Sensorimotor behaviour from an untrained brain-and-nerve-cord connectome
Authors: John R. Minnick, Jesus Gonzalez-Ferrer, Jinghui Geng, Jason K. Eshraghian, Mohammed A. Mostajo-Radji, David Haussler, Mircea Teodorescu
Organizations: Department of Electrical and Computer Engineering, University of California, Santa Cruz, CA, USA. · UC Santa Cruz Genomics Institute, University of California, Santa Cruz, CA, USA. · Department of Biomolecular Engineering, University of California, Santa Cruz, CA, USA. · Department of Computer Science and Engineering, University of California, Santa Cruz, CA, USA.
Connectomes now map the neurons and synapses of an adult fly's brain and nerve cord, but how much of behaviour follows from the wiring is unknown. Here we turn one such connectome into a spiking network of 172,433 neurons with a published brain model's untuned parameters, and run it alone and inside a simulated body. Two fixed rules join them: odours and looming objects drive sensory neurons, and motor neurons move the legs. Alone, the network passes a signal to the motor neurons it is known to drive when the pathway supplies about 2% or more of their input. In the body, the fly navigates around obstacles to 95% of the odour sources. A rewiring that keeps each neuron's connection counts and signs reaches 18%, and the fly's leg commands replayed with the senses off reach almost none, so navigation needs the measured wiring and sensory feedback. Dopamine-gated weakening of the mushroom body's output synapses turns approach to an odour into avoidance, which requires the pairing and is odour-specific once one olfactory synapse class is scaled. An untrained connectome thus supports a behaviour and its learned reversal, and the pathways it cannot drive mark what the diagram leaves out.
Figures & tables
Supplementary Fig. 1. A sensory pulse crosses the network from its input to the leg motor pools within 24 ms. Spiking activity after a 60 ms pulse at the receptor neurons ( a , 400 Hz) or at the LC and LPLC visual projection neurons ( b , 300 Hz), in the network alone with no other input. Rows, the 6 ms bins starting 0, 6, 12, 24 and 60 ms after the pulse’s onset, on the brain from the side and on the whole nervous system from above; a region’s colour is its rate in that bin, spikes per neuron per 6 ms, on a logarithmic scale from 0.001 to the input layer’s peak (2.33 in a , 1.04 in b ), over a grey outline of every region, in the colours of Fig. 1d. The time axis at the left carries each stage’s arrival, the end of the first bin in which the region’s rate reaches a tenth of its mean over the pulse (the rule of the Methods), with the region’s name and the time. The input stops after 60 ms; the last row shows the brain still active after it.
Supplementary Fig. 2. The order in which the wiring is traversed from the stimulated neurons predicts the order in which the pulse reaches neurons. Every neuron that spikes within the record after the odour pulse ( a , 6,775 neurons) or the looming pulse ( b , 3,884), its first spike after the onset, the end of the first 6 ms bin with a spike in the records of Supplementary Fig. 1, against its traversal distance from the stimulated group in the wiring. The traversal is the information-flow model of Schlegel et al. (Methods): the connection list is walked in rounds from the stimulated neurons, an untraversed neuron is traversed with probability min(1,r/0.3) , r the share of its input synapses from traversed neurons, and the distance is the round, averaged over ten seeded traversals and rounded. For each round, a bar at each 6 ms value whose length is the number of neurons with that first spike, the round scaled to its own peak, in the stimulated region’s colour of Fig. 1; black bar, the round’s median first spike; vertical line, its quartiles; the count under each round, its neurons. Each panel gives Spearman’s ρ over its neurons and the share of the neurons the traversal reaches that spike within the record: the traversal reaches almost every neuron, the pulse in the silent network 4% and 2% of them.
Supplementary Fig. 3. The signed wiring between the 24 regions, and the influence map against it. a , The synapse counts of Fig. 1e split by the sign the model gives the presynaptic neuron, excitatory minus inhibitory, on a symmetric log scale (vermillion, an excitatory excess; blue, an inhibitory excess); 42.5% of all synapses carry an inhibitory sign. b , The inhibitory share of each cell’s synapses. Rows and columns of a and b are in the order of Fig. 1e and f, with the row labels on a and each region’s colour, as in Fig. 1, along the edges. The rows of Fig. 1f that evoke nothing elsewhere are the rows of b whose output is mostly inhibitory-signed, or too small to drive a region (Results). c , Per region pair with any synapse (334), the influence of Fig. 1f against the excitatory-signed count; Spearman’s ρ=0.65 , and 0.56 against the count regardless of sign. The pairs with influence and no direct synapse are relayed routes, the receptors’ onto the mushroom body and the lateral horn among them.
Supplementary Fig. 4. The learned avoidance needs both the balance readout and reinforcement confined to the aversive compartments. The path of every fly in the post test after six episodes of odour paired with the reinforcement (odour + dopamine; n=20 flies per arena), under the four combinations of valence readout (rows) and of the connections the reinforcement reaches (columns). a , The network as read. b , The matched variant. Summed readout, the valence is the summed spike count of every MBON; balance readout, the count of the approach-driving MBONs minus that of the avoidance-driving MBONs. Every compartment, the reinforcement reaches all 15,523 Kenyon-cell-to-MBON connections; aversive compartments, the 8,377 of the compartments that the PPL1 dopamine neurons innervate. The upper left arena of each panel is the summed readout with the reinforcement in every compartment (Supplementary Discussion) and the lower right the analysis of the main text (the runs of Fig. 6 g and i); the other two change one of the two at a time. Triangle, the start; green dot, the odour source, with its 4 mm capture radius dotted; a dot ends each path. All eight arenas share one frame and one scale; bar, 10 mm. Under each arena: the valence in the post test (mean over flies), the flies that reach the source, and the median closest approach to it. With the summed readout the trained flies reach the source wherever the reinforcement acts (19 or 20 of 20). With the balance readout and the reinforcement in every compartment the valence goes to zero and the fly walks past the source, steering neither toward it nor away. It turns away only with the balance readout and the reinforcement confined to the aversive compartments. Every condition, with the tests, is in Supplementary Table 7.
Arena
Layout
Walls
Sources
Intact
Rewired
Lesioned
Random walk
Informed, stops at walls
Informed, slides along walls
i
open field
0
4
4.00
1.25
0.75
0.35
4.00
4.00
ii
single detour
1
3
3.00
0.70
0.70
0.40
3.00
3.00
iii
two-wall slalom
2
3
2.95
0.50
0.20
0.15
3.00
3.00
iv
wide gate
2
3
3.00
1.15
0.65
0.10
3.00
3.00
v
one-sided walls
2
3
3.00
0.40
0.15
0.05
3.00
3.00
vi
long S-curve
2
3
3.00
0.40
0.05
0.15
1.00
3.00
Supplementary Table 1. Sources reached in the nine arenas, beside three reference controllers. Mean sources reached over 20 seeds. Intact, rewired and lesioned are the three wirings of Fig. 4. The random walk is the chance level of Fig. 4f. The two informed controllers are given the position of every source and head straight for the nearest one that remains, at the intact fly’s median speed and with its capture radius, start and time limit; at a wall one stops and the other slides along it.
Condition
Runs
Faulted
Sources reached (of 3)
None
All
Closest approach (mm)
DNa02 (spikes per step)
Wall contact (steps)
Path (mm)
Leg motor rate (Hz)
Intact
20
3
3.00 ± 0.00
0
20
1.15 ± 0.04
1.99 ± 0.02
0.0 ± 0.0
193 ± 5
154 ± 0
Mushroom body lesioned
20
0
0.00 ± 0.00
20
0
14.55 ± 0.69
0.81 ± 0.11
1.9 ± 1.9
220 ± 11
152 ± 0
Rewired
20
0
0.10 ± 0.07
18
0
11.00 ± 0.91
0.54 ± 0.15
1.9 ± 1.7
212 ± 11
98 ± 5
Lesioned, valence pinned at 1.44
20
1
3.00 ± 0.00
0
20
1.08 ± 0.03
0.77 ± 0.03
0.0 ± 0.0
191 ± 5
154 ± 0
Rewired, valence pinned at 1.44
20
1
2.55 ± 0.25
3
17
2.64 ± 0.96
0.46 ± 0.11
25.1 ± 15.6
186 ± 13
103 ± 6
Lesioned, valence pinned at 0.7
20
0
3.00 ± 0.00
0
20
0.98 ± 0.07
0.74 ± 0.05
0.0 ± 0.0
190 ± 9
152 ± 0
Supplementary Table 2. The conditions of the forage arena. Mean ± s.e.m. over the runs used: the first 20 completed runs of each condition in seed order, a rule fixed before any spare seed ran. Faulted, runs that ended in a physics fault, excluded from every number. None and All, the number of runs that reached no source and all three. Closest approach, the mean over the three sources. DNa02, spikes per 50 ms step summed over the pair. Wall contact, steps within 0.5 mm of the wall.
0 Hz
40 Hz
120 Hz
120 Hz, rewired
Target
Turn
Speed
n
Turn
Speed
n
Turn
Speed
n
Turn
Speed
n
DNa01, left
− 219 ± 7
9.8 ± 0.1
20
86 ± 7
1.4 ± 0.1
20
116 ± 7
1.1 ± 0.1
20
55 ± 7
1.4 ± 0.2
21
DNa01, right
− 92 ± 13
4.6 ± 0.2
20
− 167 ± 5
0.0 ± 0.1
20
− 150 ± 7
− 0.4 ± 0.1
20
36 ± 9
2.0 ± 0.3
21
DNa02, left
− 269 ± 7
7.2 ± 0.1
20
− 1 ± 7
1.3 ± 0.1
20
− 16 ± 7
1.3 ± 0.1
20
− 7 ± 9
0.7 ± 0.1
21
DNa02, right
219 ± 7
4.3 ± 0.1
20
− 47 ± 6
0.4 ± 0.1
20
− 25 ± 6
0.2 ± 0.1
20
− 7 ± 9
1.7 ± 0.3
21
DNa02, both
− 85 ± 10
12.0 ± 0.1
20
− 48 ± 8
0.7 ± 0.1
20
− 38 ± 7
0.8 ± 0.1
20
–
–
0
Supplementary Table 3. Descending-neuron drive in the body at three receptor rates. Net turn over the 2 s of drive (degrees; positive, to the left) and the change in speed from the settling window to the drive window (mm/s), mean ± s.e.m.; n , completed runs. The receptor neurons of both antennae are driven at the rate shown. Rewired, three permutations with seven seeds each at 120 Hz.
Gain
Seeds
Kenyon cells active
Kenyon-cell overlap
Projection neurons active
Projection- neuron overlap
MBON spikes, odour A
MBON spikes, odour B
Depression of A
Depression of B
1
6
0.682
0.90
0.74
0.86
2368
2416
0.18
0.18
0.5
6
0.431
0.76
0.54
0.57
1026
1182
0.18
0.15
0.25
6
0.148
0.61
0.33
0.29
160
286
0.29
0.16
0.15
6
0.061
0.36
0.21
0.11
52
111
0.00
0.00
0.1
6
0.034
0.24
0.16
0.11
38
102
− 0.03
0.02
0.075
6
0.027
0.26
0.15
0.09
32
96
− 0.02
− 0.01
Supplementary Table 4. The odour code against the gain of the cholinergic local neurons. The output synapses of the local neurons predicted cholinergic are scaled by the gain (1, the network as read). Active, the fraction of cells with at least one spike during a 1 s presentation, averaged over the two odours; overlap, the fraction shared between the two odours’ 5% most active cells; MBON spikes, the total over the presentation before training; depression, 1 minus the ratio of the MBON output after training on odour A (strength 0.5) to before; a dash, an output before training under 5 spikes, which cannot be scored. Means over the seeds shown.
Type
Compartment
Cells
Transmitter
Transmitter source
Drives
Aversive compartment
MBON01
γ5β′2a
2
glutamate
verified in the table
avoidance
MBON02
β2β′2a
2
glutamate
verified in the table
avoidance
MBON03
β′2mp
2
glutamate
verified in the table
avoidance
MBON04
β′2mp bilateral
2
glutamate
verified in the table
avoidance
MBON05
γ4>γ1γ2
2
glutamate
verified in the table
avoidance
MBON06
β1>α
2
glutamate
verified in the table
avoidance
Supplementary Table 5. The sign of every MBON type under the balance readout. The balance readout (Methods) reads the valence as the spike count of the approach-driving MBONs minus that of the avoidance-driving MBONs. A type drives avoidance when its transmitter is glutamate and approach when it is GABA or acetylcholine; the transmitter is the connectome table’s verified entry where the type has one, else the published one, else the majority of the per-neuron predictions (references in the Methods). Compartments are the hemibrain names; the aversive compartments are those the PPL1 dopamine neurons innervate, the only ones the reinforcement reaches in the learning of Fig. 6. Cells are those the network simulates.
Measure
Summed readout
Balance readout
Nine arenas (Fig. 4): sources reached of 32 per fly (runs of 180 reaching none; reaching every source of the arena)
Intact
29.20 (0; 159)
30.25 (0; 161)
Rewired
3.50 (115; 0)
5.80 (92; 2)
Mushroom body lesioned
3.45 (117; 0)
3.45 (117; 0)
Correlated random walk at the fly’s speed
2.05
2.20
Forage arena (Fig. 5 b to d): sources reached of 3 (runs of 20 reaching none; reaching all three)
Supplementary Table 6. The in-body results under the two readouts of the valence. The summed readout reads the valence as the summed spike count of the MBONs divided by 107 (81 in the matched variant); the balance readout, the one of the main text, reads it as the count of the approach-driving MBONs minus that of the avoidance-driving MBONs (Methods; Supplementary Table 5). Every in-body run that reads the valence was made under each readout (Supplementary Discussion). Runs with the valence pinned do not read it. Under the summed readout the reinforcement reached every plastic connection; under the balance readout it reaches the 8,377 connections of the aversive compartments, so the weight row, a mean over all 15,523, is not comparable between the columns. Means over 20 seeds (the first 20 completed runs in the forage arena).
Summed readout, reinforcement in
Balance readout, reinforcement in
Measure
every compartment
the aversive compartments
every compartment
the aversive compartments
The network as read
Valence after training (range over flies)
odour + dopamine
0.24 (0.24 to 0.25)
0.83 (0.82 to 0.83)
0.03 (0.03 to 0.04)
− 0.33 ( − 0.34 to − 0.32)
dopamine alone
0.29 (0.28 to 0.29)
0.83 (0.83 to 0.84)
0.03 (0.01 to 0.08)
− 0.29 ( − 0.30 to − 0.28)
no plasticity
1.44 (1.44 to 1.45)
1.44 (1.44 to 1.45)
1.41 (1.40 to 1.42)
1.41 (1.40 to 1.42)
Supplementary Table 7. The learning experiment under the four combinations of valence readout and reinforcement. The four combinations of Supplementary Fig. 4 in both models: the valence read as the summed MBON count or as the balance of approach-driving and avoidance-driving MBONs, with the reinforcement reaching every Kenyon-cell-to-MBON connection (15,523) or those of the aversive compartments (8,377). The first column is the summed readout with the reinforcement in every compartment (Supplementary Discussion) and the last the analysis of the main text (Fig. 6); the two between them change one of the two at a time, their predictions written down before they ran. Each condition has 20 flies. Valence, the mean over the post-test episode. Closest approach, to the source in the post test; a fly reaches the source when it comes within 4 mm. Tests, two-sided Mann–Whitney U on the closest approach, with Cliff’s δ and its 95% bootstrap interval where the groups overlap; no correction for multiple comparisons. One of the 16 tests has P between 0.001 and 0.05: the matched variant with the sum readout and the reinforcement in the aversive compartments, odour + dopamine against dopamine alone ( P=0.014 ; medians 0.27 and 0.18 mm, every fly of both groups reaching the source). Without plasticity the reach of the reinforcement has no effect, and the two no-plasticity columns of a readout are the same simulation run twice: they agree fly for fly.
Group
Defining labels
Neurons
optic lobe
region optic_lobe
93,925
visual projection
super_class visual_projection
7,284
olfactory
cell_class olfactory_receptor_neuron
2,978
mushroom body
cell_class kenyon_cell or mushroom_body_output_neuron
4,608
lateral horn
cell_class containing lateral_horn
1,655
other central
super_class central_brain_intrinsic not matched by an earlier rule (the antennal-lobe projection and local neurons among them), and the 8,409 neurons matching no rule
30,575
Supplementary Table 8. The 24 groups of Fig. 1 and their defining labels. The groups that Fig. 1 b to f read from, with the number of simulated neurons in each. A neuron takes the first of the 13 coarse rules it matches, in this order: visual projection, olfactory, mushroom body, lateral horn, central complex, descending, ascending, leg motor, other motor, other sensory, optic lobe, VNC intrinsic, other central (the rules read the neuron table’s super_class, cell_class, cell_sub_class, cell_type and region columns; the names with underscores are the table’s own labels). The central-complex groups split the central-complex rule (cell_class containing central_complex) by cell_type prefix, the sensory groups split the other-sensory rule (super_class sensory, sensory_ascending or sensory_descending, not olfactory receptor or photoreceptor) by cell_class, and the leg motor groups split cell_class leg_motor_neuron by the segment named in cell_sub_class. Non-neuronal units (glia, trachea) count where their labels place them.
Connectome models of the fly nerve cord generate walking-like motor rhythms, but oscillation alone does not show that the specific wiring matters. Here we provide, to our knowledge, the first test of which features of motor output depend on the specific wiring. We simulated the leg motor systems of two independent Drosophila connectomes, with synapse counts as fixed weights and glutamatergic synapses treated as inhibitory, and compared each with six families of rewired networks that preserve progressively more of its structure, using pre-registered criteria. We find that rhythm is generic but antagonist coordination is not: many rewired networks were more rhythmic than the real ones, yet the real wiring coordinated antagonistic motor pools more strongly than every rewired network, most of all at the thorax--coxa joint. We trace this specificity to how premotor input is allocated between antagonistic pools. Both connectomes carry Sherrington's reciprocal innervation---neurons that excite one pool inhibit its antagonist---and no rewired network does. Reassigning premotor inputs between the pools abolished coordination even when motor neurons' typical input strength changed little (all pre-registered criteria met in one connectome; same direction in the other). Coordination, not rhythm, therefore reveals whether a connectome model's wiring matters.
Isabel Guan, Yuntian Zhao, Dingyuan Zhang +2
The Hong Kong University of Science and Technology · ZENBOT · Nanyang Technological University +1
Randomised copies of a connectome are the usual baseline for asking whether measured wiring matters, and the answer depends on what the randomisation preserves. We evolved embodied foraging agents whose brains are a compressed adult Drosophila connectome (FlyWire v783; 512 cell-type groups and 1,000 Kenyon cells) alongside agents built on randomised wiring, in pre-registered experiments with ten seeds, four ecologies and 600 generations. Two standard randomisations, a column shuffle and degree-preserving edge swaps, route 10.6 to 10.7 % of olfactory output directly onto descending motor groups, against 0.012 % in the connectome. On the registered primary endpoint, fitness averaged over the run, no difference was detected; at the last common-garden probe the connectome was behind both controls (-0.22 and -0.20 fitness units on seed means). Against controls that keep every sensory-output and motor-input edge and rewire only the interior, the seed-mean difference lay within a +/-0.10 equivalence bound (+0.002 and -0.074, unchanged under a calibration that also matches activity spread), although per ecology the interior column shuffle was ahead by 0.26 in one of four ecologies at ten seeds, a lead that ten further pre-registered seeds did not replicate. Rewiring the connectome so that it acquires the shortcut raised its fitness by 0.44 (10 of 10 seeds) and its dependence on olfaction from 0.15 to 0.99; graded doses raised both in step; at comparable swap counts the full dose was ahead of an interior-only sham by 0.53 (10 of 10 seeds); and a sham that rewired the same boundary edges without creating shortcuts matched the connectome (+0.007) while the full dose was ahead of it by 0.60. What a null preserves at the sensory-motor boundary can decide an evolutionary connectome comparison, and sensory-to-motor path statistics belong next to the degree statistics a null is said to preserve.
Neural network models are increasingly constructed from synapse-resolution connectomes, and the same circuit is now available in reconstructions from other animals and pipelines. We do not know whether such models are dependent on the reconstruction they are built from, and whether connectome differences are biological or measurement errors. We studied the Drosophila head-direction circuit in nine versions: four flies (male CNS, hemibrain, FlyWire, BANC) and five variants of the same fly varying only by thresholds, proofreading or synapse detection. Using wiring-based correspondences validated by anatomical labels, we place all differences on one scale. A change of synapse detection within one fly changed the circuit more than halfway to the smallest between-fly difference on four distance measures; proofreading differences were smaller but depended on the metric, and left and right halves of one fly differed no more than random halves. Networks trained on each wiring to hold a cued heading developed near-perfect ring attractors, but their error increased from below 1 degree to 10-28 degrees when all weights changed by 1%. Training on each wiring did not transfer to any other connectome, including other versions of the same fly; refitting thirty cell-type gains, we repaired filtering-level differences but remaining error depended on wiring distance. Training with noise lowered precision and gave only partial transfer. In imaging data, the fly compass drifted gradually in darkness. Connectome-constrained models should report the reconstruction they were built on and be tested for robustness across reconstructions.
Md Ali Ashraf Yad, Nazifa Tabassum Nabila
Wrocław University of Science and Technology · Comilla University