids can show aberrant behaviors in their ability to 70 migrate (Delmore and Irwin 2014), secure mates (Noor et al. 2001; Bridle et al. 2006), or 71 locate food (Linn et al. 2004). These aberrant and potentially maladaptive behaviors 72 indicate that hybrid behavioral dysgeneses may represent an important form of RI 73 between species. However, the evolution of dysfunctional hybrid behavioral traits has not 74 been systematically assessed across a single taxon. 75 Drosophila is a diverse genus that contains over 2,000 species and spans ~60 76 million years of evolutionary history (Tamura et al. 2004). Species of Drosophila can be 77 readily hybridized in the laboratory, and hybrids between some species can be found in 78 nature (Llopart et al. 2005; Matute and Ayroles 2014). This makes them a suitable system 79 for applying a comparative framework for studying the evolution of traits affecting RI 80 (Yukilevich 2012, 2013; Nosil 2013; Rabosky and Matute 2013). We applied such a 81 framework to test how the ability of hybrid Drosophila to locate food breaks down as a 82 function of divergence between their parental species. 83 Foraging for food is a fundamental behavior of Drosophila species. Drosophila 84 tend to feed, mate, lay eggs, and develop into adults on ephemeral and patchy resources 85 such as rotting fruit, flowers, or cacti. For mobile animals in their natural environment, 86 foraging is a complex decision-making process that integrates detecting food, assessing 87 its quality, and moving towards it. Components of this process influence the likelihood 88 that an individual will ultimately find appropriate dietary resources (Lee et al. 2004). It is 89 also an essential behavior for both viability and fertility because it directly affects the 90 energy available to an individual (extensively reviewed in Kamil et al. 1987). Aspects of 91 foraging behaviors have been studied in D. melanogaster (Barsh and Schwartz 2002; Lee 92 et al. 2004; Murakami et al. 2005; Semmelhack and Wang 2009; Carlsson et al. 2010; 93 Root et al. 2011) as well as other species (Jones 2005; McBride 2007; Linz et al. 2013). 94 However, we still lack a systematic assessment of foraging behaviors in hybrids and of 95 how aberrant behaviors might evolve as parental divergence increases. 96 Here we address this question by conducting behavioral assays with 94 unique 97 interspecific crosses between a total of 73 species of Drosophila. We then employed a 98 phylogenetic comparative framework to assess the rate of hybrid behavioral breakdown 99 as a function of the genetic distance between the parental species, looked for differences 100 between the sexes (Coyne and Orr 1989; Orr 1997; Delph and Demuth 2016), and 101 determined if the direction of the parental cross (in terms of the maternal genotype) 102 affected hybrid performance (Bolnick and Near 2005). We also measured the rate that the 103 ability of hybrids to locate food is restored using advanced-generation backcrosses. Our 104 results indicate that, as with prezygotic behavioral traits and fertility, the ability of 105 hybrids to locate dietary resources breaks down as the genetic distance between the 106 parental species increases, and male hybrids are more severely affected than females. 107 Unlike previous studies of different traits, we do not find any asymmetry between the two 108 cross directions in the ability of hybrids to find food. Our results suggest that behavioral 109 defects observed in hybrids, such as the ability to locate food, could serve as barriers to 110 genetic exchange and be important for the persistence of species boundaries in nature, an 111 hypothesis that warrants further study in natural populations. 112 113 METHODS 114 115 Species pairs 116 117 We used 94 species pairs that produced progeny when hybridized and a total of 73 118 distinct species (Tables S1-2; Figure S1). The species belonged to eight of the 12 119 Drosophila radiations, and the hybridizations were the result of crossing species with 120 divergence times spanning approximately 0.1 to 10 million years. Genetic divergence 121 between the species pairs (Nei's distance [D]) was obtained from Yukilevich (2012). To 122 confirm that Nei's D is an appropriate and robust predictor of genetic distance, we 123 regressed Nei's D to available estimates of neutral genome-wide divergence for the D. 124 melanogaster subgroup (Ks, Turissini et al. 2015) and found a strong correlation between 125 the two (Spearman’s rho = 0.5596, P = 0.0016; Figure S2). 126 127 Generating hybrids 128 129 All lines used to generate hybrid crosses and backcrosses were maintained in 200 130 mL bottles at similar densities (approximately 200 flies) with corn meal food 131 supplemented with yeast at 24oC under a 12 h light/dark cycle. The origin of each species 132 is shown in Table S3. We collected virgin males and females from pure-species stocks 133 maintained under standard conditions within 12 hours of eclosion. Flies were collected 134 under brief CO2 anesthesia and kept for three days in single-sex groups of 20 individuals 135 in 30mL, food-containing vials. On the morning of the fourth day, we placed forty males 136 and twenty females together at room temperature (21°–23°C) to mate en masse on corn 137 meal media. We used four day old flies for all assays because all of the species we 138 studied are reproductively active at this time, and although different species mature and 139 senesce at different rates, running assays on flies of the same absolute age minimized 140 confounding effects of age on their ability to find food. We set up at least 20 crosses per 141 species combination. Conspecific crosses were done in parallel to ensure that all flies 142 within an experimental run were the same age. Once we observed active larvae within the 143 vials we added a Kimwipe (Kimberly Clark, Delicate Task) dampened with 0.5% 144 propionic acid solution as a pupation substrate. When adult F1s began to emerge, we 145 collected them every 12 hours (as virgins) under CO2 anesthesia and maintained them in 146 sex-specific containers for four days before running behavioral assays. 147 148 Assaying the ability of Drosophila to locate food 149 150 We determined whether flies of a given genotype (either pure-species or hybrid) 151 could locate food by presenting them with a choice between moving to an empty 36 ml 152 polystyrene vial or to a similar vial containing fig. To conduct these assays we first 153 isolated groups of approximately 100 individuals of the same sex and genotype in an 154 empty vial, provided them with a source of water (i.e., a damp plug), and starved them 155 overnight (8-12 hours). We connected the vial containing the flies to two other vials the 156 following morning, one on either side (Figure S3), with blue 100-1000 uL pipette tips 157 that had the first 3 cm of their tip cut off and were inserted through the center of cellulose 158 acetate flugs (Genesee Scientific , San Diego, CA). This allowed flies to move out of the 159 central vial into either side vial, but did not allow them to move back into the center vial. 160 One of the side vials contained ~ 5 cm of dried organic fig (Ficus carica ‘Black 161 Mission’, Woodstock Foods, Providence, RI) and the other was empty. The position of 162 the vial that contained the fig, relative to the experimenter, was randomized across 163 replicates. Preliminary experiments showed that flies move towards fig at a similar rate to 164 other food types (banana and cornmeal; data not shown) and we have collected at least 25 165 different species of Drosophila from species of fig in the wild, suggesting it is a realistic 166 dietary resource. To facilitate rapid assembly of the apparatus, we dampened the flugs 167 that connected each of the two vials to the central vial. We then allowed the flies to move 168 between vials for three hours. We scored flies as locating food if they were present in the 169 vial containing the fig at the end of the three-hour assay and also recorded the number of 170 flies remaining in the center vial or moving to the empty vial. 171 To determine if the presence of fig generated a gradient in humidity (a cue that 172 the flies may detect rather than food), we used SHT75 relative humidity and temperature 173 sensors (www.sensirion.com; Staefa, Switzerland) to measure relative humidity (RH) at 174 four points across our experimental apparatus: one in the center of each of the empty and 175 fig vials and two in the center vial. We measured RH to the nearest 100 of a percentage, 176 cycling through each the four sensors every five seconds, from the food vial to the empty 177 vial, in two independent runs. In the first run we recorded RH over ~46 minutes (140 178 measurements per sensor) and in the second run we recorded RH over ~140 minutes (422 179 measurements per sensor). We found no detectable difference in RH across our apparatus 180 (linear models: run 1: F3, 558 = 0.0023; P = 0.9998; run 2: F3, 1687 = 0.0011; P = 1), and the 181 largest observed differences in RH were the empty vial having a RH 7.9% greater than 182 the fig-containing vial (run 1) and the fig-containing vial having a RH 0.09% greater than 183 the empty vial (run 2). However, these differences were within a single 20-second cycle, 184 and mean differences in humidity across all cycles between the empty and fig vials were 185 negligible (run 1: 0.002% [SD = 0.834%]; run 2: 0.015% [SD = 0.027%]). 186 187 The relationship between hybrid dysfunction and the genetic distance between 188 parental species 189 190 Prior to assaying hybrid flies, we assayed 73 pure-species to determine the 191 ‘baseline’ expectation for the proportion of flies that could locate food under our 192 experimental design. For pure-species offspring, we ran 10 replicates per species per sex, 193 with a minimum and mean number of individuals per replicate of 79 and 97, respectively. 194 We determined whether pure-species flies varied in their ability to locate food by fitting a 195 generalized model using the glm() function in R version 3.2.4 (R Core Team 2016) with a 196 binomial distribution and a logit link function. The response variable in this model was 197 (number of flies in the food vial) / (total number + 1) and the predictor was species. We 198 added one to the total number for each trial since all of the flies went to the food vial in 199 over 83% of the trials, and the logit of 1 is undefined. 200 After confirming that pure-species flies could readily locate food under our 201 experimental design, we turned to determining whether there was a relationship between 202 the genetic divergence between a pair of Drosophila species and the ability of their 203 hybrid offspring to locate food. We ran between 10 and 15 replicates per type of parental 204 cross, per sex, with a minimum and mean number of individuals per replicate of 64 and 205 96, respectively. We handled the replicates in two ways. Analyses described as using the 206 replicate data treated each replicate as a separate data point, and for the other analyses, 207 we used unweighted means across all of the replicates for each cross. We obtained 208 similar results when using weighted means. The average difference in the proportion of 209 flies going to the food vial between weighted and unweighted means was 1.53 × 10 with 210 a standard deviation of 0.0017, and the largest difference was 0.0092. We used Nei’s D 211 between parental species (Yukilevich 2012) as an estimate of genetic divergence and the 212 proportion of hybrid offspring (out of all individuals) moving to the vial containing fig as 213 an estimate of the ability of those hybrids to find food. Both experimental assays and 214 analyses were performed separately for each sex to minimize the effect of intersex 215 behavioral interactions. 216 Ideally, all observations in our data set would be phylogenetically independent; 217 however, our data set contains multiple instances where phylogenetic independence 218 breaks down. This occurs when the edges in a phylogenetic tree are traversed more than 219 once when connecting pairs of parental species and results from either a species being 220 crossed to multiple other species or multiple species from a clade being crossed to 221 multiple species from another clade. Phylogenetic trees are not available for all of the 19 222 clades that we used. To account for instances of phylogenetic non-independence we 223 utilized two approaches: species-level and clade-level sampling. For the species-level 224 sampling we randomly sampled crosses such that each species was only included once in 225 the data set. Each data set contained between 28 and 31 species pairs for females and 27 226 and 29 species pairs for males. For the clade-level sampling, we carried out the most 227 stringent correction possible and randomly sampled a single cross from within each of 19 228 monophyletic clades of Drosophila (Figure S1, Tables S1-2) resulting in a data set 229 composed of 19 species pairs. Each clade appears as a separate graph in Figure S1. We 230 repeated both procedures (species-level and clade-level sampling) 10,000 times and fitted 231 models (described below) to the resampled data independently for each iteration. We 232 chose 10,000 iterations so each species would be represented in at least several hundred 233 iterations with both the species-level and clade-level sampling (Tables S1, S2). 234 We investigated the relationship between the proportion of individuals that 235 located food and the genetic divergence between the parental species by comparing five 236 models: linear, logistic, exponential decay, exponential decay with an asymptote term, 237 and a four parameter dose response curve. The five models describe how a hybrid’s 238 ability to locate food decreases with parental divergence in different ways. The linear 239 model assumes a constant rate and is the only model where the proportions are not 240 bounded between 0 and 1. The logistic model is based on the assumption that each fly 241 represents a Bernoulli trial and yields a reverse ‘S’ shaped curve where no hybrids find 242 food at high parental divergences. The exponential model has a consistently slowing rate 243 of decay, and no hybrids are able to locate food at high levels of parental divergence. The 244 exponential with an asymptote model has a similar rate but allows for a nonzero 245 proportion of hybrids to locate food at high parental divergences. The dose response 246 model begins with an initial period where hybrids can retain their ability to locate food 247 followed by a swift decay and also allows for hybrids to locate food at high parental 248 divergences; it results in a reverse ‘S’ shaped curve. 249 The linear model was fitted using the lm (library 'stats') function in R version 250 3.2.4 (R Core Team 2016), and the logistic model was fitted using glm function with a 251 binomial distribution and a logit link function. The other three models were fitted using 252 the nlsLM (library 'minpack.lm') function in R. We used nlsLM instead of lm or glm 253 since some of the models had more than 2 parameters and could not be fit as a linear 254 model or generalized linear model. The exponential decay models followed the form: 255 !!""# = !!!!" where Pfood is the proportion of flies in the food vial, a is the y-intercept (the proportion 256 of flies in that vial when Nei’s D between the parental species is 0), and r is the rate of 257 decay. The exponential models with an asymptote term allow the curve to approach a 258 proportion d at large Nei’s D and were: 259 !!""# = !+ !!!!". 260 The four parameter dose response models had the form: 261 !!""# = ! + !− ! 1+ ! ! ! where a is the proportion of flies in a given vial for Nei’s D = 0, b is the rate of decay 262 with increasing genetic distance, c is the inflection point of the decay curve, and d is an 263 asymptote term for large Nei’s D. 264 To determine which of the five models best described the relationship between 265 hybrid dysfunction and genetic divergence we used Akaike Information Criteria (AIC, 266 Akaike 1974). We compared models for each iteration of sampling separately for the 267 species-level and clade-level sampling. We compared the AIC values for each model by 268 ascribing weights to each model using weighted AIC (wAIC) with the following 269 equation: 270 !! = ! !∆! ! ! !∆! ! !!! 271 Where i refers to the model, n is the total number of models, and Δi is the difference 272 between AICi and the minimum AIC. The model with the largest wi was taken as the best 273 fit for each iteration. 274 When the parental species are highly diverged, hybrid flies may lose their ability 275 to detect food or may even actively avoid it. We used two approaches to test whether the 276 proportions of male and female hybrids in the empty vial were greater than the 277 proportions in the food vial. First, we ran a Wilcoxon test (wilcox.test() function in R) to 278 see if the food empty vial proportions from the replicate data were less than zero (zero 279 is the null hypothesis of random movement across vials). Second, to account for 280 potentially low power stemming from the small numbers of replicates per cross (min=5, 281 max=18), we pooled crosses with Nei’s D greater than a threshold value. Wilcoxon tests 282 were run to see if the food empty proportions were less than zero for pooled data with 283 thresholds ranging between 0 and 1.6 with a step size of 0.05. This approach resulted in 284 larger sample sizes and enabled us to test whether the proportion of hybrids moving away 285 from food exceeded the proportion moving toward food as the divergence between the 286 parental species increased. 287 288 Sex effects 289 290 We next tested whether hybrids of one sex were better at finding food than the 291 other sex using two approaches. First, we carried out paired Mann-Whitney U tests on the 292 observed proportions of males and females that found food for each of the 10,000 293 iterations of species-level resampling. Second, we tested whether males were consistently 294 worse at finding food by comparing predicted values from the dose response and 295 exponential models at different values of Nei’s D (we focused on these two models as 296 they are favored over the other three models, see Results). Specifically, we conducted 297 10,000 iterations each for species-level and clade-level sampling as described above and 298 obtained fitted male and female curves for each iteration. Then for a given value of Nei’s 299 D, we subtracted the model-predicted female proportion from the male proportion for 300 each iteration generating a distribution of 10,000 male – female differences and tested 301 whether the resulting differences differed from 0 using the wilcox.test() function in R 302 version 3.2.4 (R Core Team 2016). We ran this test for each value of Nei’s D between 303 0.01 and 1.62 with a step size of 0.01. These analyses provided evidence that male 304 hybrids were generally worse at finding food than females. 305 We also tested whether the ability to locate food decreased more rapidly in males 306 than in females as the parental divergence increased. We investigated how the ability to 307 find food changed by taking the derivative of the formulas used to fit the dose response 308 and exponential models. The resulting rate formula for the dose response model was: 309 !!!""# !" = ! !− ! !!!! !!(1+ ( ! )!)! And the formula for the exponential model was: 310 !!!""# !" = −!"! !!" We calculated rates using the parameter estimates from each of the 10,000 resampled 311 data sets for both the species-level and clade-level sampling. Because we were interested 312 in identifying levels of parental divergences (Nei’s D) where the male rate was faster than 313 the female rate, we subtracted the female rate from the male rate and looked for values of 314 Nei’s D where the rate difference was less than zero. For a given value of Nei’s D, we 315 had a distribution of 10,000 rate differences and tested whether these differences differed 316 from 0 using the wilcox.test() function in R version 3.2.4 (R Core Team 2016). We 317 applied this approach for each value of Nei’s D between 0.01 and 1.62 with a step size of 318 0.01. 319 320 Asymmetries in reciprocal crosses 321 322 Finally, we investigated whether there were sex-linked or cytoplasmic effects on 323 the ability of hybrids to locate food by testing whether hybrids from both directions of a 324 cross (with respect to maternal genotype) differed in their ability to find food. We treated 325 each hybrid sex independently and restricted our analysis to the 65 crosses that produced 326 both males and females in both cross directions. We looked for crosses with asymmetric 327 hybrid behavior by comparing the proportions of flies finding food between the replicate 328 data for the two cross directions separately for each cross using Mann-Whitney U tests 329 with the wilcox.test() function in R (R Core Team 2016). We accounted for multiple 330 testing using separate 5% false discovery rates for males and females. 331 332 Regaining normal behavior in serial-backcrossed families 333 334 For six pairs of species we carried out serial backcrosses (BCs) to test the rate at 335 which ‘normal’ behaviors could be restored in hybrids. To produce BCs we mated virgin 336 F1 females from both directions of the parental cross to males of the two parental species. 337 This procedure resulted in four BC families per species pair: female parent 1 × male 338 parent 2 F1s backcrossed to either parent 1 or parent 2 males and female parent 2 × male 339 parent 1 F1s backcrossed to either parent 1 or parent 2 males. We crossed females from 340 these BC families to pure-species males for 10 generations and, in each generation, 341 measured the ability of BC individuals to find food in the setup described above. We 342 predicted that the higher the proportion of a parental genome in a backcross generation, 343 the greater the likelihood that individuals within that generation would be able to locate 344 and move towards food. We calculated the expected proportion of BC individuals 345 locating food (!!"# !"#$%& !""# ) as: 346 347 !!"# = 1− 1− !!! !"#$%& !""# × 0.5! 348 Where !!"# is the expected proportion of individuals in generation t that move into the 349 vial containing food and 0.5! is the genomic fraction of the backcrossed parental genome 350 in a given generation where t = 1 to 10. 351 We tested whether BC individuals regained the ability to locate food by modeling 352 the proportion of individuals locating food (PBC) as a function of backcross generation 353 (generation) and sex using a generalized linear mixed effects model (GLMM). The 354 model had the form: 355 PBC ~ generation + sex + (1 + g | direction / family) 356 where the term in parentheses allows for variable slopes for each cross type (direction) 357 across generations. This term also models the specific BC family nested within each cross 358 type as a random effect (due to both the parental species used in the cross and the 359 direction of the cross). We assessed the significance of “BC generation” and “sex” by 360 comparing nested models lacking either the generation or the sex term to the full model 361 using likelihood ratio tests. 362 We also tested whether the ability of an individual to locate food was affected by 363 mismatches between their paternally inherited nuclear genome and their maternally 364 inherited mitochondrion and cytoplasm. After 10 generations of serial introgressions, the 365 nuclear genome retains a very small proportion of hybrid ancestry (0.01% on average) of 366 the species not used for the backcrossing. Despite the regression of the nuclear genome, 367 individuals of a given BC family inherit their mitochondrial genome from their mother, 368 and this results in either matched or mismatched nuclear and mitochondrial genomes 369 within advanced-generation BCs (depending on the female species used in the initial F1 370 cross and the males used to accomplish the backcrossing). We fitted a GLMM, for each 371 of the six types of hybrids, that modeled the proportion of individuals locating food as a 372 function of the BC generation and the match between their mitochondrial and nuclear 373 genomes (matched or mismatched). Each of these models included BC family as a 374 random effect to account for non-independence of measurements taken each generation 375 due to shared ancestry and took the form: 376 !!"#~ !" !"#"$%&'(#+ !"#$%&!'()* !"#$h+ 1|!" !"#$%&. 377 378 RESULTS 379 380 Assaying the ability of Drosophila to locate food 381 382 We assessed the ability of both pure-species and hybrid Drosophila to find food 383 in a controlled laboratory experiment. Pure-species flies of 73 different species from 384 across the Drosophila phylogeny (Table S4, Figure S4) almost invariably located food 385 (mean = 99.7%, minimum = 99.0% for males and mean = 99.6%, minimum = 98.3% for 386 females). We found no effects of species (logistic regression; Wald test: χ = 50.295, d.f. 387 = 72, P = 0.9758), sex (χ = 1.091, d.f. = 1, P = 0.2962), or the interaction between 388 species and sex (χ = 35.555, d.f. = 72, P = 0.9999) on the proportion of flies locating 389 food. Our data set contains 62 generalist species (e.g. D. melanogaster and D. yakuba) 390 and 11 specialist species that are restricted to a single substrate (e.g. D. sechellia, D. 391 erecta, and D. teissieri) (Table S4). The life history of a species (i.e. specialist or 392 generalist) did not affect the likelihood that they would move towards fig (generalized 393 linear mixed model including species as a random effect: Likelihood ratio test: χ = 0.20, 394 d.f. = 1, P = 0.67), suggesting that all species readily detect fig as a source of food. The 395 proportion of pure-species flies finding food was not affected by whether a cross was 396 carried out within an (inbred) isofemale line, between isofemale lines, or between outbred 397 lines (data not shown). 398 399 The relationship between hybrid dysfunction and genetic distance of parental 400 species 401 402 We found a strong relationship between the genetic distance between the parental 403 species and the ability of their hybrid offspring to locate food (Figure 1). Sex-specific 404 regressions indicate that for both sexes, hybrids between more divergent species are less 405 likely to find food (species-level sampling: Figures 2A-B, 2D-E; clade-level sampling: 406 S5-6). We modeled the ability of hybrids to find food using five models: linear, logistic, 407 dose response, exponential, and exponential with an asymptote (Tables S5-6). Although 408 the sex specific declines in the ability to find food are similar, the data sets are best fit by 409 different models. When we employed species-level sampling to correct for phylogenetic 410 non-independence, a female hybrid’s ability to locate food was best explained by an 411 exponential function of the genetic distance between her parental species (89% of 412 iterations versus 11% where the dose response model provided a better fit based on 413 wAIC; Figures 2A-C and S5). When sampling was done at the clade-level, the dose 414 response model best explained this relationship slightly more often than the exponential 415 model (51.7% of iterations versus 48.1%, respectively; Figures S9 and S11A). For male 416 hybrids, the dose response model best explained the relationship for both species-level 417 sampling (72.5% of iterations compared to 27.6% for the exponential model; Figures 2D418 F and S6) and clade-level sampling (81.8% of iterations compared to 18.2% for the 419 exponential model; Figures S11B, S10). 420 We predicted that if hybrid flies did not reach the food because of locomotion 421 problems they would remain in the central vial, whereas, if they suffered from defects in 422 food perception, more flies would move away from the food and into the empty vial. We 423 found support for the latter, as a significantly higher proportion of both male and female 424 hybrids were found in the empty vial compared to the food vial at higher parental 425 divergences. Specifically, when we pooled crosses above a given Nei’s D threshold, we 426 found a higher proportion of hybrids in the empty vial when Nei’s D ≥ 0.7 for females 427 (Wilcoxon tests, all P < 0.05; Figure S12A) and when Nei’s D ≥ 0.8 for males (Wilcoxon 428 tests, all P < 0.05; Figure S12B). When looking at individual crosses, we found a 429 significantly higher proportion of flies in the empty vial for 6 of the 11 crosses with Nei’s 430 D ≥ 0.88 for females (Wilcoxon tests, all P < 0.05; Figure S12C) and for one cross with 431 Nei’s D = 1.61 for males (Wilcoxon test, P < 0.05; Figure S12D). 432 433 Sex effects 434 435 Hybrid inviability and sterility frequently manifest in hybrid males (or more 436 generally, the heterogametic sex) at lower levels of parental divergence than in females 437 (Haldane’s rule; Coyne and Orr 1989). Looking at crosses that produced both male and 438 female offspring (N= 82) we find that hybrid males are generally worse at finding food 439 than females (10,000 resampled species-level data sets; Paired Mann-Whitney U tests; all 440 P < .01; Figure S13). We also calculated the predicted male proportion minus the 441 predicted female proportion using both the fitted dose response and exponential curves 442 for all 10,000 iterations of species-level (Figures 3A, 3C) and clade-level (Figures S14A, 443 S14C) sampling. Negative values for the difference indicate lower male proportions. 444 With both the dose response and exponential models, male proportions were lower than 445 female proportions for all tested values of Nei’s D between 0.1 and 1.62 for the species446 level sampling (Mann-Whitney U tests; all P < 10). Clade-level sampling yielded 447 similar results for the exponential model, but for the dose response model, the male 448 proportion was only significantly lower for Nei’s D ≥ 0.07 (Mann-Whitney U tests; all P 449 < 10; Figures S14A, S14C). 450 We also compared the rates at which the ability of males and females to find food 451 declined as parental divergence increased. We calculated the rates by taking the 452 derivatives of the dose response and exponential functions and subtracting the female rate 453 from the male rate. Since we were interested in declining fly performance and the slopes 454 were negative, a negative difference indicated a faster male rate. With species-level 455 sampling, the rate at which male hybrids lose their ability to locate food is greater than 456 females at lower divergences (dose response model: Nei’s D ≤ 0.41; exponential model: 457 Nei’s D ≤ 0.78; Mann-Whitney U tests, red bars along the bottom of Figure 3B and D), 458 where most of the ability to locate food is lost. The female rate is greater at larger 459 divergences, where the curves have begun to slow down as they approach their 460 asymptotes (Mann-Whitney U tests; blue bars in Figure 3B and D). These results are 461 qualitatively the same under clade-level sampling, where the male rates are greater when 462 0.04 ≤ Nei’s D ≤ 0.38 (dose-response model, Figure S14B) or Nei’s D ≤ 0.74 463 (exponential model, Figure S14D) 464 465 Asymmetries in reciprocal crosses 466 467 Hybrid sterility and inviability are commonly asymmetric between reciprocal 468 crosses, a pattern known as Darwin’s corollary to Haldane’s rule (Turelli and Moyle 469 2007). We therefore studied whether the loss of the ability to locate food differed with 470 respect to the maternal direction of a reciprocal cross. We restricted our analysis to 471 species pairs where both directions of a cross produced offspring (N = 69 for females; N 472 = 65 for males). We found no significant asymmetry after correcting for multiple tests 473 (Figure 4). The lack of pervasive asymmetry in a hybrid male’s ability to locate food 474 rules out a differential effect of the two X-chromosomes (in reciprocal crosses), while the 475 lack of asymmetry in females indicates that there is no differential effect of cytoplasmic 476 and maternal elements on the ability of hybrids to locate food. 477 478 Regaining function in serial-backcrossed families 479 480 The results we present above suggest that F1 hybrids will have reduced fitness 481 relative to their parental species because of their inability to locate food. If this behavioral 482 defect is driven by negative interactions between parental genomes, then the ability to 483 locate food should be regained in advanced-generation backcrosses, as their genomes will 484 increasingly resemble one species. We tested this hypothesis by serially backcrossing F1 485 hybrid females from 6 different species pairs to pure-species males from each of their 486 parental species for 10 generations. We found that backcrossed individuals rapidly 487 regained the ability to locate food (GLMM: estimate = 0.17; !!= 42.84; P < 1 × 10), 488 and the rate at which this behavior was restored tended to follow the expectation based on 489 the fraction of the dominant parental genome present in each BC generation (Figure 5; 490 Figure S15). However, there was also a general tendency for BCs to slightly 491 underperform this simple expectation (compare the colored lines with the black dashed 492 line in Figure 5), indicating that allelic effects such as inter-locus variation in effect size, 493 epistasis, and/or dominance likely contribute to the ability of flies to locate food. The rate 494 at which BCs increased their ability to find food also differed between the sexes, with 495 backcrossed males regaining the ability to find food more rapidly than backcrossed 496 females (GLMM: estimate [female-male] = -0.33; !! = 680.93; P < 1 × 10). 497 For 5 of 6 species pairs, the ability of backcrossed individuals to find food was 498 not affected by the mismatch (or match) between the maternally inherited mitochondria 499 from the initial cross and the paternal backcross direction (GLMMs; all P > 0.05). For the 500 sixth species pair, the effect of BC direction was marginally significant after correcting 501 for multiple comparisons for female BC individuals (P = 0.00434; adjusted α = 0.05 / 12 502 = 0.0042). These results therefore do not support a general effect of cyto-nuclear 503 interactions affecting the ability of BC individuals to locate food. Rather, the fact that 504 backcrossed flies rapidly regained the ability to locate food – typically in as few as 4 505 generations of backcrossing – suggests that this hybrid dysfunction results from 506 polygenic interactions between parental genomes. 507 508 DISCUSSION 509 510 We have shown that Drosophila hybrids are worse at locating food than their 511 pure-species parents and aspects of the breakdown of this behavior are similar to other 512 traits frequently studied in speciation. For example, we show that the ability to find food 513 breaks down with increasing genetic distance between the parental species, as is seen for 514 traits such as mating behavior, gametic compatibility, hybrid fertility, and survival to 515 adulthood (Coyne and Orr 1989; Presgraves 1998; Sasa et al. 1998; Mendelson 2003; 516 Moyle et al. 2004; Bolnick and Near 2005; Yukilevich 2012; Singhal and Moritz 2013; 517 Owens and Rieseberg 2014). Our experimental design covers a wide range of genetic 518 divergences between parental species, from D. americana eastern (brown morph) and D. 519 americana western (black morph) (Nei’s D = 0.01) to D. buzzatii and D. richardsoni 520 (Nei’s D = 1.61), allowing us to model the rate of decay. Rather than a linear decline, we 521 find that the decreased ability to locate food is best explained by either an exponential or 522 dose response model. For females the relative support for these two models is mixed 523 (based on wAIC) and depends on the phylogenetic correction we employ. For males, the 524 dose response model consistently provides a better fit to our data than the exponential 525 model. In biological terms, the exponential model has the fastest rate of hybrid 526 breakdown at low levels of parental divergence followed by a slowing rate of decline 527 until the complete inability to locate food at higher parental divergences. This pattern has 528 been discussed in the context of a ‘slowdown’ in the evolution of RI with increasing 529 genetic divergence (Gourbière and Mallet 2009). Alternatively, the dose response curve 530 is more similar to a logistic function: at low divergences, there is initially little or no loss 531 in the ability of hybrids to locate food followed by a sharp drop in performance and an 532 eventual asymptote at higher levels of divergence. 533 Similar to other postzygotic traits such as sterility and survival to adulthood, we 534 find that the heterogametic sex (males in the case of Drosophila) is worse at locating 535 food than the homogametic sex and this ability is lost more quickly as parental 536 divergence increases. For both hybrid sterility and inviability, this pattern is frequently 537 interpreted in light of Haldane’s rule: the overexpression of recessive deleterious alleles 538 in the heterogametic sex (Orr 1997; Delph and Demuth 2016). Hybrid males are not only 539 less capable of finding food than hybrid females but this ability also deteriorates at a 540 faster rate indicating that the hemizygous X-chromosome might be involved in behavioral 541 postzygotic isolation. Recessive (or semidominant) alleles involved in this deleterious 542 trait would be masked in females that carry two X-chromosomes. Additionally, at low 543 levels of parental divergence, males lose the ability to locate food more quickly than 544 females (Figures 3B, 3D). Female hybrids may, therefore, tolerate more divergence 545 between their parental species before their ability to locate food is reduced. 546 A major difference between our results and those for traits such as sterility and 547 inviability is that the latter two traits frequently have asymmetric magnitudes between 548 reciprocal crosses (Turelli and Orr 2000; Turelli and Moyle 2007; Yukilevich 2012). We 549 find no systematic asymmetries in the ability of hybrid flies to locate food in either our 550 main data set of F1 hybrid crosses or in the six advanced-generation backcross 551 populations. Asymmetries in the breakdown of postzygotic traits are interpreted as the 552 result of nuclear–cytoplasmic interactions, X–autosome interactions, genetic maternal 553 effects, or dominance relationships for genes of large effect (Coyne and Orr 1989; Turelli 554 and Moyle 2007; Yukilevich 2012). We found no evidence of these types of interactions 555 systematically affecting the ability of hybrids to locate food. Rather, the rate at which 556 backcrossed populations regain the ability to locate food suggests that this defect is 557 caused by polygenic interactions between parental genomes. The ability of backcrossed 558 individuals to locate food tends to follow the prediction based on the fraction of the 559 parental genome present in each backcross generation (Figures 5; S15-19); however, 560 there is also a small, but consistent, deviation from this expectation, with performance in 561 backcross families tending to be below expectations. This pattern suggests a more 562 complex (and realistic) genetic architecture underlying the ability to locate food including 563 epistasis and / or variation in per-locus effect sizes. Complex polygenic control of the 564 ability to locate and move towards food is perhaps unsurprising because foraging 565 involves traits that affect chemical detection, decision mating, and locomotion (e.g. 566 Sokolowski 1980; Mast et al. 2014). 567 Finally, our backcrossing experiment also produced individuals whose nuclear 568 genome was 99.9% pure-species, on average, yet contained the cytoplasm from another 569 species. We used this fact to evaluate the effects of combining the nuclear genome of one 570 species with the cytoplasm of the other. Backcrossed individuals from these families 571 found food as well as those that had nuclear genomes and cytoplasm from the same 572 species, strongly suggesting that the cytoplasm (e.g. the mitochondria) has no effect on 573 this behavior. 574 575 Potential Causes of a Hybrid’s Inability to Locate Food 576 As we suggest above, hybrids that did not find food may suffer from locomotion 577 problems, food avoidance, or food perception problems. At high levels of parental 578 divergence, defects in locomotion would result in flies remaining in the central vial, 579 active avoidance of food would result in more flies moving towards the empty vial, and a 580 breakdown in food perception would result in equal proportions of flies in the empty and 581 food vials. We found evidence for some level of active avoidance of food, potentially due 582 to problems with food perception. Specifically, at higher parental divergences we find 583 that there is a greater proportion of flies in the empty vial compared to the food584 containing vial (Figure S12). 585 One explanation for active food avoidance would be if the neural circuitry used to 586 detect olfactory cues is disrupted in hybrids such that their excitation elicits the opposite 587 response to that in their parents. As with the general inability to locate food, a breakdown 588 in neural circuitry could result from negative epistasis between parental alleles found in 589 hybrids. The behavioral breakdown we report here could therefore represent a type of 590 behavioral Dobzhansky-Muller interaction. The Dobzhansky-Muller model posits that 591 hybrid dysfunction is the result of maladaptive allelic interactions between divergent 592 genomes. The specific genetic factors and mechanisms affecting hybrid behavioral 593 dysfunction is a question that deserves future exploration. 594 One caveat of our study is that we assayed the ability of flies to locate food in the 595 lab, raising the possibility that our assay might not accurately model the ability to find 596 food in nature. Three of our results suggest that this potential issue may not have a large 597 effect on our general findings. First, parental species flies assayed in the same way as the 598 hybrids almost invariably move into the vial containing food. Second, we assayed known 599 specialist species (e.g., D. sechellia (Jones 2005), D. erecta (Linz et al. 2013), D. teissieri 600 (Lachaise 1983)) and they all chose to move into the vial containing figs, suggesting that 601 fig was an appropriate source of food and the flies perceived it as such. Third, in 602 backcrossed families, advanced-generation backcrossed individuals rapidly regained (in 603 as few as 4 generations) the ability to locate and move towards food. Together, these 604 results suggest that our assay was a suitable and accurate measurement of the ability of 605 flies to locate food. 606 A second caveat is that the flies we assayed could have relied on a signal other 607 than food when making their decision. We measured humidity and found that there was 608 no detectable gradient in humidity across our apparatus (Methods). That said, the flies 609 could have still been cueing in on a signal other than food or have been sensitive to 610 differences in RH below the resolution that we measured (the nearest 100 of a 611 percentage). Regardless of the specific cue, our results illustrate a behavioral dysfunction 612 present in hybrids that is predicted to dramatically reduce their fitness relative to their 613 parental species. 614 A final caveat is that we only used one line per species and it is possible that 615 variation in the genetic factors underlying hybrid dysfunction segregate within species. 616 We consider this unlikely given the consistent behavior we observed across pure-species 617 flies and the dramatic and consistent breakdown in hybrid behavior we observed when 618 the parental divergence exceeded Nei’s D ~ 0.2 (Figure 2). 619 620 Conclusion 621 Understanding the types of reproductive isolating mechanisms that evolve during 622 speciation is important for understanding how species arise and are maintained (Coyne 623 and Orr 1989, 2004; Hudson and Price 2014). Until recently, the study of traits 624 responsible for RI has largely focused on mating behavior, gametic compatibility, and 625 hybrid survival and fertility. However, many additional types of traits generate 626 reproductive isolation between species. Our findings highlight how behavioral defects 627 such as the ability to locate food may affect the fitness of hybrid individuals and 628 contribute to reproductive isolation. A breakdown in the ability to locate food has been 629 observed in hybrids between host races of Rhagoletis flies (Linn et al. 2004) and warrants 630 further study in species such as Drosophila that mate and lay their eggs at feeding sites. 631 Additional examples have documented behavioral dysfunctions in hybrids (Bridle et al. 632 2006; Delmore and Irwin 2014); however, our study is the first to show that the severity 633 of hybrid dysfunction in a fundamental behavioral trait scales with increasing divergence 634 between the parental species. 635 636 Acknowledgments 637 We thank members of the Jones, Burch, and Matute lab meeting, the editors M. 638 Noor and M. Streisfeld, and two anonymous reviewers for constructive comments on 639 previous drafts of this manuscript, UNC for startup funding, and Dr. Peter Charles for 640 help with measuring humidity within our experimental apparatus. The authors do not 641 have any conflicts of interest. 642 643 Data Accessibility 644 All raw data and scripts for analyzing and generating figures from those data will 645 be deposited on Dryad upon acceptance. 646 FIGURE LEGENDS 647 648 FIGURE 1. Proportion of hybrid offspring moving to each vial. Lines represent the 649 relative proportion of hybrid offspring found in each of three vials: food (blue), center 650 (green), and empty (red). Nei’s D is the genetic distance between the parental species. 651 Proportions for individual vials can be found in Figure 2 (food), Figure S7 (center), and 652 Figure S8 (empty). A) Full data set results for female hybrids. B) Full data set results for 653 male hybrids. Not all crosses produced viable male offspring. 654 655 656 0.0 0.5 1.0 1.5 0. 0 0. 2 0. 4 0. 6 0. 8 1. 0 female Genetic distance (Nei's D) Pr op or tio n food center empty A) 0.0 0.5 1.0 1.5 0. 0 0. 2 0. 4 0. 6 0. 8 1. 0 male Genetic distance (Nei's D) Pr op or tio n food center empty B)