Characterizing changes in individual preferences in Drosophila melanogaster. A. 2 hr sample of centroid-tracking data for a fly in a circular arena. Each point is colored based on whether it is moving CCW, CW, or radially in the arena. B. Sample of 100h of continuous recording for 4 individual flies (colors). Hourly means of turning indices are shown in light lines. Dashed lines are lowpass filtered with a timescale cutoff of 24 h. Solid lines are lowpass-filtered shuffled data showing the average tendency across the experiment. C. Mean power spectrum of turning data for all flies (n=252) for actual and shuffled data. Shaded areas represent 95% confidence intervals generated via bootstrapping (n=1000). D. Schematic of Y-maze assay. Flies make either a left or right turn each time they walk through the intersection. E. Standard deviation of daily right biases vs average right bias across days for individual flies (points). Colors indicate DGRP genotype. F. Autoregressive model of individual right bias over time with parameters to estimate the initial right bias variability (σBet-Hedging) and rate of daily change in right bias (σDrift). G. Posterior estimates of σDrift, σBet-Hedging, and φ (the autoregressive parameter characterizing the rate of reversion to zero bias), for each DGRP genotype. Right: Countour plot of 2-dimensional joint posterior over σDrift and σBet-Hedging Lines represent deciles of the posterior distribution, with the outermost line representing 95% posterior density.. H. As in G-Left, posteriors estimates of right bias variability parameters for flies treated with 5-HTP, AMW, and controls. I. As in H, for control and mutant flies with a missense mutation in TrH generated by in vivo CRISPR.

Modeling adaptive effects of drift. A. In a simplified model in which both phenotypes and environments have two states, the optimal fraction of the population that should change preference (fshift) over a period of time equals the probability the environment changes (pshift). See Supplementary text REFREF. B. Model for the number of individuals with a particular continuous preference as a function of time in a fluctuating environment and individual age. nϕ,α,t, is the number of individuals with preference ϕ, age α at day t in the simulation. Two different cases determine this value, one for flies surviving from the previous timestep (α > 0), and one for flies born in a particular timestep (α = 0). The number of flies of a particular behavioral phenotype surviving on each successive day is determined by a function of how that preference is from the ideal preference on that day (orange), the total number of flies that already have, or drift into having that phenotype on that day (red), and a bounding term that stops the distribution of preferences from diffusing away from general range of what is adaptive (purple). The key behavioral strategy parameter from these terms is σd, which determines the rate at which flies’ preferences drift over time. The number of new flies born each day is given by the total number of flies above the age of reproductive maturity amin (red) times the birth rate β. New flies are born with an initial preference from a normal distribution centered on the long term environmental mean with a standard deviation given by σbet-hedging (blue) C. Example environmental fluctuations and corresponding fitness landscape showing final populations for differing amounts of drift and bet-hedging. Green dot indicates ideal strategy (ii)D. Population over time for strategies marked with roman numerals in (C). E, F. As in (C) for two additional example environmental fluctuation patterns.

Effects of amplitude of environmental fluctuation, frequency of fluctuation, and generation time on ideal amounts of bet-hedging and drift. A. Fitness landscapes over the space of environmental fluctuation amplitude and frequency. Each heatmap shows the geometric mean of the log change in population for each combination of drift and bet-hedging over 100 randomized environments. Heatmaps are normalized to their maximum and minimum values. Rows of heatmaps have the same environmental fluctuation frequency and columns have the same environmental fluctuation amplitude (as measured by the standard deviation of all timepoints σMean. The nine amplitude and frequency combinations in this panel correspond to values denoted with white boxes in (C). B. As (A), except columns of heat maps have the same generation times, as determined by amin. C. Optimal amounts of bet-hedging (warm color scale; left) and drift (cool color scale; right) for each combination of environmental fluctuation amplitude and frequency. White squares indicate values associated with heatmaps in A. Dotted blue line corresponds to an environmental fluctuation period of 20 days, which is twice the (fixed) generation time in these simulations. D. As (C), except for optimal amounts of bet-hedging and drift for each combination of environmental fluctuation frequency and amin. White squares indicate values associated with heatmaps in (B).

Optimal bet-hedging and drift strategies for real world environmental fluctuations. A. 1000 days of relative humidity data from the Arikaree River in Colorado, USA (left panels) were used to generate an environmental selection filter with low (top-left) or high (bottom-left) amplitude fluctuations (σmean in Figure 2B). Fitness landscape heatmaps over bet-hedging and drift strategies for different ages of reproductive maturity (amin). B. As in (A), using average daily temperature data from Longreach, Australia. C. Pipeline for comparing the optimal variability strategies of organisms subject to real world environmental fluctuations. Daily environmental time series from many sites were collected, normalized, and used in the model to produce fitness landscapes over σBet-Hedging and σDrift. All landscapes were then subject to principle components analysis. These simulations held amin and σmean constant. See Methods. The loadings of PC1 (97.8% of the variance; top-right) indicate that this component encodes the optimal amount of bet-hedging, while PC2 (1.9% of the variance; bottom-right) encodes optimal drift. D. Environmental time series from specific locations, plotted on PC2 vs PC1 axes, colored by optimal amount of bet-hedging. E. As in (D), except color indicates optimal amount of drift. F. As in (D), except color indicates the type of environmental measurement. G. As in (D), except color indicates the the Köppen climate classification of their location.