Olfactory investigation in the home cage

We have developed a behavioral paradigm to study volitional olfactory investigation in mice over several months. We placed odor ports in the wall of a standard cage that administer a neutral odorant stimulus when a mouse pokes its nose inside. Even though animals were fed and watered ad libitum , and sampling from the port elicited no outcome other than the delivery of an odor, mice readily sampled these stimuli hundreds of times per day. This self-paced olfactory investigation persisted for weeks with only modest habituation following the first day of exposure to a given set of odorants. If an unexpected odorant stimulus was administered at the port, the sampling rate increased transiently (in the first 20 min) by an order of magnitude and remained higher than baseline throughout the subsequent day, indicating learned implicit knowledge. Thus, this system may be used to study naturalistic olfactory learning over extended time scales outside of conventional task structures.

We devised an apparatus that operationalizes self-paced, volitional odor investigation in the home cage over long periods of time (weeks to months).We assembled a custom mouse cage to include a bank of odor ports in the wall (Fig. 1A).Odorized air is delivered under closed loop whenever the animal's snout breaks an infrared beam inside of the port.The animal eats and drinks ad libitum and the only motivation to sample the port is delivery of the odor stimulus.This system allows continuous measurement of odor investigation under ordinary living conditions and in which the animal is left undisturbed with complete control over its experience with the odor stimuli.Animals were clearly comfortable with this altered environment, often constructing their nests adjacent the odor ports and readily sampling from them.This system thus permits longitudinal observation of self-initiated odor investigation.

Results
We found that mice readily sampled odorant stimuli administered from odor ports in their home cage.We measured 43,868 odor sampling E-mail addresses: af2243@columbia.edu(A.J.P. Fink), ces2001@columbia.edu(C.E.Schoonover).events (nose pokes, "samples") from 8 mice over 14 days.This selfinitiated odor investigation followed a diurnal cycle with a peak in sampling rate during the animals' dark cycle (Fig. 1B and C).This behavior did not require administration of innately appetitive odorants, such as those associated with food or conspecifics.Rather, animals explored otherwise 'neutral' mono-molecular odorants hundreds of times per day (Fig. 1D, 392 ± 176 samples/day, N = 8 mice, mean ± s. d.).Therefore, similar to prior observations in rats over far shorter time scales (Long & Tapp, 1967), mice eagerly investigate odorant stimuli when afforded the opportunity.Importantly, the system permits precise measurement of odor sampling.Daily sampling rates varied considerably across animals (mean sampling rate ranged from 88 to 571 samples per day across N = 8 mice).Although frequent, the total duration of sampling events accounted for only 355 ± 313 s per day (0.41 ± 0.36% of total time in the apparatus, 43,868 pokes in N = 8 animals over 14 days, mean ± s.d.).This is because the duration of each sampling event was typically very brief, with a median of 83 ms (Fig. 2A)-on par with a single investigatory sniff (Findley et al., 2021).We fit sampling durations with a double lognormal distribution (modes 33.5 ms and 298 ms, R 2 = 0.95).This suggests that the animals employ at least two types of sampling.Measurements of odor kinetics with a photoionization detector showed that sufficient odor is emitted by the port to permit odor sampling even for short (30 ms) beam breaks (26 ms decay time constant following the offset of beam break).Furthermore, videography of the sampling behavior indicated that beam breaks corresponded to olfactory sampling behavior (active sniffing) as opposed to a motor behavior unrelated to olfactory sampling (e.g.forelimb in the odor port or other non-olfactory actions).More than 99% of beam breaks were triggered by animal's nose entering the port as opposed to another part of the body or a foreign object, such as bedding.Videos 1 and 2 show examples for long and short duration modes of sampling, respectively.
Notably, mice did not sample the port uniformly in time, but rather exhibited pronounced burstiness on a timescale of minutes, as evidenced by the autocorrelogram of sampling times (Fig. 2B).The decay from the peak of the sampling time autocorrelogram was well fit by a double exponential function (decay time constants of 14.3 s and 6.0 min, R 2 = 0.97).The inter-sample intervals were well captured by a mixture of two lognormal distributions (R 2 = 0.99) (Fig. 2C), suggesting that mice investigate in bouts of multiple samples.We defined a bout as a series of one or more samples separated from other samples by at least 20.mean ± s.d., Fig. 2C, inset).Samples within a bout were tightly spaced in time (Fig. 2D, within-bout inter-sample interval = 3.9 ± 3.9 s, mean ± s. d.) and on average, each bout consisted of 4.5 ± 8.0 samples (mean ± s. d.),This approach therefore affords precise quantification of odor sampling across many thousands of self-initiated sampling events.Although we did not take advantage of this here, these experiments can include simultaneous measurement of sniffing (McAfee et al., 2016;Mutlu et al., 2018).
We tested whether this volitional behavior requires the presence of an odor stimulus, or rather was simply a consequence of multimodal investigation of the odor port itself.We compared the sampling behavior of mice in which a nose poke resulted in the delivery of odorized air to the behavior of a control group (N = 4 mice) in which a nose poke resulted in the delivery of air alone, without an odorant stimulus.We found that in the presence of an odorant, the sampling rate was more than 5-fold greater (P < 10 − 5 , Wilcoxon rank-sum test) than in the presence of air alone (Fig. 3A), with only modest changes in sample duration and inter-sample interval (Fig. 3B).Absent odor, the distribution of inter-sample intervals was bimodal with a greater fraction of extended periods without a sample when odors were not presented (Fig. 3C bimodality coefficient (Freeman & Dale, 2013) for odor, 0.44, vs. air alone, 0.66, values greater than 0.56 indicate bimodality).Therefore, mice preferentially explore odor stimuli, consistent with previous reports (Tapp & Long, 1968).
Finally, we took advantage of the continuous nature of our measure of olfactory investigation to study how this behavior evolved over time.We found that animals persistently investigated the odor stimuli over a period of two weeks (Fig. 4A).We observed a modest decrease in sample frequency over time: animals sampled on average 597 ± 347 times per day over the first 24 h and 376 ± 257 times per day across days 2 through 14 (Fig. 4B, left, P < 0.05 Wilcoxon rank-sum test, mean ± s.d.).We note that this habitation in sampling likely does not only reflect habituation to the odor stimuli.We observed a similar rate of habituation for animals in the air only configuration (Fig. 4C), indicating that there is habituation of sampling rate to the contingencies of the environment itself, independent of odor.This initial habituation to odor across the first 24 h saturated: we observed near zero (r = − 0.078, P = 0.44) decrement in sampling rate across days 2 through 14 (Fig. 4B, right).Therefore, this approach permits continuous measurement of odor sampling over extended time periods (weeks to months).
We had anecdotally observed that mice intensely sample the odor port immediately following replenishment of odorant dilutions.This raised the possibility that animals can report changes they detect at the odor port by increasing their sampling rate.In a separate experiment, we first presented a different cohort of mice a single odor stimulus at a single port.Then, following 4 days with that first odor, we replaced the stimulus with a novel, unexpected molecule.We measured the animals' rate of sampling following their first sample after the odor had been switched.As only a single port was included in the home cages for these experiments, all comparisons of sampling rate before and after odor switch are between sampling at the same odor port.The animals dramatically increased their sampling (Fig. 5A): over the ensuing 20 min animals sampled 18.7-fold more frequently relative to their average sampling rate across the duration of the experiment (Fig. 5A, right, P < 10 − 5 ; maximum of acute, mean population fold increase over average sampling rates: 37.0, N = 90 mice).77% of mice (70 out of 90 animals) increased their sampling rate by greater than 5-fold.We also found that individual animals' baseline sampling rates before odor switch were predictive of the increase in sampling rate following odor switch (Fig. 5B, left, r = 0.65, P < 10 − 5 ).We then asked whether baseline sampling rate was predictive of the fold change in sampling rate (sampling rate after odor switch divided by sampling rate before odor switch).However, we found no dependence, indicating that the rate at which an animal samples the odor stimulus at baseline does not predict the relative change in sampling rate following odor switch (Fig. 5B, right, r = − 0.09, P = 0.41).In addition, mice maintained an elevated sampling rate for a prolonged period following the switch, still sampling 4.8-fold more frequently in the 12 to 16 h window following odor switch (Fig. 5C, P < 10 − 5 ).Thus, animals increase their investigation of odor stimuli that are not predicted by past experience.
In summary, animals repeatedly sample odors over long periods of time with marginal habituation after the first day, permitting long term measurement of volitional odor investigation.Moreover, mice report changes in the odor stimulus, demonstrating learned implicit knowledge of the odor stimuli.

Discussion
We have developed a naturalistic behavioral assay that permits precise quantification of self-initiated olfactory investigation in the home cage.We found that mice sample odor ports placed in the walls of their home cages hundreds of times per day, in bouts of multiple samples; that this sampling is not contingent upon the pairing of these odor stimuli with extrinsic reward such as food or water; that the presence of an odor, rather than odorless air, promotes greater than 5-fold more frequent sampling; and that this self-initiated olfactory investigation persists for weeks.
Finally, we found that the animals' sampling rate increases markedly after they encounter an unexpected odor cue, both acutely in the minutes following detection of the event, and in a persistent manner over the subsequent 24 h.This constitutes to our knowledge the first report of a transition to such an extended investigatory state; it was observable Fig. 3. Animals sample odor stimuli at higher rates than air alone.A, left, Sample counts per day, for every animal, port, and day and right, cumulative distributions across all animals, ports, and days, for odorized (red, replotted from Fig. 1D) and non-odorized (blue) air.Points, individual sampling counts per animal per day.Crosses, mean and 95% confidence interval across all animals and all days.P < 10 − 5 , Wilcoxon rank-sum test.B, sample statistics for odorized (red, replotted from Fig. 2) and non-odorized (blue) air.C, Left, distributions of inter-sample intervals for odorized (red, replotted from Fig. 2) and non-odorized (blue) air and right, cumulative distributions.Arrowhead indicates second peak in distribution for non-odorized air.
only because the assay permits continuous, 24-hour observation in contrast to traditional session-based paradigms.This long-term change suggests that the animal's overall drive to investigate the odor port may scale with uncertainty about its content.This raises the possibility that the sustained motivation animals exhibit for investigating the port may not be to gain additional information, so much as to continually assess whether the world continues to conform to the mouse's understanding of it.
This approach produces a continuous, long-term measure of odor sampling, providing a complete record of a subject's volitional engagement with a set of olfactory cues.The paradigm therefore permits the longitudinal study of exploratory behavior in a simple and readily parameterizable domain.In future longitudinal studies, the final line leading to the odor port may be split to permit measurement of the odor pulse during every sample with a photo-ionization detector.This would control against any possible depletion in odor concentration or cross odor contamination.Since animals clearly report detection of unexpected events, this system can also be used to study learning outside the context of a conventional task or trial structure.By tapping into the animal's natural drive to explore its environment, this approach dispenses with the requirement of extrinsic reinforcement such as food or water rewards.This positions the assay to access forms of learning-and their underlying mechanisms-that have traditionally been understudied in psychology and neuroscience.

Methods
All procedures were approved by the Columbia University Institutional Animal Care and Use Committee (protocol AC-AABH6557) and were performed in compliance with the ethical regulations of Columbia University as well as the Guide for Animal Care and Use of Laboratory Animals.The data in this study were taken from N = 98 male C57BL/6J mice (Jackson laboratories, Bar Harbor, ME) divided into two cohorts.For experiments described in Figs.1-4 we used 8 mice (age 32.6 ± 1.5 weeks, mean ± s.d.) and for experiments described in Fig. 5 we used 90 mice (age 9.3 ± 1.2 weeks, mean ± s.d.).
We modified standard mouse home cages (Allentown, LLC, Allentown, NJ, NexGen Mouse 500 cage) by cutting an opening in the side that could accommodate one or several odor ports.We 3D printed a port fixture that was inserted inserted into this opening.The ports were standard ports from Sanworks (Sanworks, LLC, Rochester, NY, Mouse Behavior Port, 2.2 cm diameter, Product ID: 1002) equipped with a IR beam, light sensory, and valve.When the IR beam was broken the state of the valve was changed.For the design employed in this study for data shown in Figs.1-4 a bank of four ports were used, allowing us to administer four odor stimuli.For the data shown in Fig. 5 we used a single port that delivered a single odor.Animals were acclimated to the reversed light/dark cycle for at least two weeks and then introduced to the modified home cages.Data collection began immediately upon placing animals in the modified home cage.
Odors were prepared and delivered using standard methods as previously described (Fink, Axel, & Schoonover, 2019;Schoonover et al., 2021).Briefly, monomolecular odorant stimuli (2% octanal, 2% cis-3hexen-1-ol, 4% 5-methyl-5-hexen-2-one, 2% isopentyl acetate) were dissolved in dipropylene glycol to the concentrations specified.They were then placed in bottles, as previously described (Fink, Axel, & Schoonover, 2019;Schoonover et al., 2021).The bottles were continuously bubbled with air at approximately 0.25 L per minute, with the air from the bottles routed to exhaust through the valve at the nose port.If the beam was broken, the valve would change state and the odorized air stream would be routed to a metal tube within the noseport so that that animal could sample the odor in the cage.When the beam was restored, when the animal removed its nose from the cage, the odorized air stream was routed once again to exhaust (vacuum line outside the cages).On very rare occasions (≪1 % of beam breaks) cage bedding became stuck in the odor port, triggering an extended beam break.Therefore, beam breaks longer than 10 s were discarded from the dataset.Future versions of the assay will include closed loop video to track the head of the mouse in order to ensure that odors are delivered only when the mouse's head is at the port.The cages were kept in ventilated racks and the odor in the cage was exhausted using the rack venting.Odor solutions were replenished every one to three days to prevent depletion.For experiments with blank stimuli, odor bottles were used that contained only dipropylene glycol.
The nose ports were controlled by a Bpod system (Sanworks) using custom software written in Matlab (Mathworks).Data was logged to a text file that encoded the time and duration of each sample (sample onset, sample offset).All analysis was done using custom routines in Matlab.To measure per animal changes in odor sampling following bottle switch (Fig. 5), we normalized the sampling rate in the period following odor switch by the average sampling rate over the entire experiment.).Red dotted line, linear regression over days 2 through 14; red shading, 95% confidence interval.Correlation coefficient (r) computed using Pearson's correlation.C, Sampling rate (normalized within each animal by sampling rate on day 1) on day 1 and days 2-14 for condition with odor (black) or with air alone (blue).Mean ± s.d.The difference in sampling rates on days 2-14 between the odor and air alone conditions is not significant (P > 0.05) Wilcoxon rank-sum test.

Fig. 1 .
Fig. 1.Measuring odor sampling in the home cage A, Diagram and photographs of experimental setup.B, Mean sample rate (samples per hour, measured across all days, all mice and all odor ports) as a function of time.Grey shading, standard error of the mean.Blue box, time in which room lights were off.C, Raw sample counts for the eight animals included in the analyses described in Figs.1-4 for the first seven days of measurement.D, Distribution of sample counts per day for each odor port, for each animal, for each day tested.
Fig. 2. Odor sampling statistics A. Distribution of poke durations.Grey lines, individual animals; black line, mean across animals; red line, double lognormal fit (R 2 = 0.95).B, sample time autocorrelograms across all odor ports and animals.Grey lines, individual animals; black line, mean across all animals.Inset, same data with ordinate in log scale.C, Inter-sample intervals.Grey lines, individual animals.Grey points, mean across animals.Red curve, dual Gaussian-fit.Magenta curves, individual Gaussians used for dual-Gaussian fit.Cyan dashed line, bout threshold.Inset, mean bouts per days across all animals.Grey points, individual animals.Red cross, mean and s.d.D, Distribution of within bout inter-sample intervals.Vertical dashed line, mean.

Fig. 4 .
Fig. 4. Odor sampling behavior across time.A, left, cumulative sample counts per day summed across all ports for each animal.Right, mean across animals.B, left, mean sample count across all ports and all animals on day 1 vs. mean single day sample count averaged over days 2 through 14 (mean ± s.d.; P < 0.05, Wilcoxon rank-sum test).Right, sample count per port per day across animals (mean ± s.d.).Red dotted line, linear regression over days 2 through 14; red shading, 95% confidence interval.Correlation coefficient (r) computed using Pearson's correlation.C, Sampling rate (normalized within each animal by sampling rate on day 1) on day 1 and days 2-14 for condition with odor (black) or with air alone (blue).Mean ± s.d.The difference in sampling rates on days 2-14 between the odor and air alone conditions is not significant (P > 0.05) Wilcoxon rank-sum test.

Fig. 5 .
Fig. 5. Increase in sampling rate in response to an unexpected cue.A, left, Raw sampling rate.T = 0 indicates each animal's first sample after the odor was changed.Individual animals, grey lines.Mean across all animals, black line.Right, fold change in sampling rate (compared to per animal mean sampling rate across entire experiment) beginning at the first sample following change in the odor.Red line, mean across all animals (N = 90 mice, P < 10 − 5 , Wilcoxon rank-sum test); red shading, 95% confidence interval.B, left, sampling rate after bottle change (mean sampling rate per minute in the 20 min following odor switch) vs. baseline sampling rate (mean sampling rate per minute across the entire experiment).Right, fold change in sampling rate (mean per animal sampling rate in the twenty minutes after bottle change divided by mean per sampling rate across entire experiment) vs. baseline sampling rate.Grey points, individual animals.Blue dashed line, linear regression.Blue shading, 95% confidence interval.Correlation coefficient (r) computed using Pearson's correlation.C, mean sampling rate in the 24 h after odor change (red) and over the initial three days of odor exposure (blue, baseline) normalized for each animal by the animal's mean sampling rate across the entire experiment.Shading, 95% C.I. Right, mean sampling rates (normalized) across the 12-18 h window following the time of day when the odor is changed at baseline (initial three days of odor exposure, blue) and on the day of the odor change (red, P < 10 − 5 , Wilcoxon rank-sum test).