Mice discriminate odour source distance via sub-sniff temporal features of odour plumes

13 Rodents rely on olfaction to navigate complex environments, particularly where visual cues 14 are limited. Yet how they estimate the dist ance to an odour source remains unclear. The 15 spatiotemporal dynamics of natural odour pl umes, shaped by airflow turbulence, offer 16 valuable cues for locating odour sources. Here , we show that mice can discriminate odour 17 sources placed at different distances by extrac ting information from the sub-sniff temporal 18 structure of naturalistic odour plumes. Using a wind tunnel and an olfactory virtual reality 19 system, we generated dynamic plumes and demonstrated, through high-throughput 20 automated behaviour, that mice distingu ish near from far sources based on odour 21 fluctuations operating faster than their resp iratory cycle. Two-photon calcium imaging of 22 olfactory bulb projection neurons revealed that distance-dependent responses are present in 23 a small subset of mitral and tufted cells, and that population activity encodes source 24 distance. Critically, neural responses corr elated more strongly with high-frequency plume 25 features than with mean odour concentration. Our results identify a neural basis for distance 26 estimation from odour dynamics and highlight the importance of rapid temporal processing in 27 mammalian olfaction. 28

29 Olfaction provides animals with the remarkable ability to remotely sample their environment 30 under low-visibility conditions, a vital skill for crepuscular and nocturnal species, such as 31 rodents. Rather than relying solely on direct contact, as taste does, mice detect airborne 32 odours carried by complex, often turbulent airflows, enabling them to gather information 33 about distant objects, such as potential food sources, predators, or mating opportunities 34 (Marin et al., 2021; McKissick et al., 2024; Sunil et al., 2024). From an ecological standpoint, 35 the ability to estimate how far away an odour s ource is can be crucial for a rodent’s survival. 36 For instance, detecting the scent of a predator at close range prompts immediate evasive 37 action, whereas sensing it from afar might allow continued foraging. In the same way, 38 evaluating whether a food source is realisticall y within reach helps the animal to conserve 39 energy and minimise exposure (Findley et al., 2021; Gire et al., 2016). 40 In natural environments, odour signals rarely spread as uniform gradients. Instead, they form 41 spatiotemporally complex, dynamic plumes shaped by turbulent airflow (Crimaldi et al., 42 2022). Within these plumes, pockets and filaments of higher concentration are interspersed 43 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 2 with intervals of relatively odour-free air (Celani et al., 2014; Crimaldi & Koseff, 2001; Moore 44 & Atema, 1991), producing odour events on timesca les of tens to hundreds of milliseconds, 45 often exceeding typical sniffing frequencies of 4–12 Hz in rodents (Wachowiak, 2011; 46 Welker, 1964). As a plume travels farther from its source, it expands spatially while changing 47 its temporal statistics, including odour event parameters such as peak height, onset slope, 48 and intermittency (Reddy et al., 2022; Schmuker et al., 2016; Vergassola et al., 2007). 49 These features therefore carry crucial info rmation about the distance to an odour source, 50 offering cues to guide navigation and decisi on-making (Balkovsky & Shraiman, 2002; Mafra-51 Neto & Cardé, 1994; Rigolli et al., 2022). One key open question is whether and how rodents 52 exploit these high-frequency fluctuations (Ackels et al., 2021; Dasgupta et al., 2022; Marin et 53 al., 2021; Sunil et al., 2024; Tootoonian et al., 2025). If rodents can capitalise on this 54 temporal information, it would not only enable them to pinpoint and distinguish different 55 sources from the atmospheric background (Rokni et al., 2014), but also to construct a spatial 56 map of their environment using purely ol factory cues (Jacobs, 2012; Poo et al., 2022; 57 Sterrett et al., 2024). Critically, such a strategy would let them rapidly locate and identify 58 odours without needing to physically move between sources to confirm their identity or 59 position (Gire et al., 2016). 60 Behavioural evidence indicates that rodents mi ght indeed exploit these transient cues. In 61 certain odour localisation tasks, mice commit to a directional choice shortly after detecting 62 the odour, well before they have traversed the ent ire distance to the source (Findley et al., 63 2021). This early commitment suggests that relev ant spatial information is available in the 64 odour signal itself, even at a distance. Moreov er, when mice are trained in multi-choice 65 arenas where multiple strategies such as memory-based foraging are possible, they 66 sometimes switch between systematic searching and potentially more nuanced reliance on 67 the spatiotemporal dynamics of the odour stimulus (Findley et al., 2021; Gire et al., 2016; 68 Tariq et al., 2024). The fact that they do so under conditions of high plume complexity hints 69 that rodents can integrate rapid temporal changes of the stimulus that might be missed by 70 simpler gradient-following strategies (Gire et al., 2016; Jones & Urban, 2018; Khan et al., 71 2012). 72 Growing evidence suggests that each sniff captures not just a static snapshot of odour 73 composition but can also convey intricate temporal details. In rodents, behavioural 74 experiments with optogenetic stimulation have shown that mice can discriminate between 75 inputs delivered as little as 10-20 ms apart (Li et al., 2014; Rebello et al., 2014; Smear et al., 76 2011, 2013), and asynchronous stimulation across the two olfactory bulbs can elicit different 77 responses (Kuruppath et al., 2021). Systemat ic investigations of their temporal 78 discrimination abilities have revealed that mi ce can discriminate odour correlation structures 79 at frequencies up to 40 Hz, surpassing their typical sniffing rate (Ackels et al., 2021; 80 Dasgupta et al., 2022; Warner et al., 2024). A head-restrained odour-counting task reveals 81 that mice integrate sporadic, plume-like odour pulses across dozens of sniffs, giving greater 82 w eight to inhalation-phase stimuli which might ultimately constrain behavioural accuracy 83 (Boero et al., 2025). Psychophysical work in humans shows that observers can distinguish 84 the temporal order of odour pulses spaced by as little as 120 ms within a single sniff, 85 indicating that fine-grained temporal coding extends across species (Wu et al., 2024). This 86 capacity challenges earlier assumptions that the mammalian nasal cavity and slow 87 transduction kinetics of olfactory receptor neurons (ORNs) would “wash out” high-frequency 88 features of incoming odour plumes (Duchamp-Viret et al., 1999; Kepecs et al., 2006). 89 Instead, recent physiological and modelling work indicates that the convergence of large 90 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 3 populations of ORNs in the olfactory bulb (OB), allows access to fast-changing odour signals 91 (Ackels et al., 2021). Correspondingly, OB projection neurons, mitral and tufted cells 92 (MTCs), can encode odour correlation structure at frequencies up to 20 Hz (Ackels et al., 93 2021; Dasgupta et al., 2022). Further it has been shown that odour events such as onset, 94 offset, whiffs, and blanks are tightly coupled to MTC population activity, with different MTC 95 clusters exhibiting varying degrees of correla tion (Lewis et al., 2021, 2024) and that mice 96 can discriminate odour stimuli of low and high in termittency values (G umaste et al., 2024). 97 Such high-frequency processing confers a potential advantage for resolving spatial 98 information from turbulent odour plumes. It could, for instance, enable rodents to perform 99 source separation, correctly assigning odour mo lecules to their respective origins, and help 100 tackle the “olfactory cocktail party problem” (Hopfield, 1991; Rokni et al., 2014; Tootoonian 101 et al., 2025). An olfactory system with access to rapid fluctuations allows mammals to extract 102 distance and direction cues without physically sampling multiple locations, thereby improving 103 efficiency in odour-driven spatial tasks. 104 Here, we tested whether mice can indeed extract distance-related information from the 105 spatiotemporal structure of odour plumes. By combining wind tunnel recordings in an 106 automated behavioural setup (Erskine et al., 2019) and an “olfactory virtual reality” system 107 (Ackels et al., 2021), we systematically ex plored how temporal features at various 108 frequencies contribute to distance discrimination. Our results show that odour fluctuations in 109 the sub-sniff frequency range are especially informative about space, providing more salient 110 distance cues than slower timescale averages such as mean concentration. Mice learned to 111 distinguish odour sources placed at differ ent distances purely based on these high-112 frequency cues, underscoring the temporal bandwidth of rodent olfaction. Furthermore, two-113 photon Ca 2+ imaging in the OB identified a subpopulation of MTCs that exhibit distance-114 dependent responses, suggesting a neural correlate for this spatial sensitivity. Overall, our 115 work demonstrates that rodents can make use of rapid temporal signatures of odour plumes 116 for spatial perception. 117

118 Mice can discriminate between odours originating from near and far sources 119 Natural odour plumes are shaped by airflow turbulence, resulting in high-frequency odour 120 intensity fluctuations that may contain info rmation about odour source location (Gumaste et 121 al., 2020, p. 20; Hopfield, 1991; Liu et al., 2020). Recent work has shown that mice can 122 access high frequency components of odour stimuli (Ackels et al., 2021; Dasgupta et al., 123 2022; Warner et al., 2024), and suggested this could be used by mice to gather spatial 124 information using olfaction (Ackels et al., 2021; Bhattacharyya & Singh Bhalla, 2015; Findley 125 et al., 2021; Liu et al., 2020). In this work, we investigated whether mice can extract and use 126 this information to discern odour source relative distance, and how this information is 127 represented in the OB. 128 To this end, we first created a naturalistic but controlled environment that was robust enough 129 to allow for long-term, high-throughput animal training, but complex enough to mimic 130 naturally-occurring plumes ( Fig. 1A-C ). Odour stimuli were generated with custom-built 131 distance Odour Delivery Devices (dODDs, Methods, Fig. 1B) placed inside a wind tunnel, 132 using a passive-release mechanism (liquid odour picked up by passing airflow when the 133 dODD is opened, Fig. S1.1 , S1.2 A,B ). This approach allowed for odour plumes to be 134 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 4 generated anew for every trial, thus creating highly complex temporal structures ( Fig. 1C) 135 under reproducible conditions that neverthel ess mimic the high variability occurring under 136 natural conditions. 137 138 139 Figure 1: Mice can discriminate between near and far odour sources. (A) Complex airflow is 140 generated inside a wind tunnel using a fan, a honeycomb structure and an obstacle. Odour sources 141 are placed downstream of the fan and upstream of the detector (PID), forming the distance Odour 142 Delivery Device (dODD). The dODD is formed of 8 odour boxes, 4 at each distance (near or far) from 143 the PID, arranged symmetrically across the midline of the tunnel (Left and Right positions). Mice (n = 144 24) were housed in an automated operant conditioning system (AutonoMouse), where the only water 145 source was the reward from the GNG task. (B) Odour presentation boxes consisting of a metal casing 146 surrounding a glass dish. An Arduino controlled servo motor opens the lid to allow airflow to pick up 147 odour molecules. The lid opens horizontally to the table for minimum airflow disturbance. (C) Complex 148 airflow in the wind tunnel generated by a honeycomb structure and a cylindrical obstacle placed 25 149 cm downstream of the fan. Maximum intensity projection through a 10 s video of neutrally buoyant 150 soap bubbles in the wind tunnel, filmed 30 cm downstream from the fan, immediately behind the 151 obstacle. Bubbles appear elongated due to the exposure time of the camera. See also Fig. S1.1 (D) 152 Schematic of task structure: Depicted is the reward structure for half of the animals (n = 13), where 153 plumes originating from the near source (30 cm) were associated with a reward (S+ trials) and the 154 plumes originating from the far source (80 cm) were not associated with a reward (S- trials). A lick 155 response (top) would allow for early termination of the trial if the total time licked (shaded in blue on 156 the lick trace) was >200 ms in a 2 s rolling window. The absence of a lick response (bottom) was 157 defined as total time licked <200 ms in a 2 s rolling window, and the trial would continue until its 158 maximum length was reached at 3s. These two categories of responses to the two types of trials (S+ 159 or S-) produced the outcomes listed on the right-hand column (Hit, Miss, False Alarm, Correct 160 Rejection). (E) Histograms of total time licked for an example mouse, as calculated by the 161 AutonoMouse software to decide upon delivering a water reward, using time above threshold (dotted 162 line) in the lick traces as shown in D (S+ in blue, S- in orange). (F) Discrimination index defined as the 163 ratio of total time licked (S-/S+) for each mouse below the threshold (t = 200 ms), compared to the 164 mean of shuffled values for that mouse. The measured value is significantly higher than shuffle 165 (measured>shuffled in 99% of shuffles) in 19/24 mice (filled dots). (G) Hit rate (fraction of S+ trials 166 licked) plotted against False alarm rate (fraction of S- trials licked) averaged across a period of 10 167 blocks of 100 trials each during a stable period of the experiment. Each dot is an individual mouse. A 168 perfect block would be placed in the upper left corner, while a block with continuous licking would be 169 placed in the upper right corner. A block where the mouse was not engaged in the task would be 170 placed in the lower left corner. Indiscriminate licking while engaged in the task would be placed along 171 the diagonal. 172 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 5 A home cage-based automated behavioural system (AutonoMouse (Erskine et al., 2019); 173 Methods) was connected to the downwind end of the wind tunnel, allowing the animals to 174 sample odour plumes created inside the wind tunnel ( Fig. 1A, Fig. S1.3). This allowed us to 175 simultaneously train a cohort of group-housed mice (n = 24) for a long period of time (up to 176 14 weeks), resulting in a large number of tr ials (4053.9 ± 660 trials/animal). Using this 177 system, mice were trained to discriminate ca tegories of odour plumes originating from 178 different distances using an olfactory Go/No-Go (GNG) task paradigm ( Fig. 1A-D ). By 179 harnessing the high number of trials that can be generated and presented to the mice, we 180 were not limited to predetermined stimulus structures and can thus study discrimination 181 ability in mice in a naturalistic context with ethological relevance, where the stimulus is highly 182 variable, but also highly informative of odour source distance. 183 184 Mice were trained to respond with licking to the S+ and suppress licking to S- stimuli, where 185 the odour source distance associated with a reward (S+) was “Near” for half of the animals, 186 and “Far” for the other half ( Fig. S1.5 ). The total duration of licking (total time licked) was 187 calculated for each trial and used in the analysis of discrimination ability ( Fig. 1E, Fig. S1.6). 188 Using this measure, we found the distribution of total time licked to be different between S+ 189 and S- trials for 16/24 mice (p < 0.05, Bonferroni-corrected Mann-Whitney U test). More 190 specifically, mice showed a narrower distribution of lick durations during S+ trials compared 191 to a wider distribution, often shifted to the left or with an additional peak below 200 ms for S- 192 trials ( Fig. 1E example mouse, more mice in Fig. S1.6 ). This suggests mice were more 193 engaged in licking to S+ trials, compared to S- (they spent more time licking and licked more 194 consistently over a large number of trials). 195 To quantify the difference between the distributions of lick durations in response to S+ or S- 196 trials, we split the response window at a threshold of t = 200 ms and calculated a 197 discrimination index defined as the ratio between the number of S- trials and the number of 198 S+ trials with a lick duration smaller than the threshold t ( Fig. 1F ). A ratio >1 indicated a 199 higher number of S- trials with lick durations shorter than t, indicating correct lick behaviour. 200 Conversely, a ratio value <1 indicated a pref erence for S+ trials. For a perfectly trained 201 mouse with no bias, the ratio value would tend towards infinity, while a ratio value close to 1 202 would indicate no preference (and theref ore no difference between S+ and S-). To 203 distinguish a real difference from one obtained by chance, while also taking the lick bias and 204 the difficulty of the task into account, we compared the ratios obtained for individual mice to 205 ratios calculated from their own resp onses, after shuffling the labels ( Fig. S1.6M). All mice 206 showed a Discrimination index >1 and the differences were significant for 19/24 mice (p < 207 0.001) when compared to the corresponding shuffled controls ( Fig. 1F ). Additionally, lick 208 duration distributions remained robustly diffe rent for any threshold chosen between 50 and 209 600 m s (Fig. S1.6N,O). This analysis was necessary due to a clear bias for licking present in 210 all mice. Thus, most mice displayed almost 100% accuracy in responding to S+ stimuli by 211 licking (Hits). However, when mice did refrain from licking, they predominantly did so only in 212 S- trials, as evidenced by the lower lick fraction recorded in S- trials compared to S+ trials 213 (Fig. S1.4, fraction licked split by S+/S-), resulting in fewer False Alarms while still displaying 214 a high Hit rate ( Fig. 1G, Fig S1.6P). This is what allows mice to perform with above chance 215 accuracy (Fig. S1.4). Overall, we conclude mice can discriminate odour sources placed at 216 different distances based on the different lick behaviours expressed in response to S+ 217 compared to S-. 218 The primary aim of the behaviour experiment described above was to determine whether 219 mice are capable of discriminating between odour stimuli with the same odour identity, but 220 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 6 delivered from two different locations in space, and thus arriving at the nose of the animal 221 with different temporal structures. We found that mice were capable of this discrimination, 222 but it remained unclear what information in the stimulus they were using to inform their 223 responses. In order to answer this question, we further aimed to investigate what information 224 is present in the odour plumes presented to the mouse, and what information from these 225 plumes is accessible to the mouse olfactory system. 226 What features of the odour plume allow for distance discrimination? 227 To analyse what information is available in the plume structure that would allow distance 228 discrimination, we recorded temporal struct ures of odour plumes using a photoionization 229 detector (PID). We placed the detector in the same location where a mouse’s nose would be 230 during a trial, and delivered odours as described above ( Fig. 2). This was done during the 231 same period as animal training (though not simultaneously), with the additional purpose of 232 confirming stimulus stability throughout the training period. This large plume bank allowed us 233 to get a handle on the large variability seen in nat uralistic plumes, which is a key feature of 234 our approach, as well as an experimental chal lenge. The temporal structure of recorded 235 odour plumes was different between distances. Figure 2B shows example recorded plumes 236 from a far source and near source, respectively. Additional examples ( Fig. S2.1 ) further 237 illustrate the differences between distances, as well as the variability within categories and 238 across recorded plumes overall. 239 240 241 Figure 2: Odour plume structure varies reliably with distance and sub-sniff features are the 242 most informative. (A) Schematic of the distance Odour Delivery Device (dODD) as in Fig. 1 . 243 Complex airflow is generated inside a wind tunnel using a fan, a honeycomb structure and an 244 obstacle. Odour sources are placed downstream of the fan and upstream of the detector (PID) and 245 arranged symmetrically across the midline of the tunnel (Left and Right positions). (B) Example 246 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 7 plumes recorded from the Near (30 cm) and Far (80 cm) positions. (C) Quantification of Mean odour 247 concentration (left), Maximum odour concentration (middle) and Number of peaks (right) for near and 248 far plumes. Box indicates 25th-75th percentiles, thick line is median, whiskers are most extreme data 249 points not considered outliers (see methods). (D) tSNE analysis of odour features (mean odour 250 concentration, maximum odour concentration (largest peak height), number of peaks, maximum peak 251 prominence, standard deviation, variance, kurtosis, skewness) calculated for a 5 s time window for 252 each plume, showing clear separation by distance. Each dot is a trial (n = 360 plumes, see methods). 253 (E) Same as (D) but for plume structure features. Dots are colour-coded the value of individual 254 temporal structure features to visualise any gradients preserved in the two-dimensional space. (F) 255 Schematic of temporal Olfactory Delivery Device (tODD). (G) Example plumes from near and far 256 distances. Recorded plumes in grey, reproduced plumes in blue (near) or orange (far). (H) Same as in 257 (C) but for plumes reproduced using tODD. Box indicates 25th-75th percentiles, thick line is median, 258 whiskers are most extreme data points not considered outliers. 259 To quantify the differences in the temporal st ructure of odour plumes, we compared the two 260 distance categories for eight features calculated for each plume instance ( Fig. 2B,C, Fig. 261 S2.2, S2.3, S2.5 , Methods): mean odour concentration, maximum odour concentration 262 (largest peak height), number of peaks, maximum peak prominence, standard deviation, 263 variance, kurtosis, skewness ( Fig. 2C and Fig. S2.3 with all features). The most substantial 264 differences were observed in temporal features of the plumes, such as number or height of 265 peaks. These were also the features that allowed for the highest accuracy of a linear 266 discriminant trained on individual features one at a time ( Fig. S2.4). Moreover, combining 267 different features allows for almost perfect classification of plumes into near and far 268 categories (tSNE, eight features used, Fig. 2D ) where the number of peaks is the most 269 informative axis for the discrimination (Fig. 2E ; see Fig. S2.7 for complementary 270 dimensionality reduction analysis using PCA). Thus, the temporal structure of odour plumes 271 contains enough information to distinguish stimu li from sources at different distances, with 272 some features allowing for higher discrimination accuracy than others. 273 In order to further investigate which features are most informative, we moved to a more 274 controlled setting, in which we could readily simulate odour plumes from different distances 275 using a high-speed odour delivery device: the temporal Odour Delivery Device (tODD 276 (Ackels et al., 2021); Fig. 2F). Using this device, we were able to reproduce the recorded 277 structures described above with high fidelity ( Fig. S2.8, S2.9, S2.10 ). The reproduction 278 strategy retained the discriminability between plume categories based on individual features 279 by capturing the differences and variability in features observed in recorded plumes ( Fig. 280 2G,H, S2.11, S2.12). 281 282 By capturing and replicating the spatiotemporal attributes that differentiate near from far 283 odour sources, we have shown that temporal f eatures can reliably classify plume distance 284 categories. Precise playback of these reco rded plumes preserves variability across 285 individual stimuli. This sets the stage for tightly controlled physiology experiments under 286 head-fixation, to link the observed discriminability in behavioural tasks to neural responses in 287 the OB. 288 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 8 Odour source distance is encoded in a subset of OB projection neurons 289 We have so far shown that mice can discrimi nate between odour sources placed at different 290 distances, and that temporal information pres ent in odour plumes would allow for this 291 discrimination. We next examined whethe r these high-frequency c ues are actually 292 accessible to OB neurons. Historically, the brief, rapidly fluctuating concentration peaks 293 found in turbulent plumes were assumed to exceed the temporal resolution of mammalian 294 olfaction . However, recent work suggests otherwise (Ackels et al., 2021; Dasgupta et al., 295 2022; Warner et al., 2024; Wu et al., 2024). To investigate whether the OB can indeed 296 extract distance-related information from nat uralistic plume structures, we combined a 297

for reliably recreating plume dynamics with two-photon calcium imaging of dorsal OB 298 projection neurons (MTCs) in head-fixed, anaes thetised mice (Tbet-cre:GCaMP6f, n=3), 299 while simultaneously monitoring respiration (Fig. 3A-C). Rather than repeatedly presenting a 300 single plume signature, we delivered 40 distinct “near” plumes and 40 distinct “far” plumes 301 (Fig. 3D , outer columns), capturing a br oad range of temporal features ( Fig. 2G-H ). This 302 allowed us to assess whether information about distance can be extracted by the OB from 303 highly variable, naturalistic stimuli. Additionally, the large number of distinct plumes, and the 304 resulting wide range of different features and f eature combinations covered by our stimulus 305 space ( Fig. 2G-H ), allowed us to further probe into how different features of naturalistic 306 odour plumes are represented in the OB. 307 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 9 308 309 Figure 3: Responses of MTCs to temporally structured odour plumes recorded from different 310 distances. (A) Schematic of the two-photon imaging approach. (B) GCaMP6f fluorescence from 311 mitral and tufted cells (maximum projection of 8,000 frames). Responses from ROI marked in red are 312 shown in (C). Scale bar, 100 μ m (C) Example single trials in response to an odour stimulus from a 313 near (30 cm) or far (80 cm) virtual odour source (top row). Respiration traces (inhalation pointing 314 upwards) from the same trials (middle row). Example fluorescence traces (df/f) from the red circled 315 ROI in (B). Shaded area represents 3 s odour stimulation window. (D) Individual near and far plumes 316 (outer columns) and corresponding calcium responses (inner columns) of one ROI. (E) Averaged 317 responses to all near and far plumes (n = 40 for each condition, mean ± s.e.m.). (F) Average 318 fluorescence traces (mean ± s.e.m., n = 160 trials) for 6 example distance sensitive ROIs (near: blue, 319 far: orange; ROIs i-iv: EB, v-vi: 2H). 320 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 10 To assess whether the OB is able to discriminat e between plume structures originating from 321 different distances, we recorded neuronal activity (as expressed by a change in GCaMP6f 322 fluorescence, Fig. 3D, inner columns) in response to 40 different plume structures for each 323 distance category ( Fig. 3D, outer columns). Across all trials (~90 per distance, two odours; 324 ethyl butyrate and 2-heptanone), we identified a small subset (13 out of 1062 cell-odour 325 pairs; 1.2%) that responded differently to near versus far plumes ( Fig. 3E). These “distance-326 sensitive” cells showed stronger responses to far stimuli in the majority of cases ( Fig. 3F), 327 and each cell’s sensitivity was typically specific to one odour ( Fig. S3.1, S3.2). In addition, a 328 subset of plumes was presented multiple times to capture response variability ( Fig. S3.2 ). 329 An overview of responses of all ROIs to all stimuli ( Fig. S3.3) and to intermediate distances 330 (Fig. S3.4 ) and to a subset of plumes that were presented multiple times ( Fig S3.5) 331 illustrates the variety in their response strengths and kinetics. This indicates that, despite the 332 broader heterogeneity in temporal plume featur es, a subset of MTCs can encode distance-333 related information. 334 This finding, while involving only a small number of MTCs, suggests that distance-specific 335 signals may be accessible at the population level. It therefore led us to the question of 336 whether these neural responses can be used to predict an odour source’s distance? In the 337 following section, we address this by examini ng whether the patterns of OB activity can 338 serve as a readout of plume distance. 339 Odour plume distance can be predicted from MTC responses 340 We next assessed whether distance-dependent differences in OB activity extend beyond the 341 small number of clearly distance-sensitive MTCs ( Fig. 3 ). To visualise the responses 342 underlying this classification, we plotted the average calcium signals in response to near and 343 far plumes over time (Fig. 4A, left and middle). The difference map ( Fig. 4A, right) highlights 344 that a subset of cell-odour pairs exhibited higher activity in response to far plumes (blue) or 345 near plumes (red). A linear classifier was trained on the population responses of all recorded 346 cell-odour pairs to determine whether the stimulus originated from a near or far source, 347 based on the ensemble activity ( Fig. 4B). Classifier performance steadily increased with the 348 number of cell-odour pairs included (purple line, maximum performance: 73.0 ± 9.0%), 349 substantially exceeding the shuffled control accura cy (black line). This result indicates that 350 while only a fraction of MTCs responded signific antly differentially to distance, activity 351 patterns across the population of cells carry significant information about plume origin. 352 Further classification analyses for both odours individually ( Fig. S4.1, S4.2 ) and a pairwise 353 comparison between further intermediate odour source distances (Fig. S4.3) can be found in 354 the supplement. 355 When comparing mean responses of individual cells to far versus near conditions ( Fig. 4C), 356 the majority of responses showed similar amplitude for both plume types, but a subset 357 (purple) fell above the unity line, showing str onger responses towards far stimuli. When we 358 further segregated these cells’ activity by odour identity, it became clear that distance 359 sensitivity was associated with the o dour that elicited stronger responses ( Fig. S4.4, S4.5 ). 360 Overall, these results demonstrate that ev en modest differences in MTC response strength 361 allow above-chance prediction of odour source distance from population activity. 362 In the final part of our results, we will examine which temporal features of naturalistic plumes 363 underlie this distance-dependent coding, providing a direct link between the neural 364 representation in the OB and the animal’s ability to discriminate odour sources at varying 365 distances. 366 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 11 367 368 Figure 4: Population activity and linear discriminant analysis of MTC responses. (A) Calcium 369 transients as colour maps for averaged responses to near (left) and far (middle) plumes and for the 370 difference between the two (right) for all cell-odour pairs (n = 1062). (B) Accuracy of linear classifier 371 trained on a 5 s response window (purple; black, shuffle control) to near versus far plumes (n = up to 372 1062 cell-odour pairs from 3 mice; mean ± s.d. of 50000 repetitions). (C) Responses to near vs 373 responses to far plumes plotted as mean df/f response for all trials from those distances. Distance-374 sensitive cells shown in purple. Cells above diagonal respond more strongly to far plumes while cells 375 below diagonal respond more strongly to near plumes. 376 Stimulus features informing distance-sensitivity 377 We first demonstrated that mice can discrimi nate between odour sources placed at different 378 distances. Additionally, we showed distance-related information is also discernible from the 379 activity of OB projection neurons. Next, we aimed to understand which features of the 380 stimulus inform the OB (and consequently the animal) about distance. To this end, we took 381 advantage of the wide range of plume structures we presented in the head-fixed setting. As 382 described above (and Methods), we presented 40 different plume structures for each 383 distance category, with each plume not only informing about distance, but also expressing 384 many temporal features, in different combinations. Thus, this experimental design allowed us 385 to analyse cell responses not only in terms of preference for a particular distance, but also 386 preference for specific features of the odour plume. To further leverage diversity in the 387 temporal features space, we introduced two intermediate source distances (40 cm and 60 388 cm) for a total of 160 different plume structures presented to the mouse ( Fig. 5A). For each 389 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 12 plume we defined extracted values for an array of different features. This allowed us to move 390 beyond simple near vs far comparison (broad categories) and explore how the OB encodes 391 specific temporal features across and within categories. To support consistent plume 392 presentation and alignment, respiration was recorded during the experiments and remained 393 stable across mice and conditions at ~2 Hz ( Fig. S5.1 ). Stimuli were aligned to inhalation 394 onset, ensuring consistent timing across trials. We also verified that total and inhaled odour 395 concentrations were broadly matched across distances ( Fig. S5.2), ensuring that observed 396 neural differences were not simply driven by concentration differences but solely by temporal 397 plume structure. When examining odour exposure on a per-sniff basis, we found that the 398 amount of odour inhaled during the first sniff was comparable across distances, while 399 differences in the inhaled odour amount between sniffs emerged from the second sniff 400 onwards, particularly for near versus far plumes (Fig. S5.3). 401 402 403 404 Figure 5: Features of the temporal structure of odour plumes involved in distance 405 discrimination. (A) Example plumes recorded from four different distances (30 cm, 40 cm, 60 cm 406 and 80 cm). (B) Quantification of mean odour concentration (left), number of peaks (middle) and 407 maximum odour concentration (right) for all plumes from four distances. Box indicates 25th-75th 408 percentiles, thick line is median, whiskers are most extreme data points not considered outliers (see 409 methods). (C) Response of an example distance-sensitive ROI (df/f averaged over the 5 s response 410 window) plotted against individual features of the odour stimulus (x-axis). Each dot is a trial (n = 160 411 trials), colour-coded by distance (80 cm orange, 60 cm pale orange, 40 cm pale blue, 30 cm blue). (D) 412 Correlation coefficients to distance and other stimulus features for three ROIs (i-iii). (E) Heatmap of 413 correlation coefficients as calculated in C for all cells (columns, n = 531 ROIs from 3 animals), sorted 414 by their correlation to distance (top row). Labels i, ii, iii correspond to ROIs shown in (D). 415 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 13 To investigate which stimulus properties drive the responses of distance-sensitive cells, we 416 leveraged the substantial variability in tempor al features within each distance category (Fig. 417 5B). Specifically, we examined the correlation between trial-by-trial response amplitudes and 418 individual plume features. Figure 5C illustrates the method for a representative distance-419 sensitive ROI: the calcium response is plo tted against three plume descriptors (mean odour 420 concentration, number of peaks and maximum peak concentration). This approach revealed 421 cells whose activity is strongly correlated to certain plume features. For example, the 422 response of Cell ii in Fig. 5D appears to be closely linked to the total odour inhaled. Notably, 423 for distance-sensitive cells the strongest negative correlations appeared with the plume 424 features that best indicate source distance. These features operate at sub-sniff timescales, 425 such as the number of peaks and the maximum peak concentration ( Fig. 5C,D). When we 426 sorted all recorded cells by their correlation with distance, we observed that correlation 427 between response and distance could be inversely mapped to the correlation between 428 response and sub-sniff temporal features ( Fig. 5E, bottom part of the heatmap). In contrast, 429 no pattern emerged between correlation with distance and average correlation measures 430 (Fig. 5E). 431 432 Collectively, these findings underscore that sub-sniff temporal dynamics, rather than simple 433 concentration averages, serve as key indicators of source distance for OB neurons. This 434 provides critical insight into how rodents mi ght use the complex temporal dynamics of odour 435 plumes under natural conditions. 436

437 Rodents depend on olfactory cues to navigate their surroundings, making the ability to 438 discern whether an odour source is nearby or distant critical for survival decisions such as 439 foraging or predator avoidance. To invest igate how rodents might perform such 440 discrimination, we developed an experimental setup that reliably produces temporally 441 complex odour plumes from different distan ces, striking a balance between naturalistic 442 complexity and experimental control. With high-throughput behavioural assays and in vivo 443 physiological recordings, we examined whether mice can discriminate the distance of 444 different odour sources and how distance-related temporal plume features are encoded 445 within the OB. 446 447 Our results demonstrate that mice can discrim inate odour sources placed at different spatial 448 distances, showing for the first ti me that information within odour plumes provides sufficient 449 cues for distance discrimination. We observed t hat plume features at sub-sniff timescales, 450 rather than broader measures such as mean odour concentration, carried more information 451 to discriminate odour source distances. Particularly the number of peaks and maximum peak 452 concentration provided higher accuracy in distinguishing between near and far sources. This 453 aligns with odour bout count and stimulus inte rmittency measures proposed in previous 454 research (Gumaste et al., 2024; Schmuker et al., 2016). We conducted plume feature 455 analysis at high sampling frequencies (1 kHz) which likely exceeds the temporal resolution of 456 mouse olfactory perception (approximately 40 Hz; (Ackels et al., 2021)) and thereby 457 potentially inflates peak counts and magnitude values. Future studies should systematically 458 test plume feature reliability and informativeness at various sampling frequencies to align 459 with rodent sensory capabilities (Boero et al., 2025; Gumaste et al., 2024; Lewis et al., 460 2024). The distances we chose for our ex periments provided clear and discriminable 461 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 14 statistical differences in plume features, effect ively capturing the natural variability rodents 462 might encounter in their environments (Reddy et al., 2022; Rigolli et al., 2022). Interestingly, 463 we observed that the four distances studied grouped into two distinct categories, suggesting 464 a near versus far categorisation as opposed to a gradient mapping of source distance. This 465 categorical distinction may reflect an intrinsi c boundary in rodent olfactory perception (Liu et 466 al., 2020; Rigolli et al., 2022). Yet, whether this categorization generalises beyond our 467 experimental settings remains to be determined. Expanding the range of distances and 468 systematically varying environmental parameters, such as airflow dynamics and obstacle 469 configuration, would clarify how robustly specific plume features inform distance 470 discrimination across diverse contexts. 471 472 Our findings show mice consistently displayed differential licking responses to odour sources 473 at varying distances, although achieving cons istently high discrimination accuracy proved 474 challenging. Several factors likely contributed to this moderate performance. Notably, the 475 inherent asymmetry of the Go/No-Go task, where an impulsive action (licking) is contrasted 476 against inhibitory control (withholding licki ng), might have influenced overall accuracy 477 (Carandini & Churchland, 2013). Detailed analys is of licking patterns revealed that mice 478 were more consistently responsive during re warded (S+) trials, suggesting clear reward 479 anticipation, whereas variability in responses during non-rewarded (S−) trials might reflect 480 fluctuations in motivation rather than genuine perceptual errors (Berditchevskaia et al., 481 2016). Additionally, efforts to mitigate predictabl e differences in odour arrival times between 482 distances, by incorporating variable onset delays and flexible trial termination, inadvertently 483 intensified task asymmetry. Although designed to maintain engagement, these adjustments 484 allowed mice to conclude rewarded trials prematurely while requiring them to endure the full 485 duration of non-rewarded trials for a correct response. Consequently, mice may have 486 strategically licked simply to reduce trial durations, potentially decoupling their licking 487 decisions from actual stimulus discriminat ion. Taking these behavioural and task-design 488 considerations into account, it seems like ly that mice possess greater discrimination 489 capability than was observed under our current experimental conditions. Refining the training 490 paradigm, for instance by implementing a two-alternative forced-choice (2AFC) task 491 (Bitzenhofer et al., 2022; Boero et al., 2025; Nakayama et al., 2022) to mitigate motivational 492 biases, may help shed further light onto the limits of sensory discrimination capabilities. 493 494 A key question we asked was: How is odour source distance represented in the OB? To 495 ensure naturalistic variability of temporal pa tterns, we exposed animals to a wide range of 496 plume structures recorded from near and far sources. When decoding odour source distance 497 from the population of MTC calcium responses, we found that accuracy exceeded 498 behavioural performance of mice proficient in the discrimination task. This suggests that 499 distance information carried by odour plumes c ould, in principle, support even more precise 500 discri mination. Strikingly, a small subset of individual MTCs exhibited clear differential 501 responses to near versus far plume sources, with most distance-selective MTCs responding 502 more strongly to far than to near stimuli in mo st cases. Notably, this tuning was already 503 present in naïve animals that had never been trained to judge source distance. To 504 understand what might be driving this selectivit y, we presented a large and diverse set of 505 plume structures, including intermediate distances (40 cm and 60 cm). When correlating 506 neural responses in a trial-by-trial fashion with specific stimulus features, we observed that 507 their activity tended to correlate inversely with sub-sniff plume features most predictive of 508 source distance. Furthermore, when sorting cells by their correlation with distance, we found 509 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 15 that their response tuning could be almost perfe ctly mapped to their tuning to these fast 510 temporal features. In contrast, correlations with broader, average plume statistics did not 511 show such a pattern. These findings suggest that fa st fluctuations in odour intensity, rather 512 than slow integrative cues, may play a c entral role in how OB neurons encode spatial 513 information. We acknowledge, however, that this observation is correlative, and future work 514 is needed to determine whether these features ar e used directly by the OB circuitry to 515 support behavioural discrimination. 516 What cellular or circuit mechanisms could account for these distance-dependent responses? 517 One possibility is that MTCs are not “count ing peaks” per se, but instead integrate the 518 constancy of odour concentration (Boero et al., 2025; Lewis et al., 2024). Less-peaky, more 519 continuous plumes (typical of far sources) c ould drive a larger, temporally summated input 520 from ORNs than highly intermittent near plumes (Lewis et al., 2024). Since ORNs of a given 521 receptor type are scattered across the epithelium before converging onto the same 522 glomerulus (Malnic et al., 1999; Mombaerts et al., 1996), pockets of odorised air will likely 523 recruit partially non-overlapping ORN ensembl es over time. A smoother plume might 524 therefore activate a larger part of that ensemble, whereas a smoother plume with low 525 intermittency may engage fewer ORNs within brief time windows. Additional temporal 526 summation could occur at the level of individual ORNs or of their postsynaptic MTCs, as 527 ORN spike fidelity drops at high stimulat ion frequencies or high concentrations (Ghatpande 528 & Reisert, 2011). Taken together, this could tran sform fine-scale plume fluctuations into 529 population codes that emphasise the more stabl e temporal signature of distant sources. 530 Consistent with this view, the distance-sens itive MTCs we identified were generally more 531 narrowly tuned to odour identity. This raises the testable idea that the most reliable 532 responders within a glomerular column jointly encode both chemical identity and temporally 533 derived spatial information (Arneodo et al., 2018). 534 Methodologically, our calcium imaging approach is well suited for identifying and targeting 535 specific neurons, but its temporal resolu tion limits capturing millisecond-scale plume 536 dynamics. Future work using high-density electrophysiology, faster GCaMP variants, or 537 genetically encoded voltage indicators will be essential to resolve the precise timing of ORN 538 and MTC firing relative to fine temporal plume f eatures (Lewis et al., 2024; Storace et al., 539 2015). Combining such recordings with selective activation of a single, genetically labelled 540 glomerulus (Arneodo et al., 2018; Schwarz et al., 2018), and multiplexed imaging of both 541 ORN input and MTC output (Martelli & Storac e, 2021; Storace & Cohen, 2017) could reveal 542 how lateral inhibition, sister-cell diversity (Zhang et al., 2025), and molecular specialisations 543 together give rise to distance-sensitive signalling in the OB. 544 545 Our results show that mice can discriminate the distance of an odour source by tracking 546 rapid, sub-sniff changes in a plume, and that a subset of MTCs in the OB reflects this 547 information. By tying together plume measur ements, behaviour, and single-cell activity, we 548 add a strong link between the physics of natural odours, animal behaviour and neural 549 coding. Futur e studies could further clarify wh ich temporal plume features rodents utilise for 550 spatial discrimination in different contexts by employing an olfactory virtual reality coupled 551 with precise plume control, or even in unr estrained mice. Such experiments will help to 552 pinpoint precisely, how specific olfactory bulb ci rcuits transform natural chemical signals into 553 reliable cues for navigation. 554 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 16

555 Ackels, T., Erskine, A., Dasgupta, D., Marin, A. C., Warner, T. P. A., Tootoonian, S., 556 Fukunaga, I., Harris, J. J., & Schaefer, A. T. (2021). Fast odour dynamics are encoded in the 557 olfactory system and guide behaviour. Nature, 593(7860), 558–563. 558 https://doi.org/10.1038/s41586-021-03514-2 559 Arneodo, E. M., Penikis, K. B., Rabinowitz, N., Cichy, A., Zhang, J., Bozza, T., Rinberg, D., 560 Licata, A., Cichy, A., Zhang, J., Bozza, T., & Rinberg, D. (2018). Stimulus dependent 561 diversity and stereotypy in the output of an olfactory functional unit. Nature Communications, 562 9(1), 1347. https://doi.org/10.1038/s41467-018-03837-1 563 Balkovsky, E., & Shraiman, B. I. (2002). Olfactory search at high Reynolds number. 564 Proceedings of the National Academy of Sciences, 99(20), 12589–12593. 565 https://doi.org/10.1073/pnas.192393499 566 Béra, J.-C., Michard, M., Sunyach, M., & Comte-Bellot, G. (2000). Changing lift and drag by 567 jet oscillation: Experiments on a circular cylinder with turbulent separation. European Journal 568 of Mechanics - B/Fluids, 19(5), 575–595. https://doi.org/10.1016/S0997-7546(00)00122-9 569 Berditchevskaia, A., Cazé, R. D., & Schultz, S. R. (2016). Performance in a GO/NOGO 570 perceptual task reflects a balance between impulsive and instrumental components of 571 behaviour. Scientific Reports, 6(1), 27389. https://doi.org/10.1038/srep27389 572 Bhattacharyya, U., & Singh Bhalla, U. (2015). Robust and Rapid Air-Borne Odor Tracking 573 without Casting. eNeuro, 2(6), 1–26. https://doi.org/10.1523/ENEURO.0102-15.2015 574 Bitzenhofer, S. H., Westeinde, E. A., Zhang, H. X. B., & Isaacson, J. S. (2022). Rapid odor 575 processing by layer 2 subcircuits in lateral entorhinal cortex. eLife, 11. 576 https://doi.org/10.7554/ELIFE.75065 577 Boero, L. E., Wu, H., Zak, J. D., Masset, P., Pashakhanloo, F., Jayakumar, S., Tolooshams, 578 B., Ba, D., & Murthy, V. N. (2025). Perception and neural representation of intermittent odor 579 stimuli in mice (p. 2025.02.12.637969). bioRxiv. https://doi.org/10.1101/2025.02.12.637969 580 Carandini, M., & Churchland, A. K. (2013). Probing perceptual decisions in rodents. Nature 581 Neuroscience, 16(7), 824–831. https://doi.org/10.1038/nn.3410 582 Celani, A., Villermaux, E., & Vergassola, M. (2014). Odor Landscapes in Turbulent 583 Environments. Physical Review X, 4(4), 041015. https://doi.org/10.1103/PhysRevX.4.041015 584 Crimaldi, J. P., & Koseff, J. R. (2001). High-resolution measurements of the spatial and 585 temporal scalar structure of a turbulent plume. Experiments in Fluids, 31(1), 90 –102. 586 https://doi.org/10.1007/s003480000263 587 Crimaldi, J., Lei, H., Schaefer, A., Schmuker, M., Smith, B. H., True, A. C., Verhagen, J. V., 588 & Victor, J. D. (2022). Active sensing in a dynamic olfactory world. Journal of Computational 589 Neuroscience, 50(1), 1–6. https://doi.org/10.1007/s10827-021-00798-1 590 Dasgupta, D., Warner, T. P. A., Erskine, A., & Schaefer, A. T. (2022). Coupling of mouse 591 olfactory bulb projection neurons to fluctuating odour pulses. The Journal of Neuroscience, 592 JN-RM-1422-21. https://doi.org/10.1523/JNEUROSCI.1422-21.2022 593 Duchamp-Viret, P., Chaput, M. A., & Duchamp, A. (1999). Odor Response Properties of Rat 594 Olfactory Receptor Neurons. Science, 284(5423), 2171–2174. 595 https://doi.org/10.1126/science.284.5423.2171 596 Erskine, A., Bus, T., Herb, J. T., & Schaefer, A. T. (2019). AutonoMouse: High throughput 597 automated operant conditioning shows progressive behavioural impairment with graded 598 olfactory bulb lesions. PLoS ONE, 291815. https://doi.org/10.1371/journal. pone.0211571 599 Findley, T. M., Wyrick, D. G., Cramer, J. L., Brown, M. A., Holcomb, B., Attey, R., Yeh, D., 600 Monasevitch, E., Nouboussi, N., Cullen, I., Songco, J. O., King, J. F., Ahmadian, Y., & 601 Smear, M. C. (2021). Sniff-synchronized, gradient-guided olfactory search by freely moving 602 mice. eLife, 10, 1–39. https://doi.org/10.7554/elife.58523 603 Ghatpande, A. S., & Reisert, J. (2011). Olfactory receptor neuron responses coding for rapid 604 odour sampling. The Journal of Physiology, 589(Pt 9), 2261–2273. 605 https://doi.org/10.1113/jphysiol.2010.203687 606 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 17 Gire, D. H., Kapoor, V., Arrighi-Allisan, A., Seminara, A., & Murthy, V. N. (2016). Mice 607 Develop Efficient Strategies for Foraging and Navigation Using Complex Natural Stimuli. 608 Current Biology, 26(10), 1261–1273. https://doi.org/10.1016/j.cub.2016.03.040 609 Gumaste, A., Baker, K. L., Izydorczak, M., True, A. C., Vasan, G., Crimaldi, J. P., & 610 Verhagen, J. (2024). Behavioral discrimination and olfactory bulb encoding of odor plume 611 intermittency. eLife, 13, e85303. https://doi.org/10.7554/eLife.85303 612 Gumaste, A., Coronas-Samano, G., Hengenius, J., Axman, R., Connor, E. G., Baker, K. L. 613 K., Ermentrout, B., Crimaldi, J. P., & Verhagen, J. V. (2020). A Comparison between Mouse, 614 In Silico , and Robot Odor Plume Navigation Reveals Advantages of Mouse Odor Tracking. 615 Eneuro, 7(1), ENEURO.0212-19.2019. https://doi.org/10.1523/ENEURO.0212-19.2019 616 Haddad, R., Lanjuin, A., Madisen, L., Zeng, H., Murthy, V. N., & Uchida, N. (2013). Olfactory 617 cortical neurons read out a relative time code in the olfactory bulb. Nature Neuroscience, 618 16(7), 949–957. https://doi.org/10.1038/nn.3407 619 Hopfield, J. J. (1991). Olfactory computation and object perception. PNAS, 88(15), 6462–620 6466. 621 Jacobs, L. F. (2012). From chemotaxis to the cognitive map: The function of olfaction. 622 Proceedings of the National Academy of Sciences, 109, 10693–10700. 623 https://doi.org/10.1073/pnas.1201880109 624 Jones, P. W., & Urban, N. N. (2018). Mice follow odor trails using stereo olfactory cues and 625 rapid sniff to sniff comparisons (p. 293746). bioRxiv. https://doi.org/10.1101/293746 626 Kepecs, A., Uchida, N., & Mainen, Z. F. (2006). The sniff as a unit of olfactory processing. 627 Chemical Senses, 31(2), 167–179. https://doi.org/10.1093/chemse/bjj016 628 Khan, A. G., Sarangi, M., & Bhalla, U. S. (2012). Rats track odour trails accurately using a 629 multi-layered strategy with near-optimal sampling. Nature Communications, 3, 703. 630 https://doi.org/10.1038/ncomms1712 631 Kuruppath, P., Bai, L., & Belluscio, L. (2021). Functional Relevance of Dual Olfactory Bulbs 632 in Olfactory Coding. Eneuro, 8(5), ENEURO.0070-21.2021. 633 https://doi.org/10.1523/ENEURO.0070-21.2021 634 Lewis, S. M., Suarez, L. M., Rigolli, N., Steinmetz, N. A., & Gire, D. H. (2024). The spiking 635 output of the mouse olfactory bulb encodes large-scale temporal features of natural odor 636 environments. https://doi.org/10.110 1/2024.03.01.582978 637 Lewis, S. M., Xu, L., Rigolli, N., Tariq, M. F., Suarez, L. M., Stern, M., Seminara, A., & Gire, 638 D. H. (2021). Plume Dynamics Structure the Spatiotemporal Activity of Mitral/Tufted Cell 639 Networks in the Mouse Olfactory Bulb. Frontiers in Cellular Neuroscience, 15. 640 https://doi.org/10.3389/fncel.2021.633757 641 Li, A., Gire, D. H., Bozza, T., & Restrepo, D. (2014). Precise Detection of Direct Glomerular 642 Input Duration by the Olfactory Bulb. Journal of Neuroscience, 34(48), 16058–16064. 643 https://doi.org/10.1523/JNEUROSCI.3382-14.2014 644 Liu, A., Papale, A. E., Hengenius, J., Patel, K., Ermentrout, B., & Urban, N. N. (2020). Mouse 645 Navigation Strategies for Odor Source Localization. Frontiers in Neuroscience, 14(March), 646 1–16. https://doi.org/10.3389/fnins.2020.00218 647 Madisen, L., Garner, A. R. R., Shimaoka, D., Chuong, A. S. S., Klapoetke, N. C. C., Li, L., 648 van der Bourg, A., Niino, Y., Egolf, L., Monetti, C., Gu, H., Mills, M., Cheng, A., Tasic, B., 649 Nguyen, T. N. N., Sunkin, S. M. M., Benucci, A., Nagy, A., Miyawaki, A., … Zeng, H. (2015). 650 Transgenic Mice for Intersectional Targeting of Neural Sensors and Effectors with High 651 Specificity and Performance. Neuron, 85(5), 942–958. 652 https://doi.org/10.1016/j.neuron.2015.02.022 653 Mafra-Neto, A., & Cardé, R. T. (1994). Fine-scale structure of pheromone plumes modulates 654 upwind orientation of flying moths. Nature, 369(6476), 142–144. 655 https://doi.org/10.1038/369142a0 656 Malnic, B., Hirono, J., Sato, T., & Buck, L. B. (1999). Combinatorial receptor codes for odors. 657 Cell, 96(5), 429–466. 658 Marin, A. C., Schaefer, A. T., & Ackels, T. (2021). Spatial information from the odour 659 environment in mammalian olfaction. Cell and Tissue Research, 0123456789. 660 https://doi.org/10.1007/s00441-020-03395-3 661 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 18 Martelli, C., & Storace, D. A. (2021). Stimulus Driven Functional Transformations in the Early 662 Olfactory System. Frontiers in Cellular Neuroscience, 15. 663 https://doi.org/10.3389/fncel.2021.684742 664 McKissick, O., Klimpert, N., Ritt, J. T., & Fleischmann, A. (2024). Odors in space. Frontiers 665 in Neural Circuits, 18. https://doi.org/10.3389/fncir.2024.1414452 666 Mombaerts, P., Wang, F., Dulac, C., Chao, S. K., Nemes, A., Mendelsohn, M., Edmondson, 667 J., & Axel, R. (1996). Visualizing an olfactory sensory map. Cell, 87(4), 675–686. 668 https://doi.org/10.1016/S0092-8674(00)81387-2 669 Moore, P. A., & Atema, J. (1991). Spatial Information in the Three-Dimensional Fine 670 Structure of an Aquatic Odor Plume. The Biological Bulletin, 181(3), 408–418. 671 https://doi.org/10.2307/1542361 672 Nakayama, H., Gerkin, R. C., & Rinberg, D. (2022). A behavioral paradigm for measuring 673 perceptual distances in mice. Cell Reports Methods, 2(6), 100233. 674 https://doi.org/10.1016/j.crmeth.2022.100233 675 Pachitariu, M., Stringer, C., Schröder, S., Dipoppa, M., Rossi, L. F., Carandini, M., & Harris, 676 K. D. (2016). Suite2p: Beyond 10,000 neurons with standard two-photon microscopy. 677 bioRxiv, 061507. https://doi.org/10.1101/061507 678 Poo, C., Agarwal, G., Bonacchi, N., & Mainen, Z. F. (2022). Spatial maps in piriform cortex 679 during olfactory navigation. Nature, 601(7894), 595–599. https://doi.org/10.1038/s41586-680 021-04242-3 681 Rebello, M. R., McTavish, T. S., Willhite, D. C., Short, S. M., Shepherd, G. M., & Verhagen, 682 J. V. (2014). Perception of Odors Linked to Precise Timing in the Olfactory System. PLoS 683 Biology, 12(12), e1002021. https://doi.org/10.1371/journal.pbio.1002021 684 Reddy, G., Murthy, V. N., & Vergassola, M. (2022). Olfactory Sensing and Navigation in 685 Turbulent Environments. Annual Review of Condensed Matter Physics, 13(1), 191–213. 686 https://doi.org/10.1146/annurev-conmatphys-031720-032754 687 Rigolli, N., Magnoli, N., Rosasco, L., & Seminara, A. (2022). Learning to predict target 688 location with turbulent odor plumes. eLife, 11. https://doi.org/10.7554/eLife.72196 689 Rokni, D., Hemmelder, V., Kapoor, V., & Murthy, V. N. (2014). An olfactory cocktail party: 690 Figure-ground segregation of odorants in rodents. Nature Neuroscience, 17(A ugust), 1225–691 1232. 692 Schmuker, M., Bahr, V., & Huerta, R. (2016). Exploiting plume structure to decode gas 693 source distance using metal-oxide gas sensors. Sensors and Actuators B: Chemical, 235, 694 636–646. https://doi.org/10.1016/j.snb.2016.05.098 695 Schwarz, D., Kollo, M., Bosch, C., Feinauer, C., Whiteley, I., Margrie, T. W., Cutforth, T., & 696 Schaefer, A. T. (2018). Architecture of a mammalian glomerular domain revealed by novel 697 volume electroporation using nanoengineered microelectrodes. Nature Communications, 698 9(1). https://doi.org/10.1038/s41467-017-02560-7 699 Smear, M., Resulaj, A., Zhang, J., Bozza, T., & Rinberg, D. (2013). Multiple perceptible 700 signals from a single olfactory glomerulus. Nature Neuroscience, 2013(11), 1687–1691. 701 https://doi.org/10.1038/nn.3519 702 Smear, M., Shusterman, R., O’Connor, R., Bozza, T., & Rinberg, D. (2011). Perception of 703 sniff phase in mouse olfaction. Nature, 479(7373), 397–400. 704 https://doi.org/10.1038/nature10521 705 Sterrett, S. C., Findley, T. M., Rafilson, S. E., Brown, M. A., Weible, A. P., Marsden, R., 706 Tarvin, T., Wehr, M., Murray, J. M., Fairhall, A. L., & Smear, M. C. (2024). Olfactory bulb 707 tracks breathing rhythms and place in freely behaving mice (p. 2024.11.06.622362). bioRxiv. 708 https://doi.org/10.1101/2024.11.06.622362 709 Storace, D. A., Braubach, O. R., Jin, L., Cohen, L. B., & Sung, U. (2015). Monitoring Brain 710 Activity with Protein Voltage and Calcium Sensors. Scientific Reports, 5, 10212. 711 https://doi.org/10.1038/srep10212 712 Storace, D. A., & Cohen, L. B. (2017). Measuring the olfactory bulb input-output 713 transformation reveals a contribution to the perception of odorant concentration invariance. 714 Nature Communications, 8(1), 81. https://doi.org/10.1038/s41467-017-00036-2 715 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 19 Sunil, A., Pedroncini, O., Schaefer, A. T., & Ackels, T. (2024). How do mammals convert 716 dynamic odor information into neural maps for landscape navigation? PLOS Biology, 22(11), 717 e3002908. https://doi.org/10.1371/journal.pbio.3002908 718 Tariq, M. F., Sterrett, S. C., Moore, S., Lane, Perkel, D. J., & Gire, D. H. (2024). Dynamics of 719 odor-source localization: Insights from real-time odor plume recordings and head-motion 720 tracking in freely moving mice. PLOS ONE, 19(9), e0310254. 721 https://doi.org/10.1371/journal.pone.0310254 722 Tootoonian, S., True, A. C., Stark, E., Crimaldi, J. P., & Schaefer, A. T. (2025). Quantifying 723 spectral information about source separation in multisource odour plumes. PLOS ONE, 724 20(1), e0297754. https://doi.org/10.1371/journal.pone.0297754 725 Van Loo, P. L., Mol, J. A., Koolhaas, J. M., Van Zutphen, B. F., & Baumans, V. (2001). 726 Modulation of aggression in male mice: Influence of group size and cage size. Physiology & 727 Behavior, 72(5), 675–683. https://doi.org/10.1016/s0031-9384(01)00425-5 728 Van Loo, P. L. P., Van Zutphen, L. F. M., & Baumans, V. (2003). Male management: Coping 729 with aggression problems in male laboratory mice. Laboratory Animals, 37(4), 300–313. 730 https://doi.org/10.1258/002367703322389870 731 Vergassola, M., Villermaux, E., & Shraiman, B. I. (2007). “Infotaxis” as a strategy for 732 searching without gradients. Nature, 445(7126), 406–409. 733 https://doi.org/10.1038/nature05464 734 Wachowiak, M. (2011). All in a Sniff: Olfaction as a Model for Active Sensing. Neuron, 71(6), 735 962–973. https://doi.org/10.1016/j.neuron.2011.08.030 736 Warner, T. P. A., Tootoonian, S., & Schaefer, A. T. (2024). Robust encoding of sub-sniff 737 temporal information in the mouse olfactory bulb (p. 2024.03.26.586830). bioRxiv. 738 https://doi.org/10.1101/2024.03.26.586830 739 Welker, W. I. (1964). Analysis of Sniffing of the Albino Rat. Behaviour, 22(3), 223–244. 740 Wu, Y., Chen, K., Xing, C., Huang, M., Zhao, K., & Zhou, W. (2024). Human olfactory 741 perception embeds fine temporal resolution within a single sniff. Nature Human Behaviour. 742 https://doi.org/10.1038/s41562-024-01984-8 743 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 20

744 Behaviour experiments 745 Animals 746 All animal procedures performed in this study were approved by the UK government (Home 747 Office PPL number PA2F6DA12) and by Institutional Animal Welfare Ethical Review Panel. 748 All mice used were C57/Bl6 males (n = 24 mice), evenly distributed across the different 749 groups. Long-term group housing precluded us from using mixed-sex cohorts. 750 All mice in a cohort (n = 24) were group-housed from weaning (21 days of age) to avoid 751 disruption of social hierarchy and aggression la ter in the experiment (Van Loo et al., 2001; 752 Van Loo et al., 2003). They were housed into a suitably large cage, with chew blocks, a 753 variety of houses and hiding places and multiple sources of water and food, until ready to be 754 implanted with the RFID. All mice were kept on a 12h light cycle with the light phase 755 between 7:00 and 19:00. The change in light occurred gradually over a period of 15 minutes. 756 Room temperature and humidity were checked daily and kept around 21°C and 50% 757 respectively. The group cage (and later the Au tonoMouse cage) was the only cage in the 758 room, to avoid conspecific odours disturbing the social hierarchy and causing aggression 759 within the cage. 760 RFID implantation 761 The RFID implant surgery occurred around 6 weeks of age, as previously described (Erskine 762 et al., 2019). Briefly, Mice were anaesthetised under isoflurane (induction: 5% in O 2 2 l/min, 763 maintenance: 2%) and placed on a heat pad for maintenance of body temperature during the 764 surgery. The fur around the base of the neck and scruff was shaved away and the skin 765 cleaned with chlorhexidine (1%) and then dried with a sterile swab. A pre-sterilised needle 766 (IM-200, RFID Systems Ltd., Yorkshire, UK) containing an RFID chip (ID-100B, RFID 767 Systems Ltd., Yorkshire, UK, in Chapter 3, ID-162B/1.4, RFID Systems Ltd., Yorkshire, UK 768 in Chapter 5) was then loaded onto a plunger and in serted into the loose skin at the base of 769 the neck, facing towards the tail of the mouse. The plunger was used to push the chip out of 770 the needle before removing the needle, leaving the RFID chip implanted under the skin. 771 Forceps were then used to pinch shut the incision made by the needle and medical 772 superglue (Vetbond, 3M) was applied to seal the wound. Animals were returned to an 773 individual cage for 10 minutes following the surgery to recover from anaesthesia and for the 774 superglue to dry. Once the righting reflex was regained and the wound was confirmed as 775 properly sealed the mouse was returned to the group cage with its. Following at least 5 days 776 recovery from the implant surgery, the mice were transferred to the AutonoMouse system 777 and kept for up to 18 months. 778 Automated Behaviour setup 779 In AutonoMouse, groups of mice (up to 24) implanted with an RFID chip are housed in a 780 common home cage (for detailed description see Erskine et al., 2019; Ackels et al., 2021). 781 Within the common home cage of AutonoMouse, mice have free access to food, social 782 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 21 interaction and environmental enrichment (including nesting material, chew blocks, climbing 783 frames, cardboard tubes, cardboard and red plasti c houses). Water is not freely available in 784 the system, but can be gained at any time by completion of an operant conditioning Go/No-785 Go task. To access these behavioural tasks, mice must leave the home cage and enter a 786 behavioural area. This behavioural area cont ains the odour port and a lick port through 787 which water rewards can be released. The lick port is also connected to a lick sensor, which 788 registers the animal’s response (time spent in contact with the sensor) in response to the 789 task stimuli. As animals can only gain their daily water intake by completing behavioural 790 tasks, mice are motivated to complete long sequences of trials without manual water 791 restriction. However, there is no limit on how many times mice can trigger a trial, or how 792 much time they can spend in the odour port, which could add additional constraints on using 793 timeouts for negative reinforcements. 794 Acclimatisation to the AutonoMouse system 795 Animals were weighed and their weight recorded on the day they were transferred to the 796 AutonoMouse and then every day for 2 weeks. 797 The first phase of the behavioural task assigned to all animals was a ’pre-training’ phase, 798 designed to train animals to reliably gain their water intake from the behavioural port. This 799 phase was itself split into multiple tasks. At first water was delivered as soon as an animal 800 was detected in the behaviour port (20 trials). Next, water was delivered only if the mouse 801 licked at least once after activating the behavi our port (50 trials). Following from that, the 802 percentage of total trial time (2 s) that the animal must lick to gain a water reward was 803 increased in 50 ms increments up to 10% of trial length, or 200 ms (50 trials per step). Each 804 water reward was initially 15 µl. This was adjusted to 10-30 µl depending on animal 805 performance (to ensure all mice performed roughly the same number of trials per day). 806 For the first two weeks of any AutonoMouse experiment, animal weights were checked daily 807 to ensure health status of the cohort. After two weeks, weight was checked more 808 infrequently (once a week) but total trials performed were monitored daily to ensure animals 809 had acquired sufficient water rewards (e.g. by performing >100 trials, 50% rewarded) in the 810 last 24 hours. Any animal not meeting this criterion or consistently dropping in weight (more 811 than 2 days in a row) was isolated in the behaviour port and manually given water rewards 812 from the lick port. All animals successfully learnt to acquire water by engaging the task. 813 Once reliable licking was achieved, the S+ stimulus was introduced. This often caused a 814 drop in lick reliability, but mice quickly got accustomed to the new stimuli. Once lick reliability 815 was re-established, the mice were progressed to the first stage of the Go/No-Go odour 816 discrimination task (EB vs IAA). 817 Distance discrimination stimuli 818 To probe whether mice could discriminate bet ween odours originating from two different 819 sources, odours were placed at different distances to the mouse in the wind tunnel. All tasks 820 followed a standard Go/No-Go training paradigm, where one distance was associated with a 821 water reward (S+ trials), while the other distance was not (S- trials), and licking during these 822 trials would trigger a time-out of 7 s. Reward was reversed for roughly half the experimental 823 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 22 group (13 mice trained to lick for Near stimuli, 11 mice trained to lick for Far stimuli). Before 824 being introduced to the distance discrimination tasks, mice underwent a series of pre-training 825 stages that gradually increased in difficulty: 826 Stage A - Introduction to the AutonoMouse system – 1-7 days 827 Briefly, animals were first trained to lick the wa terspout for a certain amount of time (the lick 828 threshold, usually 10% of total trial length, or 200 ms for trials of variable length) in the 829 presence of an odour. Further reinforcement was offered for animals performing fewer than 830 200 trials/day, to ensure they could successfully use the system to meet their daily water 831 requirement. 832 Stage B - Odour discrimination – 3 weeks 833 Next followed a simple odour discrimination task (EB vs IAA) where animals learnt to lick to 834 the Go cue and refrain from licking in response to the No Go cue. Half of the animals trained 835 to lick in response to EB, and half in response to IAA. Once stable performance above 80% 836 was achieved, the onset of the odour delivery wa s varied to train the mice to wait for odour 837 arrival, in preparation for the variable odour arrival of the plumes from the wind tunnel in the 838 distance discrimination stage. The delay was chosen to be between 100-950 ms, 839 randomised for each trial and starting with shorter delays before progressing to the full range 840 of delays. A switch valve control was performed at this stage, by introducing two new valves 841 that had not been used before in the task. This assessed whether animals were using odour 842 information and no other environmental cues. 843 Stage C - Valve distance discrimination – introduction to the wind tunnel – 4 weeks 844 Next, the animals performed the same odour discrimination (EB vs IAA), but this time 845 delivered from the wind tunnel. EB was delivered from the distance odour delivery devices, 846 forming odour plumes, while IAA was delivered from a pressurised airflow valve. The 847 concentration of IAA was decreased in small st eps (5% per day starting from 100%) over 4 848 weeks, such that by the end of this stage the animals were effectively performing a distance 849 discrimination task between the near EB source(s) and the far EB source(s). For roughly half 850 of all animals, the near source was S+ (rew arded) and the far source was S- (unrewarded); 851 in the other half of the group, this reward valence was reversed. 852 Stage D - Distance discrimination task – 4 weeks 853 Finally, mice were then progressed to the nex t stage of the experiment, the full distance 854 discrimination task. This was split across 4 experimental stages (27 days excluding breaks, 855 or 3981 ± 550 trials) with short pauses between stages for equipment maintenance (10 days 856 between 2 and 3, and 3 days between 3 and 4). Additionally, the experiment was paused for 857 ~30 min every 12h to replenish the odours, shuffle the spatial arrangement of the dODD, 858 and replace any servo motors that were at risk of failure due to overheating. 859 The algorithm for generating distance discrimination trials was as follows: 860 1. Choose whether the stimulus will be Near or Far. 861 2. Randomly select an onset delay of 100-950 ms. 862 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 23 3. Randomly select an odour dODD position (Left or Right) from the corresponding distance. 863 4. Randomly select a blank dODD position (Left or Right) from opposite distance. 864 5. Randomly select two mineral oil valves from the tODD to create a blank plume and anti-865 plume structure, to create a noise decoy, using a 3s long plume template from the distance 866 plume bank as described elsewhere. 867 6. Trigger the trial to have valves in (5) open from time 0 until the trial is terminated, and the 868 2 dODD positions open from time 0 + onset delay decided in (2) + 2 s. 869 Data analysis 870 AutonoMouse data was first extracted into MATLAB using the corresponding AutonoMouse 871 software feature (Erskine et al., 2019, https://github.com/RoboDoig/autonomouse-control), 872 extracting all task parameters and lick data as to tal time licked (sum of all “on” times) and 873 individual lick times (times of threshold crossing). Full lick sensor traces were also saved as 874 separate files sampled at 20 kHz for future reference. First analysis steps were performed in 875 MATLAB using custom scripts and further analysis was performed using Python 3 in Jupyter 876 Lab. 877 Imaging experiments 878 Animals 879 Three female mice aged 20 weeks expressing the calcium indicator GCaMP6f in MTCs were 880 used for this experiment. The mice were Tbet-Cre (Haddad et al., 2013) crossed with 881 GCaMP6f reporter line Ai95(RCL-GCaMP6f)-D (Madisen et al., 2015). 882 Surgical procedure 883 Prior to surgery all utilised surfaces and apparat us were sterilised with 1% trigene. Mice 884 were anaesthetised using a mixture of fentanyl/midazolam/medetomidine (0.05, 5, 0.5 mg/kg 885 respectively). Depth of anaesthesia was monitored throughout the procedure by testing the 886 toe-pinch reflex. The fur over the skull and at the base of the neck was shaved away and the 887 skin cleaned with 1% chlorhexidine scrub. Mice were then placed on a thermoregulator (DC 888 Temperature Controller, FHC, ME USA) heat pad controlled by a temperature probe inserted 889 rectally. While on the heat pad, the head of the animal was held in place with a set of ear 890 bars. The scalp was incised and pulled away from t he skull with four arterial clamps at each 891 corner of the incision. A custom head-fixation implant was attached to the base of the skull 892 with medical super glue (Vetbond, 3M, Maplewood MN, USA) such that its most anterior 893 point rested approximately 0.5 mm posterior to the bregma line. Dental cement (Paladur, 894 Heraeus Kulzer GmbH, Hanau, Germany; Simplex Rapid Liquid, Associated Dental Products 895 Ltd., Swindon, UK) was then applied around the edges of the implant to ensure firm 896 adhesion to the skull. A craniotomy over the left olfactory bulb (approximately 2 × 2 mm) was 897 made with a dental drill (Success 40, Osada, Tokyo, Japan) and then immersed in ACSF 898 (NaCl (125 mM), KCl (5 mM), HEPES (10 mM), pH adjusted to 7.4 with NaOH, 899 MgSO4.7H2O (2 mM), CaCl2.2H2O (2 mM), glucose (10 mM)) before removing the skull 900 with forceps. The dura was then peeled back using fine forceps. A layer of 2% low-melt 901 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 24 agarose diluted in ACSF was applied over the exposed brain surface before placing a glass 902 window cut from a cover slip (borosilicate glass #1 thickness [150 μ m]) using a diamond 903 scalpel (Sigma-Aldrich) over the craniotomy. The edges of the window were then glued with 904 medical super glue (Vetbond, 3M, Maplewood MN, USA) to the skull. Following surgery, 905 mice were placed in a custom head-fixation apparatus and transferred to a two-photon 906 microscope rig along with the heat pad. 907 Imaging parameters 908 The microscope (Scientifica Multiphoton VivoScope) was coupled with a MaiTai DeepSee 909 laser (Spectra Physics, Santa Clara, CA) tuned to 940 nm (~30 mW average power on the 910 sample) for imaging. Images (512 × 512 pixels, field of view 550 × 550 μ m) were acquired 911 with a resonant scanner at a frame rate of 30 Hz using a 16 × 0.8 NA water-immersion 912

(Nikon). Using a piezo motor (PI In struments, UK) connected to the objective, a 913 volume of ~300 μ m was divided into 6 planes resulting in an effective volume repetition rate 914 of ~5 Hz. The odour port was adjusted to approximately 0.5 cm away from the ipsilateral 915 nostril to the imaging window, and a flow sensor (A3100, Honeywell, NC, USA) was placed 916 to the contralateral nostril for continuous resp iration recording and digitised with a Power 917 1401 ADC board (CED, Cambridge, UK). 918 Odour stimuli 919 For this experiment a total of 752 trials were planned to be presented for each mouse. The 920 plumes used were recorded in the wind tunnel during the same session and were picked 921 from the centre of the distributions of plume features as shown in Fig. S2.3. We presented 922 40 plumes for each of the 4 distances using 2 main odours (EB and 2H), resulting in 320 923 “core” trials that were randomised before bei ng added to the trial bank. The next 384 trials 924 consisted of the same core 320 trials, together with an additional 8 more repeats of one 925 plume/distance/odour, which were again independently randomised and added to the trial 926 bank. Finally, the last 48 trials were 3 repeats of one plume/distance/odour for an additional 927 4 odours, also randomised. This trial order was then presented to each mouse, in blocks of 928 40 trials, in order to facilitate data acquisiti on. Odour delivery was triggered by inhalation 929 using a threshold-crossing algorithm. The delay between trigger and odour delivery was the 930 same for all plumes at 65ms. 931 Imaging data analysis 932 For MTC imaging, motion correction, segment ation and trace extraction were performed 933 using the Suite2p package (https://github.com/MouseLand/suite2p and (Pachitariu et al., 934 2016). Putative neuronal somata were automatically identified by segmentation and curated 935 manually. Fluorescence signal from all pixels within each ROI was averaged and extracted 936 as a time series. Δ F/F = (F − F0)/F0, in which F is raw fluorescence and F0 is the median of 937 the fluorescence signal distribution. Soma and neuropil fluorescence traces were extracted 938 and neuropil fluorescence was subtracted from t he corresponding soma trace. All further 939 analysis with custom written scripts in Python 3 and JupyterLab. 940 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 25 Responsive cells / cell-odour pairs were defined as cells/cell-odour pairs for which the mean 941 fluorescence during the response window for all trials of an individual odour was above 942 mean + 3*standard deviation of the period 3 s before stimulus onset. 943 Distance sensitive cells / cell-odour pairs were defined as cells showing a significantly 944 different (one way ANOVA) response to one distance category of the four presented. 945 Multiple comparisons arising from these test s were controlled using the Benjamini-Hochberg 946 false-discovery-rate (FDR) procedure with α = 0.05. 947 Unless otherwise specified, linear classifiers used to understand the ability of the olfactory 948 bulb to represent distance-dependent plumes were all LinearSVC, as implemented in the 949 scikit-learn package (https://scikit-learn.org/). Data was split into a training set and a test set 950 using a 75:25 split stratified by category (dis tance). Values were then scaled using Standard 951 Scaler. C parameter was set to 0.001. 952 Respiration data analysis 953 Respiration data was extracted using Spike 2 (CED, Cambridge, UK) and further analysed in 954 Python 3. Inhalation was performed using a threshold crossing approach. Traces were 955 normalised within one experiment to the maximum value over the length of the experiment, 956 then segmented into inhalation (above threshold) and non-inhalation (below threshold) 957 phases. This segmentation was used to creat e a binary signal that was overlaid on the 958 plume traces to calculate the “inhaled plume” and features that took account of inhalation. 959 Odour delivery 960 Odours used 961 All odours were obtained in their pure form from Sigma-Aldrich, St. Louis MO, USA. Odours 962 used: isoamyl acetate (IAA), 1-heptanal (1H), nonanoic acid (NA), α -terpinene (AT), ethyl 963 butyrate (EB), 2-heptanone (2H), benzyl acetate (BA). Odours used in the wind tunnel were 964 presented in their pure form (EB, IAA). For imaging experiments, odours were diluted 1:50 in 965 mineral oil and a further 1:2 in air, resulting in ~1:100 or 1% concentration reaching the 966 animal. 967 Wind tunnel 968 To produce olfactory stimuli with similar tem poral structures as those measured outdoors 969 (Ackels et al., 2021), a wind tunnel with near-laminar flow was developed. An obstacle 970 placed in the airflow introduced controlled complexity to the airflow. The wind tunnel was 971 square in cross-section and measured 165 cm (L) x 65 cm (W) x 62 cm (H). The frame was 972 constructed from aluminium profiles (MayTe c Aluminium Systemtechnik GmbH, Dachau, 973 Germany) and walled with clear acrylic panels. The floor of the tunnel was a laboratory 974 bench placed 5cm lower than the odour port that mice used for sampling. The obstacle was 975 a plastic cylinder (height: 12.7 cm, diameter: 7.6 cm), positioned on the midline 15 cm 976 downstream of the fan, inspired by the routine use of cylindrical obstacles to generate airflow 977 turbulence in fluid dynamics research (Béra et al., 2000). 978 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 26 Wind generation was achieved by a centrifugal blower (D2E133-AM47-23, ebm-papst) 979 placed in the inlet section of the tunnel, with the rear end being open for air to exit the tunnel. 980 The room ventilation system had its outlet placed at the back of the room, above and slightly 981 behind the mouse/PID end of the wind tunnel. To better control the airflow, we used a 982 honeycomb structure (Hexagonal Aluminium W/CRIII Coating 3.0 3/8 .002N 5052, 10 cm in 983 thickness, Texas Almet) placed between the fan and the tunnel, to create a largely laminar 984 airflow. This ensured that the airflow complexity we observed was primarily created by the 985 obstacle. Neutrally buoyant helium-filled soap bubbles (Sage Action, Inc.) were used to 986 visualise airflow, as filmed with a high-speed camera (Basler). 987 Distance Odour Delivery Device (dODD) 988 The distance Odour Delivery Device (dODD) was used to deliver odours into the airflow in a 989 passive fashion. The device of choice was a custom-built ensemble of an Arduino controlled 990 servo motor (TowerPro SG-5010, Adafruit) that rotates an aluminium lid covering a glass 991 petri dish (5cm diameter). Eight such devices were placed inside the wind tunnel, and 992 controlled via a compact data acquisition (cDAQ) device (National Instruments) and either 993 PulseBoy or the AutonoMouse software (Ackels et al., 2021; Erskine et al., 2019, 994 https://github.com/RoboDoig/PulseBoy and https://github.com/autonomouse-control). This 995 also allowed for integration with the temporal ODD (Ackels et al., 2021), which was adapted 996 and integrated into the system for the imaging experiments. Before each recording session 997 the petri dishes were filled with 10 ml ethyl butyrate (EB), which was the odour used for all 998 plume experiments. 999 Temporal Odour Delivery Device (tODD) 1000 Reproduction of recorded odour plumes was achieved with a temporal Odour Delivery 1001 Device (tODD) and custom control software, as previously described in Ackels et al., 2021. 1002 In brief, the odour delivery device was based on a modular design of four separate odour 1003 channels, and consisted of an odour manifold for odour storage, a valve manifold for control 1004 of odour release and hardware for controlling and directing airflow through the system. We 1005 used high-speed micro-dispense valves with cu stom electronics for pulse-width modulation 1006 to maximise bandwidth. Two manifolds of four valves were joined together using a three-way 1007 connector (TMMA3203950Z, The Lee Company) to create an eight-channel device. Pulse 1008 profiles for calibration and stimulus production were generated with custom Python software, 1009 allowing us to define pulse parameters across multiple valves using a graphical user 1010 interface. The drivers themselves were passed valve opening times via a 5 V TTL pulse from 1011 digital I/O controls via a compact data acquisition (cDAQ) device (National Instruments). 1012 Odour plume recordings 1013 The temporal structure of odour plumes was recorded using a mini-PID photoionization 1014 detector with a bandwidth of ~330 Hz (200B miniPID, Aurora Scientific, Aurora ON, Canada) 1015 and Spike2 software (Cambridge Electronic Design), sampled at 1 kHz. The tip of the 1016 detector was placed in the behaviour chamber at the centre of the odour port, to mimic the 1017 position of the nose of a mouse performing the task. For each set of recordings, the odour 1018 source was alternated between 4 different positions (near or far, left or right) inside the wind 1019 tunnel. The PID was calibrated periodically by adjusting the gain and baseline such that it 1020 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 27 would produce the same response to a series of square pulses of known amplitude (1000 1021 ppm, 100 ppm, 10 ppm isobutylene). Thus, we could compare broad ranges of concentration 1022 over different recording sessions 1023 Reproduction of recorded plumes 1024 Reproduction of recorded odour plumes was achieved using pulse-width modulation (PWM) 1025 generated with custom Python software (PyPulse, PulseBoy, daqface; 1026 https://github.com/RoboDoig and https://github.com/warnerwarner) as previously described 1027 in Ackels et al., 2021. Briefly, this was ac hieved by mapping odour signal amplitudes from 1028 recorded plume data to PWM duties, converting a time series of amplitudes to a time series 1029 of PWM duties. These duties could then be used as the input to an odour valve to reproduce 1030 the original signal. First, each PID trace was corrected to baseline drift by performing 1031 baseline subtraction, as compared to a 1 s period before each trial. The trace was then 1032 normalised to between 0 and 1, as set by the target_max variable in PulseBoy, and then 1033 converted into a series of binary opening and closing times, at a frequency of 500 Hz. To 1034 reproduce the plume using the whole dynamic range of the olfactometer, target_max would 1035 be set to 1. The length of the openings and closings relate directly to the value of the 1036 normalised signal, a value of 1 translates to a continuous opening, and a value of 0 1037 translates to continuously closed. In an offline assignment, target_max values were adjusted 1038 to group ratios and within group ratios to match mean concentration over the whole plume. 1039 Thus, there was variability in the overall conc entration presented, but that variability was 1040 largely within groups of plumes from the sa me distance, not between different distances. 1041 Generally, wherever an odour position is inactivated a blank position was activated to 1042 compensate for flow change. Thus, an inverse sequence of commands as used for the 1043 plume generation, termed “anti-plume”, was si multaneously fed to a valve connected to a 1044 mineral oil (blank) channel to produce the co mpensatory flow. This reproduction strategy 1045 allowed to precisely control stimulus onset and to fully equalise mean concentration, though, 1046 crucially, preserving variability within groups (Fig. 2H, Fig S2.11, S2.12). 1047 Plume analysis 1048 Plumes were recorded using Spike2 (Cambridge Electronic Design), sampled at 1 kHz. Trial 1049 onset was calculated using a TTL pulse passed through an additional channel on the 1050 acquisition board (Cambridge Electronic Desi gn). Further analysis was then performed 1051 initially in MATLAB and Python 3 using Jupyter Lab. Briefly, traces were segmented using 1052 the trial onsets and baseline subtraction performed using the 1 s prior to trial onset. Plume 1053 features were defined and calculated as follows: Correlation was calculated using the corr 1054 function in MATLAB or DataFrame.corr(method=’pearson’) in Python 3 using the pandas 1055 package. Plume arrival time was calculated as the time of the first point above 3*SD of the 1056 baseline (1s interval before trial onset). Mean odour concentration was calculated as the 1057 average P ID signal over a certain time window as specified. Maximum peak was calculated 1058 as the maximum value of the PID signal over a certain time window as specified. Number of 1059 peaks was calculated using the find_peaks function (scipy.signal.find_peaks) in Python 3, 1060 using the parameters height = 0.05 and prominence = 0.1 for the reproduced plumes, and 1061 height = 0.01 and prominence = 0.01 for the recorded plumes. Maximum prominence was 1062 extracted from the outputs of the find_peaks function above. Standard deviation, Variance, 1063 Kurtosis and Skewness were calculated using the corresponding functions in the scipy.stats 1064 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint 28 library for different time windows as specified. Reproduced plume integral and Odour inhaled 1065 were calculated using the integrate.trapz function in scipy. Cumulative odour presented and 1066 Cumulative odour inhaled were calculated using the cumsum function in numpy. 1067 Statistical analysis and data display 1068 To test for statistical significance between individual groups where appropriate we used 1069 either paired or unpaired student t-tests or, for non-parametric data, the Mann–Whitney U 1070 test, or the Kolmogorov-Smirnov test for the equality of probability distributions. Statistical 1071 test details and P values are provided in figures and/or legends. 1072 Data figures were plotted using Jupyter Lab using seaborn, pandas, sci-kit learn or matplotlib 1073 functions in Python 3. Unless specified otherwise, boxplots depict the median as a thick line 1074 and default maximal whisker length of 1.5 × (q3 − q1) where q3 and q1 indicate the 75th and 1075 25th percentile, respectively. If points were located outside this whisker range, they were 1076 displayed individually as outliers. Unless ot herwise specified, line plots depicting animal 1077 performance show mean (thick line) ± SD (shaded area) for a rolling window of blocks of 100 1078 trials. 1079 Data availability 1080 Source data supporting the graphs presented in the main figures are provided as a 1081 repository from https://zenodo.org/records/15390336. 1082 Code availability 1083 Analysis code to recreate the main figures is available from https://github.com/ackels-1084 lab/Odour_distance_paper. All custom analysis code supporting the findings of this study will 1085 be made publicly available through the same code repository upon publication. 1086

1087 This work was supported by the Francis Crick Institute which receives its core funding from 1088 Cancer Research UK (FC001153), the UK Medical Research Council (FC001153), and the 1089 Wellcome Trust (FC001153); a Wellcome Trust Investigator (110174/Z/15/Z) grant and the 1090 NeuroNex program “From Odor to Action” to A.T.S., a BIF doctoral fellowship to A.C.M., and 1091 a DFG postdoctoral fellowship to T.A. It was further supported by the German Research 1092 Foundation (FOR5424 “Modolfor”, A.T.S and T.A) and the European Union (ERC, 1093 “TempCOdE”, 101077017, T.A.). Views and opinions expressed are, however, those of the 1094 author(s) only and do not necessarily reflect those of the European Union or the European 1095 Research Council. Neither the European Union nor the granting authority can be held 1096 responsible for them. 1097 .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 17, 2025. ; https://doi.org/10.1101/2025.05.14.653752doi: bioRxiv preprint