Bounded rationality in C. elegans is explained by circuit-specific normalization in chemosensory pathways

Abstract

Rational choice theory assumes optimality in decision-making. Violations of a basic axiom of economic rationality known as "Independence of Irrelevant Alternatives" (IIA) have been demonstrated in both humans and animals and could stem from common neuronal constraints. Here we develop tests for IIA in the nematode Caenorhabditis elegans, an animal with only 302 neurons, using olfactory chemotaxis assays. We find that in most cases C. elegans make rational decisions. However, by probing multiple neuronal architectures using various choice sets, we show that violations of rationality arise when the circuit of olfactory sensory neurons is asymmetric. We further show that genetic manipulations of the asymmetry between the AWC neurons can make the worm irrational. Last, a context-dependent normalization-based model of value coding and gain control explains how particular neuronal constraints on information coding give rise to irrationality. Thus, we demonstrate that bounded rationality could arise due to basic neuronal constraints.

Fig. 1

C. elegans display rational decisions. a A scheme for the Independence of irrelevant alternatives chemotaxis assays. A chemotaxis index (C.I.) (number of worms in A, divided by the number of worms in A and B together) was calculated. Each plate contained ~200–400 worms. b The relative preference for 2-butanone over 2,3-pentanedione is unaffected by increasing concentration of benzaldehyde as a third attractant (two-tailed Wilcoxon signed-ranks test; n = 6). c The relative preference for 2,3-pentanedione over 2-butanone is unaffected by increasing concentration of benzaldehyde as a third attractant (two-tailed Wilcoxon signed-ranks test; n = 6). d, e Introducing AWA- sensed odorants as a third attractant, does not influence the relative preference between 2-butanone and 2,3-pentanedione (two-tailed Wilcoxon signed-ranks test; d n = 8, e n = 6). f Benzaldehyde as a third attractant, does not affect the relative preference between the two AWA- sensed odorants pyrazine and diacetyl (two-tailed Wilcoxon signed-ranks test; n = 6). Two-tailed Wilcoxon signed-ranks test. Bars represent the C.I. of odor A. Error bars represent the standard error of the mean C.I.

Fig. 2

C. elegans exhibit IIA violations when specific neuronal architectures are induced. a The effect of 2-butanone as a third attractant on the relative preference between benzaldehyde and pyrazine, and the overall preference of each attractant point in every condition (two-tailed Wilcoxon signed-ranks test, C = 1/500: W = 2, q = 0.0012; C = 10⁻²: W = 0, q = 0.0006; n = 6). b The effect of 2-butanone as a third attractant on the relative preference between benzaldehyde and 2,3-pentanedione, and the overall preference of each attractant point in every condition. (two-tailed Wilcoxon signed-ranks test, C = 1/500: W = 6, q = 0.0144; C = 10⁻²: W = 2, q = 0.0036; n = 6). c In all the violations that we described so far, 2-butanone, sensed specifically by the AWC^ON neuron, functioned as odor C, and benzaldehyde, sensed by both AWC neurons, functioned as odor A. d 2,3-pentanedione as a third attractant does not change the relative preference between benzaldehyde and pyrazine (two-tailed Wilcoxon signed-ranks test, n = 6). e 2-butanone as a third attractant does not change the relative preference between 2,3-pentanedione and pyrazine (two-tailed Wilcoxon signed-ranks test, n = 6). f 2-butanone as a third attractant significantly reduced the relative preference for isoamyl-alcohol over pyrazine (two-tailed Wilcoxon signed-ranks test, C = 1/500: W = 14, q = 0.0278; C = 10⁻²: W = 2, q = 0.0005; n = 7). Bars represent the C.I. of odor A. Error bars represent the standard error of the mean C.I. *q < 0.05, **q < 0.01, ***q < 0.001, ****q < 0.0001

Fig. 3

The AWC^ON neuron makes the worm vulnerable to IIA violations. a The influence of acetone as a third attractant on the relative preference between benzaldehyde and pyrazine, in AWC^ON/ON mutant worms. (two-tailed Wilcoxon signed-ranks test, C = 10⁻²: W = 1, q = 0.0018; n = 6). b The influence of 2-butanone as a third attractant on the relative preference between benzaldehyde and pyrazine, in AWC^ON/^ON mutant worms. (two-tailed Wilcoxon signed-ranks test, C = 1/500: W = 2, q = 0.0141; C = 10⁻²: W = 2, q = 0.0141; n = 7). c 2-butanone as a third attractant does not change the relative preference between 2,4,5-trimethylthiazole and pyrazine (two-tailed Wilcoxon signed-ranks test, n = 6). d The influence of 2-butanone as a third attractant on the relative preference between 2,4,5-trimethylthiazole and 2,3-pentanedione (two-tailed Wilcoxon signed-ranks test, C = 10⁻²: W = 0, q = 0.0258; n = 6). e 2-butanone as a third attractant changes the relative preference between 2,4,5-trimethylthiazole and pyrazine in AWC^ON/^ON mutant worms (two-tailed Wilcoxon signed-ranks test, C = 10⁻²: W = 3, q = 0.0456; C = 1/500: W = 7, q = 0.1862; C = 10⁻³: W = 0, q = 0.0132; n = 6). f Isoamyl-alcohol as a third attractant does not change the relative preference between 2,4,5-trimethylthiazole and pyrazine (two-tailed Wilcoxon signed-ranks test, n = 6). g Pyrazine as a third attractant does not change the relative preference between 2,4,5-trimethylthiazole and benzaldehyde (two-tailed Wilcoxon signed-ranks test, n = 7). h Benzaldehyde as a third attractant changes the relative preference between 2,4,5-trimethylthiazole and pyrazine (two-tailed Wilcoxon signed-ranks test, C = 10⁻²: W = 60, q = 0.003; n = 12). Bars represent the C.I. of odor A. Error bars represent the standard error of the mean C.I. *q < 0.05, **q < 0.01, ***q < 0.001, and ****q < 0.0001

Fig. 4

Sensory gain control model of chemosensation explains circuit architecture-specific IIA violations. Predicted choice behavior in a divisive normalization model of sensory gain control in C. elegans chemosensation. Odors driving the same chemosensory neuron are assumed to drive cross-odor normalization in neural representation. Simulation parameters were not fit to empirical choice data, but instead were chosen to demonstrate qualitative similarity in behavioral data under different circuit activation patterns. Each combination was simulated for n = 10⁶ repetitions. In each combination, the left panel shows the circuit activation pattern, the middle shows the preference index for odor A (relative choice of odors A vs. B), and the right shows the model-predicted preference for the odors. Our results indicate that IIA violations can occur due to an asymmetric overlap between odors “A” and “C”. a Model-predicted IIA violations in asymmetric overlap circuit architectures. When odors A and C both activate AWC^ON, increasing concentrations of odor C reduce the representation of odor A in the model via cross-odor normalization. The model captures IIA violations with (right) and without (left) preference reversals, both of which are observed in the empirical data. b Model-predicted rational choice behavior in symmetric overlap circuit architectures. In symmetric circuits where odor C activates neurons sensing both odor A and odor B, cross-odor normalization affects the neural representations of both high-value odors. Increasing concentrations of odor C affects the neural representation of odors A and B similarly, and relative preference does not vary. c Model-predicted rational choice behavior in nonoverlap circuit architectures. In these circuits, odors A and C activate distinct chemosensory neurons and no cross-odor normalization occurs in the model. Thus, the neural representations of odors A and B (and the relative choice preference of A over B) do not vary with the concentration of odor C. d Model behavior in expanded-bandwidth circuits. In expanded-bandwidth scenarios, odor A activates both AWC and both AWA neurons. In wild-type worms, cross-odor normalization only affects 25% of the neural representation of odor A and the model predicts weak IIA violations (left). In nsy-1 mutants with two AWC^ON neurons, cross-odor normalization affects 50% of the representation of odor A and the model predicts stronger IIA violations (right)

Fig. 5

Increasing concentrations of 2,3-pentanedione (AWC^OFF-sensed odor) as a third alternative can induce IIA violations. a Model-predicted IIA violation driven by AWC^OFF neuron activation by odor C. Left panel: predicted preference index for odor A (isoamyl-alcohol) relative to odor B (pyrazine) at different concentrations of odor C (2,3-pentanedione). Right panel: predicted preference for all three odors. In the model, the ability of the AWC^OFF neuron to generate IIA violations arises from cross-odor normalization and an equal weighting of AWC^ON and AWC^OFF neuron activity in the representation of odor A (isoamyl-alcohol). b The influence of 2,3-pentanedione (AWC^OFF) as a third attractant on the relative preference between isoamyl-alcohol (10⁻²) (AWC^BOTH) and pyrazine (10⁻³) (AWA) (two-tailed Wilcoxon signed-ranks test, C = 10⁻⁵: W = 17, q = 0.2721; C = 10⁻⁴: W = 12, q = 0.1424; C = 10⁻³: W = 13, q = 0.1424; C = 1/500: W = 19, q = 0. 0.1424; C = 10⁻²: W = 8, q = 0.0275; n = 6). Bars represent the C.I. of odor A. Error bars represent the standard error of the mean C.I. *q < 0.05, **q < 0.01, ***q < 0.001, and ****q < 0.0001