Abstract
Handedness in humans – better performance using either the left or right hand – is personally familiar, moderately heritable1, and regulated by many genes2, including those involved in general body symmetry3. But behavioral handedness, i.e. lateralization, is a multifaceted phenomenon. For example, people display clockwise or counter-clockwise biases in their walking behavior that is uncorrelated to their hand dominance4,5, and lateralized behavioral biases have been shown in species as disparate as mice (paw usage6), octopi (eye usage7), and tortoises (side rolled on during righting8). However, the mechanisms by which asymmetries are instilled in behavior are unknown, and a system for studying behavioral handedness in a genetically tractable model system is needed. Here we show that Drosophila melanogaster flies exhibit striking variability in their left-right choice behavior during locomotion. Very strongly biased "left-handed" and "right-handed" individuals are common in every line assayed. The handedness of an individual persists for its lifetime, but is not passed on to progeny, suggesting that mechanisms other than genetics determine individual handedness. We use the Drosophila transgenic toolkit to map a specific set of neurons within the central complex that regulates the strength of behavioral handedness within a line. These findings give insights into choice behaviors and laterality in a simple model organism, and demonstrate that individuals from isogenic populations reared under experimentally identical conditions nevertheless display idiosyncratic behaviors.
In order to investigate whether flies display individual left-right locomotor biases, we developed a simple, high throughput assay to quantify turning. Flies were placed individually in Y-shaped mazes, allowed to walk freely for two hours, with their centroids tracked in two dimensions (Fig 1a-c, Supplemental Movie 1). Each maze was symmetrical and evenly lit, so that choices were unbiased rather than stimulus-driven. The fraction of times the fly passed through the center of the maze and chose to go right defined a “turn bias score” (Fig 1d). Each fly typically performed hundreds of choices per experiment (Fig S1a). Precise quantification of the distribution of individual behaviors requires high sample sizes, so many mazes were arrayed in parallel (Fig S1b-d). Thus, our results reflect over 25,000 individual flies and 16,000,000 turn choices.
We measured the turn biases of hundreds of individual flies from seven different fly lines: Berlin-K (BK), Canton-S (CS), Cambridge-A (CA,9), two lines of CS that were independently inbred for 10 generations, CA that was inbred for 10 generations10, and w1118, the background line for many transgenic flies (Figs 1e-f, S1e). As expected, the average probability of turning right (the turning score), averaged across all individuals within each line was ∼50%. However, this belied profound individual-to-individual variability, and an individual fly's probability of turning right often diverged markedly from the population average. For example, nearly one quarter (23.5%) of CS flies turned right greater than 70% of the time or less than 30% of the time. This is an unlikely outcome indeed if all flies were choosing to turn right with identical probabilities. This null hypothesis can be modeled using the binomial distribution, with each fly performing ni choices (equal to the number it performed in the experiment) and a probability of turning right p (equal to the mean probability observed across all flies). This is statistically justified because sequential turns were essentially independent of one another (Fig S1f). Compared to this null hypothesis, biased “righty” and “lefty” individuals are vastly over-represented (p < 10−16, 10−4 by χ2 test of variance and bootstrap resampling respectively).
We were unable to identify any trivial sources of left-right turning bias. Neither the light boxes, nor the maze arrays, nor the positions of the mazes within the arrays had any effect on the observed mean turning bias (Figs S1b-d). Anosmic flies11 displayed the same variability as control flies (Fig S1g), suggesting that flies were not following odor trails within the mazes. Lastly, general health or activity level did not explain the strong biases of flies; there is no correlation between turning score and number of turns completed in the 2hour experiment (Fig S1h).
Next, we evaluated the persistence of turning biases. Individual flies were tested in the Y-mazes, recovered, stored individually, and then tested a second time, in a different maze, either 1, 2, 6, 13 or 27 days later. Individual turning scores were highly correlated across time, and range from r=0.57 for day 1 vs. day 28 to r=0.81 for day 1 vs. day 2 (all p<0.0001, Fig 2a-d). The persistence of handedness through time provides further evidence that biases are not introduced by some experimental artifact. If, for example, flies were following a wall or a trail of odors or pheromones, these results would require that they do so in a highly reproducible manner, across long time scales, and in different randomly assigned Y-mazes.
Turning bias is evidently a persistent property of individual flies – perhaps it reflects a single source of behavioral chirality. We tested this hypothesis by measuring two additional lateralized behaviors: the direction of spontaneous exploration in circular arenas (Fig S2) and the folding arrangement of the wings at rest (Fig S3). We found that individual flies demonstrate a characteristic preference in the direction in which they circle. On average, flies spend equal amounts of time moving clockwise and counterclockwise, but individuals within the population often show strong preferences to circle in one direction or the other (Figure S2d-f). Likewise, individual flies exhibit preferences in which wing is placed on top at rest. Some fold left on top of right, others right on top of left (Fig S3a). As with the Y-mazes, both arena circling bias and wing-folding bias persist across days (Figs S2f, S3b). We tested individual flies in two assays each and found that a fly's turn bias in the Y-maze correlates significantly with a clockwise circling tendency in the arena (Fig S2g). In contrast, circling bias was completely uncorrelated to wing-folding bias. From these observations, we conclude that lateralized behavior is multifaceted, even within Drosophila, and that the turning biases we see in the Y-maze likely reflect an assay-independent locomotor bias phenomenon.
There are numerous possible causes of individual turning bias. One potential source of variation is the presence of polymorphic "lefty" and "righty" alleles in the population. However, locomotor handedness showed no heritability (Fig 2e, mean h2=−0.03, standard error=0.018, Fisher selection test of heritability12), and we found no evidence that inbreeding reduces variability in locomotor bias (Figs 1f, S1e). While an individual's handedness is not heritable, the total degree of variability at the population level is under genetic control, with some lines more idiosyncratic than others (see Fig 1, Canton-S vs. w1118 and 13) Another potential source of persistent locomotor bias is morphological asymmetry. We examined whether variability in leg lengths could account for turning biases. We tested 28 metrics of leg length asymmetry and found that just one correlates with turning bias, and weakly (r2=0.11, p=0.007, p=0.18 after multiple comparisons correction, Figure S4).
Given that neither cryptic genetic variation nor morphological asymmetry are major sources of variation in turning behavior, perhaps idiosyncratic locomotor asymmetries have a neurobiological basis. The central complex (CC) is a protocerebral structure with integral roles in processing sensory information and controlling locomotor output across arthropods10,14–17. We examined whether disrupting the CC can alter a population's distribution of turning biases. First, we tested seven mutants that perturb central complex development and morphology10. Of these, No-bridge, central-complex-deranged, and central-body-defect (cbdKS96) showed a significant increase in individual variation in turning as compared to heterozygous controls (Fig 3a). cbdKS96 is an missense mutant of Ten-a18, a transmembrane protein involved in axon targeting and synapse formation19,20, and causes severe and widespread defects in the fan-shaped body (FB), ellipsoid body (EB), and noduli (No), leading to high individual-to-individual variation in the gross morphology of the CC10. Thus, variability in the structure and function of central complex circuits may give rise to variability in turn bias.
We next sought to perturb central complex function more specifically with inducible transgenes21. We selectively silenced subsets of CC neurons using a panel of GAL4 lines to express a temperature inducible inhibitor of vesicle fusion (Shibirets or Shits)22. By comparing the median absolute deviations from the mean (MADs) of the distributions of turning scores at the permissive (23°C) and non-permissive (33°C) temperatures, we identified three GAL4 drivers that regulate the amount of turn bias variability in a population (Figs 3b, S4–5). Acutely disrupting the function of c465, R16D01, or R73D06 cells by silencing them with Shits caused large increases in the variability of turning scores. A similar effect resulted from acutely silencing c465 cells with GAL80ts;Kir2.115, or by hyperactivating them with dTRPA123 (Fig 3b).
The GAL4 lines c465, R16D01, and R73D06 drive expression in subsets of columnar neurons projecting from the protocerebral bridges (PB), to the fan-shaped body (FB) and contralateral No (“PFNs,” Figs 3c,d, 4 and S6, 7), with dendritic fields in the PB and axonal fields in the FB and No24,25 (Fig 3e). c465 is also expressed in the mushroom bodies, but silencing them had no effect on turn bias variability (Fig 3b). The only cell type present in all three of these lines are the PFNs (Figs 4, S6, S7). PFNs can be sub-classified into one of three types based on the regions of innervation within the FB and No25. Our data suggests PFNs projecting to No domain 3 may specifically be the regulators of turn bias variability (Figs 4c-d). Of the six GAL4 lines in our screen that had PFN expression, the three which had no effect all share strong expression in No domain 4 (Figs S8), hinting that silencing domain 4 PFNs might counteract or gate the effect of silencing domain 3 PFNs, a possibility which has some statistical support in our data (see Supplementary discussion).
Our results suggest that genetically and environmentally matched fruit flies exhibit individual differences in the neural processing of sensory information and the execution of locomotor patterns, resulting in profound levels of idiosyncratic handedness. Specifically, columnar PFN neurons of the central complex may be involved in the integration of bilateral sensory information, or the modulation of stimulus signal-to-noise ratios, and individual asymmetries in their functions may result in asymmetric behavioral outputs (see Supplementary discussion and Fig S9). I.e., when a fly must make a left vs. right decision in the absence of an asymmetric stimulus, asymmetries within the brain predispose the animal to go one way rather than the other. Perhaps this can help the animal avoid analysis paralysis, or perhaps it is a feature of noisy biological systems. Individual variation in wiring26–28, physiology29 and behavior9,30 may prove to be a very general feature of neural circuits, with broad implications both for our basic understanding of developmental neurobiology and the emergence of behavioral phenotypes at the individual level.
Methods
All raw data, data acquisition software and analysis scripts are available at http://lab.debivort.org/neuronal-control-of-locomotor-handedness/.
Fly lines
Flies were housed on modified Cal Tech medium (either from KD Medical or the Harvard University BioLabs fly food facility) according to standard protocols. A full list of the lines used in this study is available in the Supplementary Information. Flies used for Shibirets experiments were reared at 25°C and transferred to 33°C 30 minutes prior to and during data collection. GAL80ts;Kir2.1 experimental groups were reared at 18°C, transferred to 30°C for 48 hours prior to testing, and transferred to 33°C for data collection. Controls were kept at 18°C until testing at 33°C.
Maze fabrication
Mazes were cut into 1/16” thick black acrylic using a laser engraver (Epilog). Each arm of a maze was 0.37” long and 0.13” wide. In order to inhibit the flies from flipping upside down, the floors of the mazes were lightly roughened with a random orbital sander and fine-grit sand paper, and clear acrylic lids (one per maze) were lubricated with Sigmacote (Sigma). Circular arenas were fabricated similarly and were 2 inches in diameter. A diffuser made of two sheets of 1/4" thick clear acrylic roughened on both sides by orbital sanding, placed between the LS array and the maze array, provided for uniform illumination.
Behavioral tracking and data analysis
Flies were placed into individual Y-mazes or arenas and allowed to walk freely for 2hrs. Mazes were illuminated from below with white LSs (5500K, LuminousFilm), imaged with 2MP digital cameras (Logitech, Point Grey), and the X-Y positions of the flies' centroids were automatically tracked using background subtraction and recorded with software custom written in LabView (National Instruments). Data were then analyzed with custom written scripts in MatLab (The MathWorks). Fly identity was maintained for day-to-day experiments by storing flies individually in labeled culture vials between tests.
Wing-folding
To measure the wing-folding preference of individual flies, we moved each fly into a vial singly. The vial was flicked or agitated until the fly flew, assuring that its wings were unfolded. We then anesthetized it with CO2, and examined it manually under a dissecting scope to determine which wing was on top. The animal was then returned to its vial and allowed to waken, at which point we repeated the process. Flies were examined in this way 5 times sequentially per day. To generate the day-to-day correlation (Fig S3b), we compared the aggregated wing-folding data from days 1 and 2 (10 total observations) with the aggregated data from days 3 and 4 (another 10 total observations). Similarly, the flies that were first measured for wing-folding and then turning bias in the Y-maze were scored for the former over two days (10 total observations) and tested in the Y-maze on day 3.
Statistics
Expected distributions (Figs 1e, S1e) were calculated by summing binomial distributions with ni equal to the number of choices made by fly i within the corresponding experimental group, and pi equal to the average right-turn probability of the entire population. These individual curves were interpolated into a normalized [0,1] domain before summing. Standard errors of MAD scores and correlation coefficients were calculated using bootstrap resampling, with a minimum of 1000 replicates. P-values reported based on bootstrap resampling in three different ways: 1) To compare observed MADs to known null hypotheses (Fig 1e,f), samples were drawn from the known null distribution in numbers corresponding to the data. The number of samples in which the MAD of the randomly drawn values equal or was less than that of the null hypothesis was recorded. If the bootstrap replicates produced the tested condition by chance alone at a rate of k out of n resamples, p was reported as the highest probability such that . That is, the highest value of p that would yield the tested event k times out of n or fewer at least 2.5% of the time. 2) To compute bootstrapped z-tests (e.g. Fig S4b), we determined the number of standard errors away from 0 the observed MAD was by bootstrapping the points contributing to that MAD value. 3) To compare the MADs of two experimental groups (Figs 3a-b 4d, and S6) we assumed the bootstrap estimated errors on the MADs were Gaussian and calculated the 1-tailed probabilities of that a MAD drawn from the experimental error distribution would be less than a MAD drawn from the control error distribution. Mutual information (Fig S1f) was calculated on a fly-by-fly basis between turn t and t+1. Significance asterisks in Fig 3a reflect a Bonferroni correction for multiple comparisons. Significance in Figs 3b and 4d is not corrected since comparisons are only between 23°C and 33°C experimental groups.
Immunohistochemistry and imaging
Adult nervous tissue was dissected and fixed overnight in 4% paraformaldehyde at 4°C. After fixation, tissue was counterstained with Alexa Fluor 568 conjugated to phalloidin for 24–48hrs (Life Technologies, 1:50 dilution). Stained brains were washed in PBT, and mounted on glass slides in 70% glycerol or vectaShield mounting medium (Vector Labs). Images were collected on a Zeiss LSM710 or LSM780 confocal microscope. Panels modified from FlyLight images (Figs S7d,e and f right side, and S8 bottom row) were downloaded from http://flweb.janelia.org/cgi-bin/flew.cgi and color rotated into a red-cyan palette. Depth coded images (Fig S7f) and lateral views (Figs 4c and S8) were calculated using stack functions in FIJI (http://fiji.sc/Fiji).
Modeling
Model simulations (Fig S9) were performed in MATLAB (The MathWorks) using Euler approximation, with Δt=0.01. The model was considered converged (i.e. a “decision” had been made) if variables changed by less than 0.001 between iterations, or at 1000 iterations, whichever was earlier. Tuning curves were determined empirically based on 1000 replicates of the model. Behavioral distributions were based on 1000 tuning curves. Beta fitting parameters were determined analytically from the means and variances of the behavioral distributions. Additional details are given in Extended Experimental Procedures. The model was highly robust to parameter choice, and the parameter values used for Figure 6 are αL=αR=0.01, βL=βR=0.02, δ=0.03, γR=0.01, and γL=0.01b, where b determines the intrinsic network bias. Stimulus noise was implemented as where εi ∼ N(µ,σ2), with µ=0 in all cases, σ=0.1 for the green curves, and σ=0.2 for the purple curves of Figure 6. Intrinsic bias was normally distributed with mean=1 and standard deviation deviation=0.025 in Figure S9c and 0.01 in S9d and e.
Conflict of interest:
The authors have no financial interests related to this work.
Acknowledgements:
This research was funded by the Junior Fellows Program at The Rowland Institute at Harvard. We would like to thank Mike Burns, Chris Stokes, and other members of The Rowland Institute for fruitful scientific discussions and technical assistance, as well as Frank Hirth, Tom Maniatis, Charles Zuker, and members of their labs for helpful feedback. We thank Shmuel Raz, Roland Strauss, Michael Reiser, Aravi Samuel, Sam Kunes, Chuntao Dan, Douglas Armstrong and Hiromu Tanimoto for sharing fly lines. We thank the Janelia Farm FlyLight consortium for allowing us to reuse and modify their GAL4 expression images.