Olfactory bulb tracks breathing rhythms and place in freely behaving mice

Reviewer #1 (Public review):

In this manuscript, Sterrett et al. assess whether and how the olfactory system may integrate odor-driven activity with contextual, egocentric variables such as instantaneous location in space and active odor sampling. To address this, they co-record respiration and the spiking activity of principal output neurons of the mouse olfactory bulb (OB), while mice explore a small arena in the absence of any explicit reward or task structure. The authors find that mice exploring the arena breathe in bouts, switching between discrete states of particular breathing rates that persist over varying time scales (seconds to minutes). This state-like activity is also apparent in the OB population activity. Zooming into the activity of individual OB neurons, the authors show that OB activity in this setting is primarily modulated by respiration. In general, while the response times of individual neurons remain tightly locked to the inhalation onset, the overall response amplitude is modulated by the instantaneous sniff frequency. The authors further suggest that a subset of OB neurons appear to show place-selectivity, in a manner that is not explained simply by respiratory or olfactory variables.

Overall this work addresses an important question regarding the basic temporal structuring of odor sampling behavior and activity patterns in the mouse OB. A good understanding of these features is essential to further investigate how stimulus and/or task-driven activity may add on top of this already ongoing modulation. The authors do a commendable job of analyzing the behavior and neuronal activity using a variety of analysis methods. However, in its current form, the results presented are high-level summary figures that are largely comparative (role of parameter A vs B) and hard to assess quantitatively (how well does a given parameter/model explain the responses to begin with). This makes it hard to build a clear model of the underlying mechanisms and to evaluate alternative hypotheses. These concerns can largely be addressed by some additional analyses and by presenting more intermediate-stage output of their existing analyses. In addition, the authors report that a small fraction of OB neurons show spatially selective firing patterns, akin to those observed in the Hippocampus. While this is a very exciting possibility, in my opinion, the data and analysis presented currently are not sufficient to conclude this and additional experiments would be required to test this rigorously.

Major concerns:

A) Regarding the claim about Spatial selectivity in OB neuron responses:

i) From the data presented, it is very hard to assess whether a simple modulation of sniff rate, selectively in some parts of the arena can explain apparent spatial selectivity. The authors attempt to address this concern with Figure 8 - Figure Supplement 1, but the presented combinatorial color maps are hard to interpret. A simpler format would be to show the sniff-aligned raster of the given unit in question along with a heatmap (location distribution) of the actual sniff rates in the arena (not the behavioral states).

If the authors allow the mice to explore the arena over large periods, such that the sniff rates are relatively uniform in space, are the place fields still apparent? A complementary control is to compare responses in the 'place field' with other parts in the arena with comparable sniff rate distributions.

ii) The analysis shown in Figure 8 suggests that sniff parameters are the main predictors of individual neuron responses. The authors point out that there is however a small, but significant fraction of cells that are better predicted by place than by the sniff parameters. It would be useful to provide more raw data to get a better sense of what distinguishes these cells from the rest. Are spatially selective cells typically less sniff-aligned on average? Do they tend to be less or more frequency-modulated?

iii) The authors compare the decoding performance of OB and hippocampal neurons. While it appears space can indeed be decoded from OB neurons, it would be useful to know how the performance scales with the number of neurons and number of traversals in the arena in the two brain regions. Further, the authors should provide some analysis of the robustness of these apparent 'place fields' within a session.

iv) The floor rotation control is underwhelming. First, the arena is quite small and one would generally expect this to impact much more so the 'place fields' that are biased towards the corners than in the center. Second, olfactory cues on the walls may be as important - why did the authors not rotate the entire arena?

Considering the possibility that floor rotation rules out trivial olfactory explanations, what would happen if the authors rotated the entire arena? If these are truly place fields, then one should expect that while they are robust to floor rotation, they should reformat if the distal cues change. Without these additional analyses, I find it hard to conclude the presence of spatial selectivity in the OB.

Moderate concerns:

B) Regarding the lack of state-like structure during head-fixation:

While it is clear that overall sniff rates are lower and that mice do not typically sniff at peak rates during head-fixation, it is unclear if the transitions in breathing rhythm are necessarily less structured, and further whether this can be attributed to head-fixation alone. For example, if the mice are head-fixed but in a floating-platform arena or VR that is non-static - the sniffing distributions may change dramatically.

i) The breathing patterns shown in Figure 1E, in particular during the second head-fixation phase do not appear fundamentally different from the freely moving stretch (20-30 minute window). If one subsamples the free-moving data to match overall sniff distributions, will the long-timescale autocorrelation still be more apparent in freely moving stretches than in the head-fixation periods?

ii) Are the mice on a running wheel? How does the overall distribution of sniff rates and temporal structure change if the mice are head-fixed but simply allowed to run?

Minor concerns:

C) Regarding the parsing of breathing and movement into 3 distinct behavioral states:
The authors show breathing patterns of freely exploring mice are temporally structured with extended bouts of sniffing at select rates. They use a HMM model to show that this structure can be captured by a 3 state-model wherein each state can be thought of as a joint distribution of movement and sniff rate. While the approach is interesting and the data are well presented, I have some minor concerns regarding the exact interpretation.

i) While the relationship between movement and sniffing is indeed non-trivial, it is unclear if the statelike partitioning requires the incorporation of the movement variable at all in the HMM model. The state-like patterns are also apparent if one focuses exclusively on the instantaneous sniff rate while ignoring movement velocities (Figure 1 - Figure Supplement 1) or the inferred HMM states (Figure 1E). Have the authors tried modeling the breathing activity alone using an HMM with each state just being a biased distribution of sniff rates, from which the instantaneous sniff rate is drawn? Will the authors' conclusions be fundamentally different from such a model?

ii) While it is clear that there are at least 2 distinct states a) resting (mice are generally uninterested and sniff at 2-3 Hz) and b) exploration (mice are interested in their local environment and sniff rapidly). It is hard to assess whether there is indeed a third distinct and behaviorally interpretable state that the authors call grooming or are there simply intervening periods where it is unclear what's driving the variability in sniff rates - change in movement speed, moderate curiosity, boredom, etc. From the movement velocities shown in the supplement (Figure 1 - Figure Supplement 1), it appears that the movement speed during this 'grooming' state is significantly higher than at rest. It is not obvious why a mouse should move around more while grooming. It would help if the authors provide supporting data, perhaps from behavioral pose analysis to better justify the classification of this state as grooming or alternatively choose a different name to avoid confusion.

iii) Insufficient analysis of state transition matrices: The authors do not show the transition matrices for individual sessions and/or mice. This limits what one can learn about the behavior from the 3 state modeling of breathing states. Do individual mice have stereotypical transition patterns across sessions? How well does the model perform: can one predict the expected sniff rate in one part of the session from knowing sniff patterns in another part of the session?

D) Regarding the dependence of individual neuron responses on sniff and movement parameters:

i) Could the authors report the relative proportions of sniff frequency insensitive vs. frequency sensitive neurons in their data?

ii) Could some of the striking frequency modulation the authors show in Figure 3A result from the fact that mice selectively sniffed at high or low rates in different parts of the arena? While it is unlikely that all of the modulation the authors see results from the location/presence of trace odors in different parts of the arena, it would be informative to perform the same analysis on the data recorded during head-fixation where its external environment is less variable.

iii) Comparison of SnF latency profiles between head-fixed and freely moving conditions:
The SnF latency profiles of a given OB neuron appear strikingly similar during head-fixed and freely moving conditions. It would be useful if the authors could explicitly quantify this.

iv) Comparison of SnF frequency profiles between head-fixed and freely moving conditions: The authors comment that SnF frequency profiles are different across the head-fixed versus freely moving conditions and that they do not observe the 3 distinct clusters present in the freely moving state in their head-fixed data. If true, this is an interesting observation. Together with the observation of relatively similar SnF latency profiles in both head-fixed and freely moving conditions, this implies that sniff frequency dependence is selectively enhanced during free-moving behavior perhaps through a top-down signal.

However, this is hard to conclude from the current data as the overall distribution of sniff rates is very different in the two conditions, with a clear underrepresentation of high-frequency sniffs in the head-fixed periods. To enable a fair comparison, the authors should undersample the sniffs in the freely moving period and compare sniff fields constructed from frequency-matched distributions.

v) The authors suggest that the 2 types of SnF latency profiles may putatively map onto tufted and mitral cells. While this is an interesting possibility, it would be nice to support the claim with auxiliary analysis of other features such as recording depth, baseline firing rates, spike shapes, etc that indicate that these are indeed two different cell types.

Reviewer #2 (Public review):

In this study, the authors investigate the structure of breathing rhythms in freely moving mice during exploratory behaviour in the absence of explicit cues or tasks. Additionally, they link behavioural states, derived from sniffing frequency and speed movement data, to the neural activity recorded in the olfactory bulb (OB). To further characterize OB neuronal responses, the authors introduce the concept of "sniff fields" which consider the joint distribution of sniff frequency and the latency from inhalation. Lastly, they explore how OB neurons encode spatial information, and they compare this finding with previously known spatially encoding cells in the hippocampus.

The authors successfully establish that breathing in freely moving mice is structured even in the absence of explicit olfactory cues. By simultaneously recording sniffing and movement data, they find that this structure is associated with movement in a non-linear manner and can be modelled using a Hidden Markov Model (HMM). Interestingly, they demonstrate that neuronal activity in the OB tracks this behavioural structure by showing that HMM states can effectively cluster the neural data. Additionally, they describe OB activity using sniff fields, advancing our understanding of how individual neurons encode sniffing properties such as frequency and phase. Furthermore, they report unprecedented findings showing that some OB neurons encode place independently of the sniffing field contribution. Overall, the authors provide valuable insights regarding the contribution of different behavioural variables to OB activity.

However, some of the conclusions presented by the authors are not fully supported by the data provided. Quantitative analysis and statistical tests are missing from the description of the breathing structure. Regarding spatial encoding, the authors claim in the abstract that "at the population level, a mouse's location can be decoded from olfactory bulb with similar accuracy to hippocampus". However, they show that place was significantly decoded in only 18/31 sessions from OB activity, and in 12/13 sessions from hippocampal activity. No further comparison of decoding accuracy between OB and HPC is provided. Moreover, it is unclear whether place contributes independently of movement, which was previously shown in this study to influence neuronal activity.

Additionally, there is a lack of methodological detail regarding the experimental procedures, which could affect the interpretation of the data. Specifically, information is missing on aspects such as head-fixed conditions, the number of mice used per experiment, and the number of sessions per mouse.

Studying mice behaviour in more naturalistic conditions, without explicit tasks, is a very interesting approach that provides new insights into the structure of sniffing and its neuronal representation. The fact that some OB neurons encode spatial information is highly relevant beyond the field of olfaction, even though this information was not as accessible as in the hippocampus. I believe the manuscript would benefit from a revision to ensure the text aligns more closely with the data presented in the figures.

Author response:

We thank the editor and reviewers for recognizing the value of studying neural dynamics and behavior in naturalistic, task-free conditions and the importance of linking olfactory bulb activity to movement and place. We appreciate the suggestions for analyses and edits to further quantify these relationships and clarify our interpretation.

The primary sticking point regards our result that olfactory bulb neurons are selective for place:

“analysis supporting the potentially exciting result on the encoding of place is currently incomplete”

In this paper, we report evidence for spatial selectivity in the olfactory bulb, make relative comparisons with canonical “place cells” in the hippocampus, and control for alternative hypotheses such as odor- or behavior-driven sources, to motivate future experiments which can more precisely identify the mechanistic basis of these responses. Throughout the reviews, our result on the correlation of OB activity with place is not questioned, but rather whether we can better determine how much behavior or odor explain this result. Regarding the concern about behavior, we are confident that the spatial non-uniformities of breathing rhythms do not explain OB spatial selectivity based on the analyses included in the paper. We thank the reviewers for suggestions of additional analyses with which we can further test this claim and will incorporate several, as we will detail below.

Regarding the points about odor, indeed we do not claim that we have entirely ruled out odors as an explanation of place selectivity in the bulb. Rather, our claim is that our analyses show that scent marks on the floor, the most obvious olfactory place cue, cannot fully explain place selectivity. We acknowledge that our experiments do not exclude the possibility that other odors in the environment may also contribute. Odors are invisible and difficult to measure, and the odor sensitivity of rodents vastly outstrips that of any device known to humanity. Indeed, no study of which we are aware can fully rule out odor as a cue to the animal’s internal model of place. However, encoding of place, even if explained by odor, is still encoding of place. We will clarify our interpretation of the data, and we thank the reviewers for proposing ideas for further analysis, some of which we are implementing. However, experiments such as effects of distal cues on spatially selective olfactory bulb neurons are beyond the scope of this paper.

We will further test whether neurons in the olfactory bulb are spatially selective by reporting additional statistical analyses including:

- More completely quantifying the spatial distribution of sniffing patterns (visualized in Figure 8 - Sup 1) by plotting sniff-frequency distributions across locations in the arena.

- Demonstrating independent contribution of place over speed in GLMs

- Characterizing the temporal stability of spatially selective cells across a session (1st half vs second half)

- reporting mean decoding errors for olfactory bulb and hippocampal decoders (visualized in Fig 7C)

We will add to the analyses of behavioral state models by:

- Comparing the performance of hidden Markov models fit to breathing frequency alone with those fit to breathing frequency and movement speed

- Quantifying individual differences in state-transition matrices

Further, we address the question around the use of “grooming” as a descriptor of the intermediate sniff frequency state. We used the term ‘grooming’ based on extensive video observation. During this state, ‘Speed’ is significantly non-zero because we defined speed as the movement of the head keypoint which moves substantially during grooming. We will make this point more explicit in the figures and text, and we will provide additional video documentation of these and the other behavioral states.

Lastly, we will further discuss the fact stated in the first paragraph of the Results section that mice are placed in “head-fixation on a stationary platform” and thus inhibited from running. While different breathing states than those observed in our stationary platform may occur during head-fixation with a treadmill, we believe the differences between head-fixed running and free moving running are beyond the scope of this paper. Nevertheless, it’s an important point that we will more explicitly discuss in our revision.

We appreciate these constructive comments and hope these additional analyses and textual edits will help clarify our interpretations and motivate future experiments to further test and refine them.