Sniff-invariant intensity perception using olfactory bulb coding of inhalation dynamics

For stable perception of odor intensity, there must exist a neural correlate that is invariant across other parameters, such as the highly variable sniff cycle. Previous hypotheses suggest that variance in inhalation dynamics alters odor concentration profiles in the naris despite a constant environmental concentration. Using whole cell recordings in the olfactory bulb of awake mice, we directly demonstrate that rapid sniffing mimics the effect of odor concentration increase at the level of both mitral and tufted cell (MTC) firing rate responses and temporal responses. In contrast, we find that mice are capable of discriminating concentrations within short timescales despite highly variable sniffing behavior. We reason that the only way the olfactory system can differentiate between a change in sniffing and a change in concentration is to receive information about the inhalation parameters in parallel with information about the odor. While conceivably this could be achieved via corollary discharge from respiration control centres, we find that the sniff-driven activity of MTCs without odor input is informative of the kind of inhalation that just occurred, allowing rapid detection of a change in inhalation. Thus, a possible reason for sniff modulation of the early olfactory system may be to inform downstream centres of nasal flow dynamics, so that an inference can be made about environmental concentration independent of sniff variance.

2 For consistency of perception, sensory systems must be able to stably encode the same perceptual 25 features across a wide range of situations. An example of this is the encoding of object size 26 independent of object distance in the visual system (Helmholtz, 1867) -we do not perceive a giant 27 apple when viewed at close range, and similarly we do not misperceive buildings as tiny objects when 28 viewed from great distance. In the olfactory system, studies have looked at how odor identity may be 29 the piriform cortex (Bolding and Franks, 2017). This is thought to arise since OSNs will depolarise to 45 threshold more quickly when the concentration profile in the naris is steeper. 46 In awake mice, sniffing behaviour is in continual flux (Kepecs et al., 2007;Welker, 1964; Wesson et al., 47 To determine this, we first analysed 13 cell-odor pairs with early excitatory responses recorded in 102 passive mice where only a single concentration stimulus (1% saturated vapour pressure) was 103 presented to the animal across trials. Comparing the FR response over the first 250 ms for 'fast' sniff 104 trials (>70 th percentile peak inhalation slopes), and 'slow' sniff trials (<30 th percentile), it was apparent 105 that faster inhalation could cause a latency advance of the excitatory burst ( Fig. 2A-B). Across all cell-106 odor pairs, faster inhalation caused a significant latency reduction in mean response onset (latency 107 change (fast-slow) = -16 ± 14 ms, p = 0.002 paired t-test between onsets for slow and fast inhalations; 108 8 reductions for pMCs were greater than reductions for pTCs (pMCs: latency change = -30 ± 7 ms, p = 123 7x10 -4 , paired t-test, n = 5 cell-odor pairs; pTCs: latency change = -8±10 ms, p = 0.08, paired t-test, n = 124 8 cell odor pairs; pMCs vs pTCs: p = 0.001, unpaired t-test; Fig. 2F-G), and this was largely because 125 pTCs already tended to respond with shorter latency during slow sniffs than pMCs (pTC onset median 126 = 75 ms, IQR = 55-90 ms; pMC onset median = 110 ms, IQR = 80-113 ms, p = 0.13, Ranksum; Fig. 2E). 127 Thus, the temporal shifts and cell-type specificity in the effect of faster sniffing matches that previously 128 described for concentration increases in anaesthetized mice (Fukunaga et al., 2012). 129  about sniff parameters in order to determine the concentration. Thus, even on short timescales, a 150 more rapid inhalation mimics concentration increases at the level of the olfactory bulb output. 151  that the discrimination task was not trivial (Fig. 5B). Thus, shifts in perceived concentration should be 202 overtly seen in the performance curves. To test this, we first separated trials according to whether the 203 first sniff was fast (<30th percentile inhalation duration) or slow (>70 th percentile) (Fig. 5D). This 204 resulted in a comparison of trials between which the difference in the inhalation duration matched or 205 exceeded that used in the whole cell recordings when comparing fast and slow sniff trials (Fig. 5E). 206

Mice can successfully discriminate concentrations on rapid timescales
Recalculating performance curves for each subset, there were no large or significant differences in the 207 performance curves for mice performing on either contingency ( Fig. 5F; p>0.01 paired t-tests). 208 Secondly, on a small selection of trials for 5 of the mice, the puff stimulus (as used during the 209 physiological recordings) was used to evoke fast sniffs, including the first inhalation (Fig. 5G). The 210 mean changes in first inhalation duration evoked by this puff were again highly comparable to that 211 used for analysis of fast and slow sniffs in the physiological data (Fig. 5H). While this had a minor but 212 insignificant effect on error rate likely owing to distraction (percent correct: control trials = 83 ± 8%, 213 probe trials = 77 ± 9%, p = 0.16 paired t-test, n = 5 mice), there were remarkably no gross differences 214 in the performance curves compared to control trials (p>0.01, paired t-tests; Fig. 5J). Finally, when 215 separating trials for each concentration according to the response of the mouse (either 'go' or 'no 216 go'), there was no overt differences in first inhalation between go and no-go trials ( Supplementary 217 Fig. 6). 218 Given that we have only considered the first sniff cycle, it is possible that mice take another sniff prior 219 to making a decision if the initial sniff was fast and gave rise to ambiguity about concentration. This 220 would be reflected in longer reaction times for fast compared to slow first sniff trials. Comparing trials 221 14 222 Thus, variable sniffing appears to have no overt negative impact on concentration perception. 234 235 We have so far shown that it is difficult to distinguish the effect of a change in inhalation or a change 236 in concentration via their effects on MTC responses (Fig. 1-3), yet mice are perfectly capable of fine 237 concentration discrimination in the face of fluctuating inhalations (Fig. 5). We thus conjecture that the 238 (F) Go rate as a function of concentration when splitting trials according to duration of first inhalation as in D. Dotted line shows mean go rate for sniffs with inhalation between 30 th and 70 th percentile. (G) Example sniff traces for one animal for a puff trial (a trial in which an air puff to the flank was used to evoke fast sniffing) and an adjacent control trial of the same odor. (H) Comparison of changes in mean first inhalation duration for physiological analysis (Fig. 3, n = 20) and for puff vs control trials during behaviour (n = 7 mice while others showed decreasing spike count (Fig. 6A2-B2) and more hyperpolarising membrane 256 potential (Fig. 6C2). 24% of cells showed significant relationships between spike count and inhalation 257 duration (p<0.01, linear regression; Fig. 6D) compared to only 3% in shuffle controls. R 2 for the actual 258 correlations were also significantly higher than for shuffle controls (actual R 2 median = 0.54, IQR=0.17-259 0.82; shuffled median=0.18, IQR=0.04-0.45, p=1x10 -4 , Ranksum, n=41 vs 369; Fig. 6D). Similarly, 22% 260 showed significant correlations with mean membrane potential compared to 2% of shuffle controls 261 (p<0.0, linear regression; Fig. 6D), with R 2 values for the actual data being significantly higher than for 262 shuffled data (actual R 2 median = 0.56, IQR=0.20-0.73; shuffled median=0.18, IQR=0.04-0.41, p=9x10 -7 , 263

Mitral and tufted cells respond to inhalation changes in cell type specific ways
Ranksum, n=41 vs 369). Timing of activity was also often linearly correlated with inhalation 264 17 265  showed a significant relationship (p<0.01) between inhalation duration and at least one or more of 269 these activity parameters (Fig. 6E). Thus, phase locking, which likely relates to MC and TC phenotype, determines how a cell will respond 295 to changing sniff parameters in absence of odor. Thus the large population of cells that are not directly 296 involved with the encoding of odor information could instead be utilised to encode the parameters of 297 each inhalation. 298 299 We next sought to determine whether we could read out changes in inhalation from the spiking 300 activity of cells in absence of odor, as a proxy for cells that are not responding directly to the odor. 301

Inhalation change can be detected and decoded from MTC spiking on rapid timescales
We first wanted to determine how rapidly a change in inhalation could be detected. For all cells with 302 enough sniff variation (>50 sniffs in each inhalation duration category), we calculated sequences of 303 spike histograms for different inhalation durations using random subsets of sniffs within each group 304 ( Fig. 8A; see methods). We constructed either a sequence with PSTHs calculated from three 305 consecutive sniffs of 95 ms inhalation duration, or a sequence with PSTHs calculated from 2 306 consecutive sniffs of 95 ms, with the last PSTH instead constructed from 55 ms inhalation duration 307 sniffs (Fig. 8A). Using these, it was possible to determine a change in inhalation duration (95 to 55 ms 308 inhalation duration) within only 70 ± 12ms by calculating Euclidean distances between constructed 309 population vectors of the two different sequences (Fig. 8B see methods). Smaller changes in inhalation 310 duration (95 to 75 ms) could also be detected on similarly rapid timescales (Supplementary Fig. 7). slow inhalation cycles could be achieved within 130 ms after inhalation onset (Fig. 8D). 320 Thus, even for relatively low numbers of neurons, mitral and tufted cell activity -in absence of odor 321 input -is informative of the inhalation that just occurred, such that non-odor responsive cells could 322 be utilised by the olfactory system to distinguish sniff changes versus concentration changes. 323

324
For stable perception, sensory systems must find ways of encoding of stimulus features independent 325 of fluctuating sampling behavior. Here we show that faster sniffs can evoke response changes in the 326 olfactory bulb that appear indistinguishable from those caused by increasing concentration (Fig. 1-3), 327 yet mice are highly capable of perceiving concentration on fast timescales, regardless of sniffing 328 parameters (Fig. 4-5). We reason that the only way the olfactory system can distinguish these two 329 occurrences is via information about the kind of sniff that just occurred. This could potentially be 330 achieved through corollary discharge from a motor circuit involved with breathing rhythms (such as 331 the pre-bötzinger complex). However we find that single MTC activity already correlates with 332 inhalation duration (Fig. 6), and that this is likely generated from feed-forward input in a cell type 333 specific way (Fig. 7), allowing inference about the kind of sniff that just occurred on a rapid timescale 334 (Fig. 8). Thus, the olfactory bulb itself does not appear to be the site where the sniff-invariant percept 335 of intensity is generated, but does appear to already contain the information needed to generate the 336 percept elsewhere. 337 Given the timescale of decision making for concentration (Fig. 4), it seems likely that the information 338 used by the mouse is the fast timescale temporal shifts in excitation that have been previously changes can indeed mimic the effect of concentration change at the level of both firing rates (Fig. 1), 351 and temporal shifts in spike activity (Fig. 2-3). This is not to say that OSN input is perfectly matched 352 when we compare faster sniff rates and higher concentration. In fact, since subthreshold inhibition is 353 greater for the higher concentration ( Supplementary Fig. 3 are changed, a feature which is widespread throughout MTCs (Fig. 6). We thus reason that a primary 368 function of sniff modulation is to inform the olfactory system of what kind of inhalation took place, 369 such that a change in concentration and a change in sniffing are distinguishable. Congruently we find 370 that inhalation parameters can indeed be readily and rapidly inferred from the spiking activity of MTCs 371 (Fig. 8). 372 24 Encoding of 'sniff effort' has been hypothesized previously when psychophysics showed that humans 373 could categorise concentrations well despite large changes in inhalation flow rate (Teghtsoonian et 374 al., 1978). Airway resistance is subject to continual changes, and even differs between the two nostrils 375 (Principato and Ozenberger, 1970;Sobel et al., 1999), which will naturally result in varying nasal flow 376 rates for identical respiratory motor commands. Previous work has shown that sniff modulation of the 377 olfactory bulb is generated peripherally rather than centrally, since blocking the naris abolishes sniff 378 modulation in the olfactory bulb (Margrie and Schaefer 2003). Thus reafference using mechanoceptive 379 encoding of sniff pressure, rather than efference copy of the motor commands (which would require 380 constantly updated internal models to calculate the effect on airway flow for each nostril) may be the 381 optimal strategy for encoding inhalation parameters. This could be the reason that olfactory receptors 382 evolved to respond to pressure changes as well as olfactory stimuli (