Connectivity biases generate a learning hierarchy in the Drosophila mushroom body

Connectivity biases generate a learning hierarchy in the Drosophila mushroom body | bioRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-M677548'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search New Results Connectivity biases generate a learning hierarchy in the Drosophila mushroom body Alexander John MacKenzie , Lia Beatty , Jilian Mae Ulibarri , Dua Azhar , Prisca Amematsro , Andrew R. Butts , View ORCID Profile Sophie Jeanne Cécile Caron doi: https://doi.org/10.1101/2025.10.29.684686 Alexander John MacKenzie 1 School of Biological Sciences, University of Utah , Salt Lake City, United States 2 Neuroscience Program, University of Utah , Salt Lake City, UT 84112, United States Find this author on Google Scholar Find this author on PubMed Search for this author on this site Lia Beatty 1 School of Biological Sciences, University of Utah , Salt Lake City, United States Find this author on Google Scholar Find this author on PubMed Search for this author on this site Jilian Mae Ulibarri 1 School of Biological Sciences, University of Utah , Salt Lake City, United States Find this author on Google Scholar Find this author on PubMed Search for this author on this site Dua Azhar 1 School of Biological Sciences, University of Utah , Salt Lake City, United States Find this author on Google Scholar Find this author on PubMed Search for this author on this site Prisca Amematsro 1 School of Biological Sciences, University of Utah , Salt Lake City, United States Find this author on Google Scholar Find this author on PubMed Search for this author on this site Andrew R. Butts 1 School of Biological Sciences, University of Utah , Salt Lake City, United States Find this author on Google Scholar Find this author on PubMed Search for this author on this site Sophie Jeanne Cécile Caron 1 School of Biological Sciences, University of Utah , Salt Lake City, United States 2 Neuroscience Program, University of Utah , Salt Lake City, UT 84112, United States Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Sophie Jeanne Cécile Caron For correspondence: sophie.caron{at}utah.edu Abstract Full Text Info/History Metrics Preview PDF ABSTRACT Learning and memory centers must balance maximizing coding capacity with prioritizing biologically relevant information. Expansion layers, a circuit motif common to many learning and memory centers, including the insect mushroom body, transform dense sensory representations into sparse, distributed ones, and theoretical models propose that random connectivity within these layers maximizes coding capacity by generating highly discriminable responses. Yet this solution creates a fundamental problem: purely random connectivity treats all stimuli equally, without prioritizing survival-relevant cues over neutral ones. Here, we show that the Drosophila melanogaster mushroom body resolves this capacity-selectivity trade-off through systematic biases in projection neuron–Kenyon cell connectivity. Although connectivity is random at the single-cell level, some projection neuron types connect up to 15-fold more frequently than others. These biases translate directly into function: Kenyon cell responses scales with projection neuron connectivity, and the breadth of odor-evoked responses predicts learning performance. Odors activating more than 20% of Kenyon cells drive robust associative memories, whereas those activating fewer than 10% are poorly learned. VL1 projection neurons are a notable exception: despite their weak connectivity, they elicit broad Kenyon cell activity but fail to support learning, revealing a circuit-level gate on learning. These results show that the mushroom body embeds a learning hierarchy in its connectivity architecture, prioritizing ethologically relevant odors while preserving coding capacity for diverse associations. MAIN TEXT Brains are built to learn. At any given moment, sensory systems flood the brain with information, but only a fraction of that information is used to learn and form memories 1 . This selectivity reflects a fundamental trade-off that constrains learning and memory centers: these centers, with their limited number of neurons and synapses, must maintain the capacity to encode a wide range of stimuli, yet they must also reliably prioritize recurring, survival-relevant cues an animal is likely to encounter throughout its lifetime. This trade-off is expected to constrain the circuit architecture of learning and memory centers: architectures that maximize sensory coding capacity may not be optimal for selective encoding of highly relevant cues, and vice versa 2 . Learning and memory centers must therefore be wired in ways that balance both coding capacity and biological selectivity. How this balance is achieved remains poorly understood. Expansion layers — a well-characterized circuit motif found across diverse systems with different learning and memory functions — may provide insight into how circuit architectures balance capacity with selectivity. Expansion layers are characterized by a small number of input neurons connecting to a vastly larger number of coding neurons, transforming sensory inputs into sparse, distributed representations 3 , 4 . This circuit motif appears in both vertebrates and invertebrates, and across diverse brain centers that support different forms of learning and memory. In the vertebrate cerebellum, a relatively small number of mossy fiber inputs connect to a greatly larger number of granule cells that generate sparse representations critical for motor learning and coordination 5 . In the vertebrate hippocampus, entorhinal cortical inputs connect with many more granule cells in the dentate gyrus, that generate sparse representations that support episodic memory 6 . In invertebrates, similar expansion layers are found in the insect mushroom body, an associative center that transforms olfactory information into memories 7 . Theoretical models propose that in these systems, stochastic or “random”, connectivity maximizes coding capacity by generating sparse representation and high discriminability 8 . Yet purely random connections treat all stimuli equally, failing to prioritize survival-relevant stimuli — thus maximizing capacity at the expense of selectivity. To determine whether expansion layers in biological systems deviate from pure randomness to resolve this trade-off, we harnessed the Drosophila melanogaster mushroom body, an accessible experimental system supported by extensive anatomical, genetic, and functional datasets. The D. melanogaster mushroom body comprises approximately 2,000 Kenyon cells that receive input from 51 olfactory glomeruli via projection neurons 7 . Each Kenyon cell receives input from approximately six projection neurons through specialized dendritic structures termed “claws” 9 , 10 . While connectivity appears random at the single-cell level, systematic biases exist across the population 10 – 13 . We previously found that certain projection neurons connect to Kenyon cells far more frequently than others, and that these connectivity biases do not result from sensory activity and are species-specific 11 – 13 . If the mushroom body optimally resolves the capacity-selectivity trade-off, these connectivity biases should represent an evolutionary compromise: sacrificing some coding capacity — which would be maximized by random, uniform connectivity — to selectively prioritize the representation of biologically relevant glomerular channels. This predicts that projection neurons with higher connectivity frequencies should process behaviorally critical odors and support stronger associative learning. We tested this hypothesis in three steps: first, we established that connectivity biases are stereotyped architectural features of the mushroom body; second, we determined whether connectivity biases correlate with Kenyon cell responses; and third, we examined whether connectivity biases translate into learning biases. RESULTS Projection neuron–Kenyon cell connectivity shows systematic biases Projection neurons connect individual olfactory glomeruli of the antennal lobe to the Kenyon cells of the mushroom body ( Figure 1A–D ). We previously found that projection neuron–Kenyon cell connections, while appearing random at the single cell level, show systematic biases: some projection neurons connect to Kenyon cells far more frequently than others 11 – 13 . For example, projection neurons from the DL3 glomerulus connect to Kenyon cells less frequently than expected under a uniform distribution, while DA1 projection neurons connect more frequently ( Figure 1E–F ). To establish whether these biases represent conserved architectural features of the Drosophila mushroom body, we quantified their magnitude and reproducibility across multiple independent datasets. Download figure Open in new tab Figure 1. Projection neuron–Kenyon cell connections are systematically biased. (A,B) Electron microscopy–based reconstructions of DL3 (green) and DA1 (pink) projection neurons innervating glomeruli in the antennal lobe (AL). Each projection neuron forms en passant synapses with Kenyon cells (black) in the mushroom body (MB) and terminates in the lateral horn (LH). Panels show anterior (A) and dorsal (B) views. Outlines of the antennal lobe and mushroom body are included for anatomical context (gray). Orientation arrows indicate dorsal (D), medial (M), and posterior (P) directions. (C) A Kenyon cell (black) forms five claws, one of them (red arrowhead) receiving input from a DL3 projection neuron (green). (D) A Kenyon cell (black) forms six claws, one of them (red arrowhead) receiving input from a DA1 projection neuron (pink). (E) Distribution of glomerular input frequencies to Kenyon cells in two electron microscopy–based connectomes. Each data point represents the proportion of synapses contributed by projection neurons from a given glomerulus to the entire Kenyon cell population. The dotted line represents the connectivity frequency expected under a uniform distribution (in this case, 1.96%). Symbols indicate datasets: upward triangle (Scheffer et al., 2020), downward triangle (Dorkenwald et al., 2024). Colors indicate degree of representation: pink (overrepresented glomeruli), green (underrepresented glomeruli) or gray (normally represented glomeruli). (F) Distribution of glomerular input frequencies in four large-scale connectivity matrices from single-cell dye-filling experiments. The dotted line represents the connectivity frequency expected under a uniform distribution (in this case, 1.98%). Symbols indicate datasets: circle (Caron et al., 2013), square (Hayashi et al., 2022), upward triangle (Ellis et al., 2024, female), downward triangle (Ellis et al., 2024, male). Colors indicate degree of representation: pink (overrepresented glomeruli) or green (underrepresented glomeruli). See Table S1 . We first analyzed two complete connectomes that reconstructed full mushroom bodies: the hemibrain connectome and the FlyWire whole brain connectome 14 , 15 . These datasets report the number of synapses contributed by the projection neurons associated with each glomerulus to the entire Kenyon cell population. If connections were purely random, each of the 51 projection neuron types — corresponding to the 51 glomeruli — would be expected to contribute roughly 2% of the total input. Instead, we observed a markedly non-uniform distribution, with individual glomeruli contributing between 0.34% and 4.91% of total inputs, about 15-fold difference, indicating substantial biases ( Table S1 ). Strikingly, only four glomeruli were consistently represented at near-expected levels in both connectomes ( Figure 1E ). Notably, the identities of over- and underrepresented glomeruli were consistent across both connectomes, underscoring the robustness of these connectivity patterns. Second, we validated this observation using a complementary method and analyzed four large-scale connectivity matrices generated from dye-filling and single-cell tracing experiments 11 – 13 . In these datasets, projection neuron inputs to individual Kenyon cells were systematically traced, with each matrix comprising at least 200 Kenyon cells. Unlike the electron microscopy–based connectomes, these matrices reflect the frequency with which projection neurons from specific glomeruli connect to Kenyon cells, independent of number of synapses. These datasets also revealed a non-uniform distribution of inputs, with glomerular contributions ranging from 0.19% to 4.48% ( Table S1 ). Only twelve glomeruli were consistently represented at near-expected levels in all four matrices ( Figure 1F ). Most importantly, the identities of over- and underrepresented glomeruli are consistent across both electron microscopy–based connectomes and large-scale tracing datasets. The fact that connectivity biases are both stereotyped and reproducible across independent datasets generated using distinct methods indicates that they represent a systematic departure from purely random connectivity. Bouton morphology predicts connectivity biases Our previous study suggested that connectivity biases arise from the number of synaptic boutons collectively formed by the projection neurons associated with a given glomerulus in the calyx of the mushroom body rather than the number of projection neurons associated with a given glomerulus 13 . We sought to confirm this finding. To visualize projection neurons with varying degrees of connectivity, we used publicly available split-GAL4 transgenic lines 16 . Such lines are currently available for a limited number of projection neurons types — specifically the underrepresented VL1, DL3, and VM4 glomeruli and the overrepresented DA1 glomerulus ( Figure 2A–D,G–J ). Additional lines target projection neurons innervating pairs or trios of glomeruli, including the DA4l and DC3 glomeruli, and the VM5v, VM7d, and VM7v glomeruli ( Figure 2E – F,K–L). Based on the combined connectivity frequencies of the glomeruli they target, both lines are expressed in projection neurons with a cumulative representation that exceeds that of the average glomerulus and are therefore classified as overrepresented. Download figure Open in new tab Figure 2. Morphological features of projection neurons correlate with connectivity frequency. (A–F) Distribution of connectivity frequencies from Figure 1E , with colored markers highlighting the six projection neuron types targeted by the split-GAL4 lines used in this study. Lines target either individual glomeruli (VL1, DL3, VM4, DA1) or glomerular combinations (DA4l, DC3; VM5d, VM7d, VM7v). Numbers indicate average connectivity frequencies for lines targeting a single glomerulus or cumulative connectivity frequencies for lines targeting multiple glomeruli. Colors indicate degree of representation: pink (overrepresented glomeruli) or green (underrepresented glomeruli). (G–R) Expression patterns of projection neurons labeled by the split-GAL4 lines , visualized with UAS-GFP and immunostained for GFP (green) and Bruchpilot (magenta). Scale bar: 100 μm. (G′–L′) Higher magnification views of antennal lobe glomeruli showing projection neuron arborizations and somata. Scale bar: 20 μm. (M–R) Higher magnification views of mushroom body calyx showing presynaptic boutons. Scale bar: 20 μm. (S–U) Quantification of morphological features versus average connectivity frequency from electron microscopy connectomes. (S) Number of projection neurons per line. (T) Total bouton number in calyx. (U) Total bouton volume. Each data point represents a split-Gal4 line; error bars show standard error of the mean; linear regression lines are shown with R² and p values. Symbols indicate the identity of the split-Gal4 lines: circle (VL1), square (DL3), downward triangle (VM4), upward triangle (DA1), hexagon (DA4l+DC3), diamond (VM5d+VM7d+VM7v). Colors indicate degree of representation: pink (overrepresented glomeruli) or green (underrepresented glomeruli). See Table 1 and Figure S1 . View this table: View inline View popup Download powerpoint Table 1. Morphological features of projection neurons labeled by split-GAL4 lines. The table reports projection neurons expressing each split-GAL4 transgene, their associated glomeruli, and quantified morphological features including number of neurons, number of boutons, and total bouton volume. For comparison, projection neuron counts for each glomerulus from two electron microscopy connectomes are included (Scheffer et al., 2020; Dorkenwald et al., 2024). Values are reported as means with standard error of the mean. We used these transgenic lines in combination with a UAS-GFP reporter to quantify different morphological features of these projection neurons: the number of GFP-labeled neurons, the number of boutons formed in the mushroom body calyx, and the bouton volume ( Table 1 ). We then compared each of these features to the average connectivity frequency of the corresponding glomeruli, based on data from both the electron microscopy–based connectomes and the large-scale tracing datasets ( Figure 2S–U , Figure S1 ). The number of boutons and bouton volume were strong predictors of connectivity frequency ( R ² = 0.77, p = 0.02 and R² = 0.78, p = 0.02, respectively; Figure 2T,U ). In general, projection neurons that formed more or larger boutons were more frequently connected to Kenyon cells. For example, projection neurons from the VL1 glomerulus — consistently underrepresented across datasets — had the fewest boutons and the smallest total bouton volume ( Figure 2A,G,M,T,U ). By contrast, projection neurons from the DA1 glomerulus had above-average bouton number and volume and were among the most overrepresented ( Figure 2D,J,P,T,U ). The line expressing in the VM5v, VM7d, and VM7v projection neurons formed the largest number of boutons, consistent with the combined connectivity frequencies of these glomeruli ( Figure 2F,L,R,T,U ). By comparison, the number of projection neurons per glomerulus was a weaker predictor of connectivity frequency ( R ² = 0.52, p = 0.10; Figure 2S ). For example, the DL3 glomerulus had the highest projection neuron count among lines restricted to a single glomerulus, with eleven labeled neurons, yet it was underrepresented in both connectomes and two of the four connectivity matrices ( Table 1 , Figure 2B,H,N,S ). These findings confirm that connectivity biases arise from morphological differences in projection neuron boutons — specifically bouton number and volume — rather than from projection neuron number, revealing a structural basis for biased connectivity within the mushroom body circuit architecture. Connectivity biases correlate with breadth of Kenyon cell responses We next tested whether projection neuron–Kenyon cell connectivity biases result in differences in Kenyon cell responses. To address this, we developed an imaging approach that combined optogenetic stimulation of projection neurons with calcium imaging in Kenyon cells. Projection neurons that expressed the light-gated channel Chrimson under the control of split-GAL4 transgenic lines were activated using 637 nm light, while Kenyon cells expressing GCaMP6f under LexA control were imaged following a brief light pulse across multiple optical planes using two-photon microscopy. The combination of these split-GAL4 and LexA tools enabled us to specifically activate a given type of projection neuron while recording across Kenyon cells. For each Kenyon cell, response magnitude was quantified as ΔF/F, and cells were initially categorized as either responders or non-responders ( Figure 3A–L , Figure S2 ). When plotting the distribution of response amplitudes among responders, we found it to be unimodal with a pronounced right, positive skew, indicative of a substantial population of strongly responsive cells ( Figure S3A –F). Based on this distribution, responders were further classified as either weak or strong based on the magnitude of their responses ( Figure 3A ’’–F’’, Figure S2C,D ). We quantified three parameters for each sample: the percentage of responding Kenyon cells (“responders”), the frequency of weak and strong responders and the mean maximum peak amplitude among all responders. These parameters were then compared to the average connectivity frequency of the glomeruli stimulated in each experiment, as derived from electron microscopy–based connectomes or large-scale tracing datasets ( Figure 3M–P , Figures S4A-D ). Download figure Open in new tab Figure 3. Projection neuron connectivity frequency predicts Kenyon cell responses. (A–F) Optogenetic activation of projection neurons from split-GAL4 lines and corresponding responses in Kenyon cells. For each line, panels show: a representative fluorescence image of Kenyon cells expressing GCaMP6f in a single optical plane (A–F); a heat map of ΔF responses following light stimulation (A′–F′); and segmented activation map in which non-responsive Kenyon cells are filled in white, weak responders are filled in yellow, and strong responders are filled in red (A″–F″). Scale bar: 20 μm. (G–L) Representative raster plots showing calcium responses from all recorded Kenyon cells following optogenetic stimulation obtained using each of the split-GAL4 lines: VL1 (G), DL3 (H), VM4 (I), DA1 (J), DA4l+DC3 (K), and VM5d+VM7d+VM7v (L). Each row represents one cell, each column represents one imaging frame, and color intensity indicates ΔF/F amplitude at that frame (scale: 0 to 0.5 standard deviation). The dotted vertical line marks stimulus onset. (M–T) Quantification of Kenyon cell responses plotted against connectivity frequency of stimulated projection neurons (averaged across both connectomes; Table 1 ). Panels M–P include all data; panels Q–T show the same analyses with VL1 excluded. (M,Q) Percentage of all responders. (N,R) Percentage of weak responders. (O,S) Percentage of strong responders. (P,T) Maximum peak ΔF/F amplitude. Each data point represents a split-Gal4 line; error bars show standard error of the mean; linear regression lines are shown with R² and p values. Symbols indicate projection neuron identity: circle (VL1), square (DL3), downward triangle (VM4), upward triangle (DA1), hexagon (DA4l+DC3), diamond (VM5d+VM7d+VM7v). Colors indicate glomerular representation: pink (overrepresented) or green (underrepresented). See Table S2 and Figures S2 – S13 . We found that the percentage of weak and strong responders varied across split-GAL4 lines and positively correlated with the connectivity frequency of the stimulated projection neurons, though this correlation did not reach statistical significance ( Figure 3M–O , Figure S4A–C , Figures S5 – S11 ). The proportion of strong responders was higher when stimulating projection neurons with higher connectivity frequency — for example, in lines targeting the DA1 glomerulus or multiple glomeruli ( Table S2 , Figures S8 – S11 ). In contrast, strong responders were nearly absent in lines targeting projection neurons with lower connectivity frequency, with VL1 being the only consistent exception — its stimulation generated the largest number of strong responders ( Table S2 , Figure S5 , Figure S11 ). This is a surprising pattern we would observe again with odor stimulation. When VL1 was excluded from the analysis, the correlation between connectivity frequency and strong responders became more evident and approached significance ( R² = 0.73, p = 0.07; Figure 5S ). The mean maximum peak amplitude of responders did not correlate with connectivity frequency, either with or without VL1 included in the analysis, suggesting that connectivity frequency primarily affects response likelihood once a cell is responsive ( Figure 3P,T , Figure S4D,H ). View this table: View inline View popup Download powerpoint Table 2. Predicted glomerular activation and connectivity frequencies for tested odors. The table reports the predicted olfactory receptor and glomeruli activated by each of the tested odors, based on publicly available response data (Münch & Galizia, 2016). Mean connectivity frequencies of the corresponding projection neurons are provided from electron microscopy connectomes and large-scale tracing datasets (Caron et al., 2013; Li et al., 2020; Hayashi et al., 2022; Dorkenwald et al., 2024). For multi-glomerular odors, cumulative connectivity frequency is calculated as the sum of individual glomerular connectivity frequencies across all activated receptors. To explore potential mechanisms underlying the discrepancy noted with VL1, we considered two possibilities. First, VL1 projection neurons may preferentially converge onto the same Kenyon cells, increasing the likelihood of coincident input. However, our analysis of one of the connectomes provided no support for this hypothesis: there are only two Kenyon cells receiving two inputs from the VL1 glomerulus in the FlyWire connectome, a number consistent with biased random connectivity ( Figure S12A–B ). Second, individual VL1 projection neuron–Kenyon cell connections may contain an unusually high number of synapses, allowing a single input to drive Kenyon cell responses more effectively. But again, our analysis of the FlyWire connectome provided no such support for this hypothesis, as the VL1 projection neurons had the second-lowest synapse count per Kenyon cell connection among all projection neurons ( Figure S12C ). Interestingly, a previous study reported that VL1 projection neurons differ from other projection neurons in that they form an unusually large number of synapses onto downstream inhibitory neurons 17 . Thus, the large Kenyon cell responses evoked by stimulation of the VL1 projection neurons may reflect indirect disinhibition rather than direct excitation. Together, these findings suggest that additional, unidentified circuit mechanisms contribute to the unexpectedly strong activation of Kenyon cells by VL1 projection neurons. These results demonstrate that connectivity biases directly translate into differences in Kenyon cell responses, with projection neurons having higher connectivity frequencies activating more Kenyon cells, and doing so more strongly. Odor-evoked Kenyon cell responses reflect connectivity biases Our optogenetic experiments revealed that the frequency with which projection neurons connect to Kenyon cells predict the number of Kenyon cells that respond to their stimulation, suggesting that biased connectivity could impact odor-evoked responses. We next wanted to test whether larger odor-evoked Kenyon cell responses may be more supportive of learning than smaller ones. To explore this, we sought to identify odors that activate Kenyon cells at varying frequencies. Odors differ in the number and identity of olfactory sensory neurons, and consequently, glomeruli they activate: some odors activate multiple glomeruli, which we refer to as “multi-glomerular odors,” while others primarily activate a single glomerulus, which we refer to as “mono-glomerular odors.” We hypothesized that multi-glomerular odors would activate more Kenyon cells than mono-glomerular odors. To test this, we imaged Kenyon cell responses to six odors with varying glomerular breadth. Three mono-glomerular odors — geosmin (DA2), farnesol (DC3), and valencene (DC1) — were compared to three multi-glomerular odors — pyrrolidine (two glomeruli), 4-methylcyclohexanol (three glomeruli), and 3-octanol (ten glomeruli) ( Table 2 ). Glomerular activation patterns were estimated from a publicly available database of olfactory responses 18 . Kenyon cells expressing GCaMP6f under LexA control were imaged using two-photon microscopy during a three-second odor pulse delivered via an olfactometer ( Figure 4A–L ). For each sample, Kenyon cell somata were segmented, and odor-evoked calcium responses were analyzed following a procedure similar to that described in the previous section: based on response amplitude, each Kenyon cell was classified as a non-responder, weak responder or strong responder ( Figures 4A ’’–F’’, Figure S14 ). To assess how connectivity influences odor-evoked activity, we quantified three response features for each sample: the percentage of responders, the frequency of weak and strong responders, and the mean maximum peak amplitude among responders. These parameters were then compared to the average connectivity frequency of the glomeruli activated by each odor, as determined from either electron microscopy–based connectomes or large-scale projection neuron tracing datasets ( Figures 4M-P , Figure S15A–D ). Download figure Open in new tab Figure 4. Multi-glomerular odors generate broader Kenyon cell responses. (A–F) Kenyon cell responses to stimulation with odors of varying glomerular breadth. For each odor, panels show: a representative fluorescence image of Kenyon cells expressing GCaMP6f in a single optical plane (A–F), a heat map of ΔF responses following odor presentation (A’–F’), and segmented activation map in which non-responsive Kenyon cells are filled in white, weak responders are filled in yellow, and strong responders are filled in red (A″–F″). Scale bar: 20 μm. (G–L) Representative raster plots showing calcium responses from all recorded Kenyon cells following presentation of each odor: farnesol (G), valencene (H), geosmin (I), 4-methylcyclohexanol (J), pyrrolidine (K), and 3-octanol (L). Each row represents one cell, each column represents one imaging frame, and color intensity indicates ΔF/F amplitude at that frame (scale: 0 to 0.5 standard deviation). The dotted rectangular outline marks odor onset and offset. (M–T) Quantification of Kenyon cell responses plotted against cumulative connectivity frequency of glomeruli activated by each odor ( Table 2 ). Panels M–P include all odors; panels Q–T show the same analyses with pyrrolidine excluded. (M,Q) Percentage of all responders. (N,R) Percentage of weak responders. (O,S) Percentage of strong responders. (P,T) Mean maximum peak ΔF/F amplitude. Each data point represents an odor; error bars show standard error of the mean; linear regression lines are shown with R² and p values. Symbols indicate odor identity: square (farnesol), downward triangle (valencene), lozenge (geosmin), hexagon (4-methylcyclohexanol), upward triangle (pyrrolidine), circle (3-octanol). See Table S3 and Figures S14 – S17 . Download figure Open in new tab Figure 5. Kenyon cell response breadth predicts learnability across paradigms (A–D) Associative learning performance for six odors measured across four classical conditioning paradigms varying in reinforcement type and memory duration. (A) Appetitive short-term memory assessed 30 minutes after training with sucrose reward. (B) Appetitive long-term memory assessed 24 hours after training. (C) Aversive short-term memory assessed 30 minutes after training with electric shock punishment. (D) Aversive long-term memory assessed 24 hours after training. Each data point represents mean performance index; error bars show standard error of the mean. Symbols indicate odor identity: square (farnesol), downward triangle (valencene), lozenge (geosmin), hexagon (4-methylcyclohexanol), upward triangle (pyrrolidine), circle (3-octanol). (E–L) Correlation between learning performance and Kenyon cell responses. Performance index from each paradigm (A–D) plotted against percentage of responders for each odor (from Figure 4 ). Panels E–H include all odors; panels I–L show analyses with pyrrolidine excluded. Each data point represents an odor; error bars show standard error of the mean; linear regression lines are shown with R² and p values. Symbol as in (A–D). (M–T) Correlation between learning performance and cumulative connectivity frequency of glomeruli activated by each odor ( Table 2 ). Panels M-P include all odors; panels Q–T show analyses with pyrrolidine excluded. Each data point represents an odor; error bars show standard error of the mean; linear regression lines are shown with R² and p values. Symbol as in (A–D). See Figure S14 – S17 . As expected, multi-glomerular odors generally generated broader Kenyon cell responses, with approximately 20% of Kenyon cells responding to the odor, while most mono-glomerular generated sparser responses, with fewer than 10% of the Kenyon cells ( Figure 4M ). These results align with predictions from our optogenetic experiments. Frequency of responders tended to increase with the cumulative connectivity frequency of the glomeruli activated by an odor, yielding a moderate positive correlation that did not reach statistical significance ( Figures 4M , S15A ). In other words, odors that activated more glomeruli — and glomeruli whose projection neurons are more frequently connected to Kenyon cells — generally drove broader responses. As observed in the optogenetic dataset, the distribution of amplitudes among responders was unimodal with a pronounced right, positive skew in multi-glomerular odors, indicative of a substantial population of strong responders ( Figure S3G–L ). Strong responders were indeed more frequent when multi-glomerular odors were used ( Table S3 , Figure 4O , Figure S15C , Figure S16 , Figure S17 ). By contrast, mean maximum peak amplitude showed no correlation with connectivity frequency consistent with the results we obtained using optogenetic stimulation ( Figure 4P , Figure S15D ). A notable exception was pyrrolidine, an odor that activates only two glomeruli, one of which is the underrepresented VL1 glomerulus. Consistent with our optogenetic results, pyrrolidine activates more Kenyon cells than expected based on its connectivity frequency: it elicited responses in 17.4% ± 9.7 of the Kenyon cells despite activating projection neurons with relatively low connectivity (3.73%), a response magnitude to that evoked by multi-glomerular odors that activate projection neurons with higher connectivity frequencies ( Table 2 , Figure 4M , Table S3 ). Strikingly, these broad responses closely mirrored those observed following optogenetic stimulation of VL1 projection neurons. When pyrrolidine was excluded from the analysis, the correlations between connectivity frequency and number of responders became stronger and approached or reached significance ( Figure 4Q–T , Figure S15E–H ). This finding supports the idea that, although projection neuron connectivity generally predicts Kenyon cell responses, VL1 projection neurons can bypass this relationship through alternative circuit mechanisms. These mechanisms may enable disproportionately widespread downstream activation of Kenyon cells, possibly through indirect disinhibitory mechanisms, as discussed in the previous section. These results demonstrate that odor-evoked Kenyon cell responses result from connectivity frequency, confirming that connectivity biases shape odor representation in the mushroom body. Odor-evoked Kenyon cell responses predict learning, with a notable exception Having established that different odors evoked Kenyon cell responses that vary in breadth, we next asked whether the breadth of these responses predicted learning performance. We hypothesized that odors eliciting broader Kenyon cell activation would support stronger memory formation and therefore better learning performance than those eliciting narrower responses. To test this, we used a classical conditioning paradigm where flies learned to associate odors with appetitive or aversive stimuli 19 , 20 . Learning performance was quantified as a preference index and assessed at both 30 minutes (short-term memory) and 24 hours (long-term memory) after training. We tested six odors for which Kenyon cell activity had been previously characterized. Consistent with our hypothesis, odors with broader responses generally supported higher learning indices ( Figure 5A-D ). The multi-glomerular odors 3-octanol and 4-methylcyclohexanol, which activated the largest number of Kenyon cells, produced the highest preference indices across all paradigms, independent of reward type or memory duration. Learning performance correlated positively with both breadth of responses and the cumulative connectivity frequency of activated glomeruli across paradigms ( Figure 5E-H , 5M-P ). However, not all odors conformed to this relationship. Despite evoking large Kenyon cell responses — most likely through VL1 projection neurons — pyrrolidine failed to support learning in any conditioning paradigm ( Figure 4E , Table S3 , Figure S16 , Figure S17 ). When pyrrolidine was excluded from analysis, correlations between learning performance and both breadth of responses and connectivity frequency approached or reached significance across paradigms ( Figure 5I-L,Q-T , Figure S18 ). This exception demonstrates that while breadth of responses and connectivity frequency are strong predictors of learning performance for most odors, specific channels can override these general rules through alternative circuit mechanisms. Altogether, these findings reveal a learning hierarchy in the mushroom body — a connectivity– function–learning rule — where connectivity biases shape circuit architecture to favor certain odors while maintaining overall learning flexibility. DISCUSSION Our study demonstrates that connectivity biases built into the Drosophila melanogaster mushroom body architecture generate a learning hierarchy. Projection neurons that form more presynaptic boutons connect more frequently to Kenyon cells, enabling larger odor-evoked responses and supporting stronger learning performance, whereas those with fewer boutons show reduced connectivity, smaller responses, and weaker learning. This hierarchical relationship allows learning performance for most odors to be predicted from connectivity frequencies and odor-evoked Kenyon cell responses. The VL1 projection neurons are, however, a notable exception: these neurons activate a large number of Kenyon cells despite minimal connectivity, and fail to support learning, revealing that additional circuit mechanisms can override this learning hierarchy. The relationship between connectivity, function and learning identified here might reveal a fundamental architectural feature that enables learning and memory centers to balance coding capacity with biological selectivity. Statistical considerations in analyzing circuit motifs with random connectivity Analyzing the connectivity architecture of brain centers built with expansion layers, such as the mushroom body, presents a unique statistical challenge because of the inherent randomness of their wiring. Unlike most biological systems where variability primarily reflects experimental noise, the mushroom body exhibits genuine biological randomness at the single-cell level: projection neurons connect to Kenyon cells in a stochastic manner 9 – 11 . This architectural feature introduces substantial biological variance in population-level measurements, such as the fraction of Kenyon cells responding to a stimulus or learning performance. Consequently, traditional statistical tests can be overly conservative for detecting systematic biases, creating a “statistical floor” that limits apparent correlation strength even when correlations exist. The key challenge, therefore, is to distinguish meaningful population-level structure from background variability intrinsic to randomly wired brain centers. Our approach of analyzing multiple independent datasets — including two complete connectomes and four large-scale tracing studies — provides a robust cross-validation that strengthens confidence in the observed patterns. The consistency of glomerular rankings across these independent datasets indicates that the connectivity biases we identified reflect genuine architectural features rather than sampling artifacts. We also collected large datasets such that effect sizes for key correlations when excluding the outlier VL1 projection neurons were consistently high (R² = 0.54-0.88), with most exceeding R² = 0.70, indicating that connectivity biases explain substantial proportions of the variance in strong Kenyon cell responses and learning performance across paradigms ( Figure 3S , Figure 4S , Figure 5I-L,Q-T , Figure S4G , Figure S15G , Figure S18E-H ). Several correlations reached conventional significance thresholds, while others approached significance. Given the statistical floor imposed by the intrinsic randomness of mushroom body connectivity, additional sampling would not be expected to reduce this variance substantially or increase correlation strength. Future studies examining connectivity–function–learning relationships across a broader panel of odors would, however, determine the generality of the learning hierarchy identified here. VL1 glomerular channel as an exception to the connectivity–function–learning rule The VL1 glomerular channel represents a notable exception to the connectivity–function–learning relationship we observed. Despite being among the least connected projection neuron types, stimulation of the VL1 projection neurons using optogenetic tools resulted in unusually large Kenyon cell responses. Pyrrolidine — an odor detected by the VL1 glomerulus — evoked similarly broad Kenyon cell responses yet failed to support associative learning in both appetitive and aversive paradigms. Thus, the VL1 glomerular channel decouples Kenyon cell responses from learning, indicating that additional circuit mechanisms can override the learning hierarchy imposed by biased connectivity. We can conceive of at least two non-exclusive mechanisms that may account for this discrepancy. First, analysis of the FlyWire connectome revealed that the VL1 projection neurons connect extensively to inhibitory neurons, suggesting that VL1 projection neurons may activate Kenyon cells through indirect mechanisms 17 . While this provides a clear anatomical basis for the broad pyrrolidine-evoked responses we observed, it does not explain why these activity patterns are insufficient to support learning. Second, gating could occur at the level of plasticity rather than neuronal activation: the VL1 projection neurons may activate Kenyon cells together with unknown modulatory circuits that prevent learning by blocking synaptic modifications between Kenyon cells and mushroom body output neurons. While this provides an explanation for the learning failure, it does not explain why VL1 neurons generate such broad Kenyon cell activity patterns. Further circuit dissection studies will be required to determine whether one, both, or neither of these mechanisms are at play. Despite these likely mechanisms, it remains unclear why the VL1 glomerular channel appears specialized — or at least segregated — from other channels in D. melanogaster . The VL1 glomerulus receives input from Ir75d-expressing olfactory sensory neurons, whose only known ligand is pyrrolidine, a plant secondary metabolite produced as a defensive compound against herbivores 21 , 22 . For cues that signal potential toxicity, innate, immediate responses are likely more adaptive than learned ones. A circuit motif that enables robust detection and broad activation of Kenyon cells while gating plasticity downstream would allow flies to respond rapidly to noxious odors without committing them to memory, thereby conserving neuronal resources for stimuli with greater behavioral relevance. Whether and how D. melanogaster uses pyrrolidine in its natural environment — and how the VL1 glomerular channel contributes to such behavior — remain open questions for future investigation. The VL1 exception exemplifies how systematic analysis of the connectivity architecture of a learning and memory center can generate unexpected findings with mechanistic significance and, potentially, ecological insights. What initially appeared as a statistical outlier in the connectivity– function–learning relationship revealed to be a functionally specialized channel that operates outside the general learning hierarchy and may be segregated by distinct circuit mechanisms. This finding demonstrates the worth of our approach: by establishing quantitative relationships between circuit architecture and behavior across multiple odors, we created a framework that not only predicts learning performance but also identifies exceptional cases that point to undiscovered circuit mechanisms that are most likely shaped by ecological pressures. Random-biased connectivity as a common circuit motif in expansion layers Expansion layers are a recurrent circuit motif in learning and memory centers that can maximize coding capacity by transforming dense sensory inputs into sparse, high-dimensional representations through random connections 8 . Yet in the D. melanogaster mushroom body, we found that the mushroom body architecture deviates from purely random connectivity, and that biases emphasize certain glomerular channels over others. These biases may resolve a fundamental trade-off faced by learning and memory centers: maintaining flexibility to represent a wide range of inputs while prioritizing those of greatest behavioral relevance. This architectural feature prevents meaningful information from being diluted among countless equiprobable combinations, allowing the mushroom body architecture to remain both general and selective. In this view, randomness and bias act as complementary architectural features — randomness maximizing coding capacity, and bias allocating that capacity toward ecologically important stimuli. Hints of biased random connections as an architectural feature appear in vertebrate expansion layers. In the cerebellum, granule cells receive mossy-fiber inputs that are far from uniform. Mossy fibers originating from related sensorimotor pathways cluster into discrete microzones, and specific body or movement representations occupy disproportionately large territories within the granular layer 5 . This arrangement may bias cerebellar learning toward frequently executed or ethologically critical motor patterns, favoring precision and efficiency over uniform sampling. Likewise, in the hippocampal dentate gyrus, inputs from the entorhinal cortex are unevenly distributed: medial entorhinal inputs conveying spatial information and lateral entorhinal inputs carrying object or contextual signals form distinct and unequally weighted projections 23 . Such asymmetries may predispose learning toward spatial or contextual associations, which are essential for survival in freely moving animals. Although the functional consequences of these biases remain to be tested directly, their anatomical prominence suggests that random-biased organization could be a general feature of expansion layers, tuning representation space toward high-value information while retaining capacity for novelty. Together, these observations suggest that random-biased connectivity may represent a general circuit strategy for resolving the trade-off between capacity and selectivity. By embedding structured biases within otherwise random wiring, expansion layers can balance flexibility with evolutionary tuning, ensuring that learning systems remain capable of generalization while retaining sensitivity to behaviorally salient cues. Understanding how this balance is achieved across taxa will be key to uncovering shared organizational principles that govern learning and memory architectures. Biased random connectivity might therefore be a general circuit motif used in expansion layers to solve the tradeoff between capacity and selectivity. Balancing capacity and selectivity The connectivity biases uncovered in the mushroom body, and their functional and behavioral implications, likely reflect evolutionary tuning to the ecological niche specific to D. melanogaster . Glomerular channels detecting food sources and mates are overrepresented suggesting that biases emphasize stimuli that are predictive of survival. In light of these findings, our results suggest that expansion layers are not a mere tabula rasa coding device but have a circuit architecture adapted to the environment of a species that has been shaped by ecological pressures. Each brain, and its learning and memory centers, need to operate with limited number of neurons and synapses. No brain can encode everything and therefore every brain must prioritize cues that are most relevant for the survival of an animal. It remains unclear what the full set of advantages is that could make the selectivity conferred by connectivity biases outweigh the loss in capacity. By compressing representational space around biologically relevant dimensions, such biases may enhance higher-order functions such as stability, generalization, and rapid inference 24 . A biased expansion layer, for instance, may maintain stable and reliable representations even as sensory inputs fluctuate. Additionally, because biases preserve relationships among survival-related stimuli, they may promote generalization across them. Likewise, a circuit already weighted toward biologically meaningful cues could enable faster and more efficient learning. Future behavioral investigations using the experimental framework established here will be essential to determine whether the selectivity afforded by biased connectivity, despite the loss of capacity, supports such higher-order functions. Understanding how evolution tunes this trade-off — how much randomness a brain can afford to lose — will be key to uncovering the architectural features that make learning and memory centers both adaptive and economical. AUTHOR CONTRIBUTIONS A.J.M. and S.J.C.C. conceived the project. A.J.M. performed and analyzed the functional imaging experiments. J.M.U., P.A. and A.R.B. performed and analyzed the behavioral experiments. D.B.A. and P.A. performed and analyzed the anatomical experiments under the supervision of A.J.M. L.B. and A.J.M. curated and organized all the experimental dataset. L.B. prepared all the figures under the supervision of A.J.M. and S.J.C.C. A.J.M. and S.J.C.C. wrote the manuscript with input from all authors. DECLARATION OF INTERESTS The authors declare no competing interests. METHODS Fly stocks and husbandry Flies were reared on standard cornmeal agar medium under standard conditions (25°C, 60% humidity) in incubators that maintain a 12 h light/12h dark cycle (Percival Scientific Inc, Cat#DR36VL). The stocks used and their sources were as follows: D. melanogaster : w 1118 ;[R13F02-LexA] attP40 /CyO ; (Bloomington Stock Center, 52460), w 1118 ;[13XLexAop2-IVS-GCaMP6f-p10 }su (Hw) attP5 (Bloomington Stock Center, 44277), w 1118 ;[10xUAS-IVS-myrGFP] Su (Hw) attp5 (Bloomington Stock Center, 32199), w;;[UAS-Chrimson::mVenus] attP2 (D. Anderson, Caltech University), w 1118 ;[R54A11-p65.AD] attP40 ; [VT033006-GAL4.DBD] attP2 (Bloomington Stock Center, 88835), w;[R44A02-p65ADZp] attP40 ; [VT033006-ZpGdbd] attP2 (Janelia Research Campus, SS01165), w;[VT033006-p65ADZp] attP40 ; [R26B04-ZpGdbd] attP2 (Janelia Research Campus, SS01306), w;[R65G01-p65ADZp] attP40 ; [R82D04-ZpGdbd] attP2 (Janelia Research Campus, MB546B), w;[VT033006-p65ADZp] attP40 ; [R33H11-ZpGdbd] attP2 (Janelia Research Campus, SS01318), w;[VT033006-p65ADZp] attP40 ; [R19G08-ZpGdbd] attP2 (Janelia Research Campus, SS01265);. Connectome analysis Hemibrain analyses were conducted on the hemibrain:v1.2.1 dataset using a custom query to identify all connections with a weight value of five or greater between the uniglomerular olfactory projection neurons reported in this study and their downstream Kenyon cell synaptic partners. Flywire analyses were performed on the FAFB v783 dataset using a custom script to process the connections_princeton_no_threshold table. As in the hemibrain analysis, uniglomerular olfactory projection neurons with at least five synapses onto Kenyon cells were included. Quantifying morphological features of projection neurons Fly brains were dissected at room temperature in phosphate-buffered saline (PBS; Sigma-Aldrich, P5493) and fixed in 2% paraformaldehyde (Electron Microscopy Sciences, 15710) for 45 minutes at room temperature. Following fixation, brains were washed five times in PBST (PBS containing 0.1% Triton X-100; Sigma-Aldrich, T8787) at room temperature and blocked with 5% normal goat serum (Jackson ImmunoResearch Laboratories, AB_2336990) in PBST for 30 minutes at room temperature. For primary antibody incubation, brains were incubated overnight at 4°C in a solution containing mouse anti-nc82 (1:20; Developmental Studies Hybridoma Bank, AB_2314866) and rabbit anti-GFP (1:25; Thermo Fisher Scientific, AB_221569) diluted in 5% normal goat serum/PBST. The following day, brains were washed four times in PBST and incubated overnight at 4°C with secondary antibodies: goat anti-mouse Alexa Fluor 488 (1:500; Thermo Fisher Scientific, AB_2534069) and goat anti-rabbit Alexa Fluor 546 (1:500; Thermo Fisher Scientific, AB_2534077) in 5% normal goat serum/PBST. After secondary antibody incubation, brains were washed four times in PBST and mounted on glass slides (Fisher Scientific, 12-550-143) using VECTASHIELD mounting medium (Vector Laboratories, H-1000). Immunostained preparations were imaged using an LSM 880 confocal microscope (Zeiss). Confocal imaging All confocal images were acquired using a Zeiss LSM 880 microscope equipped with Airyscan technology. Whole-brain images were collected using a Plan-Apochromat 20×/0.8 M27 objective lens with a voxel size of 0.69 × 0.69 × 0.84 μm. Antennal lobe regions were imaged using a Plan-Apochromat 40×/1.3 Oil M27 objective lens with a voxel size of 0.21 × 0.21 × 0.84 μm. High-resolution imaging of mushroom body calyx structures was performed using a Plan-Apochromat 63×/1.4 Oil DIC M27 objective lens with a voxel size of 0.13 × 0.13 × 1.0 μm. Quantifying morphological features of projection neurons All quantitative analyses of projection neurons were performed by researchers blinded to genotype. For neuronal cell counts, confocal z-stacks of the antennal lobe were analyzed using ImageJ/FIJI software (National Institutes of Health) 25 . Individual cells were manually counted by stepping through each optical section of the z-stack and tallying distinct cell bodies using the ‘Cell Counter’ plugin. Mushroom body calyx bouton counts were obtained using the same manual counting approach on confocal z-stacks, with individual boutons tallied using the ‘Cell Counter’ plugin in ImageJ/FIJI. Total bouton volume measurements were performed using FluoRender software (University of Utah Scientific Computing and Imaging Institute, version 2.30.0) 28 . Boutons were selected and segmented using the ‘Paint Brush’ function with ‘Edge Detect’ enabled and the ‘Edge STR’ parameter set to 0.512. Threshold values were adjusted individually for each sample to account for variations in background fluorescence intensity. Functional imaging using odor stimulation All functional imaging experiments were performed on two-to-five-day-old female flies. Flies were anesthetized on ice for about ten minutes and mounted on a custom platform using tape (Shurtape Technologies, DUC280068). The platform contained an opening connected to a saline reservoir. To minimize movement artifacts, contact points between the head and tape, as well as the legs, were secured using UV-activated resin (Bondic, SK8024). A small window overlying the mushroom body was created in the cuticle, and the exposed brain was continuously immersed in saline containing: 108 mM NaCl, 5 mM KCl, 5 mM HEPES, 5 mM trehalose, 10 mM sucrose, 1 mM NaH₂PO₄, 4 mM NaHCO₃, 2 mM CaCl₂, 4 mM MgCl₂, and 1.7 mM NaOH (pH 7.3). Odor stimuli — farnesol (Sigma-Aldrich, 43348), (+)-valencene (Sigma-Aldrich, 75056), geosmin (Sigma-Aldrich, UC18), pyrrolidine (Sigma-Aldrich, 69581), 3-octanol (Sigma-Aldrich, 218405), and 4-methylcyclohexanol (Sigma-Aldrich, 153095) — were each diluted to 10 −3 in mineral oil (Sigma-Aldrich, M5904). Filtered carrier air (1.6 l/min) was mixed with odors (0.4 l/min) in a flow-controlled manner using a stimulus controller (Ockenfels Syntech, CS-55). Calcium imaging was performed using an Investigator two-photon laser scanning microscope (Bruker Corporation, RRID:SCR_019807) equipped with a Chameleon Ti:Sapphire laser (Coherent) tuned to 925 nm. Laser power was modulated by Pockels cells (Conoptics, 350-80LA/BK-02) and maintained between 14-24 mW across experiments. Emitted fluorescence was detected using a GaAsP photomultiplier tube (Hamamatsu Photonics). Images were acquired using a Bruker piezo device, imaging across seven focal planes spanning the mushroom body, with planes separated by 6 μm each. Odors were presented in pseudo-random order using the following protocol: 5 seconds baseline, 3 seconds stimulus presentation, 12 seconds recovery. This sequence was repeated four times per odor with 1-minute inter-trial intervals and 2-minute inter-odor intervals. All six odors were presented to each preparation. Images were collected at 496 × 342-pixel resolution (pixel size: 0.19 × 0.19 μm) using a resonant galvanometer scanner at an effective frame rate of 5 fps per plane. Functional imaging using optogenetic stimulation Optogenetic stimulation experiments followed the same preparation protocol as odor stimulation experiments, with flies aged two to five days old mounted and imaged using identical microscopy parameters. Prior to experimentation, flies were maintained on fresh food supplemented with 10 μM all-trans-retinal (Sigma-Aldrich, R2500) and protected from light for approximately 24 hours. Optogenetic activation was delivered using 637 nm light pulses (Coherent, 1196625) configured as five repetitions of five 40 μm spirals targeting the mushroom body calyx over a 100 milliseconds duration. Stimulation protocols consisted of 2.5 seconds baseline recording, followed by optogenetic activation, then 7.5 seconds post-stimulus imaging. This sequence was performed twice at each of the seven planes spanning the mushroom body (6 μm separation), with 1-minute intervals between trials and 1-minute intervals between planes. Images were collected at 496 × 342-pixel resolution (pixel size: 0.19 × 0.19 μm) using a resonant galvanometer scanner at 45.2 fps. Functional imaging processing Image sequences were motion-corrected using CaImAn-MATLAB 27 , with parameters adjusted individually based on movement severity. Motion-corrected sequences were maximum intensity projected using ImageJ/FIJI and automatically segmented using a custom-trained cyto3 model in Cellpose 26 . Cellpose parameters were: diameter = 18.44, flow threshold = 0, cell probability threshold = -5, and minimum size = 100 pixels. All automated segmentations were manually reviewed, with regions of interest added or removed in cases of obvious segmentation errors. Preparations with poor segmentation performance due to unresolvable motion artifacts were excluded from analysis. Fluorescence changes were quantified as ΔF/F using a custom Python script, where: ΔF/F = (F(t) - F₀) / F₀; F(t) represents raw fluorescence at time point t, and F₀ represents the mean baseline fluorescence calculated from pre-stimulus frames. ΔF/F values were calculated for each trial and concatenated to generate final traces for downstream analysis. Appetitive and aversive learning paradigm Appetitive and aversive learning paradigms were conducted following previously published protocols 13 . In short, groups of 60 to 100 flies were collected several hours before experimentation. For aversive learning, flies were maintained on standard food throughout the experiment. For appetitive learning, flies were transferred to vials containing 1% agar 16 to 20 hours before training and kept on agar for the duration of the paradigm. All training was performed in a T-maze apparatus (CelExplorer Labs, TMK-501) with a constant airflow maintained of 0.7 L/min regulated by a flowmeter (Dwyer Instruments, 116011-01). For aversive conditioning, flies were first exposed to a conditioned stimulus (CS+) paired with electric shocks (12 pulses, 90 V, 0.2 Hz) delivered by a stimulator (Grass Instruments, S48) for one minute, followed by a 45-second rest in ambient air. They were then exposed to the second odor (CS−) for one minute without shock, followed by another 45-second rest period. In the short-term paradigm, flies received a single training regimen of electric shocks, whereas in the long-term paradigm, they underwent six training regimens separated by fifteen-minute rest intervals. For appetitive conditioning, flies were first exposed to the CS− in a tube containing dry paper for two minutes, followed by a 30-second rest in ambient air. They were then exposed to the CS+ in a tube lined with paper coated in dried sucrose, followed by another 30-second rest period. Conditioned stimuli consisted of odors dissolved in mineral oil or mineral oil alone (Sigma-Aldrich, M5904). The odors were farnesol (1:1000; Sigma-Aldrich, 43348), 3-octanol (1:1000; Sigma-Aldrich, 218405), 4-methylcyclohexanol (1:1000; Sigma-Aldrich, 153095), (+)-valencene (1:1000; Sigma-Aldrich, 75076), pyrrolidine (1:1000; Sigma-Aldrich, 69581), and (±)-geosmin (1:1000; Sigma-Aldrich, FR162423). Following training, flies were transferred to vials containing standard food (aversive paradigm) or 1% agar (appetitive paradigm) and held for either 30 minutes (short-term memory) or 24 hours (long-term memory). During testing, flies were given a choice between T-maze arms containing the CS+ or CS−. A performance index (PI) was calculated as: PI = (CS+ - CS-)/(CS+ + CS-), where CS+ and CS− represent the number of flies in each arm. Each data point represents the average of reciprocal experiments in which CS+ and CS− assignments were switched. Innate odor preference was measured by allowing flies to choose between tubes containing odor or mineral oil for two minutes. Shock avoidance was assessed by allowing flies to choose between chambers with electrified copper grids (90 V pulses every five seconds for one minute) and non-electrified grids. Sugar preference was measured by offering flies a choice between chambers lined with dry paper or paper coated with dried sucrose. Statistical Analysis A glomerulus was classified as under- or overrepresented in connectivity if it had a p-value less than 0.05 in a two-sided binomial test relative to the dataset’s average connectivity rate. Kenyon cells were considered “responders” if the post-stimulation mean ΔF/F exceeded twice the standard deviation of ΔF/F measured during the pre-stimulation period. Weak and strong responders were classified using the KMeans clustering function (n = 2) from the scikit-learn package, applied to the maximum ΔF/F value in the post-stimulation period of the trial-averaged trace 30. All other statistical analyses were performed using GraphPad Prism version 10.4.0 (GraphPad Software, San Diego, CA). Pearson correlation coefficients were computed using simple linear regression. Behavioral tests reporting performance index were analyzed using a one-sample t-test to assess significant differences from zero, with significance defined as p < 0.05. Significance levels are reported as follows: p < 0.05 (*), p < 0.01 (**), p < 0.001 (***), p < 0.0001 (****), and ns (non-significant). Supplemental information View this table: View inline View popup Download powerpoint Table S1. Connectivity frequencies across datasets Related to Figure 1 . The table reports connectivity frequencies for each glomerulus across electron microscopy connectomes (Scheffer et al., 2020; Dorkenwald et al., 2024) and large-scale tracing datasets (Caron et al., 2013; Hayashi et al., 2022; Ellis et al., 2024). Values are reported as means with standard error of the mean. View this table: View inline View popup Download powerpoint Table S2. Kenyon cell response frequencies for optogenetic experiments Related to Figure 3 . The table shows, for each split-GAL4 line, the total number of Kenyon cells recorded per sample, along with the percentages of non-responders, weak responders, and strong responders for individual samples. The first row provides summary statistics for the entire dataset, including the total percentage of responders across all samples. Values are reported as means with standard error of the mean. View this table: View inline View popup Download powerpoint Table S3. Kenyon cell response frequencies for odor experiments Related to Figure 4 . The table shows, for each odor, the total number of Kenyon cells recorded per sample, along with the percentages of non-responders, weak responders, and strong responders for individual samples. The first row provides summary statistics for the entire dataset, including the total percentage of responders across all samples. Values are reported as means with standard error of the mean. Download figure Open in new tab Figure S1. Morphological features correlate with connectivity frequency in large-scale tracing datasets. Related to Figure 2 . (A–C) Quantification of morphological features versus average connectivity frequency from large-scale tracing datasets. (A) Number of projection neurons per split-GAL4 line . (B) Total bouton number in calyx. (C) Total bouton volume. Each data point represents a split-GAL4 line; error bars show standard error of the mean; linear regression lines are shown with R² and p values. Symbols indicate projection neuron identity: circle (VL1), square (DL3), downward triangle (VM4), upward triangle (DA1), hexagon (DA4l+DC3), diamond (VM5d+VM7d+VM7v). Colors indicate degree of representation: pink (overrepresented) and green (underrepresented). Download figure Open in new tab Figure S2. Classification of Kenyon cell responses to optogenetic stimulation Related to Figure 3 . (A) Representative fluorescence image of Kenyon cells expressing GCaMP6f during optogenetic stimulation with split-GAL4 line targeting the expression of UAS-Chrimson in VL1 projection neurons. (A’) Heat map of ΔF responses following optogenetic stimulation. (A’’) Segmented image showing individual Kenyon cell boundaries. (A’’’) Classification map showing non-responders (white), weak responders (yellow), and strong responders (red). (B–D) Representative calcium traces (ΔF/F) from a non-responder (B), weak responder (C), and strong responder (D). Black lines indicate two stimuli presentations per sample. Download figure Open in new tab Figure S3. Distribution of response amplitudes across recorded Kenyon cells Related to Figures 3 and 4 (A–F) Histograms showing the natural logarithm of the maximum ΔF/F amplitude distributions among all Kenyon cell responders following optogenetic stimulation of each split-GAL4 line: VL1 (A), DL3 (B), VM4 (C), DA1 (D), DA4l+DC3 (E), VM5d+VM7d+VM7v (F). (G–L) Histograms showing the natural logarithm of the maximum ΔF/F amplitude distributions among all Kenyon cell responders following odor presentation: farnesol (G), valencene (H), geosmin (I), 4-methylcyclohexanol (J), pyrrolidine (K), 3-octanol (L). Download figure Open in new tab Figure S4. Optogenetic responses versus connectivity frequency from large-scale tracing datasets Related to Figure 3 (A–H) Quantification of Kenyon cell responses plotted against connectivity frequency from large-scale tracing datasets. Panels A–D include all data; panels E–H show the same analyses with VL1 excluded. (A,E) Percentage of all responders. (B,F) Percentage of weak responders. (C,G) Percentage of strong responders. (D,H) Maximum peak ΔF/F amplitude. Each data point represents a split-GAL4 line; error bars show standard error of the mean; linear regression lines are shown with R² and p values. Symbols indicate projection neuron identity: circle (VL1), square (DL3), downward triangle (VM4), upward triangle (DA1), hexagon (DA4l+DC3), diamond (VM5d+VM7d+VM7v). Colors indicate degree of representation: pink (overrepresented) and green (underrepresented). Download figure Open in new tab Figure S5. VL1 optogenetic responses across all samples Related to Figure 3 Kenyon cell responses to VL1 optogenetic stimulation across ten samples. Each imaging plane are arranged from dorsal (left) to ventral (right), with a total of 5,015 cells recorded. Cells are classified as non-responders (white), weak responders (yellow), or strong responders (red) according to the criteria shown in Figure S2 . Download figure Open in new tab Figure S6. DL3 optogenetic responses across all samples Related to Figure 3 Kenyon cell responses to DL3 optogenetic stimulation across ten samples. Each imaging plane are arranged from dorsal (left) to ventral (right), with a total of 5,059 cells recorded. Cells are classified as non-responders (white), weak responders (yellow), or strong responders (red) according to the criteria shown in Figure S2 . Download figure Open in new tab Figure S7. VM4 optogenetic responses across all samples Related to Figure 3 Kenyon cell responses to VM4 optogenetic stimulation across eight samples. Each imaging plane are arranged from dorsal (left) to ventral (right), with a total of 3,590 cells recorded. Cells are classified as non-responders (white), weak responders (yellow), or strong responders (red) according to the criteria shown in Figure S2 . Download figure Open in new tab Figure S8. DA1 optogenetic responses across all samples Related to Figure 3 Kenyon cell responses to DA1 optogenetic stimulation across nine samples. Each imaging plane are arranged from dorsal (left) to ventral (right), with a total of 4,432 cells recorded. Cells are classified as non-responders (white), weak responders (yellow), or strong responders (red) according to the criteria shown in Figure S2 . Download figure Open in new tab Figure S9. DA4l+DC3 optogenetic responses across all samples Related to Figure 3 Kenyon cell responses to DA4l+DC3 optogenetic stimulation across ten samples. Each imaging plane are arranged from dorsal (left) to ventral (right), with a total of 4,199 cells recorded. Cells are classified as non-responders (white), weak responders (yellow), or strong responders (red) according to the criteria shown in Figure S2 . Download figure Open in new tab Figure S10. VM5d+VM7d+VM7v optogenetic responses across all samples Related to Figure 3 Kenyon cell responses to VM5d+VM7d+VM7v optogenetic stimulation across twelve samples. Each imaging plane are arranged from dorsal (left) to ventral (right), with a total of 4,617 cells recorded. Cells are classified as non-responders (white), weak responders (yellow), or strong responders (red) according to the criteria shown in Figure S2 . Download figure Open in new tab Figure S11. Response proportions for optogenetic experiments Related to Figure 3 Pie charts showing proportions of non-responders (gray), weak responders (yellow), and strong responders (red) for each split-GAL4 line across individual samples or average all sample in a data set. Download figure Open in new tab Figure S12. VL1 convergence analysis Related to Figure 3 (A,B) Connectome reconstructions of the only two Kenyon cells receiving double VL1 inputs identified in the FlyWire connectome (Dorkenwald et al., 2024). (A) One Kenyon cell (black) receives inputs from two distinct VL1 projection neurons (light and dark green; red arrows). (B) Another Kenyon cell (black) receives two inputs from the same VL1 projection neuron (dark green). Orientation axes: P, posterior; V, ventral; L, lateral. (C) Probability of observing Kenyon cells with double inputs from the same glomerulus under a purely stochastic connectivity model, shown across different connectivity rates. The red marker indicates the connectivity frequency observed for VL1 in the connectome dataset (0.42%), which corresponds to one Kenyon cell with double VL1 inputs per mushroom body. Download figure Open in new tab Figure S13. Synaptic strength across projection neuron types Related to Figure 3 Number of synapses formed between Kenyon cells and projection neurons for each projection neuron type across all glomeruli identified in the FlyWire connectome (Dorkenwald et al., 2024). VL1 projection neurons are marked in red. Data points represent means; error bars show standard error of the mean. Download figure Open in new tab Figure S14. Classification of Kenyon cell responses to odor stimulation Related to Figure 4 (A) Representative fluorescence image of Kenyon cells expressing GCaMP6f. (A’) Heat map of ΔF responses following odor presentation. (A’’) Segmented image showing individual Kenyon cell boundaries. (A’’’) Classification map showing non-responders (white), weak responders (yellow), and strong responders (red). (B–D) Representative calcium traces (ΔF/F) from non-responder (B), weak responder (C), and strong responder (D). Gray shaded boxes indicate three odor stimuli presentations per sample. Download figure Open in new tab Figure S15. Odor responses versus connectivity frequency from large-scale tracing datasets Related to Figure 4 (A–H) Quantification of Kenyon cell responses plotted against cumulative connectivity frequency from large-scale tracing datasets. Panels A–D include all odors; panels E–H show the same analyses with pyrrolidine excluded. (A,E) Percentage of all responders. (B,F) Percentage of weak responders. (C,G) Percentage of strong responders. (D,H) Mean maximum peak ΔF/F amplitude. Each data point represents an odor; error bars show standard error of the mean; linear regression lines are shown with R² and p values. Symbols indicate odor identity: square (farnesol), downward triangle (valencene), lozenge (geosmin), hexagon (4-methylcyclohexanol), upward triangle (pyrrolidine), circle (3-octanol). Download figure Open in new tab Download figure Open in new tab Figure S16. Odor-evoked responses across representative samples Related to Figure 4 Kenyon cell responses to each odor across three representative samples. Each sample shows imaging planes from dorsal (left) to ventral (right). Cell counts: farnesol (617, 701 and 687 cells), valencene (602, 743 and 741 cells cells), geosmin (617, 727 and 777 cells), 4-methylcyclohexanol (617, 741 and 748 cells616), pyrrolidine (623, 685 and 723 cells), 3-octanol (597, 728 and 728 cells). Non-responders (white), weak responders (yellow), and strong responders (red) classified as in Figure S14 . Download figure Open in new tab Figure S17. Response proportions for odor experiments Related to Figure 4 Pie charts showing proportions of non-responders (gray), weak responders (yellow), and strong responders (red) for each odor across individual samples. Download figure Open in new tab Figure S18. Learning performance versus connectivity frequency from large-scale tracing datasets Related to Figure 5 (A–H) Correlation between learning performance and cumulative connectivity frequency from large-scale tracing datasets. Panels A–D include all odors; panels E–H show the same analyses with pyrrolidine excluded. Each data point represents an odor; error bars show standard error of the mean; linear regression lines are shown with R² and p values. Symbols indicate odor identity: square (farnesol), downward triangle (valencene), lozenge (geosmin), hexagon (4-methylcyclohexanol), upward triangle (pyrrolidine), circle (3-octanol). ACKNOWLEDGMENTS We thank members of the Caron laboratory for their comments on the manuscript and discussions throughout the project. We are grateful to Dr. Florian Maderspacher for thoughtful discussions, careful feedback, and constant support. We thank Dr. Yoshi Aso for sharing split-GAL4 lines prior to their publication; Adam Lin and Sylvia Yang for preparing the standard cornmeal agar medium; and Miles Jacob for assistance with general laboratory operations. We also thank the Cell Imaging Core at the University of Utah for use of the Zeiss LSM 880 microscope. We acknowledge the use of the following software and databases: Fiji 25 , Cellpose 26 , the DoOR 2.0 odor database 18 , CaImAn 27 , FluoRender 28 , Matplotlib 29 , and Navis 17 . We thank the Bloomington Drosophila Stock Center (NIH P40OD018537) for providing all the Drosophila stocks. This work has been funded by grants from the National Institute for Neurological Disorders and Stroke (R01 NS 106018 and R01 NS 1079790) and the National Science Foundation (IOS 2042397). Further financial support was provided by a Developmental Biology Training Grant (T32HD007491, J.M.U.), a Genetic Training Grant (T32GM141848, A.R.B.), the Arnold and Mabel Beckman Foundation Beckman Scholars Program (P.A.), the University Research Opportunities Program (D.B.A.) and the Georges S. and Dolores Eccles Foundation (S.J.C.C.). Funder Information Declared National Institute of Neurological Disorders and Stroke, https://ror.org/01s5ya894 , R01 NS 106018 and R01 NS 1079790 U.S. National Science Foundation, https://ror.org/021nxhr62 , IOS 2042397 Footnotes ↵ # Lead contact REFERENCES 1. ↵ Olshausen , B.A. , and Field , D.J . ( 2004 ). Sparse coding of sensory inputs . Curr Opin Neurobiol 14 , 481 – 487 . doi: 10.1016/j.conb.2004.07.007 . OpenUrl CrossRef PubMed Web of Science 2. ↵ Chklovskii , D.B . ( 2004 ). Exact solution for the optimal neuronal layout problem . Neural Comput 16 , 2067 – 2078 . doi: 10.1162/0899766041732422 . OpenUrl CrossRef PubMed Web of Science 3. ↵ Babadi , B. , and Sompolinsky , H . ( 2014 ). Sparseness and Expansion in Sensory Representations . Neuron 83 , 1213 – 1226 . doi: 10.1016/j.neuron.2014.07.035 . OpenUrl CrossRef PubMed 4. ↵ Cayco-Gajic , N.A. , and Silver , R.A . ( 2019 ). Re-evaluating Circuit Mechanisms Underlying Pattern Separation . Neuron 101 , 584 – 602 . doi: 10.1016/j.neuron.2019.01.044 . OpenUrl CrossRef PubMed 5. ↵ Dean , P. , Porrill , J. , Ekerot , C.F. , and Jörntell , H . ( 2010 ). The cerebellar microcircuit as an adaptive filter: experimental and computational evidence . Nat Rev Neurosci 11 , 30 – 43 . doi: 10.1038/NRN2756 . OpenUrl CrossRef PubMed Web of Science 6. ↵ Kesner , R.P. , and Rolls , E.T . ( 2015 ). A computational theory of hippocampal function, and tests of the theory: New developments . Neurosci Biobehav Rev 48 , 92 – 147 . doi: 10.1016/j.neubiorev.2014.11.009 . OpenUrl CrossRef PubMed 7. ↵ Cognigni , P. , Felsenberg , J. , and Waddell , S . ( 2018 ). Do the right thing: neural network mechanisms of memory formation, expression and update in Drosophila . Curr Opin Neurobiol 49 , 51 – 58 . doi: 10.1016/j.conb.2017.12.002 . OpenUrl CrossRef PubMed 8. ↵ Litwin-Kumar , A. , Harris , K.D. , Axel , R. , Sompolinsky , H. , and Abbott , L.F . ( 2017 ). Optimal Degrees of Synaptic Connectivity . Neuron 93 , 1153 – 1164 .e7. doi: 10.1016/J.NEURON.2017.01.030 . OpenUrl CrossRef PubMed 9. ↵ Aso , Y. , Hattori , D. , Yu , Y. , Johnston , R.M. , Iyer , N.A. , Ngo , T.T.B. , Dionne , H. , Abbott , L.F. , Axel , R. , Tanimoto , H. , et al. ( 2014 ). The neuronal architecture of the mushroom body provides a logic for associative learning . Elife 3 , e04577 . doi: 10.7554/ELIFE.04577 . OpenUrl CrossRef PubMed 10. ↵ Li , F. , Lindsey , J. , Marin , E.C. , Otto , N. , Dreher , M. , Dempsey , G. , Stark , I. , Bates , A.S. , Pleijzier , M.W. , Schlegel , P. , et al. ( 2020 ). The connectome of the adult drosophila mushroom body provides insights into function . Elife 9 , 1 – 217 . doi: 10.7554/eLife.62576 . OpenUrl CrossRef PubMed 11. ↵ Caron , S.J.C. , Ruta , V. , Abbott , L.F. , and Axel , R . ( 2013 ). Random convergence of olfactory inputs in the Drosophila mushroom body . Nature 497 , 113 – 117 . doi: 10.1038/NATURE12063 . OpenUrl CrossRef PubMed Web of Science 12. Hayashi , T.T. , MacKenzie , A.J. , Ganguly , I. , Ellis , K.E. , Smihula , H.M. , Jacob , M.S. , Litwin-Kumar , A. , and Caron , S.J.C . ( 2022 ). Mushroom body input connections form independently of sensory activity in Drosophila melanogaster . Curr Biol 32 , 4000 – 4012 .e5. doi: 10.1016/J.CUB.2022.07.055 . OpenUrl CrossRef PubMed 13. ↵ Ellis , K.E. , Bervoets , S. , Smihula , H. , Ganguly , I. , Vigato , E. , Auer , T.O. , Benton , R. , Litwin-Kumar , A. , and Caron , S.J.C . ( 2024 ). Evolution of connectivity architecture in the Drosophila mushroom body . Nat Commun 15 . doi: 10.1038/S41467-024-48839-4 . OpenUrl CrossRef 14. ↵ Scheffer , L.K. , Xu , C.S. , Januszewski , M. , Lu , Z. , Takemura , S.Y. , Hayworth , K.J. , Huang , G.B. , Shinomiya , K. , Maitin-Shepard , J. , Berg , S. , et al. ( 2020 ). A connectome and analysis of the adult drosophila central brain . Elife 9 , 1 – 74 . doi: 10.7554/ELIFE.57443 . OpenUrl CrossRef PubMed 15. ↵ Dorkenwald , S. , Matsliah , A. , Sterling , A.R. , Schlegel , P. , Yu , S. , McKellar , C.E. , Lin , A. , Costa , M. , Eichler , K. , Yin , Y. , et al. ( 2024 ). Neuronal wiring diagram of an adult brain . Nature 634 , 124 – 138 . doi: 10.1038/S41586-024-07558-Y . OpenUrl CrossRef PubMed 16. ↵ Shuai , Y. , Sammons , M. , Sterne , G.R. , Hibbard , K.L. , Yang , H. , Yang , C.-P. , Managan , C. , Siwanowicz , I. , Lee , T. , Rubin , G.M. , et al. ( 2025 ). Driver lines for studying associative learning in Drosophila . Elife 13 . doi: 10.7554/ELIFE.94168 . OpenUrl CrossRef 17. ↵ Bates , A.S. , Manton , J.D. , Jagannathan , S.R. , Costa , M. , Schlegel , P. , Rohlfing , T. , and Jefferis , G.S.X.E . ( 2020 ). The natverse, a versatile toolbox for combining and analysing neuroanatomical data . Elife 9 , e53350 . doi: 10.7554/ELIFE.53350 . OpenUrl CrossRef PubMed 18. ↵ Münch , D. , and Galizia , C.G . ( 2016 ). DoOR 2.0--Comprehensive Mapping of Drosophila melanogaster Odorant Responses . Sci Rep 6 . doi: 10.1038/SREP21841 . OpenUrl CrossRef 19. ↵ Krashes , M.J. , and Waddell , S . ( 2011 ). Drosophila aversive olfactory conditioning . Cold Spring Harb Protoc 2011 . doi: 10.1101/PDB.PROT5608 . OpenUrl CrossRef 20. ↵ Krashes , M.J. , and Waddell , S . ( 2011 ). Drosophila aversive olfactory conditioning . Cold Spring Harb Protoc 2011 . doi: 10.1101/PDB.PROT5608 . OpenUrl CrossRef 21. ↵ Prieto-Godino , L.L. , Rytz , R. , Cruchet , S. , Bargeton , B. , Abuin , L. , Silbering , A.F. , Ruta , V. , Dal Peraro , M. , and Benton , R . ( 2017 ). Evolution of Acid-Sensing Olfactory Circuits in Drosophilids . Neuron 93 , 661 – 676 .e6. doi: 10.1016/j.neuron.2016.12.024 . OpenUrl CrossRef PubMed 22. ↵ Hartmann , T. , and Ober , D . ( 2008 ). Defense by Pyrrolizidine Alkaloids: Developed by Plants and Recruited by Insects . Induced Plant Resistance to Herbivory , 213–231. doi: 10.1007/978-1-4020-8182-8_10 . OpenUrl CrossRef 23. ↵ Knierim , J.J. , Neunuebel , J.P. , and Deshmukh , S.S . ( 2013 ). Functional correlates of the lateral and medial entorhinal cortex: objects, path integration and local-global reference frames . Philos Trans R Soc Lond B Biol Sci 369 . doi: 10.1098/RSTB.2013.0369 . OpenUrl CrossRef 24. ↵ Zavitz , D. , Amematsro , E.A. , Borisyuk , A. , and Caron , S.J.C . ( 2021 ). Connectivity patterns that shape olfactory representation in a mushroom body network model . bioRxiv , 2021.02.10.430647. doi: 10.1101/2021.02.10.430647 . OpenUrl Abstract / FREE Full Text 25. ↵ Schindelin , J. , Arganda-Carreras , I. , Frise , E. , Kaynig , V. , Longair , M. , Pietzsch , T. , Preibisch , S. , Rueden , C. , Saalfeld , S. , Schmid , B. , et al. ( 2012 ). Fiji - an Open Source platform for biological image analysis . Nat Methods 9 , doi: 10.1038/nmeth.2019 . 10.1038/NMETH.2019. OpenUrl CrossRef PubMed Web of Science 26. ↵ Stringer , C. , Wang , T. , Michaelos , M. , and Pachitariu , M . ( 2021 ). Cellpose: a generalist algorithm for cellular segmentation . Nat Methods 18 , 100 – 106 . doi: 10.1038/S41592-020-01018-X;SUBJMETA . OpenUrl CrossRef PubMed 27. ↵ Giovannucci , A. , Friedrich , J. , Gunn , P. , Kalfon , J. , Brown , B.L. , Koay , S.A. , Taxidis , J. , Najafi , F. , Gauthier , J.L. , Zhou , P. , et al. ( 2019 ). Caiman an open source tool for scalable calcium imaging data analysis . Elife 8 . doi: 10.7554/ELIFE.38173 . OpenUrl CrossRef 28. ↵ Wan , Y. , Otsuna , H. , Holman , H.A. , Bagley , B. , Ito , M. , Lewis , A.K. , Colasanto , M. , Kardon , G. , Ito , K. , and Hansen , C . ( 2017 ). FluoRender: joint freehand segmentation and visualization for many-channel fluorescence data analysis . BMC Bioinformatics 18 . doi: 10.1186/S12859-017-1694-9 . OpenUrl CrossRef 29. ↵ Hunter , J.D . ( 2007 ). Matplotlib: A 2D graphics environment . Comput Sci Eng 9 , 90 – 95 . doi: 10.1109/MCSE.2007.55 . OpenUrl CrossRef PubMed View the discussion thread. Back to top Previous Next Posted October 30, 2025. Download PDF Email Thank you for your interest in spreading the word about bioRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Connectivity biases generate a learning hierarchy in the Drosophila mushroom body Message Subject (Your Name) has forwarded a page to you from bioRxiv Message Body (Your Name) thought you would like to see this page from the bioRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Connectivity biases generate a learning hierarchy in the Drosophila mushroom body Alexander John MacKenzie , Lia Beatty , Jilian Mae Ulibarri , Dua Azhar , Prisca Amematsro , Andrew R. Butts , Sophie Jeanne Cécile Caron bioRxiv 2025.10.29.684686; doi: https://doi.org/10.1101/2025.10.29.684686 Share This Article: Copy Citation Tools Connectivity biases generate a learning hierarchy in the Drosophila mushroom body Alexander John MacKenzie , Lia Beatty , Jilian Mae Ulibarri , Dua Azhar , Prisca Amematsro , Andrew R. Butts , Sophie Jeanne Cécile Caron bioRxiv 2025.10.29.684686; doi: https://doi.org/10.1101/2025.10.29.684686 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Neuroscience Subject Areas All Articles Animal Behavior and Cognition (7618) Biochemistry (17633) Bioengineering (13856) Bioinformatics (41841) Biophysics (21399) Cancer Biology (18529) Cell Biology (25422) Clinical Trials (138) Developmental Biology (13352) Ecology (19860) Epidemiology (2067) Evolutionary Biology (24282) Genetics (15582) Genomics (22462) Immunology (17700) Microbiology (40295) Molecular Biology (17140) Neuroscience (88419) Paleontology (666) Pathology (2823) Pharmacology and Toxicology (4813) Physiology (7632) Plant Biology (15107) Scientific Communication and Education (2042) Synthetic Biology (4284) Systems Biology (9808) Zoology (2267)