Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 73 of the 73 references it cites were found in a registry or an index.

Transforming a head direction signal into a goal-oriented steering command

Open access

Published version of a preprint

Abstract

To navigate, we must continuously estimate the direction we are headed in, and we must correct deviations from our goal 1 . Direction estimation is accomplished by ring attractor networks in the head direction system 2,3 . However, we do not fully understand how the sense of direction is used to guide action. Drosophila connectome analyses 4,5 reveal three cell populations (PFL3R, PFL3L and PFL2) that connect the head direction system to the locomotor system. Here we use imaging, electrophysiology and chemogenetic stimulation during navigation to show how these populations function. Each population receives a shifted copy of the head direction vector, such that their three reference frames are shifted approximately 120° relative to each other. Each cell type then compares its own head direction vector with a common goal vector; specifically, it evaluates the congruence of these vectors via a nonlinear transformation. The output of all three cell populations is then combined to generate locomotor commands. PFL3R cells are recruited when the fly is oriented to the left of its goal, and their activity drives rightward turning; the reverse is true for PFL3L. Meanwhile, PFL2 cells increase steering speed, and are recruited when the fly is oriented far from its goal. PFL2 cells adaptively increase the strength of steering as directional error increases, effectively managing the tradeoff between speed and accuracy. Together, our results show how a map of space in the brain can be combined with an internal goal to generate action commands, via a transformation from world-centric coordinates to body-centric coordinates.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41586-024-07039-2

Cites

73 of 73 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Comparing model predictions with behaviour. a, A rotational servomechanism works to keep the angle θ of some device close to a goal value θg. The output is a rotational velocity command that depends on the system’s error (θ − θg). Rotational velocity is close to zero around the goal (θ = θg) and the anti-goal (θ = θg + 180°). Whereas the goal is a stable fixed point, the anti-goal is an unstable fixed point. b, In the Drosophila brain, head direction is represented in Δ7 cells as a sinusoid over two spatial cycles. c, PFL3L, PFL3R and PFL2 populations extract spatially shifted copies of the head direction representation. These three populations are aligned in the fan-shaped body, where they share inputs from putative goal cells (Extended Data Fig. 1c). d, Model: each PFL population adds its head direction input with a shared input from goal cells. This is passed through a nonlinearity and then integrated over space. e, Model: activity of each PFL population versus directional error. f, Data: path of a fly in a virtual environment with a visual head direction cue (a bright bar). Dots indicate 90° and 180° jumps of the environment; here the fly is correcting for all these jumps with rapid turns. g, Mean head direction θ in 10 min epochs with periodic jumps. Radial length denotes the consistency of head direction over time ρ, which ranges here from 0 to 0.8 in n = 56 epochs from 56 flies; 0° is towards the cue. h, Data: mean rotational speed versus directional error, the s.e.m. across flies (n = 46 flies). i, Model: PFL populations have shifted head direction inputs that tile the space of compass directions. Each population detects overlap between its shifted head direction vectors and a shared goal vector. The PFL3L population drives left turning, whereas the PFL3R population drives right turning and PFL2 drives increased rotational speed. Scale bar (f), 30 mm.
Fig. 1 Comparing model predictions with behaviour. a, A rotational servomechanism works to keep the angle θ of some device close to a goal value θg. The output is a rotational velocity command that depends on the system’s error (θ − θg). Rotational velocity is close to zero around the goal (θ = θg) and the anti-goal (θ = θg + 180°). Whereas the goal is a stable fixed point, the anti-goal is an unstable fixed point. b, In the Drosophila brain, head direction is represented in Δ7 cells as a sinusoid over two spatial cycles. c, PFL3L, PFL3R and PFL2 populations extract spatially shifted copies of the head direction representation. These three populations are aligned in the fan-shaped body, where they share inputs from putative goal cells (Extended Data Fig. 1c). d, Model: each PFL population adds its head direction input with a shared input from goal cells. This is passed through a nonlinearity and then integrated over space. e, Model: activity of each PFL population versus directional error. f, Data: path of a fly in a virtual environment with a visual head direction cue (a bright bar). Dots indicate 90° and 180° jumps of the environment; here the fly is correcting for all these jumps with rapid turns. g, Mean head direction θ in 10 min epochs with periodic jumps. Radial length denotes the consistency of head direction over time ρ, which ranges here from 0 to 0.8 in n = 56 epochs from 56 flies; 0° is towards the cue. h, Data: mean rotational speed versus directional error, the s.e.m. across flies (n = 46 flies). i, Model: PFL populations have shifted head direction inputs that tile the space of compass directions. Each population detects overlap between its shifted head direction vectors and a shared goal vector. The PFL3L population drives left turning, whereas the PFL3R population drives right turning and PFL2 drives increased rotational speed. Scale bar (f), 30 mm.CC BY · PubMed Central
Fig. 2 Dynamics around the anti-goal. a, Example experiment. ATP (red shading) depolarizes PFL2 cells expressing P2X2 (top), evoking an increase in the absolute value of rotational velocity, that is, rotational speed (middle). It also evokes a decrease in forward velocity (bottom). This fly turns right in response to the first two pulses but left in response to the last two pulses. b, Summary data for flies where PFL2 cells expressed P2X2 and genetic controls (mean ± s.e.m. across flies, n = 12 P2X2+ flies and 11 control flies). Results are shown for two ATP pulse durations (100 ms and 500 ms). See also Extended Data Fig. 4. c, PFL2 activity (ΔF/F) across the horizontal axis of the fan-shaped body over time. During this epoch, the fly is walking relatively straight. The fly’s mean head direction is taken as its goal (θg). After the environment is jumped by 180° (blue arrowhead), the fly makes a compensatory turn to reorient towards θg. We fit a sinusoid to ΔF/F at each time point to extract bump phase and amplitude. d, Change in PFL2 bump phase versus change in directional error. The phase of PFL2 activity moves right when the fly turns left. Each symbol denotes one time point (Pearson’s r = −0.63, P = 9 × 10−13), with the line of unity in grey. Shown here are data for one example fly; Extended Data Fig. 9 shows two other examples and shows the effect of z-scoring ΔF/F. e, PFL2 bump amplitude versus directional error (mean ± s.e.m. across flies, n = 33 flies). f, PFL2 bump amplitude versus the fly’s rotational velocity (mean ± s.e.m. across flies, n = 33 flies). g, PFL2 bump amplitude versus the fly’s forward velocity (mean ± s.e.m. across flies, n = 33 flies). Scale bars, 5 s (b), 10 s (a,c).
Fig. 2 Dynamics around the anti-goal. a, Example experiment. ATP (red shading) depolarizes PFL2 cells expressing P2X2 (top), evoking an increase in the absolute value of rotational velocity, that is, rotational speed (middle). It also evokes a decrease in forward velocity (bottom). This fly turns right in response to the first two pulses but left in response to the last two pulses. b, Summary data for flies where PFL2 cells expressed P2X2 and genetic controls (mean ± s.e.m. across flies, n = 12 P2X2+ flies and 11 control flies). Results are shown for two ATP pulse durations (100 ms and 500 ms). See also Extended Data Fig. 4. c, PFL2 activity (ΔF/F) across the horizontal axis of the fan-shaped body over time. During this epoch, the fly is walking relatively straight. The fly’s mean head direction is taken as its goal (θg). After the environment is jumped by 180° (blue arrowhead), the fly makes a compensatory turn to reorient towards θg. We fit a sinusoid to ΔF/F at each time point to extract bump phase and amplitude. d, Change in PFL2 bump phase versus change in directional error. The phase of PFL2 activity moves right when the fly turns left. Each symbol denotes one time point (Pearson’s r = −0.63, P = 9 × 10−13), with the line of unity in grey. Shown here are data for one example fly; Extended Data Fig. 9 shows two other examples and shows the effect of z-scoring ΔF/F. e, PFL2 bump amplitude versus directional error (mean ± s.e.m. across flies, n = 33 flies). f, PFL2 bump amplitude versus the fly’s rotational velocity (mean ± s.e.m. across flies, n = 33 flies). g, PFL2 bump amplitude versus the fly’s forward velocity (mean ± s.e.m. across flies, n = 33 flies). Scale bars, 5 s (b), 10 s (a,c).CC BY · PubMed Central
Fig. 3 Dynamics around the goal. a, ΣPFL2 activity (ΔF/F) versus directional error (mean ± s.e.m. across flies, n = 33 flies). Shown here is the summed activity of the right and left PFL2 axons, where they terminate near DNa03 dendrites in the lateral accessory lobe. Model prediction is shown for comparison. b, ΣPFL3 activity (ΔF/F) versus directional error (mean ± s.e.m. across flies, n = 23 flies). As in a, the activity is summed across the right and left lateral accessory lobe, where PFL3 cells terminate onto DNa03 and DNa02. Here we used a mixed split-Gal4 line that targets PFL2 and PFL3 cells together; because our results are opposite for what we observe for PFL2 cells alone, and because our results match the predictions of our PFL3 model (shown below), we treat this as measurement of PFL3 activity (Extended Data Fig. 5). c, ΣPFL3R and ΣPFL3L activity in the right and left lateral accessory lobe, respectively, plotted versus directional error (mean ± s.e.m. across flies, n = 23 flies). Signals are imaged from our mixed split-Gal4 line but are likely dominated by PFL3, as noted above. Model predictions are shown for comparison. d, Right–left difference in PFL3 activity versus directional error (mean ± s.e.m. across flies, n = 23 flies) and model prediction. e, Right–left difference in PFL3 activity versus the fly’s rotational velocity (mean ± s.e.m. across flies, n = 23 flies) and model prediction.
Fig. 3 Dynamics around the goal. a, ΣPFL2 activity (ΔF/F) versus directional error (mean ± s.e.m. across flies, n = 33 flies). Shown here is the summed activity of the right and left PFL2 axons, where they terminate near DNa03 dendrites in the lateral accessory lobe. Model prediction is shown for comparison. b, ΣPFL3 activity (ΔF/F) versus directional error (mean ± s.e.m. across flies, n = 23 flies). As in a, the activity is summed across the right and left lateral accessory lobe, where PFL3 cells terminate onto DNa03 and DNa02. Here we used a mixed split-Gal4 line that targets PFL2 and PFL3 cells together; because our results are opposite for what we observe for PFL2 cells alone, and because our results match the predictions of our PFL3 model (shown below), we treat this as measurement of PFL3 activity (Extended Data Fig. 5). c, ΣPFL3R and ΣPFL3L activity in the right and left lateral accessory lobe, respectively, plotted versus directional error (mean ± s.e.m. across flies, n = 23 flies). Signals are imaged from our mixed split-Gal4 line but are likely dominated by PFL3, as noted above. Model predictions are shown for comparison. d, Right–left difference in PFL3 activity versus directional error (mean ± s.e.m. across flies, n = 23 flies) and model prediction. e, Right–left difference in PFL3 activity versus the fly’s rotational velocity (mean ± s.e.m. across flies, n = 23 flies) and model prediction.CC BY · PubMed Central
Fig. 4 Navigation dynamics at cellular resolution. a, Head direction tuning in an example PFL2 cell and an example PFL3 cell. Preferred direction is θp. b, Each PFL2 and PFL3 cell is predicted to receive synaptic inhibition that varies sinusoidally with head direction. c, Whole-cell recordings from PFL2 and PFL3 cells showing changes in IPSP frequency when we impose a rotational jump on the virtual environment, emulating a change in θ. d, Change in IPSP frequency versus change in θ (relative to θp, mean ± s.e.m. across cells, n = 12 PFL3 and 10 PFL2 cells in 22 flies, Pearson’s r = 0.53). The effect of θ is significant (P = 8 × 10−3, two-way ANOVA, with θ and fly identity as the two factors). This analysis uses time points when the fly was standing still, because this makes individual IPSPs more clearly detectable. e, Model: a nonlinearity transforms Vm into firing rate for each model cell. Each cell receives head direction input that is cosine tuned to (θ − θp). The goal cell input to each cell represents a bias that does not change with head direction. This bias moves the cell’s input along the nonlinear function, changing the amplitude of the firing rate tuning curve. f, PFL2 cells are divided into bins based on (θp − θg). For each cell, we subtract the minimum y-axis value in the tuning curve, then we compute the mean of cells in the bin, for both firing rate and Vm. Model output (top) is compared with data (bottom, n = 11 cells, mean ± s.e.m. across cells). g, Same but for PFL3 neurons (n = 15 cells, mean ± s.e.m. across cells). Here we combine results from PFL3R and PFL3L (after reversing the left–right order of the five bins for the PFL3L cells, so that the model outputs are identical for R and L). Scale bar, 2 s.
Fig. 4 Navigation dynamics at cellular resolution. a, Head direction tuning in an example PFL2 cell and an example PFL3 cell. Preferred direction is θp. b, Each PFL2 and PFL3 cell is predicted to receive synaptic inhibition that varies sinusoidally with head direction. c, Whole-cell recordings from PFL2 and PFL3 cells showing changes in IPSP frequency when we impose a rotational jump on the virtual environment, emulating a change in θ. d, Change in IPSP frequency versus change in θ (relative to θp, mean ± s.e.m. across cells, n = 12 PFL3 and 10 PFL2 cells in 22 flies, Pearson’s r = 0.53). The effect of θ is significant (P = 8 × 10−3, two-way ANOVA, with θ and fly identity as the two factors). This analysis uses time points when the fly was standing still, because this makes individual IPSPs more clearly detectable. e, Model: a nonlinearity transforms Vm into firing rate for each model cell. Each cell receives head direction input that is cosine tuned to (θ − θp). The goal cell input to each cell represents a bias that does not change with head direction. This bias moves the cell’s input along the nonlinear function, changing the amplitude of the firing rate tuning curve. f, PFL2 cells are divided into bins based on (θp − θg). For each cell, we subtract the minimum y-axis value in the tuning curve, then we compute the mean of cells in the bin, for both firing rate and Vm. Model output (top) is compared with data (bottom, n = 11 cells, mean ± s.e.m. across cells). g, Same but for PFL3 neurons (n = 15 cells, mean ± s.e.m. across cells). Here we combine results from PFL3R and PFL3L (after reversing the left–right order of the five bins for the PFL3L cells, so that the model outputs are identical for R and L). Scale bar, 2 s.CC BY · PubMed Central
Fig. 5 Modulating the scale of network activity. a, Direct and indirect pathways. S adjusts the magnitude of total input to PFL2 and PFL3 cells. b, Nonlinear activation function. c, Top, model PFL2 bump amplitude and (ΣPFL3R − ΣPFL3L) versus directional error. Bottom, rotational velocity produced by the direct or indirect pathway alone. With both pathways, results are similar to the indirect pathway alone. d, Model: directional error over time. As S increases, the network brings head direction towards the goal (red line). If the indirect pathway is omitted, high S produces overshooting. e, Data: example path during four jumps of the virtual environment, separated by 60 s. The fly corrects for the first jump, but not the rest. The probability of correction typically did not change over time. f, Change in PFL2 membrane potential (ΔVm) before and after each 180° jump, comparing corrected jumps with high ρ (n = 31 of 276 jumps) or uncorrected jumps with low ρ (n = 27 of 276 jumps). Variance in ΔVm is higher for corrected versus uncorrected jumps (P = 0.01363, Brown–Forsythe test). See also Extended Data Fig. 7. g, Same but for PFL3 (n = 60 of 348 corrected, 17 of 348 uncorrected, P = 0.02776). h, Absolute ΔVm and rotational speed during corrected jumps. Mean ± s.e.m., n = 157 of 701 (90°) and n = 91 of 624 (180°), pooling data from PFL2 and PFL3 cells. i, Path of two flies in a virtual environment over 10 min, one with high consistency of head direction (high ρ) and the other with low ρ. j, Spatial profile of PFL2 activity, divided into four bins based on head direction, relative to the directions associated with the highest and lowest PFL2 bump amplitude (darkest and lightest traces, respectively). Data (top) are from the two paths in i. Model results (bottom) are generated by setting S = 0.8 or S = 0.2, producing high or low ρ, respectively, as shown in d. k, Data (top): for each 10 min trial we computed ρ and also analysed the spatial profile of PFL2 activity as in j, taking the difference between the maximum and minimum bump amplitudes. Across trials (symbols), bump amplitude modulation is correlated with ρ (Pearson’s r = 0.37096, P = 1.7 × 10−4, n = 33 flies). Model (bottom): same analysis on model output. Here we obtained a range of model outcomes by varying S and using different random seeds. Scale bars, 5 cm (e,i), 1 s (f). In k, Max., maximum; min., minimum.
Fig. 5 Modulating the scale of network activity. a, Direct and indirect pathways. S adjusts the magnitude of total input to PFL2 and PFL3 cells. b, Nonlinear activation function. c, Top, model PFL2 bump amplitude and (ΣPFL3R − ΣPFL3L) versus directional error. Bottom, rotational velocity produced by the direct or indirect pathway alone. With both pathways, results are similar to the indirect pathway alone. d, Model: directional error over time. As S increases, the network brings head direction towards the goal (red line). If the indirect pathway is omitted, high S produces overshooting. e, Data: example path during four jumps of the virtual environment, separated by 60 s. The fly corrects for the first jump, but not the rest. The probability of correction typically did not change over time. f, Change in PFL2 membrane potential (ΔVm) before and after each 180° jump, comparing corrected jumps with high ρ (n = 31 of 276 jumps) or uncorrected jumps with low ρ (n = 27 of 276 jumps). Variance in ΔVm is higher for corrected versus uncorrected jumps (P = 0.01363, Brown–Forsythe test). See also Extended Data Fig. 7. g, Same but for PFL3 (n = 60 of 348 corrected, 17 of 348 uncorrected, P = 0.02776). h, Absolute ΔVm and rotational speed during corrected jumps. Mean ± s.e.m., n = 157 of 701 (90°) and n = 91 of 624 (180°), pooling data from PFL2 and PFL3 cells. i, Path of two flies in a virtual environment over 10 min, one with high consistency of head direction (high ρ) and the other with low ρ. j, Spatial profile of PFL2 activity, divided into four bins based on head direction, relative to the directions associated with the highest and lowest PFL2 bump amplitude (darkest and lightest traces, respectively). Data (top) are from the two paths in i. Model results (bottom) are generated by setting S = 0.8 or S = 0.2, producing high or low ρ, respectively, as shown in d. k, Data (top): for each 10 min trial we computed ρ and also analysed the spatial profile of PFL2 activity as in j, taking the difference between the maximum and minimum bump amplitudes. Across trials (symbols), bump amplitude modulation is correlated with ρ (Pearson’s r = 0.37096, P = 1.7 × 10−4, n = 33 flies). Model (bottom): same analysis on model output. Here we obtained a range of model outcomes by varying S and using different random seeds. Scale bars, 5 cm (e,i), 1 s (f). In k, Max., maximum; min., minimum.CC BY · PubMed Central
Extended Data Fig. 1 Model predictions. a, A resolver measures the current angle of some object (θc, e.g., the angular position of a shaft) and resolves that angle into its Cartesian components, xc and yc. The goal angle θg is similarly resolved into its Cartesian components, xg and yg. These components are cross-multiplied, and their difference is used to generate a rotational velocity command dθ/dt ∝ xcyg - xgyc. We treat positive velocity values as clockwise (CW) rotations. In this example, the current angle is rotated CW relative to the goal, meaning a positive directional error. This drives a CCW rotation. But because the error at this point (●) is almost 180°, rotational speed will be small. Mittelstaedt suggested that a similar process might be implemented in the brain’s navigation centres to control an organism’s heading, and thus its path through the environment; this is known as “Mittelstaedt ‘s bicomponent model” of steering control7. Arrowhead denotes the system’s stable fixed point. b, Any vector can be represented as a sinusoidal function whose amplitude represents the magnitude of the vector, and whose phase represents the angle of the vector. Although it is convenient to represent these sinusoids as continuous functions, they can also be discretized into spatial activity patterns over neural space15,16. Adding these sinusoids is equivalent to performing vector addition. c, Model elements shown in Fig. 1d, here schematized as spatial activity patterns over neural space. The horizontal axis of this space represents the horizontal axis of the fan-shaped body. d, Model: goal input to PFL2&3 cells (left). When this spatial pattern is shifted leftward, this produces a clockwise shift in the model’s rotational velocity as a function of head direction (right). Arrowheads denote the system’s stable fixed point. e, Model: shifts in the spatial phase of goal input to PFL cells produce equal shifts in the head direction corresponding to the system’s stable fixed point. This is true for all values of S > 0. f, Model: The effect of silencing PFL2 cells on rotational velocity is similar to the effect of removing the indirect pathway (compare with Fig. 5c). In both cases, the rotational velocity function becomes equally steep around the goal and the anti-goal. g, Model: The effect of silencing PFL2 cells on steering dynamics is similar to the effect of removing the indirect pathway (compare with Fig. 5d). In both cases, the system oscillates when S is high.
Extended Data Fig. 1 Model predictions. a, A resolver measures the current angle of some object (θc, e.g., the angular position of a shaft) and resolves that angle into its Cartesian components, xc and yc. The goal angle θg is similarly resolved into its Cartesian components, xg and yg. These components are cross-multiplied, and their difference is used to generate a rotational velocity command dθ/dt ∝ xcyg - xgyc. We treat positive velocity values as clockwise (CW) rotations. In this example, the current angle is rotated CW relative to the goal, meaning a positive directional error. This drives a CCW rotation. But because the error at this point (●) is almost 180°, rotational speed will be small. Mittelstaedt suggested that a similar process might be implemented in the brain’s navigation centres to control an organism’s heading, and thus its path through the environment; this is known as “Mittelstaedt ‘s bicomponent model” of steering control7. Arrowhead denotes the system’s stable fixed point. b, Any vector can be represented as a sinusoidal function whose amplitude represents the magnitude of the vector, and whose phase represents the angle of the vector. Although it is convenient to represent these sinusoids as continuous functions, they can also be discretized into spatial activity patterns over neural space15,16. Adding these sinusoids is equivalent to performing vector addition. c, Model elements shown in Fig. 1d, here schematized as spatial activity patterns over neural space. The horizontal axis of this space represents the horizontal axis of the fan-shaped body. d, Model: goal input to PFL2&3 cells (left). When this spatial pattern is shifted leftward, this produces a clockwise shift in the model’s rotational velocity as a function of head direction (right). Arrowheads denote the system’s stable fixed point. e, Model: shifts in the spatial phase of goal input to PFL cells produce equal shifts in the head direction corresponding to the system’s stable fixed point. This is true for all values of S > 0. f, Model: The effect of silencing PFL2 cells on rotational velocity is similar to the effect of removing the indirect pathway (compare with Fig. 5c). In both cases, the rotational velocity function becomes equally steep around the goal and the anti-goal. g, Model: The effect of silencing PFL2 cells on steering dynamics is similar to the effect of removing the indirect pathway (compare with Fig. 5d). In both cases, the system oscillates when S is high.CC BY · PubMed Central
Extended Data Fig. 2 Example images for mixed and PFL2 split-Gal4 lines. a, Left panel shows a max z-projection of the mixed PFL2/3 split-Gal4 line expressing GFP. Right panel shows a max z-projection from an MCFO clone with 3 PFL3 neurons (PFL3 identity confirmed by counting axons) along with unidentified neurons outside of the central complex. Across 7 brains, we counted 7 PFL3 neurons and 4 PFL2 neurons, no expression from other cell types was observed in the region of the lateral accessory lobes where PFL2 and PFL3 axon terminals are found. b. Example cell fills obtained from electrophysiology experiments. The left 3 columns show an example PFL2 cell fill while the right 3 columns show an example PFL3 cell fill. It was common for the axonal arbors in the lateral accessory lobe (LAL) to exhibit bright fluorescence, while the dendritic arbors in the protocerebral bridge and fan-shaped body exhibited relatively dim fluorescence. The identity (PFL2, PFL3, or other) of every recorded cell was confirmed by comparing the morphology of the filled cell to the known morphology of PFL2 and PFL3 neurons; of the 30 cells recorded, 12 were verified in this manner as PFL2 cells, 15 were verified as PFL3 cells, and 3 were found to represent other cell types. c. Left panel shows a max z-projection of the PFL2 split-Gal4 line expressing GFP. The right three panels each show a max z-projection for three brains, each containing one PFL2 neuron. The left and rightmost of these panels also show an unidentified neuron ventral to the central complex. Across 13 brains we counted 15 PFL2 neurons, and 0 other cell types in the central complex.
Extended Data Fig. 2 Example images for mixed and PFL2 split-Gal4 lines. a, Left panel shows a max z-projection of the mixed PFL2/3 split-Gal4 line expressing GFP. Right panel shows a max z-projection from an MCFO clone with 3 PFL3 neurons (PFL3 identity confirmed by counting axons) along with unidentified neurons outside of the central complex. Across 7 brains, we counted 7 PFL3 neurons and 4 PFL2 neurons, no expression from other cell types was observed in the region of the lateral accessory lobes where PFL2 and PFL3 axon terminals are found. b. Example cell fills obtained from electrophysiology experiments. The left 3 columns show an example PFL2 cell fill while the right 3 columns show an example PFL3 cell fill. It was common for the axonal arbors in the lateral accessory lobe (LAL) to exhibit bright fluorescence, while the dendritic arbors in the protocerebral bridge and fan-shaped body exhibited relatively dim fluorescence. The identity (PFL2, PFL3, or other) of every recorded cell was confirmed by comparing the morphology of the filled cell to the known morphology of PFL2 and PFL3 neurons; of the 30 cells recorded, 12 were verified in this manner as PFL2 cells, 15 were verified as PFL3 cells, and 3 were found to represent other cell types. c. Left panel shows a max z-projection of the PFL2 split-Gal4 line expressing GFP. The right three panels each show a max z-projection for three brains, each containing one PFL2 neuron. The left and rightmost of these panels also show an unidentified neuron ventral to the central complex. Across 13 brains we counted 15 PFL2 neurons, and 0 other cell types in the central complex.CC BY · PubMed Central
Extended Data Fig. 3 Jumps of the virtual environment during imaging experiments. a,. PFL2 bump amplitude, forward velocity, and rotational speed and behavior for 180° jumps and ±90° jumps of the virtual environment (mean ± s.e.m across jumps). b, Absolute difference in right-left PFL3 signals (|ΣPFL3R - ΣPFL3L|), forward velocity, and rotational speed and behavior for 180° jumps and ±90° jumps of the virtual environment (mean ± s.e.m across jumps). c, Difference in right-left PFL3 signals (ΣPFL3R - ΣPFL3L) and rotational speed, comparing +90° and −90° jumps. Clockwise (positive) jumps of the virtual environment produce higher activity in PFL3R, and rightward (positive) turns, as expected (mean ± s.e.m across jumps).
Extended Data Fig. 3 Jumps of the virtual environment during imaging experiments. a,. PFL2 bump amplitude, forward velocity, and rotational speed and behavior for 180° jumps and ±90° jumps of the virtual environment (mean ± s.e.m across jumps). b, Absolute difference in right-left PFL3 signals (|ΣPFL3R - ΣPFL3L|), forward velocity, and rotational speed and behavior for 180° jumps and ±90° jumps of the virtual environment (mean ± s.e.m across jumps). c, Difference in right-left PFL3 signals (ΣPFL3R - ΣPFL3L) and rotational speed, comparing +90° and −90° jumps. Clockwise (positive) jumps of the virtual environment produce higher activity in PFL3R, and rightward (positive) turns, as expected (mean ± s.e.m across jumps).CC BY · PubMed Central
Extended Data Fig. 4 Additional data on PFL2 chemogenetic stimulation. a, Expanded summary data for flies where PFL2 cells expressed P2X2 showing PFL2 voltage responses to 100-ms, 200-ms, 300-ms, and 500-ms pulses of ATP, as well as simultaneously recorded locomotor activity (mean ± s.e.m across jumps). Here we show forward velocity, absolute rotational velocity (i.e., rotational speed), and absolute sideways velocity (sideways speed). b, Same but for genetic controls where PFL2 cells did not express P2X2 (mean ± s.e.m across jumps). c, Response of one example PFL2 cell to all four pulse durations over a 120-s period.
Extended Data Fig. 4 Additional data on PFL2 chemogenetic stimulation. a, Expanded summary data for flies where PFL2 cells expressed P2X2 showing PFL2 voltage responses to 100-ms, 200-ms, 300-ms, and 500-ms pulses of ATP, as well as simultaneously recorded locomotor activity (mean ± s.e.m across jumps). Here we show forward velocity, absolute rotational velocity (i.e., rotational speed), and absolute sideways velocity (sideways speed). b, Same but for genetic controls where PFL2 cells did not express P2X2 (mean ± s.e.m across jumps). c, Response of one example PFL2 cell to all four pulse durations over a 120-s period.CC BY · PubMed Central
Extended Data Fig. 5 Comparing the PFL2-specific line with the PFL2&3 mixed line. a. Right minus left activity in the lateral accessory lobe (LAL) versus directional error, for both the PFL2-specific line and the PFL2&3 mixed line (mean ± s.e.m. n = 33 flies and n = 23 flies). b. Right minus left activity in the LAL versus rotational velocity, for both the PFL2-specific line and the PFL2&3 mixed line (mean ± s.e.m. n = 33 flies and n = 23 flies). c. Left panel: summed right and left activity in the LAL versus rotational velocity, for both the PFL2-specific line and the PFL2&3 mixed line (mean ± s.e.m. n = 33 and n = 23 flies). The dissimilarity in results for these two lines is further evidence that PFL2 cells make a minimal contribution to the mixed line. Therefore, we treat the results from the mixed line as a readout of PFL3 activity.
Extended Data Fig. 5 Comparing the PFL2-specific line with the PFL2&3 mixed line. a. Right minus left activity in the lateral accessory lobe (LAL) versus directional error, for both the PFL2-specific line and the PFL2&3 mixed line (mean ± s.e.m. n = 33 flies and n = 23 flies). b. Right minus left activity in the LAL versus rotational velocity, for both the PFL2-specific line and the PFL2&3 mixed line (mean ± s.e.m. n = 33 flies and n = 23 flies). c. Left panel: summed right and left activity in the LAL versus rotational velocity, for both the PFL2-specific line and the PFL2&3 mixed line (mean ± s.e.m. n = 33 and n = 23 flies). The dissimilarity in results for these two lines is further evidence that PFL2 cells make a minimal contribution to the mixed line. Therefore, we treat the results from the mixed line as a readout of PFL3 activity.CC BY · PubMed Central
Extended Data Fig. 6 Changes in single cell head direction tuning curves depending on the distance between the cell’s preferred head direction and the goal head direction. a, Tuning curves for PFL2 and PFL3 cells across different distances between the cell’s preferred head direction and the goal head direction (mean ± s.e.m across jumps). The vertical offset of the cell’s head direction tuning curve changes depending on this distance but the amplitude of the tuning curves only change at the level of firing rate. b, The same as (a) but for model generated output. These are the same results as in Fig. 4f, g; the only difference is that we have not subtracted the minimum value of each tuning curve.
Extended Data Fig. 6 Changes in single cell head direction tuning curves depending on the distance between the cell’s preferred head direction and the goal head direction. a, Tuning curves for PFL2 and PFL3 cells across different distances between the cell’s preferred head direction and the goal head direction (mean ± s.e.m across jumps). The vertical offset of the cell’s head direction tuning curve changes depending on this distance but the amplitude of the tuning curves only change at the level of firing rate. b, The same as (a) but for model generated output. These are the same results as in Fig. 4f, g; the only difference is that we have not subtracted the minimum value of each tuning curve.CC BY · PubMed Central
Extended Data Fig. 7 Jumps of the virtual environment during electrophysiology experiments. a, PFL2 recordings: absolute change in membrane potential, forward velocity, and rotational speed, for 180° jumps and ±90° jumps of the virtual environment (mean ± s.e.m across jumps, N is the number of jumps). b, Same but for PFL3. c, As expected, 180° jumps produce significantly larger changes in membrane potential, as compared to 90° jumps (PFL2 p = 0.041, PFL3 p = 0.0064, 2-sample, 2-tailed t-tests). d, Change in the difference between the fly’s head direction (θ) and the cell’s preferred direction (θp) resulting from each cue jump in the corrected and uncorrected categories. There was no significant difference in the variance of these values between the two categories (PFL2 p = 0.68793, PFL3 p = 0.99764, Brown-Forsythe test). There was also no difference between the corrected and uncorrected categories in the mean absolute ∆(θ-θp) (PFL2 p = 0.59721, PFL3 p = 0.99723, 2-sample, 2-tailed t-tests). Thus, we might expect the two types of jumps to produce similar changes in membrane potential. The fact that we see a larger membrane potential response following a corrected jump suggests that the state of the network is different before a corrected jump, and this contributes to the behavioral response. e, Correlation between membrane potential and rotational speed, as a function of lag time (PFL2: n = 54, PFL3 n = 85, mean ± s.e.m across jumps). The maximum correlation is seen when we compare membrane potential with rotational speed 150−200 ms later. This is what we would expect if PFL2&3 cells are exerting a causal influence on behavior.
Extended Data Fig. 7 Jumps of the virtual environment during electrophysiology experiments. a, PFL2 recordings: absolute change in membrane potential, forward velocity, and rotational speed, for 180° jumps and ±90° jumps of the virtual environment (mean ± s.e.m across jumps, N is the number of jumps). b, Same but for PFL3. c, As expected, 180° jumps produce significantly larger changes in membrane potential, as compared to 90° jumps (PFL2 p = 0.041, PFL3 p = 0.0064, 2-sample, 2-tailed t-tests). d, Change in the difference between the fly’s head direction (θ) and the cell’s preferred direction (θp) resulting from each cue jump in the corrected and uncorrected categories. There was no significant difference in the variance of these values between the two categories (PFL2 p = 0.68793, PFL3 p = 0.99764, Brown-Forsythe test). There was also no difference between the corrected and uncorrected categories in the mean absolute ∆(θ-θp) (PFL2 p = 0.59721, PFL3 p = 0.99723, 2-sample, 2-tailed t-tests). Thus, we might expect the two types of jumps to produce similar changes in membrane potential. The fact that we see a larger membrane potential response following a corrected jump suggests that the state of the network is different before a corrected jump, and this contributes to the behavioral response. e, Correlation between membrane potential and rotational speed, as a function of lag time (PFL2: n = 54, PFL3 n = 85, mean ± s.e.m across jumps). The maximum correlation is seen when we compare membrane potential with rotational speed 150−200 ms later. This is what we would expect if PFL2&3 cells are exerting a causal influence on behavior.CC BY · PubMed Central
Extended Data Fig. 8 Modeling changes in behavioral state and PFL2 dynamics. a, Path of three flies in a virtual environment over 10 min, one with high consistency of head direction (high ρ) and two with low ρ. The third path is shown in an inset with high magnification. The first two paths are reproduced from Fig. 5i. b, Same as Fig. 5j, but now including data and model for the third (purple) path. Here, to model, the third path, we adjust A rather than S. Whereas S scales the total synaptic input to PFL2 and PFL3 cells, A specifically scales the amplitude of the goal signal: PFL2 = f(S·(cos(θ-θ0-h + 180°) + A·cos(θg-θ0-h))) PFL3R = f(S·(cos(θ-θ0-h + 67.5°) + A·cos(θg-θ0-h))) PFL3L = f(S·(cos(θ-θ0-h-67.5°) + A·cos(θg-θ0-h))) where f is a nonlinear function, θ is head direction, θ0 is the angular position of the EPG bump at a head direction of 0°, θg is the goal angle, and h is a vector with entries that tile the full 360° of angular space (equation (9), Methods). Reductions in S decrease the overall scale of PFL2&3 activity, without changing the dependence of bump amplitude on head direction. By contrast, reductions in A cause bump amplitude to become more invariant to head direction. The examples shown here suggest that changes in behavioral state may arise in some cases from changes in S, and in other cases from changes in A.
Extended Data Fig. 8 Modeling changes in behavioral state and PFL2 dynamics. a, Path of three flies in a virtual environment over 10 min, one with high consistency of head direction (high ρ) and two with low ρ. The third path is shown in an inset with high magnification. The first two paths are reproduced from Fig. 5i. b, Same as Fig. 5j, but now including data and model for the third (purple) path. Here, to model, the third path, we adjust A rather than S. Whereas S scales the total synaptic input to PFL2 and PFL3 cells, A specifically scales the amplitude of the goal signal: PFL2 = f(S·(cos(θ-θ0-h + 180°) + A·cos(θg-θ0-h))) PFL3R = f(S·(cos(θ-θ0-h + 67.5°) + A·cos(θg-θ0-h))) PFL3L = f(S·(cos(θ-θ0-h-67.5°) + A·cos(θg-θ0-h))) where f is a nonlinear function, θ is head direction, θ0 is the angular position of the EPG bump at a head direction of 0°, θg is the goal angle, and h is a vector with entries that tile the full 360° of angular space (equation (9), Methods). Reductions in S decrease the overall scale of PFL2&3 activity, without changing the dependence of bump amplitude on head direction. By contrast, reductions in A cause bump amplitude to become more invariant to head direction. The examples shown here suggest that changes in behavioral state may arise in some cases from changes in S, and in other cases from changes in A.CC BY · PubMed Central
Extended Data Fig. 9 ∆F/F compared to z-scored ∆F/F. a, Left: ∆F/F binned into 1-s increments across an entire 10-min trial for a fly that exhibited low goal fixation, before and after z-scoring, compared with head direction. Right: ∆F/F is divided into 4 bins based on the fly’s head direction relative to the head direction associated with the lowest PFL2 bump amplitude. This is done separately for raw ∆F/F (top) and z-scored ∆F/F (bottom). Here, z-scoring reduces but does not eliminate the difference in bump amplitude across bins. b, The same as (a) but for a fly that exhibited high goal fixation during the 10-min trial. Here, there is a strong variation in bump amplitude that persists across bins, even after z-scoring. These data are shown in Fig. 5j.
Extended Data Fig. 9 ∆F/F compared to z-scored ∆F/F. a, Left: ∆F/F binned into 1-s increments across an entire 10-min trial for a fly that exhibited low goal fixation, before and after z-scoring, compared with head direction. Right: ∆F/F is divided into 4 bins based on the fly’s head direction relative to the head direction associated with the lowest PFL2 bump amplitude. This is done separately for raw ∆F/F (top) and z-scored ∆F/F (bottom). Here, z-scoring reduces but does not eliminate the difference in bump amplitude across bins. b, The same as (a) but for a fly that exhibited high goal fixation during the 10-min trial. Here, there is a strong variation in bump amplitude that persists across bins, even after z-scoring. These data are shown in Fig. 5j.CC BY · PubMed Central
Extended Data Fig. 10 Path segmentation. To obtain an accurate estimate of the fly’s current goal direction (θg), and thus an accurate estimate of directional error (θ - θg), we needed to identify moments when θg might switch. We reasoned that a switch in θg, would coincide with a dip in head direction consistency. Therefore, we looked for moments when p crossed a threshold value, and we broke the path into segments at those moments of threshold-crossing. This allowed us to segment a path into straight segments and to identify points where goal direction seemed to have switched. We used a threshold value of p = 0.88 because this produced results that corresponded to our visual impression of when the fly’s goal direction seemed to have changed, but we also confirmed that our conclusions are similar for a range of threshold values. Here we show an example path recorded over 10 min. Time points belonging to different segments are alternately shaded gray and black. Dotted line shows the threshold used to define these segments.
Extended Data Fig. 10 Path segmentation. To obtain an accurate estimate of the fly’s current goal direction (θg), and thus an accurate estimate of directional error (θ - θg), we needed to identify moments when θg might switch. We reasoned that a switch in θg, would coincide with a dip in head direction consistency. Therefore, we looked for moments when p crossed a threshold value, and we broke the path into segments at those moments of threshold-crossing. This allowed us to segment a path into straight segments and to identify points where goal direction seemed to have switched. We used a threshold value of p = 0.88 because this produced results that corresponded to our visual impression of when the fly’s goal direction seemed to have changed, but we also confirmed that our conclusions are similar for a range of threshold values. Here we show an example path recorded over 10 min. Time points belonging to different segments are alternately shaded gray and black. Dotted line shows the threshold used to define these segments.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 69 references

Research works printed but not linked: 1

  1. Schöne, H. Spatial Orientation: The Spatial Control of Behavior in Animals and Man (Princeton Univ. Press, 2014).

Cited by

See all 86

Metadata

Authors
Westeinde, E. A., Kellogg, E., Dawson, P. M., Lu, J., Hamburg, L., Midler, B., Druckmann, S., & Wilson, R. I.
Affiliations
  • Harvard University (US)
  • Stanford Medicine (US)
Corresponding author
Wilson, Rachel I.
Venue
Nature
Published
2024-02-07
Volume
626
Issue
8000
Pages
819-826
ISSN
  • 0028-0836 (ISSN-L)
  • 1476-4687
OpenAlex
W4391597906
Cited by
85
References
69
Citations by year
2026: 29 · 2025: 25 · 2024: 24
FWCI
13.60
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by-nc-nd)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Digital Access to Scholarship at Harvard (DASH) (Harvard University) (submittedVersion, cc-by)
  • RePEc: Research Papers in Economics (submittedVersion)
Funders
  • Howard Hughes Medical Institute
  • National Institutes of Health
  • National Institute of Neurological Disorders and Stroke (U19NS104655)
Topics
  • Life Sciences › Neuroscience › Cellular and Molecular Neuroscience › Neurobiology and Insect Physiology Research
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Memory and Neural Mechanisms
  • Social Sciences › Psychology › Experimental and Cognitive Psychology › Spatial Cognition and Navigation
Keywords
head direction system, Drosophila, reference frames, locomotion control, navigation, chemogenetic stimulation, speed-accuracy trade-off, brain atlas
MeSH terms
Orientation, Spatial, Animals, Brain, Drosophila melanogaster, Goals, Head, Locomotion, Neurons, Time Factors, Connectome, Spatial Navigation

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Transforming a head direction signal into a goal-oriented steering command. https://app.cochat.ai/papers/10.1038/s41586-024-07039-2
The paper
Westeinde, E. A., Kellogg, E., Dawson, P. M., Lu, J., Hamburg, L., Midler, B., Druckmann, S., & Wilson, R. I. (2024). Transforming a head direction signal into a goal-oriented steering command. Nature. https://doi.org/10.1038/s41586-024-07039-2

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.