Uncategorized

The in-print version of David Brann’s tour-de-force describing a map of the nose is now out, on the cover of Cell! The final version includes many new experiments not included in the initial bioRxiv post, so definitely check it out for the latest. At one level, this paper shows that there is a olfactory map in the nose – this map is built by the pattened expression of ~1100 odor receptors, each of which adopt a stereotyped and precise *mean* position along the dorsoventral extent of the epithelium.  On the one hand, this is amazing, as it makes the nose arguably the most patterned tissue in the nervous system. On the other hand, it is important to appreciate how this works — we show that olfactory stem cells have incredibly subtle biases in terms of which receptors their progeny express (and so there is a great deal of stochasticity in the underlying mechanisms that choose receptors), and yet macroscopically this probabilistic process yields a highly determined mean position for each receptor. David’s paper goes far beyond mere description, as it proposes a mechanism through which space is translated into identity (a dorsoventral retinoic acid gradient), identifies a transcriptional program present in both stem cells and mature sensory neurons that reflects spatial position, and reveals transcription factors and axon guidance genes that coordinately bias OR choice and specify axonal targeting in the olfactory bulb. The important question this paper poses: why are the receptors in the spatial order they are in? We have some crazy ideas about why that might be, which we are testing in the lab right now….more soon!

 

We’ve been wondering that too! Check out this new paper from Caleb Weinreb and friends in Neuron on a new version of MoSeq called (what else?) shMoSeq, which gives you access to the behavioral states that organize mouse behavior on timescales of seconds to minutes.

OK, so what is shMoSeq, why did we build it, and what does it give us? shMoSeq is designed to address an important limitation of MoSeq, the unsupervised machine learning platform we’ve been developing to characterize behavior.  MoSeq is powerful but frustrating. On the one hand, it converts video data into information about the structure of behavior; it identifies the set of syllables (sub-second, stereotyped action motifs like rears or turns) out of which mouse behavior is organized, as well as the sequence with which they occur in any experiment. This has yielded important information about how the brain structures the microarchitecture of behavior on a moment to moment basis (see here and here). But in the end, MoSeq basically gives you a statistical description of behavioral dynamics – it tells you nothing about *why* a mouse is doing what it is doing. It also doesn’t comport with our intuition that mice often organize their behavior over timescales of seconds-to-minutes, which is the timescale at which humans typically describe behavior (and develop “ground truth” labels for supervised behavioral classifiers). Consistent with this, if we do some math (see the paper) it is clear that there are longer timescales that are organizing mouse behavior, even in a 30 minute open field experiment.

Caleb decided to tackle the longer timescale challenge by building a new, hierarchical version of MoSeq that has another layer on top of the layers corresponding to syllables (i.e., pose dynamics) and poses. This new top layer encodes behavioral states; mathematically, each state is a unique transition matrix, which describes the particular order in which syllables unfold over time.  This definition means that all syllables can be used in all states — what differs is how they are sequences into coherent patterns of behavior. Applying shMoSeq to open field behavior yields a small number of higher order states that last seconds to minutes, and which occur in series over the entire experiment; applying shMoSeq to an open field with novel objects adds a couple of extra states, as does applying it to mice engaged in social interaction.

What are these states? Turns out, each corresponds to the mouse engaging with either some affordance in the arena (the walls, the floor, new objects, a social partner) in various ways, or with itself (by engaging in grooming and other care-related behaviors). In other words, it looks like during each one of these states the mouse is engaged in a self-directed task that reflects the specific affordances available in a given context. This gave Caleb the sense that maybe spontaneous behavior – as the title says – is a series of self-directed tasks. But how to prove this? Well we can’t directly (we are inferring something about the internal meaning of an external behavior after all), but Caleb generated evidence that this idea isn’t entirely unreasonable by asking the brain (in this case the pre-frontal cortex) what it seemed to care about….and it turns out during each of these states, the PFC encodes task-like variables reflecting those things that are most important for the animal to know to achieve a particular self-directed goal. For example, if a mouse is running along the arena wall, what matters most is where it is and how it is moving with respect for the wall, which turns out to be what is emphasized in the brain; if the mouse enters a different behavioral state, representations for those variables fade out and are replaced with representations for whatever variables matter in the new state.

There is all sorts of interesting and provocative stuff in this paper, but the key take-home is that with the right sorts of tools, we can now start to think about self-directed or spontaneous behavior as really being goal- or task-oriented; this aligns the study of natural behavior with long-established paradigms for studying how the brain solves problems after training. Congrats to Caleb and all the other authors!

A major mystery in sensory neurobiology relates to the missing map for olfaction. All of our other senses have peripheral maps, like the map of visual space in the retina, and the frequency map in the cochlea. The existence of these peripheral maps supports the generation of e.g., the retinotopic map in V1, the tonotopic map in A1. But olfaction……not so much. In fact, for the last 30 years, since the cloning of the receptors, the main model in the field has been that each mature olfactory sensory neuron in your nose randomly chooses which of 1000 receptors to express. This randomness means there is no map.

Until now. David Brann in the lab has used a combination of single cell methods, spatial transcriptomics and a lot(!) of incredible work and thinking to reveal a structure and stereotyped map of odor receptors in the nose, and an associated molecular logic that aligns odor maps in the nose and the brain. Check it out here: https://www.biorxiv.org/content/10.1101/2025.05.02.651738v1

From all of us to all of you, we wish everyone happy holidays for 2025. May 2026 bring stability, good sense and great science for all!

Then check out Bob’s recent Special Lecture at SfN on exploring the neural basis for natural behavior using computational ethology – three parts below:

 

The Motion Sequencing algorithm developed by the lab has been applied to a host of questions and across many labs, but early on, adoption was slow because the underlying code was bespoke, academic and brittle. Over the past several years we have taken many steps to make access and use of the MoSeq algorithm easier – you don’t need to sign a license, the code is free to download and often updated, we have stood up user communities that propose improvements, catch bugs, and help each other. Much of this information lives at the MoSeq2 webpage and linked Github repo here. Sherry Lin, a senior computer scientist, did a hero’s work in making many of these changes, and many in the community have benefitted from her efforts. She has now (with help from friends in the lab) codified much of her wisdom about the depth MoSeq pipeline into this paper, recently published in Nature Protocols. Many thanks to Sherry for her work on this and on supporting MoSeq – as a community it is really hard to find ways to make academic code usable and sustainable by others, and Sherry has done a spectacular job at this – check out the paper to learn more!

 

Did we bowl? Mos def we bowled! Happy Holidays, all!

As he noted at his goodbye party, no one has published more Datta lab papers than Win. He contributed to basically every project in the lab (see a smattering of his contributions here, here, here and here), and for years has been a part of the lab DNA. We all went out for beer to celebrate Win and to commiserate about our loss, and ended up perhaps drinking too much….which is why there is only this one terrible picture of Win before he started speechifying. We will all miss you Win – you are the best, and the lab won’t be the same without you!!

A couple of years ago, David Brann wrote a lovely review (link here) that focused on what brain scientists might learn from understanding olfaction. Now Kara Fulton, together with our colleagues David Zimmerman, Aravi Samuel and Katrin Vogt – all experts on the invertebrate side – has put together a similarly beautiful review (link here) on the current state of our understanding of smell. Kara goes deep both on mechanisms and on the relationship between the mammalian and insect senses of smell – it is a really lovely synthesis both of what we know and what we don’t, and a nice introduction to the field for newcomers.

 

Depth MoSeq is great, but it has one major limitation – it depends upon using videos taken by somewhat janky, slow and often noisy 3D cameras, like the discontinued Microsoft Kinect. What would be ideal is to instead use fast and standard 2D cameras (either alone or in combination) to track keypoints using standard algorithms like DeepLabCut or SLEAP, and then to feed those keypoints to MoSeq to identify the behavioral motifs or “syllables” our of which behavior is organized. There is a problem, though – unlike most unsupervised behavioral clustering algorithms, by design MoSeq tries to identify the exact moment when one behavioral syllable switches to the next. It does so in part by looking for discontinuities in behavior, moments when behavior abruptly changes from one mode to another. Because the output of keypoint tracking code is often pretty flickery (with individual tracked keypoints bouncing from one location to another), MoSeq breaks when fed raw keypoints – each flicker becomes a new behavioral transition. In this new paper just published in Nature Methods, the amazing Caleb Weinreb (together with our longtime friend and collaborator Scott Linderman) figured out a solution to this problem — have MoSeq infer keypoint positions based upon the premise they participate in syllable-like dynamics. Modifying MoSeq in this way – to create what we are calling keypoint-MoSeq – enables any keypoint data from any camera type to be fed to MoSeq. There is *so* much validation in this paper – we compare depth to 2D and 3D keypoints, show that this works on both mice and rats, demonstrate it works across environments (e.g., in the open field and in a complex home cage), and more! One important point – we think specifically modifying MoSeq to accept keypoints was especially worth it for folks interested in moment-by-moment relationships between neural activity and behavior. We compare depth and keypoint MoSeq at their ability to detect the systematic fluctuation in dopamine associated with syllable transitions described in this paper from the lab published last year – we find that both methods work about as well at detecting this fast rhythm in dopamine associated with syllable switching. BUT…when we used alternative behavioral clustering methods that don’t privilege time in the same way as MoSeq, this well-established (and causal) neurobehavioral relationship disappeared. There are likely many settings in which these alternatives are superior to MoSeq (largely for practical reasons), but if ultimately your goal is to ask how brain activity tracks (not on average, but in each instance) with behavior, time-sensitive methods like MoSeq remain the best current bet. Congrats to Caleb (and our many collaborators!) on this big step forward to behavioral characterization.