Large-scale assessment of olfactory preferences and learning in Drosophila melanogaster: behavioral and genetic measures

In the Evolve and Resequence method (E&R), experimental evolution and genomics are combined to investigate evolutionary dynamics and the genotype-phenotype link. This approach requires many replicates with large population sizes, which imposes severe restrictions on the analysis of behavioral phenotypes. Aiming to use E&R for investigating the evolution of behavior in Drosophila, we have developed a simple and effective method to assess spontaneous olfactory preferences and learning in large samples of fruit flies using a T-maze. We tested this procedure on (a) a large wild-caught population and (b) 11 isofemale lines of Drosophila melanogaster. Compared to previous methods, this procedure reduces the environmental noise and allows for the analysis of large population samples. Consistent with previous results, we show that flies have a spontaneous preference for orange vs. apple odor. With our procedure wild-derived flies exhibit olfactory learning in the absence of previous laboratory selection. Furthermore, we find genetic differences in the olfactory learning with relatively high heritability. We propose this large-scale method as an effective tool for E&R and genome-wide association studies on olfactory preferences and learning.

methodological issues apply to behavioral and other traits. Despite the success in identifying some 48 causative genes [e.g. 12,13], theoretical [6,14] and empirical evidence [11,15] has clarified that many 49 of the significantly changed variants are in fact false positives derived by short or long-distance 50 linkage disequilibrium. Another limit that E&R shares with other genome-wide approaches, is low 51 6 in a large population of D. melanogaster originally caught in South Africa and in a population of 130 inbred lines originally caught in Portugal. Our procedure reduces the impact of undesired selective 131 pressures and the effort in propagation and phenotyping. Furthermore this method is sensitive 132 enough to detect learning, spontaneous preferences and heritability of these traits. We discuss the 133 relevance of this T-maze based procedure for E&R and genome-wide association studies. The individual-line experiments were ran on 11 inbred lines derived from a D. melanogaster 146 population originally collected in Povoa de Varzim (Portugal) in July 2008. Before inbreeding, 147 B101, B192 and B211 were maintained in the laboratory as isofemale lines. R1-R10 were derived 148 from the same original population after being exposed to an experimental evolution procedure at 149 collected male were allowed to mate and from their offspring another virgin female and a random 156 male were used to create the next generation. After inbreeding, these lines were kept as isofemale 157 lines until the experimental assays. In each trial we used a group of 40 flies (males and females 2 158 days old or older) of the same line. 159 160

Apparatus and stimuli 161
The T-maze (31 x 17.5 cm) used for the experimental assays (Fig. 1A) consisted of a starting 162 chamber and a central chamber (12 x 8 x 1.5 cm) connected on each side to a food chamber. The 163 starting chamber (9.5 x 2.5 cm) contained the flies at the beginning of each experimental phase. 164 Food chambers (9.5 x 2.5 cm) were filled with experimental food. In each experimental phase flies 165 begun the exploration of the apparatus from the starting chamber. The central chamber was 166 connected to the food chambers with a funnel that prevents flies to re-enter the central chamber 167 once they have approached the food. A similar trapping technique has been previously used for 168 fruit flies. 169 Experimental media prepared with juice fruit (either orange or apple juice from 100% concentrate) 170 and agar (14 g/l). Aversive food was obtained adding 8 g/l of quinine to the experimental medium. 171 172 Procedure 173

Unconditioned olfactory preferences 174
We assessed unconditioned preferences for apple and orange odor by using the same procedure 175 described for the learning assays (see below), with the only difference that no food supplemented 176 with quinine was provided during the exposure phases. This similarity enables us to investigate 177 the role of the conditioning procedure on spontaneous preferences. 178 179 180

Olfactory learning assays 181
We used CO 2 anesthesia to collect flies and starve them 15-16 hours before the beginning of the 182 conditioning procedure. After starvation flies were moved to the starting chamber for Exposure 1. 183 During Exposure 1, for two hours flies were exposed to the odor associated with the aversive 184 flavor and the aversive flavor (e.g. orange odor and orange juice supplemented with quinine). 185 Flies who entered the food chambers during Exposure 1 were moved to the starting chamber for 186 Exposure 2. During Exposure 2, for two hours flies were exposed to the odor associated with the 187 palatable flavor and the palatable flavor (e.g. apple odor and apple juice). Flies who entered the 188 food chambers during Exposure 2 were starved for four hours prior to the Test. 189 In half trials we conditioned flies on apple odor associated with aversive food and orange odor 190 Flies began the Test from the starting chamber. Differently from the exposure phases, during the 197 Test the odor associated with the aversive flavor and the odor associated with the palatable flavor 198 were presented simultaneously, each on a different food chamber (Fig. 1C). We alternated the 199 right/left side in which the two odors were presented. No food was supplemented with quinine 200 during this phase. We counted flies that chose to enter either the orange odor side or the apple 201 odor side.

Data analysis 210
In the test for spontaneous preferences between the orange and apple odor we compared the 211 proportion of flies that across 28 trials chose the orange odor vs. the random choices level using a 212 t-test single sample against the random choices proportion of 0.5. Beforehand we controlled for 213 deviations from the normal distribution of the data using the Shapiro-Wilk normality test. To Before testing flies, we exposed them to both odors/flavors: in half trials flies were exposed first 251 to orange then to apple (O/A), in half trials first to apple then to orange (A/O). We have derived 252 the order effect score o to investigate the effect of the order in which the orange/apple stimuli had 253 been presented. We observed a significant order effect score (t 13 =3.09, p=0.009; Fig. 2B), 254 indicating that A/O flies (flies first exposed to Apple, then to Orange) had a significantly higher 255 preference for orange odor than O/A flies (flies first exposed to Orange, then to Apple).

265
In the conditioning experiment the overall population showed a preference for the orange odor 266

Inbred lines: spontaneous preferences and learning assays 289
To study olfactory preferences and learning in 11 inbred lines of D. melanogaster derived from a 290 population collected in Portugal we used the same procedure adopted for the large population 291 using 40 flies from the same isofemale line in each trial. Line R6 consistently did not enter the 292 food chambers in both experiments, so we excluded this line and run the analyses in the ten 293 remaining lines. 294 In the spontaneous preference assay the overall distribution of the orange odor choices was 295 significantly different from the normal distribution (Shapiro-Wilk normality test: W=0.98, p=0.03) 296 and we analyzed the data using non-parametric tests (Wilcoxon signed-rank test and Kruskal-297 Wallis test). Overall, the group of ten responsive lines showed a spontaneous preference for the 298 orange odor (mean=0.56; V=7862, p<0.001; Fig. 4A). We did not observe significant differences 299 across lines (Kruskal-Wallis Chi squared 9 =14.14, p=0.12; see

310
In the learning assay, the overall distribution of the orange odor choices was significantly different 311 from the normal distribution (Shapiro-Wilk normality test: W=0.98, p=0.005) and we analyzed the 312 data using non parametric tests (Wilcoxon signed-rank test and Kruskal-Wallis test). Overall, after 313 the conditioning procedure the ten responsive lines showed a preference for orange odor 314 %A), and we documented a significant learning effect (Fig. 5B). We also detected significant 320 differences in the proportion of orange choices between lines (Kruskal-Wallis Chi square: 23.45, 321 p=0.005). 322   . 6A and B). After using the Bonferroni-Holmes correction for multiple comparisons, 332 we found that in the A-/O procedure three lines (R10, R7, R9) had a preference significant at 5% 333 level for orange and four lines (R1, R2, R3, R5) had a preference significant at 10% level, whereas 334 in the O-/A procedure only R5 had a preference significant at 10% level. These results indicate that 335 most of the tested inbred lines are able to discriminate between apple and orange odor. Bonferroni-Holmes correction for multiple comparisons we found that three lines (R3, R7, R9) 344 showed a learning score significant at 5% level and three lines (R1, R2 and R5) that were 345 significant at 10% level. These results suggest that most of the tested lines are able to learn 346 through our conditioning procedure.

Inbred lines: heritability of olfactory behavior 353
We derived an estimate of the genetic heritability of olfactory preferences and olfactory learning 354 using the variance between (Vb) and the variance within (Vw) lines and calculating the intraclass 355 correlation t as a proxy for heritability in inbred lines [41,42]. 356 In the olfactory preferences, the variability between lines (Vb=0.09) was larger than the variability 357 within lines (Vw=0.06) and the intraclass correlation is t=0.6. The same pattern holds true for 358 olfactory learning: the variability between lines (Vb=0.017) is much higher than the variability 359 within lines (Vw=0.004), thus leading to t=0.80. The high intraclass correlations show a moderate 360 to high heritability of olfactory preferences and learning and suggest that our method is suitable to 361 investigate these traits. This procedure though is not optimal for E&R due to several drawbacks: the effort required for 388 propagation (e.g. eggs have to be rinsed and/or individually displaced on culture media), the fact 389 that selection is imposed only on half of the propagated subjects (females) and males or