Neuronal circuitry underlying female aggression in Drosophila

Aggressive social interactions are used to compete for limited resources and are regulated by complex sensory cues and the organism’s internal state. While both sexes exhibit aggression, its neuronal underpinnings are understudied in females. Here, we describe a set of connected neurons in the adult Drosophila melanogaster central brain that drive female aggression. We identified a population of sexually dimorphic aIPg neurons whose optogenetic activation increased, and genetic inactivation reduced, female aggression. Analysis of GAL4 lines identified in an unbiased screen for increased female chasing behavior revealed the involvement of another sexually dimorphic neuron, pC1d, and implicated pC1d and aIPg neurons as core nodes regulating female aggression. pC1d activation increased female aggression and electron microscopy (EM) connectomic analysis demonstrated that aIPg neurons and pC1d have strong reciprocal connections. Our work reveals important regulatory components of the neuronal circuitry that underlies female aggressive social interactions and provides tools for their manipulation.


29
Aggressive behaviors are important for gaining access to resources, including food and territory, 30 and are exhibited by both sexes in multiple species (Anderson, 2016;Kravitz and Huber, 2003;31 Zwarts et al., 2012). As aggressive actions carry the risk of injury, strict regulation of aggression 32 is needed to facilitate survival. Sensory information about the presence of other individuals and 33 the nature of the surrounding environment strongly modulate aggressive social interactions 34 (Chen and Hong, 2018;Hoopfer, 2016). However, understanding the neuronal mechanisms by 35 which such stimuli influence aggression has been hindered by a lack of knowledge about the 36 structure of the underlying neuronal circuits, particularly in females. 37

38
Centers mediating, or conveying the information necessary for, aggression have been identified 39 in the medial hypothalamus through classic experiments using electrical stimulation in cats and 40 rodents (Albert et al., 1979;Bandler et al., 1972;Berntson, 1973;Chi and Flynn, 1971; Gregg,41 2003; Kruk et al., 1983;Lammers et al., 1988;Siegel et al., 1999;Takahashi and Miczek, 2015;42 Woodworth, 1971). Such key regions are thought to perform a different role than other brain 43 areas that facilitate aggressive interactions by altering the overall level of social behavior (Siegel 44 et al., 1999). Recent work using opto-and chemo-genetic techniques have narrowed down these 45 key regions to small populations of cells in mice, including those expressing estrogen receptor 46 alpha (Esr1) and progesterone receptor (PR) in the ventrolateral part of the ventromedial 47 Aggressive behaviors have multiple components. We began by quantifying three such behaviors, 154 chasing, touching, and walking, using a set of previously created and validated automatic 155 behavior classifiers (Robie et al., 2017). We found that flies increased touching compared to the 156 empty split-GAL4 control during a 30-second stimulation (Figure 1-figure supplement 7A -B). 157 A low level of chasing was also detected upon activation (Figure 1-figure supplement 7C -D). 158 Although the average walking velocity of the flies over the course of the trial did not differ 159 compared to controls, a sharp decrease in the percent of flies walking followed stimulus onset 160 shared between males and females, there are sex-specific aspects, including head butting and the 171 way in which behavioral patterns progress during an encounter (Nilsen et al., 2004). To examine 172 female-specific attributes, we generated and validated a new JAABA classifier for female 173 aggression (Supplementary Table 1). An aggressive event was defined as either an instance of 174 fencing ( Figure 1B) and/or head butting ( Figure 1C), as these behaviors were not always 175 with the synaptic inhibitor tetanus toxin (Sweeney et al., 1995). A significant reduction in the 224 time spent performing aggressive behaviors, as measured using both manual and automated 225 behavioral analyses, was observed with three different split-GAL4 lines (

Identifying additional cell types involved in mediating female aggression 263
Having established a role for aIPg neurons in female aggression, we used two complementary 264 methods to discover additional cells involved in regulating this behavior. First, we used 265 behavioral screens to identify other cell types that could drive female aggression when activated. 266 Second, we used the aIPg neurons as an entry point for EM-based circuit mapping. As described 267 below, both approaches converged on the same set of cells. 268 269 aIPg and pC1d are two key groups of neurons involved in female aggressive behaviors 270 It is reasonable to expect that other neurons in the circuit or parallel pathways could also induce 271 aggression upon activation. To identify such neurons, we took a strategy analogous to that used 272 by geneticists to answer the question of how many distinct genes produce a particular phenotype 273 when mutated. In their experimental approach, individual mutations are placed into 274 complementation groups after performing a genetic screen large enough to sample all the genes. 275 In this way, the number of different genes that can give rise to the phenotype under study when 276 mutated can be estimated (see for example, Nüsslein-Volhard and Wieschaus, 1980). To identify 277 further cell types that could also increase female aggression, we analyzed the results of a 278 previous screen of over 2,000 GAL4 lines, in which cell types are expressed in multiple lines 279 (Robie et al., 2017). We generated split-GAL4 hemi-driver lines using the enhancers from this  . The images shown were taken from the database at www.janelia.org/gal4-gen1, where the expression patterns of the other lines listed in A can also be found. (D) The expression pattern of a split-GAL4 line made by intersecting these two enhancers; a cell with the morphology of pC1d can be seen.

289
The five cell types that make up the pC1 group express dsx and have been collectively implicated 290 in female receptivity, oviposition, male courtship, and both male and female aggression (  and 2). We also used a split-GAL4 line that labels pC1a -c (provided by K. Wang and B. 306 Dickson). No expression was observed in males in the majority of the lines used (    Taken together, our results suggest that pC1d, but not pC1e or pC1a -c, acts as a significant 353 facilitator of female aggression. 354  (A -B) Images of an area of the left and right hemisphere of an EM section from the FAFB dataset containing the fiber tracts of the 32 putative aIP-g neurons described by Cachero et al. (2010). A dot has been placed in each axon, color-coded to reflect the degree of similarity of its morphology, revealed by manual tracing and visual inspection, to the aIPg neurons contained in our split-GAL4 lines. Grey represents neurons whose morphology clearly differed, magenta represents neurons whose morphology were similar but differed in one or more branches, and green represents neurons that we judged to correspond to those in our split-GAL4 lines. Note that, as we often observe in our split-GAL4 lines (see Figure 1-figure supplement 1), the number of neurons differed between hemispheres; in this case, the left hemisphere had 17 green cells, while the right hemisphere had only 12. (C -D) Skeleton rendering of the traced aIPg neurons, colored based on their similarity to the aIPg neurons identified in our split-GAL4 lines. Panel C shows only cells we judged to correspond to those in our split-GAL4 lines, while D shows all traced cells. White lines in D indicate the approximate plane of A -B. Tracing of grey neurons was stopped when it was clear they did not match our split-GAL4 lines; therefore, their arbors are likely to be incomplete in these images. (A) Interconnectivity between pC1d, pC1e, aIPg Type 1, aIPg Type 2, aIPg Type 3, and PVL04om/lm neurons thresholded at 10 synapses. PVL04om/lm neurons also derive from the aIP-g lineage. Synapse number is noted on each arrow and dashed lines represent interconnectivity within the aIPg Type 1, Type 2 or Type 3, PVL04om/lm or pC1 neurons. Number within circle represents the number of neurons within the cell type; circles without numbers have only one neuron per brain hemisphere. Colored arrows depict synapses from pC1d (purple) and pC1e (blue) to aIPg types. (B) Electron microscopy rendering of all 11 Type 1, 2 and 3 aIPg neurons seen in the hemibrain with color-coded dots at the sites of synapses from pC1d (purple) and pC1e (blue); these synapses are concentrated in the lateral junction above the peduncle of the mushroom body. See Video 7 for more detail on interconnectivity between aIPg neuron types and Video 8 for detail on the connections between aIPg and pC1 neurons.

378
From analyzing the connectivity of aIPg neurons, we identified two pC1 cell types, pC1d and 379 pC1e, among their top six presynaptic inputs. We are confident that these pC1 cells correspond 380 to those observed in our split-GAL4 lines ( Another group of cells in the hemibrain volume, PVL04om/lm, is also heavily innervated by 388 pC1d ( Figure 8A). These cells bear a morphological resemblance to aIPg neurons, but they have 389 distinct projections in the SMP. Moreover, PVL04om/lm neurons do not directly connect with 390 the aIPg type 1, type 2, or type 3 neurons and two of their top inputs include pC1a and pC1c, 391 which do not provide input to the aIPg neurons ( Figure 8A and  Among the aIPg neurons' six strongest pre-synaptic inputs, as judged by synapse number, we 408 found that pC1d and pC1e rank first and sixth, respectively ( Figure 9 and Video 10). A neuron 409 previously implicated in oviposition, oviIN (Wang et al. 2020), ranked eleventh among aIPg 410 inputs, based on synapses number (Figure 9 and Supplementary Table 2). The second and third 411 strongest inputs, SCB014 and the ADM07t, also provide presynaptic input to pC1d, although it is 412 a lower percentage of their total output at 1.89% from SCB014 and 0.22% from the ADM07t 413 group to pC1d, compared to 7.7% and 8.6% to the aIPg neurons, respectively ( Figure 9A, Video 414 10, Supplementary Table 2 and 3). OviIN also forms connections with both aIPg and pC1d, but 415 aside from these three cell types aIPg and pC1d do not share any strong pre-synaptic inputs 416 ( Figure 9A -B, 10A -B, Video 10 and 12; note that ADM07t falls below the cutoff for being 417 displayed and for being considered a significant connection in Figure 10A and in Supplementary 418 Table 3). Only one of the top downstream targets of the aIPg neurons, ADM01r, is also a major 419 downstream target of pC1d ( Figure 9C -D and 10C -D, Video 11 and 13, Supplemental Table 2  420 and 3). However, again we found that the connectivity strengths differ considerably: 1.13% of 421 ADM01r's input comes from pC1d, while 14.42% is provided by aIPg neurons ( Figure 9C and 422 10C, Video 11, Supplemental Table 2 -3).  The circuits that govern aggression in Drosophila are known to be sexually dimorphic and are 435 poorly understood in females. In this paper, we described female aggressive behaviors, uncover 436 key components of the underlying neuronal circuits, develop genetic reagents to manipulate 437 these neurons, and map their connections using EM-level connectomics. Specifically, we 438 discovered the involvement of a subset of the aIPg lineage, a collection of cell types not 439 previously implicated in social behaviors, in mediating female aggressive social interactions. devoting ~13% of its output synapses to them. Behavioral tests using split-GAL4 lines cleanly 447 labeling pC1d demonstrated its ability to increase female aggression. In contrast, we found no 448 evidence for the involvement of the other four pC1 cell types, pC1a -c and pC1e, in aggression; 449 however, we found that pC1e makes synapses onto aIPg neurons raising the possibility that this 450 cell type plays a role that was not revealed by our behavioral assays.

Figure 11. Proposed circuit underlying female aggressive behaviors.
A circuit diagram showing the key neuronal pathways we uncovered in our behavioral and connectomics analyses. The numbers in arrows represent synapse number. We propose that pC1d facilitates female aggressive behaviors through aIPg type 1 (orange), type 2 (green), and type 3 neurons (purple), which act as a mediator of these social interactions. The aIPg type 1 neurons, based on their connections with the other two types and feedback onto pC1d, appear to be important for the recurrency within the circuit. We propose that visual information enters the circuit through interneurons (black outline) that innervate aIPg neurons (see Figure 11-figure supplement 1) or at later nodes in the circuit. Like pC1d, other neuronal populations (color-coded by the proportion of their synapses on each aIPg type) also provide different proportions of their output to the aIPg type 1, aIPg type 2 and aIPg type 3 subtypes. The EM dataset further identified differences in the post-synaptic targets of aIPg type 1, aIPg type 2 and aIPg type 3 neurons, which are shown color-coded by the relative proportion of their synaptic input they receive from each aIPg type. Additionally, while only aIPg type 3 neurons directly connect with a descending interneuron (DN), we identified four post-synaptic targets of the aIPg type 1, aIPg type 2 and aIPg type 3 neurons that form synapses onto DNs, which may function as motor outputs of the circuit. The top four inputs to pC1d and three other neurons of interest are shown. These include neurons implicated in other female behaviors (blue outline), including oviposition (oviIN, pC1a) and mating (vpoDN), suggesting communication between the circuits underlying these interactions. Inputs to the aIPg neurons were thresholded at 6 synapses per aIPg neuron, resulting in collective thresholds of 24, 30, and 12 synapses to aIPg type 1, aIPg type 2, and aIPg type 3, respectively. Outputs of the aIPg neurons were similarly adjusted based on the aIPg type, with 48, 60, and 24 synapses used as the collective thresholds for aIPg type 1, aIPg type 2, and aIPg type 3, respectively. Connections with DNs were thresholded at 30 synapses. on the circuit and behavioral dynamics will be needed to evaluate pC1e's potential role. As 474 circuits can perform more than one behavioral output, pC1e and the rest of this circuit could also 475 be involved in other interactions that will need to be studied using additional assays. 476 477 aIPg type 1, type2 and type3 neurons differ in their pre-and post-synaptic connections 478 The aIPg neurons we studied are three distinct cell types that differ in morphology and 479 connectivity. Our efforts to derive split-GAL4 lines specific for each of these cell types have 480 been unsuccessful, which has limited our ability to explore their individual roles. Nor have we 481 yet performed physiological experiments that might reveal distinct features of the responses of 482 these neurons when the circuit is activated. Nevertheless, we propose that the three aIPg types 483 found in our split-GAL4 lines make distinct contributions to the phenotypes we observed, as 484 there are many differences in their inputs and outputs ( Figure 11). For example, while pC1d 485 makes strong connections onto all three aIPg types, it only receives strong recurrent feedback 486 from type 1 neurons. Similarly, SCB014, which is the third top input to the aIPg neurons, 487 predominately sends its projections to type 2 neurons. There are likewise many differences in 488 downstream targets, some of which connect to descending neurons ( Figure 11). Generating split-GAL4 lines that specifically target such cell types will allow us to perform experiments to 490 explore their roles in aggressive behaviors. Changes in brain state, such as those occurring during aggressive encounters, also influence 527 sensory processing. For example, olfactory cues paired with a male fly winning an aggressive 528 encounter become associated with reward (Kim et al., 2018). We observed that females quickly 529 altered their speed and direction following aIPg stimulation, appearing to approach other flies. 530 These behaviors raise the possibility that visual signals influence the "decision" to be aggressive. 531 For example, certain features of the target, including its size and distance, might be important to 532 drive aggression through aIPg neurons. Another non-exclusive hypothesis is that aggression 533 includes behaviors that rely on vision, such that aIPg activation recruits orientation circuits and 534 In addition to the split-GAL4 lines (see Table 1  599 600

Optogenetic activation behavioral testing 602
Groups of 13 -18 group-housed virgin female flies (5 -10 days post-eclosion) were tested at 603 25°C and 50% relative humidity in a 127 mm circular arena with a center depth of 3.5 mm as 604