SEARCH

SEARCH BY CITATION

FilenameFormatSizeDescription
gbb12146-sup-0001-Appendix S1.docWord document1969KAppendix S1: Prediction of relaxation to linkage equilibrium.
gbb12146-sup-0002-Figure S1.docWord document100KFigure S1: Proof of principle for learning selection. Groups of white-eyed rut+ and red-eyed rut1 mutant animals were mixed in equal proportions and trained in the olfactory learning assay (see Methods). This mixed population was trained and tested in the standard T-maze (Tully & Quinn 1985; see Methods). The fraction of animals that avoided the CS+ odor was then re-trained and re-fractionated at the choice point of the T-maze. This fractionation procedure was iterated, and the ratio of white (wild type, high learning) vs. red-eyed (rut1 mutant, low learning) animals was determined after each round. Using this procedure, we are able to enrich for rut+ flies so that after just four choices, 85% of the animals are rut+. Proportions of red- and white-eyed individuals are shown after one, two, three or four choices. The number of animals at start of each experiment was 200 (one choice), 400 (two choices), 800 (three choices) and 2000 (four choices).
gbb12146-sup-0003-Figure S2.docWord document38KFigure S2: Relaxation times (to 1/e) for all pairs of loci on chromosomes II and III in the collection of 60 mutants described in the screen by Dubnau et al. (2003). The median is 20.95 generations. Relaxation times are calculated as τ = 1 − rij, where rij = 4nfd, nf is the proportional fraction of females in the population, and d is the distance between i and j in centimorgans.
gbb12146-sup-0004-Figure S3.docWord document83KFigure S3: Selected improvement in short-term memory does not impact long-term memory performance. Consolidated long-term memory was tested for one of the selected populations by testing the performance 24 h after a standard 10× spaced training procedure. Memory performance of Morgan was compared with that of rut+ wild-type control and of rut1 mutant animals. Performance of Morgan is not significantly different from that of rut1 (Tukey HSD).
gbb12146-sup-0005-Figure S4.docWord document517KFigure S4: Loci dosage by populations at generation 11. Mean dosage vs. loci, with SEs, sorted by mean dosage for each of the six populations in generation 11 (selected – MOR11, MUL11 and LEW11; unselected – BRI11, DOB11 and STU11). All populations show significant differences across many loci (see Table S4).
gbb12146-sup-0006-Figure S5.docWord document522KFigure S5: Loci dosage by populations at generation 25. Mean dosage vs. loci, with SEs, sorted by mean dosage for each of the six populations in generation 25 (selected – MOR25, MUL25 and LEW25; unselected – BRI25, DOB25 and STU25). All populations show significant differences across many loci (see Table S5).
gbb12146-sup-0007-Figure S6.docWord document38KFigure S6: Singular values as obtained from the SVD. The spread in data points at each mode are the singular values obtained from each of 100 imputations. The small variation suggests that the imputation method does not significantly affect the decomposition results.
gbb12146-sup-0008-Figure S7.docWord document103KFigure S7: Discriminant vector is robust to dropping samples. (a) Heat map displays the frequency with which the amplitude ranking for each allele is the same as in the original discrimination vector (Fig. 2c) when 90% of the points are dropped. (b) Effect of dropping 10%, 50% or 90% of samples on the direction and length of the discriminant vector. The angle θ11–25 is the direction between the vector w discriminating selected vs. unselected populations in generations 11 and 25, and |w11|, |w25| are the lengths of the discrimination vectors in generations 11 and 25, respectively. The values in the table are averages over 100 imputations of the genotype matrix.
gbb12146-sup-0009-Figure S8.docWord document200KFigure S8: Effects on rut1 memory performance of single alleles identified in by SVD/LDA. Effects are shown for the five single alleles identified by the SVD/LDA that were not tested in Fig. 3, as well as for E1847, whose amplitude opposed the separation between groups (Fig. 2c). These five loci and E1847 showed no effects on performance of rut1 heterozygous females or rut1 hemizygous males (Tukey HSD). rut1 hemizygous performance is shown for comparison. N > 7 for all groups. We also tested three additional alleles that were not identified in the SVD/LDA (D0753, E4294 and E0361) and found that none of them suppressed rut1 either on their own or in di-allele combinations with the top eight identified alleles (data not shown).
gbb12146-sup-0010-Figure S9.docWord document76KFigure S9: Percentage of suppressed genotypes necessary to explain population-level response to selection. Percentage of genotypes necessary for observed response to selection at generation 41 shown as a function of the performance of a suppressing genotype (blue curve). Vertical dashed lines indicate the mean performance (42.0) of unselected populations, the mean performance (52.0) of E3272/rut trans-heterozygous crosses and the mean performance (66.9) of selected populations at generation 41. The horizontal line indicates mean frequency of E3272 alleles in selected populations. The blue curve was calculated by keeping mutant performance (Xmut) constant at 0.42, keeping the overall performance level constant at 66.9, and solving for the necessary suppressing genotypes (Xsup). This allowed us to determine the necessary suppressing genotypes as Xsup ranged from 55% to 100%.
gbb12146-sup-0011-Figure S10.docWord document98KFigure S10: Effects on rut1 memory performance of E3272 homozygote. rut1; E3272 double homozygous animals exhibit no difference in short-term memory performance (Tukey HSD) than rut1. N = 6 for all groups.
gbb12146-sup-0012-Figure S11.docWord document70KFigure S11: Numerous combinations of three to six loci must contribute to selection response. Simulated selection experiments, using parameters similar to biological selection experiments, estimate that up to six heterozygous loci, working in combination, contribute to the observed selection response. (a) A clear response to selection in simulated populations (N = 104 populations, for each N-mer combination) is observed within 40 generations for fully rescuing heterozygous combinations of less than seven loci. Selected alleles (red, mean frequency 1 SD of all N selected alleles from all populations) reach a frequency in the population of 0.5 within 40 generations for N-mer combinations of N < 6. When N = 6, selected allele frequencies have not reached 0.5 in most populations, though it is possible. Unselected alleles in the same populations are subject to random drift (black, mean frequency 1 SD of all 23 – N alleles). (b) The probability of an individual population converging on a solution of N heterozygous loci is estimated as the proportion of simulated populations that reached a mean frequency of selected alleles 0.48 over generations 38–40. For N = 2,3, P = 1. For N = 4, P = 0.8605. For N = 5, P = 0.4067. For N = 6, P = 10−4. For N = 7–10, P < 10−4.
gbb12146-sup-0013-Figure S12.docWord document68KFigure S12: Selection model validation. Monte Carlo simulation provides a realistic model of selection experiments. In order to simplify the simulations we considered the case where a single fully heterozygous allele combination confers selective advantage. When selection is maximally strong, such suppressing combinations appear and reach optimal frequency (0.5 for each of the underlying loci) within 40 generations for combinations of less than six loci. With genetic solutions involving six loci, the underlying alleles do not reach the frequency of 0.5 within the timecourse of our experiment and for combinations involving seven or more loci, solutions are never found within 40 generations (with 10 000 simulated populations). When the model is initialized with no selective pressure, alleles behave as expected for a model of drift (a). In some cases, allele frequencies drift to 0 (example shown in black), while most alleles fluctuate around the starting frequency 0.14 (examples shown in red, green and blue). When selective advantage is given to the heterozyote at a single locus (b), the allele frequency reaches an optimum 0.5 with a timecourse dependent on selective advantage. With selective advantage of 0.01 (blue curve, mean across populations 1 SD), 0.05 (green curve) or 0.28 (red curve), frequency reaches 0.5 in approximately 90, 20 and 2 generations, respectively. A combination of one homozygous allele with two heterozygous alleles given an advantage of 0.28 behaves as expected (c). The homozygous allele frequency goes to fixation at 1 (blue curve), while the frequency of the heterozygous alleles is maintained around 0.5 (green curve). All simulations model population size (200 breeding pairs), number of loci (23) and starting allele frequencies (0.14) of the experimental selection (see Methods). N = 103 populations for all simulations shown.
gbb12146-sup-0014-Table S1.xlsExcel spreadsheet69KTable S1: Sequences used to make probes and primers for genotyping of P-element insertion alleles. The locus for which primers and probes were designed is in column 1 (ABI). Column 2 shows the sequence to which gene specific primers were designed (ABI). The location of the P-element insertion is denoted with a ‘P’.
gbb12146-sup-0015-Table S2.xlsExcel spreadsheet53KTable S2: Genotype proof of principle. A two-fluorophore assay was developed in which FAM (F) detects wild type and VIC (V) detects mutant. DNA was extracted from homozygote mutant animals and from wild-type animals. ‘Heterozygote’ animals were mimicked by mixing wild type and mutant animals in a 50/50 mixture. The results for 10 alleles are shown in column 1. Column 2 shows genotype calls for a wild-type animal. Wild-type FAM fluorphores ‘FF’ are detected at all 10 loci. Odd columns 3–21 show genotype calls for animals that were homozygous for a single allele. For mutant animals, the VAM fluorophore ‘VV’ is detected using the allele-specific probes and primers and the wild-type FAM fluorophore ‘FF’ at the other nine loci. Genotype calls for DNA samples from mixed ‘heterozygous’ DNAs are shown in even-numbered columns 4–22. For these simulated heterozygote animals, both fluorophores ‘VF’ are detected for the locus in question only. Genotype calls identified ‘FF’ (wild type) at the other nine loci.
gbb12146-sup-0016-Table S3.xlsExcel spreadsheet662KTable S3: Raw genotype data for all samples in the experiment. −1, No Call; 0, homozygous wild type; 1, heterozygous; 2, homozygous mutant.
gbb12146-sup-0017-Table S4.docWord document118KTable S4: Generation 11. Number of loci with a significantly lower mean than the listed allele, as determined from a one-way anova within each population.
gbb12146-sup-0018-Table S5.docWord document112KTable S5: Generation 25. Number of loci with a significantly lower mean than the listed allele, as determined from a one-way anova within each population.
gbb12146-sup-0019-Table S6.docWord document29KTable S6: Effect of shuffling labels on the properties of the discrimination vector w. θ11–25 is the direction between the vector w discriminating selected vs. unselected populations in generations 11 and 25, and |w11|, |w25| are the lengths of the discrimination vectors in generations 11 and 25, respectively. The values in the table are averages over 100 imputations of the genotype matrix.
gbb12146-sup-0020-Table S7.docWord document31KTable S7: The effect of dropping single populations from the data set on the robustness of the discrimination vector w. θ11–25 is the direction between the vector w discriminating selected vs. unselected populations in generations 11 and 25, and |w11|, |w25| are the lengths of the discrimination vectors in generations 11 and 25, respectively. The values in the table are averages over 100 imputations of the genotype matrix.
gbb12146-sup-0021-Table S8.xlsExcel spreadsheet47KTable S8: Co-occurrence of E3272 with each of the other 22 loci. The expected frequencies [P(X), row 1) are calculated as the product of E3272 and a given second allele's frequency. The observed frequency [P(X I E3272) is shown in row 2. A chi-square test was used to identify significant effects (chi-squared value assuming chi-squared distribution with 1 degree of freedom shown in row 3); the P-value based on this chi-square value is shown in row 4. To account for multiple (22) comparisons, Bonferroni correction was used. A difference is significant only if it has a P-value of >0.0023. Significantly ‘higher’ or ‘lower’ combinations are shown (row label ‘SIGNIFICANT’).
gbb12146-sup-0022-Table S9.xlsExcel spreadsheet649KTable S9: Genotypic diversity in each population. The summary spreadsheet shows an overview of how many genotypes are unique in each population, the total number of flies in that population and the percentage of unique genotypes. Additional spreadsheets show actual unique genotypes ranked by count (the last column on these sheets shows the count). Unique genotypes calculated for three different cases: missing data completely removed, using only a single imputation (see Appendix S1), and combining all 100 imputations.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.