Drosophila melanogaster grooming possesses syntax with distinct rules at different temporal scales

Abstract

Mathematical modeling of behavioral sequences yields insight into the rules and mechanisms underlying sequence generation. Grooming in Drosophila melanogaster is characterized by repeated execution of distinct, stereotyped actions in variable order. Experiments demonstrate that, following stimulation by an irritant, grooming progresses gradually from an early phase dominated by anterior cleaning to a later phase with increased walking and posterior cleaning. We also observe that, at an intermediate temporal scale, there is a strong relationship between the amount of time spent performing body-directed grooming actions and leg-directed actions. We then develop a series of data-driven Markov models that isolate and identify the behavioral features governing transitions between individual grooming bouts. We identify action order as the primary driver of probabilistic, but non-random, syntax structure, as has previously been identified. Subsequent models incorporate grooming bout duration, which also contributes significantly to sequence structure. Our results show that, surprisingly, the syntactic rules underlying probabilistic grooming transitions possess action duration-dependent structure, suggesting that sensory input-independent mechanisms guide grooming behavior at short time scales. Finally, the inclusion of a simple rule that modifies grooming transition probabilities over time yields a generative model that recapitulates the key features of observed grooming sequences at several time scales. These discoveries suggest that sensory input guides action selection by modulating internally generated dynamics. Additionally, the discovery of these principles governing grooming in D. melanogaster demonstrates the utility of incorporating temporal information when characterizing the syntax of behavioral sequences.

Fig 1. Applying the ABRS classifier to video data generates large-scale ethogram data sets.

A: The ABRS classifier is trained to annotate video data with labels corresponding to one of five grooming actions observed in D. melanogaster (front leg rubbing = f, head cleaning = h, abdomen cleaning = a, back leg rubbing = b, wing cleaning = w) or two non-grooming actions (walking = wk, standing = s, not shown). Color bar (right) displays the corresponding color code used for visualization of grooming action sequences. B: Flies are covered with dust, placed in a chamber, and recorded from above. Video data is then passed through the ABRS classifier, which locates the fly in each frame and applies a behavioral label. Here, the fly is shown performing a front leg rub. C: After classification, video data can be represented by a vector. Each vector element represents one frame and contains a behavioral label, visualized here using a color code. Without loss of information, the real time ethogram (left) is recast as a discrete bout ethogram, defined by discrete transitions between individual bouts, which are labeled by both action and duration. We refer to this process as “discretizing” the ethogram. This results in a vector (right) containing both an ordered list of grooming action labels (color) and grooming action durations (number). For example, the first bout shown here consisting of consecutive frames of head grooming would be discretized such that its corresponding action label would be “h”, indicated by the purple block. For illustrative purposes, we choose bout durations, shown below the discrete bout ethogram, that are representative of the observed distributions. We refer to instances of sustained identical grooming actions as bouts. D: Real time ethograms for 92 dusted flies (left) are shown as a matrix containing 92 rows and 50,000 columns. After discretization, ethograms contain variable numbers of bouts but retain information about bout order and bout duration (right).

Fig 2. Schematic overview of the multiple time scales present in Drosophila melanogaster grooming ethograms.

Grooming ethograms possess at least three distinct time scales. The grooming progression (top), which takes many minutes to observe in its entirety, occurs at the longest time scale we consider here. We refer to this phenomenon as a progression because we observe that, on average, flies begin grooming with anterior-oriented actions and end with posterior-oriented actions and more walking. At an intermediate time scale, ethograms consist of grooming motifs (middle). Motifs are composed of consecutive grooming bouts which use the same pair of legs (either anterior or posterior). At the shortest time scale, we observe individual grooming bouts (bottom), which consist of sustained grooming movements that range from several hundreds of milliseconds to a few seconds in duration.

Fig 3. Binning grooming bouts by duration allows for analysis of temporal dependence.

Ethograms are modified by introducing a binning scheme based on grooming bout duration. Here, we show the binning strategy used to create three duration categories for grooming actions. Grooming actions are classified as short, medium, or long bouts, indicated by the red, green, and blue regions of the probability density functions, respectively. Region boundaries are chosen such that each duration category contains an equivalent number of samples. Notably, anterior motif actions (f, h) possess strikingly similar duration distributions, suggesting coupling between these actions.

Fig 4. Statistical null model hypotheses (schematic).

By shuffling bout duration or bout order and comparing the resultant null hypothesis transition matrices to the maximum likelihood estimate, we identify which features are statistically significant in the data and contribute to sequence structure. Shown here are schematics of the permutations we perform on the observed ethograms to generate null hypothesis transition matrices. In the duration permuted hypothesis, bout order is preserved but duration categories are randomly permuted (h = head, wk = walking, f = front leg, s = short, m = medium, l = long). In the order permuted hypothesis, bout order is permuted but the durations associated with each bout remain the same.

Fig 5. Drosophila melanogaster grooming progresses from anterior motif to posterior motif actions.

A: Average grooming progression of D. melanogaster population (N = 92) after exposure to an irritant. Colored lines show the percentage of time spent in each action (f = front leg, h = head, a = abdomen, b = back leg, w = wing, wk = walking) across a 16.7 second sliding window (500 frames). Front leg rubbing and head cleaning proportions track closely with one another for the entire course of grooming. Additionally, back leg rubbing tracks closely with the sum of abdomen and wing cleaning. Dashed gray vertical line indicates the boundary between early and late grooming phases, which occurs after approximately thirteen minutes, or nearly half of the recording duration. B: Proportion of time spent in each action during the early phase (left) and late phase (right) of grooming. During the early phase, D. melanogaster spend the majority of time performing anterior grooming movements. In the late phase, they spend a relatively larger proportion of time engaged in posterior grooming movements and walking.

Fig 6. Anterior motifs are composed of highly correlated amounts of body-directed and leg-directed actions.

A: Schematic of an anterior motif. Anterior motifs are defined as continuous consecutive bouts of head grooming (h) or front leg rubbing (f) flanked by non-anterior motif actions on either side. Posterior motifs are defined similarly, but consist of abdomen grooming, back leg rubbing, and wing grooming (a, b, and w). Both motifs contain body-directed actions (h, a, and w), which clear irritant from the body. Leg-directed actions (f and b) clear irritant from the legs, which collect irritant during body-directed grooming actions. In this example, the anterior motif consists of six grooming bouts of varying duration. B: Anterior motifs exhibit a strong linear relationship between the total amount of time spent performing body and leg-directed actions (left). Each point corresponds to an observed motif consisting of four or more actions. Posterior motifs (right) display a weaker linear trend with greater amounts of leg-directed grooming.

Fig 7. Within-motif transitions dominate grooming syntax.

A: The maximum likelihood transition matrix, $\widehat{\mathbf{M}}$ (Eq 2), provides the best estimate of the Markov chain dynamics observed in grooming behavior (Eq 1). The network representation of $\widehat{\mathbf{M}}$ (left) illustrates transition probabilities using edge thickness (thicker edges indicate higher probabilities). Shown on the right is the matrix representation of discrete time transition probabilities (fit to the discrete time ethogram from Fig 1 with no duration information), with probability magnitude indicated by color. Here, the anterior motif, consisting of front leg rubbing and head cleaning, is indicated by the red circle (left) and red square (right). The posterior motif, consisting of abdominal grooming, back leg rubbing, and wing cleaning, is delineated by the blue circle (left) and blue square (right). B: Maximum likelihood transition probabilities for ethograms binned according to the schema shown in Fig 3. Matrices fit to data with two (left) and three (right) duration categories show similar within-motif structure with increased resolution (s = short, m = medium, l = long). For example, both matrices illustrate that anterior motif transitions between long bouts dominate syntax.

Fig 8. Grooming bout duration contributes to syntax at the scale of individual transitions.

A: Probabilities from the maximum likelihood transition matrix (top) differ significantly from null hypothesis transition matrices (bottom) to varying degrees, indicating that both grooming action order and bout duration contribute to grooming syntax. Shown on the top left is the network visualization of anterior motif transitions, with edge withs proportional to transition probabilities. Notice that transitions between actions belonging to the same duration category possess relatively high probabilities and appear nearly symmetric, suggesting a coupling mechanism between anterior motif actions. BIC values (right of matrices) provide validation that the maximum likelihood model captures statistically significant features of the data. The maximum likelihood transition matrix has a lower BIC value than the null hypotheses used for comparison, indicating that both bout order and duration contribute to sequence syntax. B: The residual values, or the difference between the maximum likelihood matrix and the null model matrices, illustrate that specific transitions differ from what the null hypotheses would predict. The duration permuted null model matrix exhibits block structure but fails to capture the temporal relationship between bouts, as illustrated by the red values in several “long-to-long” transitions, for example. The order permuted null model differs even more severely, indicating that, while duration dependence plays a role in sequence structure, action order is still the primary determinant.

Fig 9. Transition probabilities are largely stationary across the entire grooming progression despite changing sensory conditions.

Shown here are the population average transition probabilities for the early phase (left) and late phase (right) of grooming. The border between these phases is indicated by the dashed gray line in Fig 5. In the early phase, flies prioritize anterior grooming motifs, indicated by the thick red outline. In the late phase, flies perform more posterior grooming, indicated by the thick blue outline. We observe that transitions between front leg rubbing and head cleaning bouts exhibit consistent duration dependence regardless of when they occur in the sequence. Posterior motif transitions display similarities as well, but transitions between long abdomen grooming bouts and long back leg rubbing bouts are significantly more likely late in grooming. Overall, the relative stationarity of these transition probabilities despite changing sensory conditions suggests the existence of an internal mechanism that dictates bout durations. However, sensory stimuli also appear to play a role in modulating grooming transitions on long time scales, as transitions from posterior motif actions to anterior motif actions become less likely in the late phase.

Fig 10. A nonstationary Markov renewal process recapitulates grooming progression and bout structure.

A: Illustration of the time-varying transition matrix, M(t), used for generating synthetic ethograms. To approximate changing sensory conditions in the simplest possible manner, two transition matrices, M_early and M_late, are fit to the first and last 200 actions in the discrete time ethogram with 3 duration categories, respectively. As time evolves in the synthetic ethogram simulation, M(t) changes as described in Eq 6. At the beginning of the simulation, M(t) is identical to M_early and after 13 minutes, it is identical to M_late. Between those times, it is a linear combination of the two matrices. B: Synthetic ethograms display the characteristic progression from anterior to posterior grooming, as seen from comparison with Fig 5. C: Synthetic ethograms reproduce observed anterior motif composition. Synthetic anterior motifs (left) exhibit a similar, though slightly weaker, trend as observed in our data (Fig 6). Posterior motifs are less similar, indicating that other factors may be necessary to explain posterior motif structure.