Tim Pierce on olfaction and chemical sensing

  • cover play_arrow

    PLAY EPISODE


Season 2013
Season 2013
Description arrow_drop_down

Description

Why can you smell a molecule you have never encountered before, and how does the nose use antagonism, binding proteins, and chemotopic maps to decode the chemical world? Tim Pearce explores the engineering principles behind biological olfaction.

Subscribe for more from the Convergent Science Network podcast series.

Tim Pearce provides a comprehensive tour of natural olfaction, from the molecular interactions at the receptor sheet to the computational principles that enable detection of thousands of diverse chemical compounds. He highlights the system’s remarkable foreignness property: unlike vision with its handful of receptor types, olfaction deploys over one percent of the genome to create a broad, relatively unbiased sampling of chemical space, capable of responding to novel molecules never previously encountered by the species.

The episode reveals several layers of molecular complexity that precede neural processing. Odorant binding proteins in the nasal mucosa act as selective transporters, shifting sorption spectra to capture hydrophobic compounds that would otherwise resist the liquid phase. Odor degrading enzymes terminate signals in a timely fashion. Most surprisingly, recent evidence shows that receptor-ligand interactions involve not just affinity but also efficacy, and that widespread antagonism between molecules means the neural response to mixtures is far from a linear sum of individual components. These nonlinear competitive interactions at the receptor level fundamentally shape the olfactory code.

Pearce describes a chemotopic organization of the receptor sheet where molecular features like carbon chain length, functional groups, and hydrophobicity map onto different spatial zones, driven partly by differential sorption along the airflow path and partly by the zonal expression of receptor families including ancient fish-derived class 1 receptors. Analysis of molecular descriptors reveals that despite hundreds of possible chemical features, the effective dimensionality of odor space is surprisingly low, with principal components capturing much of the perceptual variance.

The discussion also covers retronasal olfaction, where volatile compounds from food reach the receptor sheet through the back of the nose and produce qualitatively different percepts than the same compounds delivered orthonasally, even at the level of receptor sheet activation patterns. Pearce connects these biological insights to engineering principles for artificial olfactory systems, arguing that the conserved architectural motifs found across species from insects to mammals provide a blueprint for building chemical sensing systems.

Tagged as:

About the author call_made

CSN Podcasts

Both the triumphs of humanity and its most evil deeds have resulted from collaboration. In a time where humanity is required to aspire to the former and minimize the latter, the question arises of how collaboration arises and why it fails. Surprisingly, this phenomenon, so central to who we are, is not well understood. Hence, a collaborative effort is required to understand collaboration in its full biological, psychological, sociological, cultural, and economic complexity and to translate this understanding into operational impact. This series of podcasts is one step toward achieving these complementary goals. The Collaboration Podcast presents interviews with people who are central orchestrators of collaboration in various domains including business, government, science, art, health, sustainability, and the military. The discussions were conducted by Prof. Dr. Paul F.M.J. Verschure and members of the Program Advisory Committee of the Ernst Strungmann Forum on Collaboration (https://www.esforum.de/forums/ESF32_Collaboration.html) during 2021 and had the goal to sketch a map of opportunities, challenges, and obstacles in human collaboration. The forum took place in May 2022, and now we would like to share this series of interviews with a broader audience. The full report of the Forum will be published in 2023 by MIT Press. The podcast was produced by the Convergent Science Network (https://www.convergentsciencenetwork.org/). Context: The stability of social systems depends critically on realizing sustainable methods of “collaboration,” yet how and by which means collaboration is achieved is not clearly understood; neither are the conditions or processes that lead to its breakdown or failure. Collaboration can be understood as cooperation between agents toward mutually constructed goals. Part of the reason for our lack of understanding is that the phenomenon of collaboration is, by nature, a highly multidisciplinary problem, and effective research into its complexities has been difficult to achieve across the broad range of scientific and technical disciplines involved. The need for a fundamental understanding of collaboration, however, has become increasingly important. Not only does humankind demand answers as it attempts to address critical challenges at multiple scales (e.g., climate change, migration, enhanced automation, social and economic inequality), but ever-increasing technological and economic means of interconnecting people and societies are disrupting long-established, familiar patterns of how we interact. Radical technological changes that are ongoing have the potential to reshape collaboration in ways that are currently hard to predict or influence (e.g., by altering configurations in interaction, information creation, and modes of communication). On one hand, such changes could disrupt hitherto stable forms of collaboration by affecting critical communication channels and traditional roles, as can be observed in the rapidly changing patterns in governance, commerce, and social interaction. Conversely, technology could lead to the emergence of novel, successful forms of collaboration that deviate from traditional “hierarchical” architectures. Evidence of this can be seen in areas as diverse as highly automated manufacturing plants, the open science movement, collaborative software repositories, user-centered services, and the sharing of economy-based modes of organization. Without a fundamental understanding of the mechanisms, processes, and boundary conditions of collaboration, it is not possible to evaluate or predict which of these possible scenarios are sustainable or even plausible. The Forum “How Collaboration Arises and Why it Fails” (May 8–13, 2022, Location: Frankfurt am Main, Germany) Chairs: Andreas Roepstorff and Paul Verschure Program Advisory Committee: Jenna Bednar, Julia R. Lupp, Bhavani R. Rao , Andreas Roepstorff, Ferdinand von Siemens, and Paul Verschure

More posts

Timestamp

  • fast_forward00:00:03 - This is the Convergent Science Network podcast. Leading researchers in the domain
  • fast_forward00:00:10 - of neuroscience, brain theory and technology are interviewed by Paul Verschure and Tony Prescott.
  • fast_forward00:00:25 - This is Paul Verschure with the Convergent Science Network podcast.
  • fast_forward00:00:30 - And I'm here with Tim Pearce, one of the speakers in our summer school, the 2013 version of it.
  • fast_forward00:00:36 - And Tim is a great specialist in olfactory systems, both their biological versions
  • fast_forward00:00:43 - and artificial versions.
  • fast_forward00:00:46 - So, Tim, what's so special about natural olfaction?
  • fast_forward00:00:51 - Great. Well, I was very pleased to be here. It's very exciting talks.
  • fast_forward00:00:55 - So, natural olfaction. Yeah, well, one of the most amazing features, I guess.
  • fast_forward00:01:01 - Well, there are many, but you have to consider pretty amazing the ability to
  • fast_forward00:01:08 - detect literally thousands of molecules.
  • fast_forward00:01:11 - Molecules and beyond that even to the
  • fast_forward00:01:14 - point where if you made a new molecule this
  • fast_forward00:01:17 - week you'd be able to if it
  • fast_forward00:01:20 - was within certain mass criteria which is very
  • fast_forward00:01:22 - wide anyway between 30 daltons and
  • fast_forward00:01:25 - 300 daltons in molecular mass then
  • fast_forward00:01:29 - you would almost certainly have some kind
  • fast_forward00:01:32 - of response to this so it has
  • fast_forward00:01:36 - this well i think one of the most amazing properties
  • fast_forward00:01:39 - of the olfactory system is this semi-biotic or foreignness
  • fast_forward00:01:42 - property that it's designed from the
  • fast_forward00:01:45 - ground up to detect uh all
  • fast_forward00:01:49 - sorts of uh foreign signals that uh
  • fast_forward00:01:52 - may not be uh have some priors right
  • fast_forward00:01:55 - so your prior in the system is very flat you don't know necessarily in advance
  • fast_forward00:02:01 - which uh chemical signals out there in the world are going to be uh extremely
  • fast_forward00:02:06 - relevant for you behaviorally and
  • fast_forward00:02:09 - you need to have as broad a sampling as possible on this without making.
  • fast_forward00:02:14 - Necessarily prior assumptions.
  • fast_forward00:02:16 - Right, but now as a functional feature, that's not specific to olfaction, right?
  • fast_forward00:02:22 - I mean, also my visual system can detect novel stimuli or my auditory system and.
  • fast_forward00:02:27 - So, that's not necessarily outstanding, is it?
  • fast_forward00:02:32 - No, but maybe in the sense of just the diversity of the basic stimuli, right?
  • fast_forward00:02:39 - I mean, and also, of course, I mean, this is also inherent in the architecture of the system, because,
  • fast_forward00:02:45 - I mean, basically you've got three receptors or four receptors for vision in
  • fast_forward00:02:50 - certain animals and humans,
  • fast_forward00:02:52 - humans um and you know which has
  • fast_forward00:02:54 - a certain cost uh in terms of the genetic real
  • fast_forward00:02:57 - estate but when you look at olfaction it's more
  • fast_forward00:03:00 - than one percent of the genetic real estate
  • fast_forward00:03:03 - is dedicated just for this reason so there has
  • fast_forward00:03:06 - to be a good reason to deploy all of that uh diverse coding uh so that you can
  • fast_forward00:03:14 - have a very broad sampling if there was a simple answer to it i'm sure it would
  • fast_forward00:03:19 - have been being found in the sort of genetic architecture, yeah.
  • fast_forward00:03:25 - So now in your talk, you mentioned five outstanding features like concentration
  • fast_forward00:03:31 - dynamics, temporal dynamics, the specificity and sensitivity.
  • fast_forward00:03:37 - Also the ability to exploit olfaction in active senses, an active perceptual
  • fast_forward00:03:43 - system, and to deal with object recognition in a high-dimensional space.
  • fast_forward00:03:49 - Right so these were the five um outstanding issues
  • fast_forward00:03:53 - and we will discuss those in a little bit more more detail
  • fast_forward00:03:56 - but the other thing that was sort of surprising is that you also emphasize this
  • fast_forward00:04:01 - so-called retronasal olfaction so what what what's the difference sure yeah
  • fast_forward00:04:06 - so i mean most of us are are used to sniffing things in the world through so-called
  • fast_forward00:04:10 - orthonormal uh olfaction which which is through the normal route through your external naras,
  • fast_forward00:04:16 - but maybe less people are aware that a lot of your olfactory signals is coming
  • fast_forward00:04:22 - when you're eating through volatile compounds going up through the back of the nose,
  • fast_forward00:04:28 - interacting with the nasal mucosa that way.
  • fast_forward00:04:33 - And, well, the interesting point is that depending on the flow,
  • fast_forward00:04:37 - it seems that you get a different experience, whether the compounds are coming
  • fast_forward00:04:42 - from internally during ingestion or whether they're coming externally.
  • fast_forward00:04:45 - Even if you control for the fact that obviously eating things has a gustatory
  • fast_forward00:04:51 - input as well, which is combined.
  • fast_forward00:04:53 - But even if you control for that, then you still have a very different olfactory
  • fast_forward00:04:58 - percept depending upon the direction of the odor.
  • fast_forward00:05:03 - So does this retronasal olfaction,
  • fast_forward00:05:06 - does it give it another quality of other sensation or other processing or does
  • fast_forward00:05:13 - it fall the same back of molecular recognition and processing as the orthonormal version.
  • fast_forward00:05:22 - Yeah, it's a good question. There have been many studies to show that the percept is very different.
  • fast_forward00:05:28 - So during retronasal olfaction, you will perceive different qualities.
  • fast_forward00:05:33 - It hasn't really been specified precisely how that changes.
  • fast_forward00:05:38 - But we also know that the neural dynamics, even at the receptor sheet,
  • fast_forward00:05:45 - is very different patterns that are elicited by exactly the same chemicals.
  • fast_forward00:05:50 - Just from the fact whether they're occurring retronasally or orthonasally.
  • fast_forward00:05:55 - So the obvious explanation for this must be to do with detecting whether things
  • fast_forward00:06:02 - you're eating are good things to eat or externally whether they're things that
  • fast_forward00:06:07 - you need to find to eat or avoid, I guess, right?
  • fast_forward00:06:10 - So there does seem to be good reasons why this might be the case. Right.
  • fast_forward00:06:15 - So now, one interesting observation that you put forward is also that in some
  • fast_forward00:06:24 - sense human olfaction might not be of equal sensitivity as many other animals,
  • fast_forward00:06:29 - but you do believe that the basic forms of processing and strategies in which
  • fast_forward00:06:34 - we use olfaction might be rather similar.
  • fast_forward00:06:36 - Is that fair to say? sure yeah i mean i it's pretty clear people like nick strausfeld
  • fast_forward00:06:42 - and so on they've gotten into incredible
  • fast_forward00:06:44 - detail in the fossil records for different olfactory systems from.
  • fast_forward00:06:48 - Mammals all the way down to uh very extremely simple uh forms of forms of life
  • fast_forward00:06:55 - have shown that there are a number of key uh generic architectural features
  • fast_forward00:07:01 - which are crucial to to building olfactory systems you know one of these being
  • fast_forward00:07:05 - uh glomeruli where there's these
  • fast_forward00:07:07 - convergence of sensory signals at the first stage of processing,
  • fast_forward00:07:12 - which is completely common to all animals achieving this.
  • fast_forward00:07:15 - There also seems to be a genuine necessity for a first stage of processing that
  • fast_forward00:07:21 - has lateral inhibition at an early stage,
  • fast_forward00:07:26 - which somehow is thought to
  • fast_forward00:07:28 - sharpen tuning and also impose all sorts of dynamical properties on top.
  • fast_forward00:07:33 - And um there's also this feature in most animals of a very uh a mixed specificity of certain,
  • fast_forward00:07:41 - receptors being highly tuned to certain compounds and other receptors being
  • fast_forward00:07:45 - very widely tuned to large groups of compounds it seems to be this also seems
  • fast_forward00:07:49 - to be common so yeah there are a number of properties that it seems if you're
  • fast_forward00:07:53 - going to build um it's a nice thought experiment if you say take a blank piece
  • fast_forward00:07:57 - of paper and think if we're going to build um.
  • fast_forward00:08:00 - An olfactory system without any prior knowledge of what animals do,
  • fast_forward00:08:04 - given that you want to detect huge diversity of compounds with certain high
  • fast_forward00:08:10 - specificity and high sensitivity to certain groups of compounds that are important to you behaviorally,
  • fast_forward00:08:15 - you know, what sort of strategies would you have to do to achieve that?
  • fast_forward00:08:20 - And I think these common architectural sort of motifs that you you see uh from
  • fast_forward00:08:26 - the fossil record sort of underpin how you might approach that also from an
  • fast_forward00:08:31 - engineering sense so but then and so that basically means whether we whether
  • fast_forward00:08:36 - we go from drosophila to humans,
  • fast_forward00:08:39 - in your mind we will find common design principles behind this olfactory system
  • fast_forward00:08:44 - right so that that's also interesting so it's how it would suggest it's rather strongly conserved.
  • fast_forward00:08:51 - So now in the human case so here we go, we have
  • fast_forward00:08:54 - our nasal cavity I have my epithelium in
  • fast_forward00:08:57 - there through which I have the sensilla sticking from
  • fast_forward00:09:01 - the olfactory receptor neurons these olfactory receptor neurons are then projecting
  • fast_forward00:09:06 - their axons through the skull into the olfactory bulb where they form these
  • fast_forward00:09:12 - glomeruli that you described and then these glomeruli in turn get read out by
  • fast_forward00:09:17 - these mitral cells and then and sends this information to other parts of the brain.
  • fast_forward00:09:21 - Now, in the mammalian case,
  • fast_forward00:09:27 - What do we know about the levels of processing along this hierarchy?
  • fast_forward00:09:31 - So, for instance, would you say the olfactory percept in the human case is already
  • fast_forward00:09:36 - defined in the olfactory bulb?
  • fast_forward00:09:38 - Or does it require higher levels of processing? Does it require interaction
  • fast_forward00:09:43 - between different levels?
  • fast_forward00:09:46 - Yeah, this is a really good question. I mean, in humans, it's pretty clear that
  • fast_forward00:09:51 - the first stage of processing the olfactory bulb is thought to do a number of things.
  • fast_forward00:09:55 - One of the most key features seems to be sharpening the representation,
  • fast_forward00:09:59 - so decorrelating the signals at an early stage is obviously very important for
  • fast_forward00:10:05 - discrimination of different odor compounds.
  • fast_forward00:10:11 - But I'm not sure I would say this is the place where the percept comes from,
  • fast_forward00:10:15 - because it clearly has to be an active process.
  • fast_forward00:10:18 - And it's also demonstrated that in
  • fast_forward00:10:21 - higher processing centers like the piriform cortex
  • fast_forward00:10:24 - they have much more sparse representations of
  • fast_forward00:10:28 - these odors which again is sharpening the
  • fast_forward00:10:32 - tunings even further that has
  • fast_forward00:10:35 - been already demonstrated quite clearly to be
  • fast_forward00:10:38 - related to odor memories and also learning and
  • fast_forward00:10:42 - the other important aspect is that these higher centers as we see in many other
  • fast_forward00:10:45 - sensory systems in the brain are enforcing a top-down processing on the olfactory
  • fast_forward00:10:54 - bulb itself on the earliest stage to change the gain control and do various things.
  • fast_forward00:10:59 - So it's also been clear that, for instance, in orbitofrontal cortex,
  • fast_forward00:11:05 - that this is also an important aspect of the perceptual aspect.
  • fast_forward00:11:09 - So I would say it's really at these more cortical levels that if you were to
  • fast_forward00:11:14 - think about percepts, this is where you'd probably look.
  • fast_forward00:11:17 - So in the human case, how many of these receptor neurons do we have sticking their sensilla in my...
  • fast_forward00:11:24 - Yeah, so the total real estate at its peak in rats was about a thousand different
  • fast_forward00:11:29 - diversity of receptor GPCR types.
  • fast_forward00:11:34 - And the story is that humans, basically our sense of smell is deteriorating in evolutionary terms.
  • fast_forward00:11:42 - It's not under sufficient selective pressure.
  • fast_forward00:11:45 - Basically and the story is that uh something like
  • fast_forward00:11:49 - about 600 of those thousand receptors have become
  • fast_forward00:11:51 - fairly defunctional now they've got introns within them junk dna and so on that
  • fast_forward00:11:56 - means that they're they're they're not functioning as olfactory receptors anymore
  • fast_forward00:12:00 - so we're left with in humans the story is somewhere between three and four hundred
  • fast_forward00:12:05 - um functional is there any idea when in evolution this sort of started to happen um,
  • fast_forward00:12:11 - Yeah, it's a good question. I don't know the answer to that in evolutionary terms.
  • fast_forward00:12:14 - Yeah, but there are very nice studies that have been done which show phylogenetic.
  • fast_forward00:12:20 - So you can, I'm not an expert on this, but you can look at the phylogenetic similarities.
  • fast_forward00:12:26 - There have been a number of papers where effectively you're looking at sort
  • fast_forward00:12:31 - of dendrogram of similarity of receptors between different animals.
  • fast_forward00:12:35 - Now, for instance, primates, do they also show the same deterioration?
  • fast_forward00:12:39 - Okay. So, primates are in a similar class to us.
  • fast_forward00:12:42 - So, in the human case, I have, let's say, 300 receptor neuron types expressed.
  • fast_forward00:12:49 - Then how many individual receptor neurons would I have in my epithelium?
  • fast_forward00:12:56 - Yeah, so the epithelium has between 10 and 100 million in total. For humans?
  • fast_forward00:13:00 - Yeah. So, you've got a large number of the same type expressed many times. And red, how many?
  • fast_forward00:13:09 - They probably have similar numbers, slightly more with a higher density because
  • fast_forward00:13:12 - it's obviously crammed into a smaller area.
  • fast_forward00:13:14 - Things like dogs and pigs would have at least an order of magnitude greater
  • fast_forward00:13:19 - than that in terms of numbers.
  • fast_forward00:13:21 - And probably slightly more different types as well.
  • fast_forward00:13:25 - Right. Okay, so now we have these 300 different types of receptor neurons.
  • fast_forward00:13:32 - As a population, they're active. We also know that they are all highly specific
  • fast_forward00:13:37 - in how they get integrated at the next stage in this glomeruli.
  • fast_forward00:13:41 - Glomeruli are very much specific to the olfactory receptor neuron types, roughly.
  • fast_forward00:13:48 - And now with that, also in the case of humans, let's say we can at least detect
  • fast_forward00:13:52 - about 10,000 different molecules with this.
  • fast_forward00:13:56 - Well, yeah, it's not so much molecules. There's this number of 10,000 sort of
  • fast_forward00:14:01 - flies around olfaction.
  • fast_forward00:14:03 - You're always hearing this number of 10,000. It has kind of a funny story,
  • fast_forward00:14:06 - actually, that originally this number of 10,000 came from perfumers,
  • fast_forward00:14:12 - I think, who were asked how many different, say,
  • fast_forward00:14:16 - different perceptual qualities or different perceptual nuances could you have for a trained perfumer?
  • fast_forward00:14:24 - And some perfumer just sort of plucked this number of 10,000.
  • fast_forward00:14:28 - They thought there might be a possibility of 10,000 different perceptions that
  • fast_forward00:14:32 - you might be able to have.
  • fast_forward00:14:33 - Um, and that's about the limit of the evidence for it really.
  • fast_forward00:14:37 - And that number has been sort of bandied, bandied about in, uh, olfaction ever since.
  • fast_forward00:14:42 - So it doesn't necessarily correspond to the number of chemicals. Uh, and in fact.
  • fast_forward00:14:48 - You can have almost an infinite variety of molecules that you could have a smell response to.
  • fast_forward00:14:57 - But I'm sure in the ethology of olfaction, people must have made some sort of
  • fast_forward00:15:03 - taxonomy of the whole collection of molecules that are detectable in principle
  • fast_forward00:15:08 - by our olfactory system.
  • fast_forward00:15:10 - So how large would that set be? Sure, yeah. How is that organized?
  • fast_forward00:15:13 - We know at least a few thousand different chemically active,
  • fast_forward00:15:18 - olfactory active molecules that would give you a response in some way.
  • fast_forward00:15:23 - And how big a subset is that of all molecules?
  • fast_forward00:15:28 - Well, all molecules, of course, that's including going way up to,
  • fast_forward00:15:32 - you know, massive proteins, 30,000, you know, Daltons, you know,
  • fast_forward00:15:37 - 30 kilodaltons, enormous molecular machines, right?
  • fast_forward00:15:41 - Because we're only looking at quite a narrow bandwidth.
  • fast_forward00:15:45 - There's actually an interesting story about that. At our recent NICE Odor Maps workshop,
  • fast_forward00:15:52 - we heard that someone did an analysis of the lower end of this quite narrow
  • fast_forward00:15:57 - spectrum between 30 and 300 Daltons of what we can appreciate.
  • fast_forward00:16:02 - And they looked at all the molecules down at the bottom
  • fast_forward00:16:05 - end in terms of um uh in
  • fast_forward00:16:10 - terms of our response them and they they looked
  • fast_forward00:16:12 - at the molecules just under the threshold so just under 30 30 daltons and they
  • fast_forward00:16:18 - called these um infra smells okay and then they looked at the molecules just
  • fast_forward00:16:23 - above the 300th uh threshold and they called these ultra smells you know in
  • fast_forward00:16:28 - line with vision as you can imagine and um,
  • fast_forward00:16:31 - And it showed some interesting things. In fact, there seems to be a kind of
  • fast_forward00:16:38 - story that these infra-smells may share with them a principle of kind of toxicity.
  • fast_forward00:16:44 - That maybe these lighter molecules are getting towards an area of toxicity.
  • fast_forward00:16:49 - And this may also be a guiding principle in terms of how our perception of odors may be organized.
  • fast_forward00:16:56 - That we're obviously detecting these to be safe or not.
  • fast_forward00:17:00 - Right. But it should be telling us something about also the specific niche for
  • fast_forward00:17:06 - which we have been optimized, right?
  • fast_forward00:17:09 - So why – because it's interesting that we're not sensitive to all possible molecules
  • fast_forward00:17:14 - because actually in the big picture, it's actually a very small subset of all
  • fast_forward00:17:18 - molecules we could ever encounter.
  • fast_forward00:17:20 - So why is this subset then behaviorally and also from a perspective of fitness,
  • fast_forward00:17:26 - evolutionary fitness? It's really relevant.
  • fast_forward00:17:28 - Really a good question. i mean i think i think it's probably limited if
  • fast_forward00:17:31 - uh by by most things is that once you get
  • fast_forward00:17:34 - beyond 300 daltons things uh don't become
  • fast_forward00:17:37 - volatile anymore right they're heavier molecules uh they
  • fast_forward00:17:40 - become less volatile so basically your chances of
  • fast_forward00:17:43 - being able to uh get these into the air to even be able to smell them at all
  • fast_forward00:17:48 - starts to become very limited right so you're sort of limited by the physics
  • fast_forward00:17:51 - now below uh below 30 daltons i'm not quite sure why that lower limit is right
  • fast_forward00:17:58 - because because you certainly can have lighter molecules than that.
  • fast_forward00:18:03 - Could it be a limitation of biophysics of the receptor neuron?
  • fast_forward00:18:08 - For the lighter molecules? Yeah, well, I'm not sure.
  • fast_forward00:18:12 - It may also be that many of these lighter molecules,
  • fast_forward00:18:17 - you don't necessarily want to be detecting, like extremely light molecules,
  • fast_forward00:18:21 - oxygen, CO2, and so on, are so common in the world that potentially maybe they
  • fast_forward00:18:27 - would flood the whole system.
  • fast_forward00:18:29 - So maybe you need to be more selective to these compounds.
  • fast_forward00:18:32 - Although the common story in olfaction is that carbon dioxide doesn't have a smell.
  • fast_forward00:18:37 - In fact, it does have a slight smell, and that may be through interacting with
  • fast_forward00:18:41 - other molecular things.
  • fast_forward00:18:42 - But probably these extremely light molecules, maybe there just isn't a behavioral
  • fast_forward00:18:48 - reason for responding to them on the basis that they're there all the time.
  • fast_forward00:18:52 - Yeah, but that's of course with the circular argument, right?
  • fast_forward00:18:53 - Because you're saying, look, well, we don't smell them because we don't smell
  • fast_forward00:18:55 - them. I mean, it's illogical in that way.
  • fast_forward00:18:58 - But at the end of, but no, the argument may be, I mean, if they're there all
  • fast_forward00:19:02 - the time, they don't give you any information, right? So in the end, it's okay.
  • fast_forward00:19:06 - But also that argument is not so convincing, right?
  • fast_forward00:19:09 - But for instance, is it possible that we have molecules binding to our receptors
  • fast_forward00:19:15 - but not leading to sufficient activation to create percepts?
  • fast_forward00:19:20 - That certainly seems to be true. I mean, part of the story in my talk was that
  • fast_forward00:19:27 - there really can be two parameters for a molecule interacting with a receptor.
  • fast_forward00:19:33 - Receptor, for a long time everyone's considered that the main parameter for
  • fast_forward00:19:38 - a ligand is its affinity to a receptor, which actually tells you how strongly it wants to bind to it.
  • fast_forward00:19:45 - And we know that ligands and receptors have a very wide distribution of these affinities.
  • fast_forward00:19:52 - But it's also very clear that in more recent evidence is that there's a second parameter,
  • fast_forward00:19:57 - which in pharmacology has been called efficacy and
  • fast_forward00:20:01 - this tells you once a ligand has
  • fast_forward00:20:04 - bound to a receptor how much kind of activity does it give downstream
  • fast_forward00:20:07 - to an orn right and so that's related
  • fast_forward00:20:10 - very much to your question that if you have for instance uh molecules
  • fast_forward00:20:13 - that are binding there uh but they don't necessarily contribute much to uh to
  • fast_forward00:20:18 - a um to a uh to a neural response then this is a classic so-called antagonist
  • fast_forward00:20:25 - which is effectively filling up a space and sort of creating a blind spot, right?
  • fast_forward00:20:33 - And up until now, until the last five years, we didn't really know about these antagonisms.
  • fast_forward00:20:43 - And as you can imagine, they're kind of a bit scary experimentally.
  • fast_forward00:20:50 - Because in order to be able to see them, you have to look at all possible mixtures,
  • fast_forward00:20:55 - and you have to see how certain molecules may be blocking other molecules.
  • fast_forward00:21:00 - And it's only really in the last five years that we've started to appreciate
  • fast_forward00:21:04 - that, in fact, these types of antagonisms could indeed be really widespread in a faction.
  • fast_forward00:21:11 - That basically our neural response is nowhere near the linear addition of just
  • fast_forward00:21:17 - adding on lots of different odors.
  • fast_forward00:21:19 - There's all sorts of these nonlinear competitions and interactions that are
  • fast_forward00:21:24 - taking place, and this is probably what defines olfaction.
  • fast_forward00:21:27 - So that might also mean that the response of glomeruli,
  • fast_forward00:21:32 - mitral cells, and so on, is reflecting a much more complex flux of binding and
  • fast_forward00:21:38 - unbinding of very broadly shaped ligands.
  • fast_forward00:21:44 - Certainly. And how they interact. And we haven't even said anything about the
  • fast_forward00:21:50 - so-called odorant binding properties which live in the mucosa.
  • fast_forward00:21:54 - And these OBPs are doing two things.
  • fast_forward00:21:58 - One thing they're doing is, in quite a clever way, shifting the sorption spectra
  • fast_forward00:22:03 - for odors out there in the air phase so that we can detect odors that are hydrophobic.
  • fast_forward00:22:08 - They don't want to be in the liquid phase. Yes.
  • fast_forward00:22:11 - We need to be able to detect those as well. So that's a solution from nature
  • fast_forward00:22:15 - to get those into the liquid phase so that we can interact with them.
  • fast_forward00:22:19 - And another thing they do, they're very large proteins. They themselves have
  • fast_forward00:22:24 - selective bindings to different chemicals.
  • fast_forward00:22:26 - And so it's actually an extra layer of building specificity and tuning into
  • fast_forward00:22:33 - the system to create more information.
  • fast_forward00:22:35 - It's really quite an exquisitely layered sequence of events that deliver the
  • fast_forward00:22:41 - certain subsets of molecules in certain places in the epithelium to be processed.
  • fast_forward00:22:46 - But these odor-binding molecules, is there a belief that they in turn set up
  • fast_forward00:22:53 - a local dynamic in the mucus layer, or are they just like transporters?
  • fast_forward00:22:59 - I think you can think of them more like transporters because the mucosal layer
  • fast_forward00:23:02 - is very thin. It's only a few microns.
  • fast_forward00:23:04 - And their main job is to basically shift the sorption spectrum to get those
  • fast_forward00:23:08 - hydrophobic compounds into the liquid phase and transport them to the binding
  • fast_forward00:23:14 - site so that they can be… You see, they're probably not traveling lengthways.
  • fast_forward00:23:18 - There probably isn't time really for them to, say, move them to different portions
  • fast_forward00:23:23 - of the sheet, for instance, or something complicated like this, right? Right.
  • fast_forward00:23:26 - So there are other mechanisms for that. For instance,
  • fast_forward00:23:30 - the fact that you've got preferential sorption means that some odors are staying
  • fast_forward00:23:35 - in longer in the air phase and some shorter,
  • fast_forward00:23:38 - which means that you've already got a nice sort of almost like postal service
  • fast_forward00:23:43 - selective delivery of molecules in certain places depending upon how much they want to be sorbed.
  • fast_forward00:23:49 - Yeah, but I was more thinking you could also imagine that these binding molecules
  • fast_forward00:23:55 - in turn, for instance, set up a competitive relationship.
  • fast_forward00:23:59 - For instance, if you have a binding with a molecule of a certain kind that it
  • fast_forward00:24:04 - leads to, let's say, signaling systems becoming active.
  • fast_forward00:24:09 - That would change again probabilities to bind to other molecules or not.
  • fast_forward00:24:13 - Yeah, it's quite possible that there's some sort of interaction between those
  • fast_forward00:24:17 - at the receptors. I don't think it's really well understood.
  • fast_forward00:24:19 - And for instance, even antagonism is not understood at all well because the
  • fast_forward00:24:25 - simple-minded naive theory would be that an antagonist sort of fills the binding
  • fast_forward00:24:30 - site on a receptor so that another ligand with higher efficacy cannot bind in that same place.
  • fast_forward00:24:36 - But it seems quite likely that, in fact, there are more complicated mechanisms
  • fast_forward00:24:44 - which are called dimers, which effectively means that they may not fill the main site.
  • fast_forward00:24:50 - The main site may still be available for preferentially or high affinity binded
  • fast_forward00:24:56 - compounds, but that these antagonists may actually bind onto the receptor in different places.
  • fast_forward00:25:01 - And by doing that, actually change how the conformation of the protein.
  • fast_forward00:25:05 - So in extremely complex ways that the protein can change its folding dynamics
  • fast_forward00:25:11 - according to the binding to signal different stuff.
  • fast_forward00:25:13 - So there's all sorts of complexities there. But in terms of your point about the dynamics,
  • fast_forward00:25:18 - that's well taken because there's also this other group of compounds in the
  • fast_forward00:25:22 - mucosa called odor degrading enzymes because a very good question is how do
  • fast_forward00:25:27 - you get rid of these odors after binding?
  • fast_forward00:25:29 - There needs to be a whole set of other molecules out there that are effectively
  • fast_forward00:25:35 - basically removing these signals and terminating the signals.
  • fast_forward00:25:39 - So it's a very exquisite balance of all of these molecular processes going on
  • fast_forward00:25:44 - to both initiate the signal and then terminate it as well in a timely fashion.
  • fast_forward00:25:50 - It's actually, it's already amazing to see, right, that here we're looking at,
  • fast_forward00:25:53 - let's say, a very advanced variation on, in some sense, the most primitive form of sensation we know.
  • fast_forward00:26:00 - Because for single cellular organisms, you have, you know, mechanosensing or
  • fast_forward00:26:05 - chemical sensing, and that's what it started with, right?
  • fast_forward00:26:08 - Yeah. So, and if you just look at it already at the level of this mucus layer,
  • fast_forward00:26:13 - the very first point where olfactory release starts, the exquisite orchestration
  • fast_forward00:26:19 - of all these different pathways and interactions is very impressive.
  • fast_forward00:26:23 - But then now, of course, we want to go step up and start to think about,
  • fast_forward00:26:28 - okay, how does this in the end lead to the encoding and representations and
  • fast_forward00:26:33 - detection of odors in the environment?
  • fast_forward00:26:36 - And so the first thing, of course, we want to understand is what's really this molecular language?
  • fast_forward00:26:42 - How would an olfactory system in the end really, if you want, decompose molecules?
  • fast_forward00:26:47 - What are the key features of molecules our olfactory system is sensitive to?
  • fast_forward00:26:52 - Yeah, so there's some very nice data sets. I think finally olfaction is coming
  • fast_forward00:26:58 - into the 21st century of sort of data sharing and there's some very nice data
  • fast_forward00:27:03 - sets with many different odor conditions and optical imaging on the receptor surface.
  • fast_forward00:27:09 - Using these methods, we can directly see actually what the, for instance,
  • fast_forward00:27:14 - calcium levels of activity are, different portions of the receptor sheet.
  • fast_forward00:27:19 - And large data sets which have shown how different portions of the sheet are
  • fast_forward00:27:26 - responding in different ways to different groups of chemicals.
  • fast_forward00:27:29 - And it's very clear when you look at these sorts of databases that different
  • fast_forward00:27:34 - portions of the receptor sheet are responding to different groups of compounds.
  • fast_forward00:27:40 - So there are sort of, you can see common properties of molecules that may be
  • fast_forward00:27:49 - preferentially tuning or activating different parts of the receptor sheet.
  • fast_forward00:27:54 - And so this is
  • fast_forward00:27:57 - leading to the idea that there is a kind of chemotopic mapping
  • fast_forward00:28:01 - of the receptor sheet rather than the receptor sheet
  • fast_forward00:28:05 - being mapped onto physical space as
  • fast_forward00:28:08 - it is in say vision on the retina the
  • fast_forward00:28:11 - mapping seems to be a chemotopic mapping so we talk about different molecular
  • fast_forward00:28:16 - features being mapped to different areas and this may be for a couple of reasons
  • fast_forward00:28:19 - a number of reasons actually one reason is this sorption thing that I talked
  • fast_forward00:28:24 - about that compounds with higher sorption parameters,
  • fast_forward00:28:29 - so they like to be in the liquid phase, they're more hydrophilic.
  • fast_forward00:28:34 - Those will basically absorb into the earlier parts of the receptor sheet as you sniff.
  • fast_forward00:28:42 - So they'll be sort of more delivered to the front end of the receptors where they live in the mucosa.
  • fast_forward00:28:48 - And other compounds that are more hydrophobic would be delivered later,
  • fast_forward00:28:54 - further back in the receptor sheet as you sniff.
  • fast_forward00:28:58 - And there are other reasons is that this family of receptors that I talked about,
  • fast_forward00:29:03 - in fact, it turns out that there's at least four classes of these,
  • fast_forward00:29:07 - although this is currently under debate, but there's two very, very clear classes,
  • fast_forward00:29:13 - class 1 and class 2, which correspond to fishoid receptors, which in fact are
  • fast_forward00:29:19 - leftover that we have from when we were fish.
  • fast_forward00:29:23 - They're dedicated to detecting non-volatile compounds largely,
  • fast_forward00:29:28 - and they are exclusively expressed in certain zones of the receptor sheet.
  • fast_forward00:29:32 - And so, as you can imagine, this leads to another form of chemotopic mapping
  • fast_forward00:29:37 - where you get a preferential response.
  • fast_forward00:29:40 - And these FISH-derived receptors wouldn't go for the higher Dalton molecules
  • fast_forward00:29:45 - because these are the ones that are less volatile?
  • fast_forward00:29:49 - No, not necessarily because whether a compound is hydrophilic or hydrophobic
  • fast_forward00:29:55 - depends more on actually the charge on the molecule. So it's more to do with the sort of asymmetry.
  • fast_forward00:30:03 - No, but you said they went for the non-volatiles or the less volatile.
  • fast_forward00:30:07 - Yeah, yeah, yeah. So I was wondering then whether these fish-derived receptors
  • fast_forward00:30:14 - are also sensitive to molecules with higher dose. Well, there's two things that
  • fast_forward00:30:20 - determine sort of volatility.
  • fast_forward00:30:22 - One is the molecular mass, and the other thing is whether they like the air
  • fast_forward00:30:25 - phase or the liquid phase, which is the sorption profile. So it's really those
  • fast_forward00:30:29 - two things together that determine.
  • fast_forward00:30:31 - So in a fishoid receptor, they would be exclusively compounds that they have
  • fast_forward00:30:38 - sorption parameters that they can never make it into the air phase really at
  • fast_forward00:30:42 - room temperatures unless you started boiling them or had a lot higher energy.
  • fast_forward00:30:47 - Okay, but so now we have a zoning, if you want, of the epithelium.
  • fast_forward00:30:52 - And the zoning might correspond to different properties of molecules like their
  • fast_forward00:30:56 - cyclization, their carbon number, bond saturation, branching,
  • fast_forward00:31:01 - substitution pattern, functional groups.
  • fast_forward00:31:05 - So are these the key chemical features of these molecules that you would map onto these zones?
  • fast_forward00:31:10 - These are at least a number of key features. Those ones you've just mentioned are very important.
  • fast_forward00:31:18 - This has been another area of a crucial study.
  • fast_forward00:31:21 - There's databases. databases uh there's some very nice software that you can
  • fast_forward00:31:25 - download for public use called dragon,
  • fast_forward00:31:29 - and for pretty much any molecule you might care about you can go to this package
  • fast_forward00:31:36 - and it will give you something like um.
  • fast_forward00:31:39 - I think it's about 400 or 500 different exotic descriptors for a molecule,
  • fast_forward00:31:44 - branching, all sorts of things that you can't imagine.
  • fast_forward00:31:47 - So a huge number of possible descriptors, because, of course,
  • fast_forward00:31:50 - you can describe a molecular structure in an almost infinite set of ways that
  • fast_forward00:31:58 - you might choose to describe that, right?
  • fast_forward00:32:01 - So these databases and chemists
  • fast_forward00:32:04 - have been very carefully characterizing different descriptor
  • fast_forward00:32:07 - or what you might call molecular determinants and then
  • fast_forward00:32:11 - of course you can play very interesting games like um
  • fast_forward00:32:14 - look at all those odorous compounds and look
  • fast_forward00:32:17 - at all the descriptors are there are there certain descriptors that
  • fast_forward00:32:20 - may be more important for certain areas of the receptor sheet
  • fast_forward00:32:23 - and others that aren't and so on you can
  • fast_forward00:32:25 - play games like this and you find that um indeed there are
  • fast_forward00:32:28 - um but you also find that um all of
  • fast_forward00:32:31 - these descriptors are highly redundant and what i mean by that is that um you
  • fast_forward00:32:35 - know you don't find for instance that there's one and two one or two descriptors
  • fast_forward00:32:40 - that sort of uh that sort of tell you very orthogonal or decorrelated information
  • fast_forward00:32:46 - they all tend to be measuring a similar thing.
  • fast_forward00:32:51 - And uh and so if you take for instance a pca
  • fast_forward00:32:54 - or principal component analysis of the contribution of all
  • fast_forward00:32:57 - of these discriminants to what you might think of as a perceptual description
  • fast_forward00:33:00 - of an odor you find that there are large numbers
  • fast_forward00:33:03 - you know because they're highly redundant so right exactly you
  • fast_forward00:33:07 - can no but so this is very interesting right because
  • fast_forward00:33:09 - so here we go if you want i
  • fast_forward00:33:12 - mean this is a little bit proto-science right where we just develop descriptors
  • fast_forward00:33:17 - of reality so the same holds for chemistry so here we are for this huge collection
  • fast_forward00:33:21 - of descriptors um but now on the other hand you'll see also you see two things
  • fast_forward00:33:27 - right so on the one there's a zoning a zoning in the olfactory system that would suggest,
  • fast_forward00:33:33 - that the properties of these molecules are in some sense also mapped onto this chemotopic map.
  • fast_forward00:33:40 - And that, of course, would also give you, let's say, this should give you some
  • fast_forward00:33:45 - sort of hierarchy of which descriptors actually are helpful or are,
  • fast_forward00:33:49 - from a biological perspective, relevant and which are indeed,
  • fast_forward00:33:53 - in that sense, redundant. That's one thing.
  • fast_forward00:33:56 - But then you showed, which is also very interesting, Interesting.
  • fast_forward00:33:59 - And if you analyze in more detail these descriptors and the odor space that
  • fast_forward00:34:03 - they define, that actually it's much lower dimensional than these descriptors would suggest.
  • fast_forward00:34:10 - Yeah, well, that's another very interesting story. Although that one that I
  • fast_forward00:34:13 - showed in the talk was actually about analyzing perceptual descriptors for a corresponding odor.
  • fast_forward00:34:19 - So you can do another thing, which is quite a lot of fun, which is to take a
  • fast_forward00:34:25 - large number of molecules, such as in the Dravnik's database,
  • fast_forward00:34:30 - where about 150 different molecules were taken.
  • fast_forward00:34:33 - And they then used a large pool of trained olfactory specialists,
  • fast_forward00:34:40 - and they asked those specialists to describe each of these 150 odors using a
  • fast_forward00:34:51 - panel of approximately 150 different descriptors,
  • fast_forward00:34:55 - such as musky,
  • fast_forward00:34:58 - grassy, nutty, so on.
  • fast_forward00:35:01 - So large numbers of these and then you can also look at the space for similarity
  • fast_forward00:35:06 - of those for different molecules so you can plot individual,
  • fast_forward00:35:09 - molecules in a high dimensional space and you can look how similar or dissimilar
  • fast_forward00:35:14 - are these discriminators so you can make a sort of odour map.
  • fast_forward00:35:17 - Of perceptual odor map in terms of
  • fast_forward00:35:20 - what you would expect is that you'd expect those points
  • fast_forward00:35:24 - uh close in this map to have similar perceptual
  • fast_forward00:35:27 - properties and therefore you'd expect them to have similar molecular properties
  • fast_forward00:35:31 - right so it enables you to play this nice game where
  • fast_forward00:35:34 - you can look at different portions of this map and see are
  • fast_forward00:35:37 - is there a nice continuum here in certain molecular
  • fast_forward00:35:40 - properties that are moving this perception in
  • fast_forward00:35:43 - this direction or whatever But what
  • fast_forward00:35:46 - you find which is very interesting when you do this Which was studied by Kolokoff
  • fast_forward00:35:49 - paper about three years ago Was
  • fast_forward00:35:53 - that when you analyse all of these 150 descriptors You
  • fast_forward00:35:57 - find that they're placed in a
  • fast_forward00:36:00 - much lower dimensional space
  • fast_forward00:36:04 - So it means that effectively they're highly
  • fast_forward00:36:08 - correlated right so and and um
  • fast_forward00:36:11 - they should they found that there was a uh at
  • fast_forward00:36:15 - least in a two-dimensional manifold was able
  • fast_forward00:36:18 - to explain something like um 60 or
  • fast_forward00:36:21 - 70 percent of the variance of this of this 150 dimensional structure can actually
  • fast_forward00:36:26 - be explained quite adequately reasonably well in this two-dimensional manifold
  • fast_forward00:36:32 - which is spread it looks a bit like a potato chip sort of shape within this.
  • fast_forward00:36:38 - Higher dimensional space.
  • fast_forward00:36:40 - And so I think it's very exciting actually. I think we will find in the future,
  • fast_forward00:36:46 - olfactory studies, we're going to find that there are all sorts of these.
  • fast_forward00:36:51 - Lower dimensional manifolds, which is kind of the answer to solving the olfactory
  • fast_forward00:36:57 - code to find out there has to be some underlying similarities and structures in these things.
  • fast_forward00:37:02 - This is of course where we want to get to, right? One thing that's,
  • fast_forward00:37:05 - of course, interesting here is that now, with this last analysis,
  • fast_forward00:37:09 - you could argue, well, maybe human experience of olfaction is relatively low-dimensional.
  • fast_forward00:37:15 - But what's complex are all these descriptions we glue onto.
  • fast_forward00:37:19 - It's also the language, right? Because this is limited by language.
  • fast_forward00:37:22 - The language just complexifies the experience.
  • fast_forward00:37:26 - But what's amazing about olfaction is that we have no other way to describe
  • fast_forward00:37:29 - certain olfactory cues other than it's like other things.
  • fast_forward00:37:34 - So this is kind of the interesting restrictions in the language because we say
  • fast_forward00:37:38 - it's like grass or it's like this or like that.
  • fast_forward00:37:42 - It doesn't seem that we have any kind of independent, really independent descriptions
  • fast_forward00:37:48 - of owners other than just likening them to other things. Right.
  • fast_forward00:37:52 - So there are no intrinsic qualities that we can use. Well, it turns out when
  • fast_forward00:37:57 - you look at this map, because you can look at this manifold and then you can
  • fast_forward00:38:01 - look at what are the actual qualities and how does this manifold actually represent the odors.
  • fast_forward00:38:07 - When you do that, you find some interesting things. So you see that the main
  • fast_forward00:38:11 - dimension of this map very strongly correlates to pleasantness and unpleasantness.
  • fast_forward00:38:16 - So maybe this is the only true underlying parameter, whether we actually have
  • fast_forward00:38:22 - a hedonic or whatever the valence is for the actual. Exactly right, yes.
  • fast_forward00:38:28 - Which would be interesting because then you would have, let's say, valence and intensity.
  • fast_forward00:38:32 - And this would be very compatible with how we can describe the emotional states.
  • fast_forward00:38:38 - We would think of valence and arousal, something like this.
  • fast_forward00:38:41 - And then within such a two-dimensional space you can
  • fast_forward00:38:44 - find all sort of very complex emotional descriptors yeah but but so but the
  • fast_forward00:38:49 - point i want to get to is then um how what does this now mean for this idea
  • fast_forward00:38:54 - of zoning of of of the receptor neurons in the epithelium would that match would
  • fast_forward00:39:00 - it match in any way the structure of this descriptor space or not,
  • fast_forward00:39:05 - um yeah that's a really good question um.
  • fast_forward00:39:10 - Yes, I think it probably will. I mean, there are another approach to describing
  • fast_forward00:39:18 - the chemotopic mapping,
  • fast_forward00:39:20 - and this is work by Kaniezo Mori, who's been very influential in the field,
  • fast_forward00:39:25 - has made a very interesting argument that this space,
  • fast_forward00:39:30 - although it's a chemotopic space,
  • fast_forward00:39:33 - may also be structured behaviorally.
  • fast_forward00:39:36 - So certain behaviorally relevant compounds may be represented in certain parts
  • fast_forward00:39:42 - of the bulb and other differently behaviorally relevant compounds in other parts.
  • fast_forward00:39:47 - And you would expect those compounds to have similar perceptual kind of descriptions.
  • fast_forward00:39:52 - Right so it does seem that the
  • fast_forward00:39:56 - the current thinking of uh mapping of
  • fast_forward00:39:59 - the first stage of odor representation in the olfactory system comes down to
  • fast_forward00:40:04 - both chemotopic in terms of molecular properties uh and probably uh behaviorally
  • fast_forward00:40:10 - behavioral relevance in different ways uh which will then of course have some
  • fast_forward00:40:15 - type of relationship to perceptual quality right exactly,
  • fast_forward00:40:18 - and this is cool right so now we've had a bit an idea of the structuring of
  • fast_forward00:40:21 - this of this sort of The receptor layer, there's a structure.
  • fast_forward00:40:25 - And now if you measure, and these are classic experiments that you also showed,
  • fast_forward00:40:30 - if you measure the response of single receptor neurons to different kinds of
  • fast_forward00:40:35 - odorants, different kinds of molecules, you see that actually they're not so specifically tuned.
  • fast_forward00:40:41 - So what's the deal there exactly? Yeah, no, this is a very good point.
  • fast_forward00:40:46 - What you find when you look in detail at each one of these receptor types is
  • fast_forward00:40:51 - that they tend to have what we call broad tuning, as you say,
  • fast_forward00:40:56 - to large numbers of compounds.
  • fast_forward00:40:59 - But yet there are also a sort of smaller subset, I guess, which are more narrowly tuned.
  • fast_forward00:41:10 - So it seems that the current story is that in fact the olfactory system has
  • fast_forward00:41:15 - a variety of broadly tuned and more specific tuning.
  • fast_forward00:41:18 - And one of the things I showed in the talk, which was some work that I did with
  • fast_forward00:41:22 - Manuel Sanchez quite a while ago, actually 10 years,
  • fast_forward00:41:25 - where we applied an information theory approach to basically describe the accuracy
  • fast_forward00:41:34 - that a neuronal code at the epithelium would be able to reconstruct the original stimulus.
  • fast_forward00:41:40 - And when you do this for very high dimensional stimulus like we have here with
  • fast_forward00:41:45 - large numbers of compounds,
  • fast_forward00:41:46 - you find that in fact this is exactly the perfect strategy that you need because
  • fast_forward00:41:51 - you want to have certain receptors that are very broadly discriminating between
  • fast_forward00:41:58 - groups of compounds. So they're very broadly tuned.
  • fast_forward00:42:01 - And then you want other receptors which are doing more fine things.
  • fast_forward00:42:06 - Uh discriminations between you know more specific sets
  • fast_forward00:42:09 - of compounds and that if you have this
  • fast_forward00:42:12 - sort of two layers and some continuum between them uh
  • fast_forward00:42:16 - in information theory terms this gives you the best possible ability to reconstruct
  • fast_forward00:42:22 - the stimulus so you you get in a high dimensional space now if your problem
  • fast_forward00:42:28 - the reason why this is interesting is that uh if you restrict the problem to
  • fast_forward00:42:33 - a lower a dimensional space,
  • fast_forward00:42:35 - such as I know Bill Hanson, Peter Mombats were here last week.
  • fast_forward00:42:40 - And for instance, Bill will be working, or the animals that Bill studies will
  • fast_forward00:42:45 - be working in much more restricted dimensional space because ecologically, maybe there's fewer.
  • fast_forward00:42:53 - Compounds which are more relevant to, you know, ethologically and ecologically for the animal.
  • fast_forward00:42:58 - Then in those types of olfactory systems
  • fast_forward00:43:01 - you tend to find that there are more specific receptors
  • fast_forward00:43:04 - because you don't get this advantage of
  • fast_forward00:43:07 - having lots of broad tune because you don't have the similar very
  • fast_forward00:43:11 - high dimensionality so your your prediction is that
  • fast_forward00:43:13 - the ratio of broad or specifically tuned
  • fast_forward00:43:16 - receptor neurons would scale with the outer space
  • fast_forward00:43:19 - you have to classify absolutely and the complexity of
  • fast_forward00:43:23 - the world within you within which you have to operate right because that this
  • fast_forward00:43:27 - Just coming back to the point I made at the beginning that if you need this
  • fast_forward00:43:30 - very xenobiotic property where you have to be able to basically associate any
  • fast_forward00:43:37 - potential molecule out there in the world with a particular behavior or whatever,
  • fast_forward00:43:42 - then you very much need this ability to deal with very high dimensional stimuli.
  • fast_forward00:43:48 - Because if you're an insect, there's probably no real requirement for these
  • fast_forward00:43:52 - sorts of things in certain settings. Right.
  • fast_forward00:43:56 - But now, let's go back to this theoretical study you just described.
  • fast_forward00:44:01 - And sometimes you have a fairly linear decomposition of the olfactory system.
  • fast_forward00:44:08 - A perception problem, right? Because you basically, you assumed,
  • fast_forward00:44:10 - look, I have a bunch of, I have a stimulus, stimulus comes in,
  • fast_forward00:44:14 - I have my receptor neurons, they feed directly into some sort of response.
  • fast_forward00:44:21 - Presumably in the glomeruli or the mitral cells.
  • fast_forward00:44:24 - And then I'm going to use some estimator. You don't need to make any commitments
  • fast_forward00:44:29 - where this estimator resides in the brain because now we're doing it for an
  • fast_forward00:44:32 - information theory game.
  • fast_forward00:44:34 - And now I get an estimated stimulus.
  • fast_forward00:44:38 - And then the game is that you really want to understand the parameters under
  • fast_forward00:44:41 - which this estimated stimulus is as equal or similar as possible to the real
  • fast_forward00:44:46 - stimulus, so you have a minimum. The error is minimal. Exactly. All right?
  • fast_forward00:44:50 - So then, of course, the usual game is, well, the response will have some noise,
  • fast_forward00:44:55 - and this noise will have certain properties.
  • fast_forward00:44:59 - So because of this noise, I cannot really inversely map anymore my response
  • fast_forward00:45:04 - to the stimulus. Exactly.
  • fast_forward00:45:06 - Right? But okay, when the noise is big. It corrupts the original. Exactly.
  • fast_forward00:45:10 - So now you use this Fisher information approach to decompose this problem and
  • fast_forward00:45:16 - understand what's the optimal strategy.
  • fast_forward00:45:18 - Exactly. So how does this Fisher information approach help you with this?
  • fast_forward00:45:23 - Well, Fisher information effectively characterizes and quantifies very precisely
  • fast_forward00:45:30 - two things, which is both your sensitivity to a stimulus.
  • fast_forward00:45:34 - So the sensitivity that a receptor
  • fast_forward00:45:38 - has to a stimulus and also how that sensitivity is corrupted by noise.
  • fast_forward00:45:47 - And then it also gives you another very nice thing, which is how does that add across a population?
  • fast_forward00:45:54 - Because you might know that for one receptor, but how do you know how all that
  • fast_forward00:46:00 - sort of information adds when you have a population? And so Fischer information
  • fast_forward00:46:04 - precisely quantifies that effectively.
  • fast_forward00:46:06 - So for an array of receptors, it quantifies very precisely for each receptor
  • fast_forward00:46:13 - what the contribution is to the sensitivity for each one of the dimensions and
  • fast_forward00:46:18 - also how that is corrupted by its own noise.
  • fast_forward00:46:21 - And effectively, when you characterize that, there are some very nice results,
  • fast_forward00:46:28 - particularly by Kramer-Rayo, which tells you that the inverse of this matrix,
  • fast_forward00:46:35 - which contains all of this information about the noise and the sensitivities,
  • fast_forward00:46:39 - basically the inverse of this directly tells you what the best estimator can
  • fast_forward00:46:48 - do in terms of being able to,
  • fast_forward00:46:52 - reconstruct the original stimulus.
  • fast_forward00:46:54 - And so therefore it tells you that it doesn't matter what the biological system
  • fast_forward00:46:58 - is, it will not be able to exceed this limit and so it's nice very nice I think in the sense that it,
  • fast_forward00:47:06 - provides a very defined limit on
  • fast_forward00:47:09 - a psychophysical test that you might do to tell
  • fast_forward00:47:12 - you for instance if you had a stimulus and you changed it a small amount so
  • fast_forward00:47:17 - the just noticeable difference the just noticeable difference or the jnd in
  • fast_forward00:47:22 - the psychophysics experiment would relate directly to this uh inverse or this
  • fast_forward00:47:27 - kramer around which comes so it makes a very nice firm prediction
  • fast_forward00:47:31 - for the architecture of a neural system and uh and and what the limits would
  • fast_forward00:47:37 - be for an animal right have you been able to to validate any of this any of these predictions.
  • fast_forward00:47:42 - Um well in terms of relating it to psychophysics
  • fast_forward00:47:46 - um that's something
  • fast_forward00:47:49 - that's something we should we should push forward but
  • fast_forward00:47:52 - in this experiment we're only really doing it as a game rather than because
  • fast_forward00:47:56 - it doesn't really matter what the precise value is because all we were really
  • fast_forward00:48:01 - interested to know is what are the actual features of the receptors which either
  • fast_forward00:48:07 - give you a good or bad ability to do this.
  • fast_forward00:48:09 - Right, but now you do assume that in this case you only have a linear interaction
  • fast_forward00:48:15 - between your receptor and your actual linear combination, right?
  • fast_forward00:48:18 - And is that not a very strong assumption?
  • fast_forward00:48:21 - It is quite a strong assumption. I mean, what you're saying is that a receptor
  • fast_forward00:48:26 - is giving a linear response to a set of combinations.
  • fast_forward00:48:31 - So, for instance, things like antagonism are kind of out. As we just discussed, right?
  • fast_forward00:48:36 - Yeah. So, you have to go to a sort of second order. Yeah.
  • fast_forward00:48:45 - You could extend it to a second order, which we haven't done.
  • fast_forward00:48:49 - But really it was a quick study to just show what the features are.
  • fast_forward00:48:52 - And I think the main sort of interest in it is that the simple properties of
  • fast_forward00:48:58 - the simple-minded approach is that you get this sort of mixture of diversity
  • fast_forward00:49:02 - in the outputs together with specific receptors that very clearly matches what
  • fast_forward00:49:09 - you see in all experiments from insects and mammals.
  • fast_forward00:49:12 - In some sense, you try to make sense of this idea of population coding with
  • fast_forward00:49:16 - broadly tuned receptor neurons that have, again, variability in this tuning.
  • fast_forward00:49:21 - This is the key point of this study.
  • fast_forward00:49:23 - Exactly. It's just to show that what the biology is doing is a sensible thing
  • fast_forward00:49:29 - in terms of when you're facing a very high-dimensional input.
  • fast_forward00:49:33 - Right, which is interesting, right? Because very often you hear people making
  • fast_forward00:49:38 - these wild claims about how suboptimal, inefficient natural systems are.
  • fast_forward00:49:43 - But I think, certainly with this example, you'll be hard-pressed to do it better.
  • fast_forward00:49:48 - Would you agree with that? Yeah, within your space, within your defined space
  • fast_forward00:49:52 - of the problem you're dealing with.
  • fast_forward00:49:54 - Right, exactly. What would be interesting to compare, which we haven't done,
  • fast_forward00:49:58 - is to then compare these measures maybe across animals, right?
  • fast_forward00:50:01 - Sure. To show that different levels of specificity,
  • fast_forward00:50:05 - like we were discussing, processing may be more or less relevant to
  • fast_forward00:50:09 - the dimensionality of what you're dealing with in your in your environment
  • fast_forward00:50:12 - but now also the other thing though with your with your
  • fast_forward00:50:14 - model are there two things on the one hand you do assume that
  • fast_forward00:50:18 - after let's say one stage of of processing
  • fast_forward00:50:21 - one of transformation you can reconstruct your stimulus right that's your quality
  • fast_forward00:50:27 - measure well we don't assume that it actually does that all we're saying is
  • fast_forward00:50:31 - that this places a bound on what any subsequent neural processing would not
  • fast_forward00:50:35 - be able to go beyond this effectively any estimators.
  • fast_forward00:50:40 - Provided that any subsequence step is fully isolated from any preceding step.
  • fast_forward00:50:46 - It has to be a cleanly modular system.
  • fast_forward00:50:49 - No, I mean it just provides a limit on that. But what it's really saying is
  • fast_forward00:50:53 - it provides a limit given the noise that was introduced.
  • fast_forward00:50:56 - So it's based on the assumption of what the noise was at the periphery.
  • fast_forward00:51:00 - But for instance, if I would have a subset of projections that would also percolate
  • fast_forward00:51:05 - up this processing hierarchy, bypassing noise stages or following a different kind of noise dynamics.
  • fast_forward00:51:13 - Oh, okay. Well, the noise we're thinking about is only at this stage the noise
  • fast_forward00:51:17 - at the receptors, because all we're characterizing is the receptor population.
  • fast_forward00:51:21 - We're not saying anything about subsequent processing. So all we're really saying
  • fast_forward00:51:25 - is the question we wanted to answer was what should the tuning of receptors be,
  • fast_forward00:51:30 - not necessarily the subsequent neural processing, which is
  • fast_forward00:51:33 - a a different story the reason why you can't really do that with
  • fast_forward00:51:36 - this measure is that the subsequent neural processing has all
  • fast_forward00:51:39 - sorts of complex time dynamics as you say quite rightly
  • fast_forward00:51:42 - there's all sorts of maybe noise properties there in complex
  • fast_forward00:51:45 - ways there's all sorts of attentional processing and
  • fast_forward00:51:48 - so on so uh at least in
  • fast_forward00:51:51 - this formulation of uh let's say
  • fast_forward00:51:54 - thinking about the fisher information it would not make a
  • fast_forward00:51:57 - whole lot of sense to okay try to extend it beyond
  • fast_forward00:52:00 - uh subsequent neural processing thing we're just
  • fast_forward00:52:03 - really trying to limit it to what the receptors themselves
  • fast_forward00:52:06 - are doing right but now what so what's the signal to
  • fast_forward00:52:08 - noise level you find in these receptors signal to
  • fast_forward00:52:12 - noise um uh well obviously it's proson we find proson firing uh quite close
  • fast_forward00:52:18 - uh to sort of cortex uh sort of proson i mean as you know in cortex i think
  • fast_forward00:52:25 - the fano factor is slightly greater than in Poisson, so they show more variability.
  • fast_forward00:52:30 - The story with the variability of firing, you sort of see that also in receptors.
  • fast_forward00:52:35 - So you see large amount of variability in the firing, beyond,
  • fast_forward00:52:39 - slightly beyond Poisson firing.
  • fast_forward00:52:41 - Okay, but the other thing. That's the noise, as far as the receptor population's
  • fast_forward00:52:47 - concerned, it's both the noise and the firing. Right, exactly.
  • fast_forward00:52:51 - But now the other thing is that in this case,
  • fast_forward00:52:55 - This is under, I assume, a constant drive with a single molecule.
  • fast_forward00:53:00 - But the real system will be operating in the face of a continuous fluctuating
  • fast_forward00:53:04 - stream of molecules bombarding these different receptors with varying binding kinetics and so on.
  • fast_forward00:53:11 - So would this statement on this upper bound of processing capability also hold
  • fast_forward00:53:20 - if we start to generalize to this more realistic input condition?
  • fast_forward00:53:24 - Sure i mean it may not uh there's all
  • fast_forward00:53:27 - sorts of complex dynamical processes going on there and
  • fast_forward00:53:30 - the fact that it's a first order description is is
  • fast_forward00:53:33 - just the simplest minded thing you can do in order to right in
  • fast_forward00:53:36 - order to uh just look at the properties of
  • fast_forward00:53:39 - the system so i don't think we make any claim that um that
  • fast_forward00:53:44 - maybe it has a great expectation
  • fast_forward00:53:47 - to match very precisely with the
  • fast_forward00:53:50 - psychophysics right because also you don't
  • fast_forward00:53:53 - know how much uh the information either gets
  • fast_forward00:53:56 - selected or not selected downstream right because
  • fast_forward00:53:59 - you have all sorts of selection processes as well right so when you come to
  • fast_forward00:54:03 - perceptual processes um you've got all sorts of selection of the information
  • fast_forward00:54:07 - going on so um yeah and you don't have complete information at the periphery
  • fast_forward00:54:14 - to to be able to say the exact state of the system to ever really be able to test very precisely
  • fast_forward00:54:19 - how exactly accurately this would be characterized.
  • fast_forward00:54:25 - Right. But then the other issue that actually we haven't touched upon at all
  • fast_forward00:54:29 - is that, okay, we have sort of talked about these molecules.
  • fast_forward00:54:34 - The ability to detect them, the reliability of the system.
  • fast_forward00:54:40 - But actually a key issue for olfaction is concentration, right?
  • fast_forward00:54:46 - So how does an olfactory system really deal with varying levels of concentration?
  • fast_forward00:54:53 - I mean, it can go from really minute, that's a homeopathic quantity, to full saturation.
  • fast_forward00:54:59 - So it has to operate in a huge dynamic range. So how does it manage to do that?
  • fast_forward00:55:04 - And what kind of sensitivity do we get?
  • fast_forward00:55:07 - Yeah, I think there's a number of tricks for this.
  • fast_forward00:55:11 - This uh the the system as you say is in a it's about the most remarkable i think in the,
  • fast_forward00:55:18 - in the nervous system and having to deal with such incredible uh dynamic range
  • fast_forward00:55:22 - you've got about probably at least 10 orders of magnitude in concentration which
  • fast_forward00:55:29 - is phenomenal phenomenal uh.
  • fast_forward00:55:32 - Uh degree of variability uh and
  • fast_forward00:55:36 - it's pretty clear that the olfactory system has to use all sorts
  • fast_forward00:55:39 - of tricks to deal with this one of the tricks is going
  • fast_forward00:55:42 - back to the original point if you have a very wide variety of affinity distributions
  • fast_forward00:55:46 - then if you do this um you can still have some receptors binding um or their
  • fast_forward00:55:54 - binding is changing even when the concentration is very high because you might
  • fast_forward00:55:58 - Even though you probably have, for a given odor,
  • fast_forward00:56:02 - you'll have some high affinity receptors, which will get bound very quickly, lower concentrations,
  • fast_forward00:56:08 - and they will quickly become saturated in their response, which doesn't inform
  • fast_forward00:56:12 - you anymore and get any more information out of those because they're just saturated.
  • fast_forward00:56:17 - But then, if you've also got this population with lower affinity receptors,
  • fast_forward00:56:24 - then even though the concentration is increasing a lot, you're still getting
  • fast_forward00:56:28 - some information from those lower affinity receptors.
  • fast_forward00:56:33 - So by using this very wide range of receptor affinities, you can still get information
  • fast_forward00:56:41 - through to the system that there's a change,
  • fast_forward00:56:44 - which you can, that then gives you choices in terms of the coding to still be
  • fast_forward00:56:54 - able to at least receive the information over a large range.
  • fast_forward00:56:57 - Now that's one strategy.
  • fast_forward00:56:59 - Our study also shows by looking at these antagonisms, and we made a,
  • fast_forward00:57:05 - again with Manuel, this was work with Manuel in Madrid,
  • fast_forward00:57:11 - we looked at these receptor antagonisms.
  • fast_forward00:57:15 - And there are some very nice and simple
  • fast_forward00:57:18 - pharmacological equations that we could apply to olfaction which hasn't really
  • fast_forward00:57:23 - been done at all before using these efficacy models and there's something called
  • fast_forward00:57:29 - an operational model in pharmacology which is used effectively to describe drug
  • fast_forward00:57:35 - binding on cells and so on.
  • fast_forward00:57:38 - So in these studies with Manuel Sanchez-Montanés you started to develop more
  • fast_forward00:57:44 - this notion of a ratio code.
  • fast_forward00:57:46 - Yeah, exactly. So what does it exactly mean? So what you find is when you apply
  • fast_forward00:57:50 - these very simple description.
  • fast_forward00:57:55 - Then what you get is that you get effectively two parameters for each ligand,
  • fast_forward00:58:01 - both an efficacy, which tells you how much it contributes to a cellular response,
  • fast_forward00:58:05 - and its affinity of binding.
  • fast_forward00:58:07 - And when you operate in this way or think about olfaction in this way,
  • fast_forward00:58:13 - you get this nice property that something with a low efficacy could effectively
  • fast_forward00:58:18 - block a space and you can get, when the receptors start to become filled,
  • fast_forward00:58:24 - you can get effectively a competitive binding regime.
  • fast_forward00:58:29 - Where you get certain sites are being filled but not contributing,
  • fast_forward00:58:33 - for instance, or other sites that are filled and contributing a lot.
  • fast_forward00:58:38 - And what you find when you look at the equation is that it very simply drops
  • fast_forward00:58:41 - out that instead of the response of the cell being determined by the distribution
  • fast_forward00:58:49 - of the concentrations of the different ligands within a mixture.
  • fast_forward00:58:54 - In the high regime when there's this competition, it actually turns out that
  • fast_forward00:58:58 - it then becomes depending upon the ratio of the concentration.
  • fast_forward00:59:03 - So it's no longer concentration-dependent, and
  • fast_forward00:59:07 - it's in this mode that we call ratio mode operation of the neuron when it's
  • fast_forward00:59:12 - operating in this high occupancy regime that you lose the concentration dependence
  • fast_forward00:59:20 - and you gain a sort of concentration invariance property,
  • fast_forward00:59:24 - which we've shown holds over very large numbers, orders, and magnitudes.
  • fast_forward00:59:30 - So although you might think it needs some very complex neural processing to
  • fast_forward00:59:36 - maybe solve this problem,
  • fast_forward00:59:38 - it's kind of a nice result in the sense that maybe just directly at the periphery
  • fast_forward00:59:42 - in terms of competing for sites may actually solve this problem for you almost for free, right?
  • fast_forward00:59:48 - Well, nothing for free, right? Well, for free in terms of the neural processing, right?
  • fast_forward00:59:52 - Because you don't have to. Okay, that's fair enough, but this works under certain
  • fast_forward00:59:58 - assumptions of binding and receptor distribution and so on, right?
  • fast_forward01:00:01 - So you could also imagine that in these highly saturated regimes,
  • fast_forward01:00:06 - you actually have so few unoccupied receptors left that the probability to actually
  • fast_forward01:00:12 - exploit this invariant or concentration invariant coding, that probability is then very low.
  • fast_forward01:00:19 - Um how do you mean exploit well that
  • fast_forward01:00:22 - so you're saying look the trick is
  • fast_forward01:00:25 - that um at some point you have saturated your
  • fast_forward01:00:28 - receptor sheet but still there are few receptors available that you can get
  • fast_forward01:00:32 - you can use yeah and those can then tell you uh about the presence or absence
  • fast_forward01:00:37 - of a certain odor independent of the concentration exactly right but now i'm
  • fast_forward01:00:41 - saying well that's that's fine But if I have saturated my receptor sheet for, let's say, 99%,
  • fast_forward01:00:47 - then the probability for these unbound molecules to still hit those receptors
  • fast_forward01:00:52 - might be so low that actually I will not be able to exploit my concentration invariant encoding.
  • fast_forward01:00:58 - Yeah, no, that's a good point. But you have to keep in mind that you have a
  • fast_forward01:01:02 - very wide variety of affinities as well.
  • fast_forward01:01:05 - So whilst there may be a subpopulation of receptors that are forced into this high occupancy regime,
  • fast_forward01:01:12 - gene it's quite likely that it's the
  • fast_forward01:01:15 - case it's probably only a small subpopulation out of
  • fast_forward01:01:17 - the total and that there will be large
  • fast_forward01:01:21 - numbers of other receptors with lower affinities that
  • fast_forward01:01:24 - will not therefore be in a they will be in a relatively low.
  • fast_forward01:01:27 - Occupancy and so they are still very much sensitive
  • fast_forward01:01:30 - to chemical changes so if this.
  • fast_forward01:01:34 - Is true and we're still look we've fitted it to the data.
  • fast_forward01:01:37 - That we've received and it matches the data that we've received.
  • fast_forward01:01:41 - On antagonism for instance from tahara and an earlier
  • fast_forward01:01:44 - paper by anderson on insect receptors
  • fast_forward01:01:47 - all the data we fitted works but if
  • fast_forward01:01:50 - it's true it predicts that there's a
  • fast_forward01:01:53 - very strong prediction that there would be likely for any
  • fast_forward01:01:58 - natural odors at
  • fast_forward01:02:01 - naturally relevant or behaviorally relevant
  • fast_forward01:02:04 - conditions there will be a subpopulation in
  • fast_forward01:02:08 - this high occupancy regime that that
  • fast_forward01:02:11 - will give you uh if you like the relational features
  • fast_forward01:02:14 - between a complex odor so for instance in a coffee uh
  • fast_forward01:02:18 - they'll be you you know the quality of the coffee is very much
  • fast_forward01:02:20 - about what is the ratio of of
  • fast_forward01:02:23 - this peak of this compound compared to another compound as
  • fast_forward01:02:26 - to whether that's a pleasant coffee for you and you don't
  • fast_forward01:02:29 - necessarily need to need a certain concentration to
  • fast_forward01:02:33 - know that that sort of that works across many orders of
  • fast_forward01:02:36 - concentration but on the other hand you
  • fast_forward01:02:39 - want probably other parts of the olfactory system that are able
  • fast_forward01:02:42 - to tell you about concentration right and it's already been demonstrated that
  • fast_forward01:02:45 - in fact the olfactory system this is one of
  • fast_forward01:02:48 - its amazing properties right that in fact that you can separate um you can separate
  • fast_forward01:02:54 - both concentration information for certain ligands and also it's sort of what
  • fast_forward01:02:59 - we call configurable properties which is basically the sort of combined percepts
  • fast_forward01:03:04 - so So on the one hand, you've got to sort of gestalt.
  • fast_forward01:03:08 - Sort of uh odor or flavor which is
  • fast_forward01:03:11 - to do with all these sort of ratio properties which are invariant across
  • fast_forward01:03:14 - concentrations and on the other hand you've got this sort
  • fast_forward01:03:17 - of opposite of gestalt which is able to sort of tease them
  • fast_forward01:03:20 - apart and look at the separate component concentrations and
  • fast_forward01:03:24 - with this mechanism you you then have a population when you combine this mechanism
  • fast_forward01:03:29 - with a wide range of affinities which we know is in the olfactory system you
  • fast_forward01:03:34 - have this very nice feature that you You would have these sort of subpopulations
  • fast_forward01:03:38 - doing different things to either support either gestalt processing or non-gestalt processing.
  • fast_forward01:03:43 - But now, with that separation, would you imagine that there is a possibility
  • fast_forward01:03:49 - for the brain to control this in a more top-down fashion, so you get something
  • fast_forward01:03:53 - like an olfactory attention?
  • fast_forward01:03:56 - This is a very good point. I think probably not at the binding level, right? Right.
  • fast_forward01:04:01 - But we do know that certainly the feed forward gain in these different pathways
  • fast_forward01:04:06 - from the different receptors, we know very well that this is...
  • fast_forward01:04:08 - That's a level of receptor neurons though.
  • fast_forward01:04:10 - Not necessarily the receptor. Well, there is some evidence that there's even
  • fast_forward01:04:13 - feedback actually to the receptors.
  • fast_forward01:04:15 - Okay. It's still not really well studied, but we know very, very surely that
  • fast_forward01:04:22 - the readout of the glomeruli themselves,
  • fast_forward01:04:23 - which is effectively at an early stage of that processing, is reflecting the
  • fast_forward01:04:28 - excitatory drive of the common set of olfactory receptors.
  • fast_forward01:04:33 - We know that those are under very complex inhibitory control and gain control,
  • fast_forward01:04:39 - top-down imposed from the piriform cortex and through centrifugal feedback coming.
  • fast_forward01:04:44 - I mean, coming back to the system, we know that that has a very strong influence
  • fast_forward01:04:49 - on the type of codes that go forward to the higher level.
  • fast_forward01:04:52 - So it's very likely that there's an additional sorts of then selection mechanisms.
  • fast_forward01:04:57 - If you could imagine there are subsets of these different receptors,
  • fast_forward01:05:04 - you could imagine that then downstream you could select between those that are
  • fast_forward01:05:08 - either in a ratio mode or in a concentration mode, for instance.
  • fast_forward01:05:14 - So that might be another way to deal with that problem that's not only feed
  • fast_forward01:05:18 - forward and dependent on the receptors themselves. So this could be synergistic mechanisms.
  • fast_forward01:05:23 - All right. So now we have a bit of an idea how we get our first responses in
  • fast_forward01:05:29 - and then in the modeling work that you have been doing.
  • fast_forward01:05:32 - You try to also get a clear idea about the actual encoding of now these odors
  • fast_forward01:05:36 - at the level of olfactory bulb.
  • fast_forward01:05:38 - So what can you tell us about the encoding that then happens of these odors?
  • fast_forward01:05:41 - Sure, yeah. So the main feature in the olfactory bulb, which is really the first
  • fast_forward01:05:45 - site of neuronal processing,
  • fast_forward01:05:47 - of all this stuff coming in from 10 million receptors in the olfactory mucosa,
  • fast_forward01:05:55 - as it comes through the skull,
  • fast_forward01:05:57 - the cribriform plate, to make contact with these glomeruli.
  • fast_forward01:06:00 - And the first thing that happens is they converge into these glomeruli structures,
  • fast_forward01:06:05 - and they are innervated by these mitral tufted cells.
  • fast_forward01:06:10 - And basically the story is that one single mitral tufted cell is innervating a single glomerulus.
  • fast_forward01:06:16 - So they're acting as a readout, if you like, of a single chemotopic feature,
  • fast_forward01:06:22 - because they're only innervating those common GPCR or receptor types. Okay.
  • fast_forward01:06:29 - And then all of those mitral receptor population,
  • fast_forward01:06:33 - which acting as the main readout of the bulb to the higher centers like piriform
  • fast_forward01:06:37 - cortex and so on, are under very quite complex lateral control.
  • fast_forward01:06:44 - So for instance, a mitral cell is under control from what are called granule
  • fast_forward01:06:53 - cells, which make lateral connections across the bulb.
  • fast_forward01:06:56 - And it's even been demonstrated that granule cells may even connect across the
  • fast_forward01:07:02 - whole length of the bulb, with quite specific connections, in fact.
  • fast_forward01:07:05 - So there's this network of
  • fast_forward01:07:08 - quite specific connections that are going laterally they're
  • fast_forward01:07:12 - forcing an inhibitory drive on the mitral cells
  • fast_forward01:07:15 - and there's at least two effects of this well three really one effect is that
  • fast_forward01:07:22 - it's been demonstrated quite clearly that this lateral inhibition has a sort
  • fast_forward01:07:29 - of on-center-off-surround type property So because it's also,
  • fast_forward01:07:35 - it's a complex type of inhibition and it's a unique synapse called a dendro
  • fast_forward01:07:42 - dendritic synapse, which is actually pretty unique in the whole brain in these granule cells.
  • fast_forward01:07:47 - You don't really see it anywhere else. And the effect of this dendrodendritic
  • fast_forward01:07:52 - synapse is that a mitral cell will activate one of these lateral granule cells,
  • fast_forward01:07:57 - which will then in turn inhibit a distant mitral cell target,
  • fast_forward01:08:02 - but it will also excite itself back because it's a two-way process.
  • fast_forward01:08:07 - Didactic synapse and the
  • fast_forward01:08:10 - effect of all of this anyway is that you get a sort of on center
  • fast_forward01:08:13 - of surround type property because it's sort of inhibiting with
  • fast_forward01:08:16 - a sustain with some nice time
  • fast_forward01:08:19 - dynamics exactly and it's been demonstrated that
  • fast_forward01:08:23 - this very nicely decorrelates and sharpens the representation
  • fast_forward01:08:26 - of these odors at this stage in
  • fast_forward01:08:29 - the olfactory bulb so that's one clear thing that happens another clear
  • fast_forward01:08:33 - thing that happens is because of this um very nice
  • fast_forward01:08:36 - balance of excitation drive coming in from
  • fast_forward01:08:39 - the receptors via the glomeruli with this
  • fast_forward01:08:42 - lateral inhibition that you're getting some very complex
  • fast_forward01:08:45 - dynamics right so in fact the olfactory bulb was the i think it's fair to say
  • fast_forward01:08:52 - actually the very first neuronal recordings in in the brain by lord adrian at
  • fast_forward01:08:57 - cambridge measured also the first oscillations in the olfactory bulb.
  • fast_forward01:09:02 - And oscillations are the very key feature and they come about due to this balance
  • fast_forward01:09:09 - of lateral inhibition and excitatory drive.
  • fast_forward01:09:15 - So there's these very complex sort of dynamics going on together with these
  • fast_forward01:09:19 - sort of sharpening things.
  • fast_forward01:09:21 - And the third feature is that we, as we mentioned before, that these
  • fast_forward01:09:24 - granule cells also act as targets for the centrifugal feedback coming back from
  • fast_forward01:09:29 - the higher centers to selectively change this lateral inhibitory mechanism to
  • fast_forward01:09:35 - effectively selectively change
  • fast_forward01:09:38 - the gain in different channels so that you may listen, for instance,
  • fast_forward01:09:41 - more to some types of chemical information and suppress others, right?
  • fast_forward01:09:45 - Right, exactly. So this has already been shown to be very important.
  • fast_forward01:09:47 - But now, do you see these mitral cells and linked to this complex network of
  • fast_forward01:09:52 - granule cells and they are sampling the glomeruli,
  • fast_forward01:09:57 - are they just adding up activity and emitting spikes or is the encoding of these odors more involved?
  • fast_forward01:10:04 - Like, for instance, does it depend on population responses? Does it depend on
  • fast_forward01:10:08 - the temporal structuring of these responses?
  • fast_forward01:10:11 - Yeah, I think evidence shows, for instance, Detlef Schildert-Göttingen has shown
  • fast_forward01:10:16 - that the latency of the mitral cell responses is very crucial to different compounds,
  • fast_forward01:10:22 - which you might expect right because if you've got
  • fast_forward01:10:25 - this dendro dendritic thing you can you have
  • fast_forward01:10:27 - these periods of inhibition so what can
  • fast_forward01:10:30 - happen is that some mitral cells can selectively shut
  • fast_forward01:10:33 - down other mitral cells in these channels for certain periods of time and so
  • fast_forward01:10:39 - what you seem to find is that if you look at the population of um olfactory
  • fast_forward01:10:44 - bulb mitral cell responses that you get these different latencies in the different
  • fast_forward01:10:48 - channels and it's been demonstrated that these latencies are very important
  • fast_forward01:10:51 - for coding the different odors, right?
  • fast_forward01:10:53 - Do these latencies also express the binding kinetics?
  • fast_forward01:10:58 - You could imagine that a receptor with high affinity to a ligand will trigger
  • fast_forward01:11:03 - a short latency response in its target.
  • fast_forward01:11:06 - Yeah, I think that's quite possible, but I think it's also reflecting the timing
  • fast_forward01:11:13 - of these lateral inhibitions and the timing of the dendritic synapses,
  • fast_forward01:11:19 - so I think it's probably both of those together.
  • fast_forward01:11:20 - The other very interesting feature, which was shown very nicely by Rainer Freydig.
  • fast_forward01:11:27 - Was to show that the spike timing of individual mitral cells in this oscillation
  • fast_forward01:11:36 - in the olfactory bulb may well be very important for coding different aspects of a complex odor.
  • fast_forward01:11:43 - So, for instance, if you're firing early on in the cycle with another set of
  • fast_forward01:11:50 - receptors, this may have one meaning.
  • fast_forward01:11:52 - And if you're firing with another subnetwork within this olfactory bulb space
  • fast_forward01:11:58 - later on in the oscillatory cycle, this may mean another thing.
  • fast_forward01:12:02 - And so he was crucial in introducing this idea of multiplexed odor codes for mixtures.
  • fast_forward01:12:09 - So maybe you have a mixture where you have different synchronous firing between
  • fast_forward01:12:15 - different subpopulations at different points in time.
  • fast_forward01:12:17 - And that these may well be telling higher centers all sorts of information about
  • fast_forward01:12:23 - grouping or binding of different chemical properties to tell you about,
  • fast_forward01:12:29 - you know, much more complex chemical stimuli that we know after all is like
  • fast_forward01:12:34 - coffee with 400 compounds or whatever.
  • fast_forward01:12:37 - Okay, so now we have a bit of an idea of the biology of olfaction.
  • fast_forward01:12:41 - In sort of a parallel existence, you also worry a lot about building artificial
  • fast_forward01:12:46 - olfaction, artificial noses, olfactory machines, olfactory robots,
  • fast_forward01:12:51 - and so on. So what are the challenges there exactly?
  • fast_forward01:12:54 - Yeah, so it's a kind of parallel life to kind of build technology.
  • fast_forward01:13:00 - I mean, similar to you, actually, that I think we both follow this idea that
  • fast_forward01:13:06 - if we really understand it, we can build it, right?
  • fast_forward01:13:09 - So let's go and see the proof by, can we take these principles,
  • fast_forward01:13:13 - put them into some type of operational system, test what they do.
  • fast_forward01:13:18 - And it introduces a whole bunch of challenges that a lot of them actually have
  • fast_forward01:13:24 - very little to do with the biology.
  • fast_forward01:13:26 - Because, for instance, just getting a good set of sensory signals to start with
  • fast_forward01:13:31 - and reliable sensory responses is already a big challenge in chemical sensing.
  • fast_forward01:13:37 - The idea of so-called building chemical analysis systems along the lines of
  • fast_forward01:13:45 - following the biological system is not a new idea.
  • fast_forward01:13:49 - There was a paper in Nature in 1982 by Krishna Persaud and George Dodd,
  • fast_forward01:13:57 - who introduced this idea of putting together groups of different tuning chemical
  • fast_forward01:14:04 - sensors in arrays and to look at population codes of these responses,
  • fast_forward01:14:08 - which is clearly similar to how this is solved in the biological system.
  • fast_forward01:14:15 - System um and you know
  • fast_forward01:14:18 - so then what you need to do is you need to you know you face a challenge of
  • fast_forward01:14:21 - finding different chemical technologies that give you nicely orthogonal and
  • fast_forward01:14:26 - decorrelated sort of sensor responses um and that are sort of robust and give
  • fast_forward01:14:32 - you reliable signals um that have sort of have reasonable noise properties,
  • fast_forward01:14:38 - can be easily deposited.
  • fast_forward01:14:42 - Can maybe have a wide range of different molecular interactions and have sort
  • fast_forward01:14:49 - of broad tunings to different things.
  • fast_forward01:14:52 - Maybe reversible responses.
  • fast_forward01:14:54 - So you don't really necessarily want sensors that don't ever recover from an interaction, right?
  • fast_forward01:14:59 - So there's a whole bunch of issues with chemical sensors.
  • fast_forward01:15:03 - And I And I think whilst things are improving, there's actually a massive range
  • fast_forward01:15:11 - of chemical sensor technologies you can use.
  • fast_forward01:15:14 - There are clearly many issues. For instance, one of the main issues,
  • fast_forward01:15:17 - I think, in chemical sensors is this concept of sensor drift.
  • fast_forward01:15:23 - So we know that pretty much any chemical sensor, its properties will be quite
  • fast_forward01:15:28 - non-stationary over time.
  • fast_forward01:15:30 - So it may have one set of tunings at one particular point in time.
  • fast_forward01:15:33 - You may come back there in a month's time and have a completely different set
  • fast_forward01:15:37 - of tunings, baseline parameters and so on.
  • fast_forward01:15:40 - These can change in very complex and difficult to predict ways. Uh-huh.
  • fast_forward01:15:47 - On one level, okay, so this can be quite frustrating perhaps when you try and build these systems.
  • fast_forward01:15:51 - But on another level, we also have to appreciate that every individual person
  • fast_forward01:15:57 - has pretty much a unique, we haven't discussed this, but we all have a unique
  • fast_forward01:16:01 - genetic fingerprint of the subset of thousand olfactory receptors that we express individually.
  • fast_forward01:16:07 - So coffee actually smells completely different to you than it does to me.
  • fast_forward01:16:12 - Well, but maybe not because earlier
  • fast_forward01:16:13 - you said that there's actually a low dimensional manifold. default.
  • fast_forward01:16:17 - Maybe we map it onto something that we can describe in similar terms,
  • fast_forward01:16:21 - but what you can guarantee is that the receptor information that comes into
  • fast_forward01:16:26 - your system is completely different to that that comes into mine, right?
  • fast_forward01:16:29 - But you're right that this has to be somehow mapped.
  • fast_forward01:16:33 - That's kind of my point really, is that somehow the olfactory system has to
  • fast_forward01:16:37 - take care of this and map it all onto a stable representation of something that
  • fast_forward01:16:42 - means coffee. And one that's It's unitary, of which you can say that's coffee,
  • fast_forward01:16:45 - not that you say, oh, it's 400 compounds.
  • fast_forward01:16:47 - Yeah, exactly. And the other point is that actually in a month's time,
  • fast_forward01:16:52 - because these olfactory receptors are basically the only receptors that are
  • fast_forward01:16:57 - in direct contact with the outside world,
  • fast_forward01:16:59 - they're really at the tough end of things, unlike most neurons,
  • fast_forward01:17:04 - which are very nicely protected in the neural systems, that they basically suffer damage.
  • fast_forward01:17:10 - And so in a month's time, you and I will both have a completely different set
  • fast_forward01:17:16 - of olfactory receptors than those that we have now.
  • fast_forward01:17:19 - Now they'll be similarly genetically determined, we'll have a sort of similar
  • fast_forward01:17:23 - subset of the total, because that's fairly determined, even that may change,
  • fast_forward01:17:28 - but somehow the olfactory system has to deal with all of this,
  • fast_forward01:17:31 - and in a sense that's kind of like drift, right?
  • fast_forward01:17:33 - Because we already know that, for instance, as the olfactory receptors are developing,
  • fast_forward01:17:40 - that their tuning is changing.
  • fast_forward01:17:42 - They actually become more broadly tuned over time. They start more specific.
  • fast_forward01:17:45 - They become more broadly tuned. And so there's a whole drift process actually
  • fast_forward01:17:48 - going on in the receptors as well.
  • fast_forward01:17:50 - And, okay, that drift process is probably very different from that that we see in the receptors.
  • fast_forward01:17:54 - But it's still a challenge that the neuromorphic system might nicely be able
  • fast_forward01:18:00 - to solve. Because if you look at a classical engineering solutions to this,
  • fast_forward01:18:04 - then it hasn't quite totally solved all of these issues.
  • fast_forward01:18:09 - Tell me, what are the different approaches people have taken to solve this problem, technically?
  • fast_forward01:18:14 - Of drift? No, and chemical sensing. Let's start with that. Okay.
  • fast_forward01:18:18 - Okay, well, yeah, I mean, that's a very broad question, right?
  • fast_forward01:18:22 - You can go all the way from developing biosensors through to developing highly
  • fast_forward01:18:29 - specific sensors for particular ligands that you're interested in with a problem,
  • fast_forward01:18:34 - all the way through to developing sort of electronic noses, using,
  • fast_forward01:18:41 - you know, arrays of broadly tuned compounds, compounds
  • fast_forward01:18:44 - through to using sort
  • fast_forward01:18:47 - of mass spectrometry ideas where you're trying to
  • fast_forward01:18:50 - basically measure specific molecular features directly like we were talking
  • fast_forward01:18:55 - about originally there might be mass features or through to using gas chromatography
  • fast_forward01:19:02 - effects where you're trying to select between different molecules due to how
  • fast_forward01:19:05 - much they sorb into different materials or not.
  • fast_forward01:19:09 - There's a whole bunch of different strategies uh so
  • fast_forward01:19:12 - what's the sensitivity of these systems compared to the biological system
  • fast_forward01:19:15 - this is a great question so um well actually
  • fast_forward01:19:19 - we did a study right a long time ago which we're together remember that um we
  • fast_forward01:19:23 - we we we had this question in mind uh when we first came to this nice artificial
  • fast_forward01:19:29 - moth project that we ran that we were wondering well okay so we know that the
  • fast_forward01:19:33 - moth is very sensitive overall it's like the world specialist in olfactory detection,
  • fast_forward01:19:39 - might we expect that, in fact, its receptors are, you know, orders of magnitude
  • fast_forward01:19:46 - more sensitive than, say, for instance, a metal oxide sensor?
  • fast_forward01:19:49 - Like a commercial one you just buy. Yeah, which is the most common one that
  • fast_forward01:19:54 - tends to be used in these kinds of arrays.
  • fast_forward01:19:56 - And indeed, what we found when we took those measurements were that,
  • fast_forward01:19:59 - in fact, no, you know, that these sensitivities were kind of more comparable
  • fast_forward01:20:03 - than you might expect. So that was kind of a surprise.
  • fast_forward01:20:07 - But you also, of course, in doing any comparison like that, you have to take
  • fast_forward01:20:11 - into account that we're not necessarily giving that receptor its ideal ligand.
  • fast_forward01:20:17 - We may be particularly one extreme end of the affinity for that particular ligand or efficacy. Yeah.
  • fast_forward01:20:25 - Parameter you know it may have other ones that it's much more sensitive to
  • fast_forward01:20:28 - but it's very difficult to tell unless you measure everything which
  • fast_forward01:20:31 - is impossible no but wait but but there's an important point here
  • fast_forward01:20:34 - right that sort of intuitively you would say like well you know if i have to
  • fast_forward01:20:37 - increase the sensitivity i should be just optimizing at the sensor front end
  • fast_forward01:20:42 - of this whole system if i just have high affinities there i'll be just fine
  • fast_forward01:20:46 - in my detection but maybe but maybe that's not the right philosophy no you really
  • fast_forward01:20:50 - can't do that because then everything binds to everything.
  • fast_forward01:20:53 - Right and then you don't have any information anymore so
  • fast_forward01:20:56 - already in the story that we saw with the antagonism as
  • fast_forward01:20:59 - we talked about before is that it really to get this
  • fast_forward01:21:02 - ability to do this over a range and to
  • fast_forward01:21:05 - to also be able to get the combination in a
  • fast_forward01:21:08 - single olfactory system of sensitivity together
  • fast_forward01:21:12 - with high specificity to get those
  • fast_forward01:21:15 - two things together and to get them over a
  • fast_forward01:21:17 - broad dynamic range you cannot just be sensitive to everything
  • fast_forward01:21:20 - right you have to you have to have this
  • fast_forward01:21:23 - range of binding so that you can basically you
  • fast_forward01:21:26 - have to you have to have it so that you can have a kind of attentional mechanism
  • fast_forward01:21:29 - to right to be able to focus in particular but it's not respect that the point
  • fast_forward01:21:35 - is maybe if you want to get a highly sensitive olfactory system the real tricks
  • fast_forward01:21:40 - sit more at the processing end of things than at the sense of content yeah i
  • fast_forward01:21:45 - would I would agree with that.
  • fast_forward01:21:46 - I mean, it's certainly clear that there's so many tricks that are played,
  • fast_forward01:21:50 - you know, by the biology in terms of getting a sensitive, specific.
  • fast_forward01:21:56 - Robust, you know, signal representation out of this that the more I look at
  • fast_forward01:22:04 - every level of the system, it's completely optimized to achieving just that.
  • fast_forward01:22:08 - But now the amazing thing is that in one of your projects with your collaborators.
  • fast_forward01:22:12 - You built an artificial nose where you specifically looked at the impact of
  • fast_forward01:22:18 - the mucus layer in processing of olfaction.
  • fast_forward01:22:22 - So why did you add that feature to an artificial chemical sensor?
  • fast_forward01:22:27 - Yeah, it's a good question. Well, in existing electronic noses up to now,
  • fast_forward01:22:32 - really there are two forms of information that have been exploited.
  • fast_forward01:22:37 - Uh, the first, uh, mechanism is this, uh, population code.
  • fast_forward01:22:41 - So for any particular complex or simple molecule, you've got a particular fingerprint
  • fast_forward01:22:47 - of, um, receptor responses. Mm-hmm.
  • fast_forward01:22:52 - This is clearly, this is what you would call a spatial code of receptor population
  • fast_forward01:23:00 - tuning that give you stimulus-dependent information, enable you to tell what's out there in the world.
  • fast_forward01:23:07 - You've got a second mechanism, clearly, which is that each of those receptors
  • fast_forward01:23:11 - or sensors is giving you, in itself, specific temporal information.
  • fast_forward01:23:16 - So this gives you sort of, en masse, gives you a sort of temporal code,
  • fast_forward01:23:21 - temporal population code as you might call it.
  • fast_forward01:23:24 - But then there's this very much less studied aspect of olfaction,
  • fast_forward01:23:29 - which is that not only are the sensors themselves maybe giving out temporal
  • fast_forward01:23:35 - information, depending upon what they're binding with.
  • fast_forward01:23:38 - That in fact the stimulus delivery itself may have some nice temporal properties.
  • fast_forward01:23:44 - And so we went about building what
  • fast_forward01:23:47 - we called an artificial mucosa and this was with collaborators at
  • fast_forward01:23:51 - university of warwick including julian gardner who build microsystems so i don't
  • fast_forward01:23:57 - build any sensors myself i build these sort of strategies for processing architectural
  • fast_forward01:24:01 - ideas for these types of systems and so we specifically you know we it was a
  • fast_forward01:24:07 - very nice example of where you see a few neuroscience papers,
  • fast_forward01:24:11 - and this basically inspires you to then go and build a technology that you think
  • fast_forward01:24:16 - can do better than what we have at the moment.
  • fast_forward01:24:18 - And what we ended up building was effectively a microchannel or an artificial
  • fast_forward01:24:24 - mucosa where you have a sort of sniff.
  • fast_forward01:24:28 - So rather than in many electronic noses, you're sampling for long periods of
  • fast_forward01:24:32 - time with very controlled flow rates.
  • fast_forward01:24:35 - This is a very different strategy, where you just effectively have a pulse of
  • fast_forward01:24:39 - odour that flows through a microchannel.
  • fast_forward01:24:42 - And then what you have inside that microchannel, similar to as we talked about
  • fast_forward01:24:46 - in the nose, is you have a thin mucosal layer, maybe a few hundred microns thick.
  • fast_forward01:24:52 - And this can be of different materials, but you can take ideas,
  • fast_forward01:24:56 - for instance, from gas chromatography that use what are called stationary phases.
  • fast_forward01:25:00 - And these stationary phases are specific materials you can go and buy off the shelf.
  • fast_forward01:25:04 - And they have very beautiful selective properties partitioning
  • fast_forward01:25:08 - properties of certain groups of compounds such as hydrophobic or hydrophilic
  • fast_forward01:25:13 - compounds would like to be in those stationary phases or like that they would
  • fast_forward01:25:17 - like to stay away from them and so what you find is that as you have an odor
  • fast_forward01:25:21 - pulse going through a column with these kind of materials you get very beautiful temporal
  • fast_forward01:25:27 - profiles in the stationary phases at different points in time that depend in
  • fast_forward01:25:33 - very complex ways on these different sorption properties and other properties
  • fast_forward01:25:39 - of the molecules that impose an additional time dynamics on the stimulus.
  • fast_forward01:25:45 - And particularly when you've got a complex mixture, like a coffee,
  • fast_forward01:25:49 - then imagine you've got, what you've really got is
  • fast_forward01:25:51 - you've got 400 of these pulses going through simultaneously and you
  • fast_forward01:25:55 - get a beautiful sort of spectrum you can think of it like a kind of
  • fast_forward01:25:58 - prism system where you take in a complex move and it
  • fast_forward01:26:01 - splits it in some way right exactly in a spatial temporal
  • fast_forward01:26:04 - code and it imposes this upon the receptor
  • fast_forward01:26:07 - population and what you find when you look at the receptor sensor responses
  • fast_forward01:26:12 - when you distribute them across this type of physical arrangement is that you
  • fast_forward01:26:16 - get absolutely beautiful and stunning very complex extemporal information that's
  • fast_forward01:26:22 - exquisitely stimulus dependent, right?
  • fast_forward01:26:25 - Because it depends upon exactly where that molecule was at a particular point
  • fast_forward01:26:29 - in time. With how much did this improve?
  • fast_forward01:26:31 - So basically you translate this whole idea of zoning of the receptor sheet and
  • fast_forward01:26:37 - a mucus layer that again controls the binding dynamics of your ligands.
  • fast_forward01:26:42 - So you translate that to a technology also basically to test its impact on processing,
  • fast_forward01:26:51 - but what was really the improvement in classification that you now observed?
  • fast_forward01:26:55 - Yeah, I mean, what we found was very interesting. So if you had,
  • fast_forward01:26:59 - it's kind of an interesting result.
  • fast_forward01:27:01 - We found in all cases, this obviously improved discrimination.
  • fast_forward01:27:06 - Purely because you've got three mechanisms in total, and the correct comparison
  • fast_forward01:27:12 - to do is to compare it to just when you have the first two mechanisms without
  • fast_forward01:27:16 - this nice, funky, spatiotemporal delivery concept.
  • fast_forward01:27:20 - So you do the direct comparison of just saying, well, when you deliver the molecules
  • fast_forward01:27:24 - directly onto the senses without any spatiotemporal sorting of any type,
  • fast_forward01:27:30 - how does your information compare?
  • fast_forward01:27:32 - And so again with Manuel, who very cleverly extended the Fisher information
  • fast_forward01:27:36 - formalism to include time so that we can actually now see not only when different
  • fast_forward01:27:43 - stimulus dependent features and the noise properties are telling you a particular point in time,
  • fast_forward01:27:48 - you can actually accumulate that information over time to tell you within particular
  • fast_forward01:27:52 - time windows how much information you've collected about the stimulus and how
  • fast_forward01:27:56 - accurate your reconstruction is of it.
  • fast_forward01:27:59 - You can actually use this formulism then to compare in a very precise way,
  • fast_forward01:28:04 - in a very quantitative way,
  • fast_forward01:28:06 - how you would have fared if you tried to discriminate just from the first two
  • fast_forward01:28:11 - mechanisms compared to when you also use this additional sorting.
  • fast_forward01:28:15 - And what you find, in fact, there's a theorem in our paper that shows that there's
  • fast_forward01:28:19 - no situation ever, there's no possibility that you can ever exceed the three
  • fast_forward01:28:25 - mechanisms with the first two.
  • fast_forward01:28:27 - So there's no situation in which you can ever do better than the three,
  • fast_forward01:28:30 - which is kind of common sense, right?
  • fast_forward01:28:33 - Because if you're adding more stimulus-dependent information.
  • fast_forward01:28:36 - So you can never do worse. So that's already good news.
  • fast_forward01:28:41 - It's more like a lower bound. It's like a lower bound. So what's your upper bound?
  • fast_forward01:28:45 - So the other very interesting point is how, so then the other question is how
  • fast_forward01:28:49 - much better you do totally depends on the complexity of the task.
  • fast_forward01:28:53 - So if you set your olfactory system, then a very simple task,
  • fast_forward01:28:58 - which is maybe, I don't know, imagine a very, very simple task of you just have
  • fast_forward01:29:03 - a single component and you just at one point in time, you push that through
  • fast_forward01:29:08 - your microchannel, you get a set of receptor responses and you discriminate that as one odor.
  • fast_forward01:29:13 - Then you separately do that for a different odor. This is a very simple discrimination
  • fast_forward01:29:17 - task with two classes. So you've got a 50% chance.
  • fast_forward01:29:24 - And even the array by itself, depending upon the chemicals you use,
  • fast_forward01:29:28 - will trivially separate this, then you'll find that by adding that space to
  • fast_forward01:29:33 - the temporal thing, you're not really going to improve it, right?
  • fast_forward01:29:35 - Because you already almost certainly probably had 100% success anyway, right?
  • fast_forward01:29:39 - So what we found in our study was that when you push it more and more,
  • fast_forward01:29:44 - then the effect of this third mechanism makes more and more of a difference.
  • fast_forward01:29:50 - So for instance, if you give it a very complex task, which you could imagine
  • fast_forward01:29:54 - one of the hardest tasks might be put through the system of,
  • fast_forward01:30:01 - a coffee mixture with 500 compounds simultaneously.
  • fast_forward01:30:06 - And the task of the chemical receptor array and the subsequent readout is to
  • fast_forward01:30:11 - detect the presence or not the presence of one of those compounds in the 500, right?
  • fast_forward01:30:16 - Right. When there's a massive overlapping spectra of all of these.
  • fast_forward01:30:20 - This is a phenomenally difficult problem because you've got all of these receptors,
  • fast_forward01:30:25 - responses are convolved over 500 components in a relatively short period of
  • fast_forward01:30:30 - time of responses over just a few seconds and.
  • fast_forward01:30:33 - Somehow over this relatively small sensor population you may I think in our
  • fast_forward01:30:38 - micro I think we only had 30 sensors right now.
  • fast_forward01:30:41 - We only have 30 channels compared to 10 million in there in the in the biological
  • fast_forward01:30:46 - system Out of all of this you've got to somehow Detect one of these components.
  • fast_forward01:30:53 - This is a very challenging problem and what
  • fast_forward01:30:57 - you find when you apply this
  • fast_forward01:31:00 - separation at the front end with this third mechanism is that
  • fast_forward01:31:03 - you find that this third mechanism becomes
  • fast_forward01:31:06 - more and more important so we found
  • fast_forward01:31:09 - that at least one or two orders of magnitude improvement in
  • fast_forward01:31:12 - the classification and the error reduction in
  • fast_forward01:31:16 - the error was two orders of magnitude and we
  • fast_forward01:31:19 - i think we only tested at a couple of
  • fast_forward01:31:21 - either an easy and a complex task and i
  • fast_forward01:31:24 - think there's a lot of interesting ideas
  • fast_forward01:31:27 - to push that to harder and harder tasks for instance attentional tasks
  • fast_forward01:31:30 - where maybe the target odor is changing over time things like this okay but
  • fast_forward01:31:35 - now compared compared to let's say a standard sensor like like a like a cmos
  • fast_forward01:31:41 - based sensor so what what's your performance if i if i go classify of our coffee
  • fast_forward01:31:46 - with a standard off-the-shelf sensor,
  • fast_forward01:31:48 - and now this one with its artificial mucus layer.
  • fast_forward01:31:51 - Yeah, I mean, you know, we've got in our first paper just to show that when
  • fast_forward01:31:56 - you do it with and without, you get an improved performance.
  • fast_forward01:32:00 - And we have this other theoretical study to show that in certain situations
  • fast_forward01:32:05 - you do about two orders of magnitude better in terms of the error.
  • fast_forward01:32:12 - And that's the sort of level of what we've quantified so far.
  • fast_forward01:32:15 - So in terms of practical… So that still has to happen. But how do you fabricate this mucus?
  • fast_forward01:32:20 - Uh yeah so in fact it's quite a challenge
  • fast_forward01:32:22 - because a lot of these stationary phase materials are quite poisonous so
  • fast_forward01:32:26 - they had a big challenge in the lab of how to get these stably inside a microchannel
  • fast_forward01:32:31 - array um the warwick people made a very beautiful there effectively that used
  • fast_forward01:32:36 - a more advanced micro lithography method of a 3d printer effectively to grow
  • fast_forward01:32:41 - grow a microchannel as you can
  • fast_forward01:32:44 - imagine we have a great interest in say also growing
  • fast_forward01:32:47 - a rat nose structure so maybe of course maybe there are
  • fast_forward01:32:50 - well it's almost certain actually that there are uh very
  • fast_forward01:32:54 - beautiful aspects of the elaborate structure of
  • fast_forward01:32:57 - the nasal turbinates of a rat nose which
  • fast_forward01:33:00 - give even more tricks in there for for instance how it controls turbulence and
  • fast_forward01:33:06 - laminar flow and all sorts of other stuff that i haven't even talked about right
  • fast_forward01:33:10 - that can be stuff in the future perhaps but right now Now you just basically
  • fast_forward01:33:15 - grow in a 3D printer a very long microchannel.
  • fast_forward01:33:19 - And you introduce this material under high pressure to get it to uniformly pass
  • fast_forward01:33:25 - along this microchannel to deposit as a thin layer.
  • fast_forward01:33:28 - Have you considered also to use mucus from biological systems?
  • fast_forward01:33:31 - We haven't yet, but of course this is a fascinating idea, right?
  • fast_forward01:33:35 - Because they already have things like OBPs in there. Exactly right.
  • fast_forward01:33:38 - And so on, and other things, and odor-degrading enzymes. times but
  • fast_forward01:33:41 - of course they're also not necessarily particularly stable right so these
  • fast_forward01:33:45 - are other things to take into account but yeah it's
  • fast_forward01:33:48 - a fascinating area of all sorts of different materials that you can potentially
  • fast_forward01:33:51 - put in there probably anything you can use pretty much any chemical material
  • fast_forward01:33:55 - with a liquid aspect will have some sort of selective properties for groups
  • fast_forward01:33:58 - of compounds and they will change the sort of array receptor properties in different
  • fast_forward01:34:03 - ways right so there's all sorts of possibilities to have sort of parallel versions
  • fast_forward01:34:07 - of these noses right It doesn't have to be one nose.
  • fast_forward01:34:10 - You can have arrays of these noses with, I don't know, peanut butter coatings,
  • fast_forward01:34:15 - snot coatings, various different stationary phases from GC columns,
  • fast_forward01:34:21 - God knows what, right? And they're all giving you different selective answers.
  • fast_forward01:34:24 - You can think of it as different windows on the chemical world. Right, exactly.
  • fast_forward01:34:28 - Sort of seeing a unique little portion of a chemical world.
  • fast_forward01:34:32 - Right. But now, do you believe, I mean, you foresee also the robots of the future sure.
  • fast_forward01:34:39 - Largely will be equipped with with artificial noses yeah
  • fast_forward01:34:43 - it's a good question there's been a lot
  • fast_forward01:34:46 - of efforts actually i mean not just on robots so for at
  • fast_forward01:34:49 - least 20 years people have talked about things like will there be smell sensors
  • fast_forward01:34:52 - on your mobile phone and uh in fact
  • fast_forward01:34:55 - at some point there was a lot of talk about that particularly in japan people were
  • fast_forward01:34:59 - very keen to know you know how is my uh breath smelling
  • fast_forward01:35:03 - today you know maybe there's a need for having
  • fast_forward01:35:06 - these kind of very cheap sensors in a in a in a mobile phone um there's as you
  • fast_forward01:35:14 - know very well there's there's uh all sorts of domains for this in terms of
  • fast_forward01:35:19 - robot robots for environmental sensing and so on and um,
  • fast_forward01:35:24 - uh there are all sorts of aspects in terms of human
  • fast_forward01:35:27 - emotion that are obviously important in in
  • fast_forward01:35:30 - odors so if you're going to ever going to
  • fast_forward01:35:33 - ever make a robot that understands humans probably
  • fast_forward01:35:36 - needs to understand odor as well because there's some sort of estimate of something
  • fast_forward01:35:40 - like 95 or 99 percent of our um uh olfactory uh consciousness is subconscious
  • fast_forward01:35:49 - so it has all of this sort of background biasing on our state of mind at a particular point in time,
  • fast_forward01:35:56 - probably a robot would need to know about this right if it needed to understand humans.
  • fast_forward01:36:02 - Um and there's all sorts of applications in flavor
  • fast_forward01:36:05 - industries and so on so there are
  • fast_forward01:36:08 - clearly a lot of applications but i i think
  • fast_forward01:36:11 - there it's always going to be a niche okay so to to finish up i mean so you're
  • fast_forward01:36:16 - you're you have the one foot in the biology of affection the other one in in
  • fast_forward01:36:21 - technology of affection and also you use the technology to understand the biology
  • fast_forward01:36:25 - and the biology to advance the technology so it's a very unique position you're in so in
  • fast_forward01:36:30 - the study of the brain and in our in our attempts to synthesize
  • fast_forward01:36:34 - brains what should be tim's law that we have
  • fast_forward01:36:36 - to follow uh well
  • fast_forward01:36:40 - it may be a bit obvious but uh to construct is
  • fast_forward01:36:43 - to prove so i think um looking at
  • fast_forward01:36:46 - the biology uh and making making sense of
  • fast_forward01:36:50 - those principles by having them in reality in
  • fast_forward01:36:53 - front of you operating in concrete ways
  • fast_forward01:36:56 - uh is is
  • fast_forward01:36:59 - crucial because uh i know too many models of too many different phenomena that
  • fast_forward01:37:05 - uh you never know how robust they are going to be in a realistic setting or
  • fast_forward01:37:10 - whatever so that would always be my my advice okay and the another thing So five years from now,
  • fast_forward01:37:18 - I'm going to come to Leicester to visit you in your lab,
  • fast_forward01:37:21 - and I'm going to confront you with the hypothesis you're going to generate today.
  • fast_forward01:37:27 - So the question is really, what's the one hypothesis you really want to commit yourself to today?
  • fast_forward01:37:32 - And five years from now, I can come and see if you actually were able to validate
  • fast_forward01:37:36 - it and what the outcome was.
  • fast_forward01:37:41 - Yeah i i think an under an underrepresented aspect uh for the future is to prove that um,
  • fast_forward01:37:49 - attentional processing in olfaction is
  • fast_forward01:37:52 - very important so at the moment we don't really
  • fast_forward01:37:55 - know much about attentional processing we don't we don't really know about it
  • fast_forward01:37:59 - much in biology and we know even less about it in say machines or deploying
  • fast_forward01:38:04 - this and so the prediction i would I would hope that in five years I could take
  • fast_forward01:38:10 - you to my lab and prove to you,
  • fast_forward01:38:13 - demonstrate that I can make olfactory machines that attend to different parts
  • fast_forward01:38:20 - of this beautifully complex molecular world that we have.
  • fast_forward01:38:24 - And that depending upon operational demands at particular points in time,
  • fast_forward01:38:30 - it would be able to give you, say, unique reports or windows on this very complex universe.
  • fast_forward01:38:39 - And I would like to be able to prove to you that it's not just making it up,
  • fast_forward01:38:45 - but it's looking at different facets of a very complex signal.
  • fast_forward01:38:49 - And that hopefully this could be done in a responsive sort of way to a particular task.
  • fast_forward01:38:57 - And that doesn't really exist at the moment. Exactly. All right,
  • fast_forward01:39:00 - Tim Pearce, thank you very much for this conversation. Thanks a lot.
  • fast_forward01:39:04 - Music.
  • fast_forward01:39:09 - The CSN Podcast was produced by the Convergent Science Network of Biometrics
  • fast_forward01:39:15 - and Biohybrid Systems, a project funded by the European 7th Research Framework Programme.
  • fast_forward01:39:23 - For more interviews, recorded lectures or upcoming conferences in the field
  • fast_forward01:39:28 - of biometrics and biohybrid systems, go to csnnetwork.eu.
  • fast_forward01:39:35 - Music.

Be the first to leave a comment

Leave a comment

Your email address will not be published. Required fields are marked *

Exploring the convergence of neuroscience, robotics, and AI through conversations with leading researchers since 2010.

A project of the Convergent Science Network Foundation.

© CSN Podcasts. Developed by IMCreativeWEBC

0%

Login to enjoy full advantages

Please login or subscribe to continue.

Go Premium!

Enjoy the full advantage of the premium access.

Stop following

Unfollow Cancel

Cancel subscription

Are you sure you want to cancel your subscription? You will lose your Premium access and stored playlists.

Go back Confirm cancellation