A Drosophila larval premotor/motor neuron connectome generating two behaviors via distinct spatio-temporal muscle activity

Animals generate diverse motor behaviors, yet how the same motor neurons generate distinct behaviors remains an open question. Drosophila larvae have multiple behaviors – e.g. forward crawling, backward crawling, self-righting and escape – and all of the body wall motor neurons (MNs) driving these behaviors have been identified. Despite impressive progress in mapping larval motor circuits, the role of most motor neurons in locomotion remains untested, the majority of premotor neurons (PMNs) remain to be identified, and a full understanding of proprioceptor-PMN-MN connectivity is missing. Here we report a comprehensive larval proprioceptor-PMN-MN connectome; describe individual muscle/MN phase activity during both forward and backward locomotor behaviors; identify PMN-MN connectivity motifs that could generate muscle activity phase relationships, plus selected experimental validation; identify proprioceptor-PMN connectivity that provides an anatomical explanation for the role of proprioception in promoting locomotor velocity; and identify a new candidate escape motor circuit. Finally, we generate a recurrent network model that produces the observed sequence of motor activity, showing that the identified pool of premotor neurons is sufficient to generate two distinct larval behaviors. We conclude that different locomotor behaviors can be generated by a specific group of premotor neurons generating behavior-specific motor rhythms.


Introduction
PMN inputs to the A1 MN population. 145 Neurotransmitter expression has been characterized for only a small fraction of the PMNs described here  Table 2), and we did not identify any neurons co-expressing two fast 153 neurotransmitters. 154 Thus, we have identified 67 bilateral PMNs which innervate the MNs in segment A1, mapped all PMN-155 MN synapses within segment A1, and determined neurotransmitter expression for the majority of the PMNs. We 156 conclude that PMNs target the majority of their pre-synapses to the dorsal neuropil, where they connect to other conclude that forward and backward locomotion are two distinct motor patterns, not simply the same pattern in 223 reverse, and that pre-motor/motor circuitry has the ability to drive two distinct patterns of rhythmic muscle 224 activity during the two different behaviors. Similar to forward locomotion, CMuGs during backward locomotion 225 do not match SMuGs.

227
Motor neurons can generate CMuGs without clustering of post-synaptic inputs 228 Next, we wanted to understand the underlying mechanisms for how CMuGs become temporally segregated 229 during locomotion. One possibility is that MNs in a specific CMuG target their post-synaptic sites to a common  Figure 6E). We conclude that co-activity of CMuGs can arise from widely distributed and 247 overlapping MN post-synaptic sites. connectome. There are six distinct SMuGs, and four CMuGs for forward and backward locomotion. We asked if 253 there are groups of PMNs that are dedicated to individual SMuGs or CMuGs. We observed that many PMNs 254 provided highly enriched synaptic input to MNs innervating a single SMuG ( Figure 7A). For example, the PMNs 255 shown in gray strongly prefer to form synapses with MNs innervating DL muscles, whereas PMNs shown in blue 256 prefer to form synapses with MNs innervating LT muscles ( Figure 7A). Despite this bias, no PMN exclusively 257 formed synapses onto MNs in a single SMuG, nor did we observe any PMN that was strongly connected to MNs 258 in all SMuGs ( Figure 7A). We also observed a few PMNs that preferentially innervated a single forward CMuG 259 ( Figure 7B). For example, PMNs in magenta strongly preferred MNs innervating CMuG F2, PMNs in green were Neural circuit motifs predicted to generate distinct motor behaviors 274 In the previous section, we identified PMNs with enriched connectivity to specific forward and/or backward 275 CMuGs. Here we further focus on PMN-MN connectivity and identify circuit motifs that are consistent with the 276 observed CMuG timing, providing candidate motifs for functional studies. We highlight intrasegmental motifs 277 that could produce the observed phase delays between the four CMuGs in a single segment, intersegmental 278 motifs that could produce the sequential activation of a specific CMuG in adjacent segments, motifs that could 279 change the relative activation order of a MN/muscle between forward and backward locomotion, and a motif that 280 could drive motor output initiating escape behavior.

282
Intrasegmental phase delays during forward or backward locomotion 283 Interactions between PMNs are likely to establish the phase delay between the four CMuGs as the motor 284 wave moves across a segment. We used connectome and neurotransmitter data to identify PMN circuit motifs 285 consistent with the observed CMuG intrasegmental phase relationships. First, we identified a disinhibition motif 286 that could generate a phase delay between CMuGs F2 and F3/4. A02i and A14a (preferentially connected to 287 CMuG F3/4) synapse onto A02e (preferentially connected to CMuG F1/F2). All of these PMNs are inhibitory, 288 and thus this motif may disinhibit F1/F2 while inhibiting F3/4, producing a phase delay between F2 and F3/4 289 ( Figure 8A). This confirms and extends previous work showing that A14a creates a phase delay between LO1 290 (CMuG F2) and LT1 (CMuG F4) . We also observed a feedforward excitatory motif that could 291 help synchronize MN activity within individual CMuGs ( Figure 8B) and a feedforward inhibitory motif that could 292 generate a phase delay between early CMuGs (F1/F2 or B1/B2) and late CMuGs (F4 or B4)( Figure 8B).

293
Furthermore, we found another feedforward excitatory motif that could explain how MNs innervating different 294 CMuGs show overlapping peak activity later in the contraction cycle within a segment. A18b2 and A18b3, hence ensuring continued excitation to earlier CMuGs F1/F2 ( Figure 8C). These motifs 299 provide testable hypotheses for how specific phase relationships between CMuGs are generated by PMNs.

301
Intersegmental phase delays during forward or backward locomotion 302 We identified both feedforward excitation and feedforward inhibition motifs that could explain the 303 sequential activation of a specific CMuG in adjacent segments during peristaltic motor waves. The excitatory 304 PMN A27k (preferentially connected to CMuG B4) is involved in a feedforward inhibitory circuit in which it 305 excites the inhibitory local PMNs A02e and A02g (preferentially connected to CMuG B1/B2). This motif could 306 terminate B1/B2 activity and allow B3/4 activity as the contraction wave moves posteriorly ( Figure 8D). A27k 307 also synapses in the next anterior segment with the trio of excitatory neurons described above (A01c1, A01c2, and 308 A18j) which are preferentially connected to CMuG B4, as well as with the inhibitory A02e PMN connected to 309 CMuGs B1/B2. Thus, when an anterior-to-posterior backwards wave stimulates A27k, it results in A27k 310 activating PMNs that trigger CMuG B4 as well as the PMN A02e that inhibits CMuGs B1/B2 in the next anterior 311 segment ( Figure 8E). Furthermore, we found feedforward excitatory and inhibitory motifs that could explain how 312 different CMuGs in the adjacent segments are coordinated. A27h excites A18b3 in the next anterior segment to 313 move the contraction wave forward, while A18b3 excites the inhibitory neurons A06c/A14a to prevent 314 premature activation of neurons in CMuG F3/4 in the next adjacent segment ( Figure 8F). propagation in backward and forward locomotion, MN targets of A03a5 will receive excitatory inputs earlier 331 during backward than forward locomotion ( Figure 8I). We conclude that PMN/MN morphological asymmetry 332 may contribute to the differential timing of muscle activation during between forward and backward locomotion. triggers the escape motor program. Here we find a candidate motor circuit that may initiate C-bending. We find 340 that gorogoro connects via A10f to the PMN A03a1, which specifically innervates dorsal body wall muscles 341 ( Figure 8K). The polarity of A10f and A03a1 are unknown, but if they are both excitatory, then gorogoro could 342 specifically activate dorsal muscles, which we have previously shown is sufficient to induce larval bending (Clark    Genetic inhibition of proprioceptor function results in slower crawling due to prolonged muscle contraction 353 during each peristaltic wave, which has led to the model that proprioceptors send a "mission accomplished" signal 354 to terminate muscle contraction (Hughes and Thomas 2007). 355 We examined the relationship between proprioceptors and PMNs to identify circuit motifs that could 356 generate a "mission accomplished" signal (from ddaE, ddaD, vpda, dmd1, and vbd) or terminating the signal 357 (from dbd). We found strong connectivity between proprioceptor neurons and PMNs, but surprisingly little direct 358 connectivity to MNs ( Figure 9A). Note that less than 20% of the proprioceptor pre-synaptic sites are targeted to 359 the PMNs we have characterized ( Figure 9B), indicating that there may be additional PMNs yet to be 360 characterized, and/or that proprioceptors preferentially connect to pre-PMNs. We find that vbd gives excitatory 361 input to the inhibitory PMNs A02e/A02g ( Figure 9C) or A02k ( Figure 9D), which connect to MNs active 362 throughout the contraction cycle. Thus, this circuit motif would contribute to a "mission accomplished" signal 363 terminating muscle contraction and speeding locomotion. Similarly, dmd1/ddaD activation of the inhibitory 364 PMN A27j could also send a "mission accomplished" signal ( Figure 9E). In support of the functional significance 365 of this motif, we find that A27j is rhythmically active during both forward and backward fictive locomotion in 366 isolated brains, although we were unable to determine its phase-relationship with proprioceptors in this reduced   where muscle length was measured on a frame by frame basis. Calcium imaging data was also analyzed using 555 custom MATLAB scripts. Due to movement artifacts, ROIs were updated on a frame by frame basis to track the 556 muscle movement. ROIs that crossed other muscles during contraction were discarded. In no single preparation 557 was it possible to obtain calcium traces for all 30 muscles. Instead, we used only preparations in which at least 558 40% of the muscles could be recorded. In order to align crawl cycles that were of variable time and muscle 559 composition, we first produced a 2 dimensional representation of each crawl cycle using PCA. Crawl cycles were maximally active, which we defined as the midpoint of a crawl cycle. It should be noted that the muscles used to 565 generate two dimensional representations of crawl cycles were different for each crawl. While this means that each 566 PCA trajectory is slightly different for each crawl cycle, we reasoned that because each experiment contained 567 muscles from every CMuG, the peak amplitude in PCA space should still correspond to a good approximation of 568 the midpoint of the crawl cycle. We defined the width of a crawl cycle as the width of this 2D peak at half-height 569 ( Figure 5 -figure supplement 1G). We aligned all crawl cycles to the crawl onset and offset (which we call 25%      Table 3. Co-activated muscle groups during forward or backward locomotion. 680 There are four co-activated muscle groups during backward and forward locomotion, but the muscles in each 681 group differ in forward versus backward locomotion. Note that backward locomotion is not simple a reverse of 682 the pattern seen in forward locomotion. This represents the most common activation sequences, although there is 683 some variation, particularly during the fastest locomotor velocities.