| |

Integrating Metabolic Phenotyping with Behavioral Neuroscience

Dr. John Lighton and Dr. Daniel Lark discuss how to integrate metabolic phenotyping with behavioral paradigms, the importance of temporal resolution, and how to avoid common pitfalls when executing behavioral and metabolic tests.

VIDEO NOT LOADING? Viewing videos requires acceptance of all cookies. To change consent settings, open the MANAGE CONSENT tab at the bottom right of this window and click ACCEPT.

Haley: Welcome everybody and thank you for joining us today for our webinar titled “Integrating Metabolic Phenotyping with Behavioral Neuroscience.” This is Haley McCaffrey from Inside Scientific, and I will be your host for today’s event. Our session is sponsored by Sable Systems International and will feature scientists discussing how to integrate metabolic phenotyping with behavioral paradigms, the importance of temporal resolution, and how to avoid common pitfalls when performing behavioral and metabolic tests. First, we will be joined by Dr. John Lighton, President and Chief Scientist for Sable Systems International and the author of Measuring Metabolic Rates, a Manual for Scientists. Today he will show how to properly plan and execute combination studies integrating high resolution metabolic measurements and behavioral data. Dr. Lighton will continue with the important physiological behavioral linkages that can be uncovered and qualified using the rich data streams from Sable Systems’ Promethion Synchronized Metabolic and Behavioral System in conjunction with its Ethoscan Behavioral Analysis Utility. Following, we will hear from Dr. Daniel Lark, a postdoctoral fellow at Vanderbilt University. Dr. Lark is interested in understanding how molecular metabolism, particularly at the mitochondrion, impacts health and disease. Today, Dr. Lark will present his findings on compensatory metabolic and behavioral responses to voluntary exercise in mice.

John Lighton: Okay, so we’re going to be looking at using a metabolic phenotyping system for behavioral analysis, and the basic thing is that behind-the-scenes behavior and energy balance are inseparable, and you really need behavior data, if it’s available, to put energy expenditure and energy balance questions into focus. So, the typical metabolic phenotyping systems out there will measure energy expenditure through water intake and what have you, but generally speaking, they will leave behavior pretty much alone. The timing of the information coming from such systems is dictated by the energy expenditure measurements, which typically will be anything from about 10 to 45 minutes between measurements. And the other thing is if you have a behavior system, it will be able to cope with the fast data stream from behavior, but on the other hand, generally speaking, they don’t measure energy expenditure or actually quantify food and water intake. So why the systems that are out there mostly don’t do behavior, they simply have a low bandwidth because energy expenditure is, you know, reasonably enough the overall focus of the system, whereas behavior changes are very rapid. And you have very extensive preprocessing to create a nice, easy-to-use spreadsheet, which is sampled fairly infrequently. And, in fact, the limited information and the small file size are generally regarded as advantages. And behavior pretty much tends to be regarded by many people as a separate phenomenon assayed by behavior-specific systems. So, what I’m hoping to show is that actually it’s possible to combine the two in a reasonably efficient way.

So, the story really begins here at the University of Cape Town, where I studied science and biology, zoology, microbiology. And when I started, when I moved up to do a master’s degree, I still remembered some of the undergraduate lectures that I had at the University of Cape Town, and in particular, one of the behavior pioneers, Jean Alpen, who’s probably emeritus right now at Princeton, really wrote a very, very nice paper, which affected me a lot. And she, in fact, has become something of an icon in the sort of scientific popularization community. And her most influential paper is one published in 1974 on observational study of behavior. and I really found this paper fascinating. We’ll come back to the influence of this paper in a short while, but in any event, I studied for my master’s, which would later turn into a PhD, in ecological physiology under Gideon Lowe. Anyway, George Bartholomew from UCLA actually came to do some work at the lab and stole me, and I became his last student at UCLA. And one of his particular, he was a very holistic physiologist and believed that physiology and morphology and behavior are pretty much inseparable things. Now, I went on to do a conventional academic career for a while. I’ve published about 90-odd papers concentrating on various metabolic measurements in a variety of animals and comparative physiology, and then wrote the book on measuring metabolic rates. Here’s a nice quote that Ted Garland at UCR was kind enough to give Oxford University Press when they asked for quotes: “John Lighton has probably done more to modernize and consolidate the field of whole animal respirometry than any single person.”

In any event, I’m now a recovering academic, and I started the company Sable Systems in 1987, and then we got so many requests to generate a metabolic phenotyping system that we finally decided, okay, this is the time to do it. We had the system ready in 2010. It’s very different. It’s designed from a physiological scientist’s perspective, not an engineering perspective, very fast sampling, relatively speaking, from all of the sensors in the system simultaneously, and then storage of absolutely raw data with no processing at all initially, so that you have basically a complete picture of the entire experiment. High resolution is a bit of a fetish of mine, and so I went for very, very high resolution on all of the sensors. And a lot of technical innovations to speed the response times of the gas analyzers. And needless to say, the analysis of the information from the system is fully traceable. You can go all the way from the raw readings of the sensors to the final published data, drill down as much as you want, and everything is there. Nothing is hidden. This is what’s called the Promethion system. And these are some of the people, not all, this is somewhat out of date, who are using it worldwide.

And your question at this point, quite reasonably, is, well, okay, whatever. This doesn’t really seem to have anything to do with behavior. So let me show you how behavior actually comes into the picture. So, this is the cage that the system uses. And notice, I’m going to be bringing up some texts, and the red text is relevant to behavior quantification. So, we have mass monitor modules that measure the mass of anything suspended from them very accurately and with very high resolution. We have a water dispenser here shown without a shroud for water intake, food hopper for food intake. We have an XYZ array which locates the animal in two-space very accurately. We have an access control if that’s required for the experiment, a running wheel within the cage itself, and we have a body mass habitat or an enrichment habitat into which the mouse can wander and every time it does so it gets its mass measured. So then also we have a metabolic measurement intake system in the cage. It’s an unusual design and has proven to be very effective, but you can do the behavior studies without needing metabolic measurement.

So, let’s have a look at some of the raw data from, let’s say, food intake. Here we have the hopper undisturbed, disturbed by a feeding event, undisturbed another feeding event. From the change of mass of the hopper from before to after the feeding event, we can obtain, of course, the amount of food eaten by the animal. This is a blow-up of one particular feeding episode. You can also quantify the exact amount of force that the animal exerted. The nice thing about the food intake is that we have very, very high resolution. We can measure reasonably confidently a food intake event down around 2 milligrams. And the older systems out there can only manage maybe 10 to 15 milligrams as the minimum, and it turns out that these micro-intake events, as we call them, account for about 30 or 40 percent of total intake events and they’re simply invisible to the old systems. So, then we can do intake pattern analysis. So, we’re looking at ingestive behavior. The highlighted regions would be invisible to conventional systems and all each feeding event is quantified in terms of the probability of it occurring by chance, which is very, very low for most of these.

Now, what about water intake? Exactly the same story. You can do all of the same analyses with water intake. What about body mass? Well, if you look at the raw body mass trace here, it looks like, you know, just trash. But if you blow it up, you can see here the animal is outside the body mass habitat. Here it goes inside, outside, inside, outside, inside, outside, inside and so on. So, you can tell exactly when the animal is in that habitat and each time it enters and each time it leaves, it leaves behind an accurate body mass measurement. Voluntary exercise in the running wheel is binned every second so we have second by second granularity on running wheel behavior, which is actually really important because the animals tend to get on and off the running wheel very rapidly. You can do cumulative distance if you wish, but as you’ll see later, it feeds beautifully into the behavior information.

Then we have total activity, so we have the position and activity of the mass also recorded on a second-by-second basis. The physical beam spacing is a centimeter. We can do easily a 2.5-millimeter calculated centroid in real time. And here you can see a heat map; this is actually generated in R from one of our cages. And, of course, you can use this kind of data for behavioral assays of position preference for depression, anxiety, and what have you. And you can see in this particular case the animal is clearly avoiding these areas of the periphery of the cage. Now let’s move on to metabolic measurement just briefly. This is our gas analyzer system. It looks very different from anything else out there because we designed and made it specifically for very high-resolution metabolic measurement. You can have up to 500-odd cages within a single system using these analyzers.

Now, what about the cage in terms of metabolic measurement? Well, most cages in the older systems are sealed. This cage is not. There is an imperfect seal around the top of the cage. And you notice the steel tube over here that runs around the bottom of the cage, And that is perforated by hundreds of tiny laser-punched holes and give a very distributed even air stream going down from the cage lid to a very distributed area around the periphery of the cage, pulling all of the air that the masses exhaled with it. And we use a very high flow rate, about five times higher than conventional systems. And we’re able to do that because we have gas analyzers that have the resolution to pick up the delta O2s, delta CO2s, even though the flow rate is very high. So, this is how we select in a multiplex system from switched cages. This is just a small 4-cage system. And here we have oxygen concentration, incurrent oxygen concentration at the top, 20.94%. And you can see we get a really nice accurate resolution of individual readings with a 15-second dwell time. Dwell time means that that’s the amount of time taken to get the reading. And we can easily do that in 15 seconds, and that translates to whipping through all of the cages very, very rapidly, because each cage, each analyzer has eight cages assigned to it. And so, with 15 seconds dwell time, we have a two-and-a-half-minute cycle time. And notice over here the data for cage four is interpolated. And you’ll also notice that because of the cage time constant, which we’ll come to later, there isn’t that much change between individual readings for each cage.

Okay, so let’s look at the relation here between activities. Notice these spikes here at the bottom; orange is pedestrian activity; blue is running wheel activity. Notice that each of these spikes corresponds to a really nice, well-defined peak of energy expenditure. These peaks were measured using a continuous system that we also make. And then we simulated a cycle time of 20 minutes, which would be pretty average for most metabolic phenotyping systems. And you can see it misses these peaks all together, it kind of amalgamates these, misses these, amalgamates these, and what have you. In other words, it’s hopelessly inaccurate. Going down to five minutes, we’re starting to get a pretty close resemblance to the underlying data. And at two and a half minutes, which Promethion is easily capable of, we have really, really close resemblance. And in fact, at that point, there isn’t a whole lot of point in going to a continuous system.

Now, cycle time is only part of the story. The other major limitation with regard to temporal resolution of metabolic signals is the time constant of the cage, which is the cage volume divided by flow rate. Typical cage volume for metabolic phenotyping with a home cage system will be about eight or nine liters STP. Legacy systems cannot pull more than about 400 or 500 mils per minute from these cages until they basically are unable to get accurate readings of the delta between the incurrent and excurrent air from the cage. So their time constant is about 20 to 25 minutes. So, if the mouse starts doing something, it will have to continue doing it for quite a long time in order to make a measurable difference to the signal from the cage. Promethion is about five times faster. So, we can actually get good resting energy expenditure and behavior energy expenditure from the system. And again, this is possible because we have our own gas analyzers, which are very, very high resolution. And one of the things that I concentrated on during my academic career was doing ultra-high resolution gas analysis, because I was working with really small signals. So, the combination of the fast cycle time and the fast time constant is actually synergistic.

Now, you might say, well, what difference does this make in practice? Let me show you. Here is data from an old system, a typical system in use. Lots and lots of these are used worldwide. And you can see here the blue trace is V02, which is kind of a stand-in for energy expenditure. And you’ll see here is activity. It’s very difficult to see really what the exact relation between activity and energy expenditure actually is. Certainly, the energy expenditure increases during the SCOTO phase, and the activity also increases, but you can’t really see the individual energy signals from the ambulatory activity. So, you don’t really have the ability to look to determine resting energy expenditure and activity expenditure without using some mathematical techniques of, you know, looking at activity on the x-axis, energy expenditure on the y, and then extrapolating back to zero activity and praying.

Okay. Here is the Promethion system, and it’s very, very, very different. Notice over here you have very, very good resolution of resting energy expenditure, so you do not have to engage in any mathematical equation in order to obtain a resting energy signal. It’s simply right there. Likewise, a very tight correlation between activity and energy expenditure. The moment cross-bridge cycling begins because of activity, you have the whole chain reaction resulting in oxygen consumption, CO2 production from which the energy expenditure can be calculated. Notice also the RQ or RER, if you wish, is very accurate because we actually measure water vapor and correct for its dilution effect mathematically. So, a very, very different looking signal from what most people are used to seeing.

So, the result is we have a very rich data stream coming in with a lot of information, all of which is synchronized to a one-second heartbeat. And so, we have wheel running, we have ambulatory locomotion, these are behavior signals, we have food intake, water intake also behavior, body mass also behavior. So, the behavior data that the system gives are derived from these bits of information, data streams, and others as well. But you might wonder, well, you know, stress has a major effect on behavior and it’s well known that metabolic phenotyping systems induce a lot of stress in animals and the reason you know this is because the animals tend to lose weight and eat very low amounts of food for the first few days that they’re in the system.

So here we have 21 body mass measurements from eight C57 black sixes and as you can see, the animals begin their acquaintance with the cage. In this particular case, it was straight out of a little communal box coming from Jackson Labs. And you can see that the body mass, generally speaking, increases at first and then plateaus. So, there’s not really any real evidence for a stress effect here, which is really important for getting reasonably accurate behavior information. So now we can actually make really good use of the temporal resolution. For example, we can look at activity energy expenditure of the kind that you would normally need a treadmill to measure. So, if you see here on this graph, we have in blue the wheel running speed, and in red we have the energy expenditure. And this is all perfectly synchronized. You can see, for example, here the wheel running and energy expenditure. And so, we can take these individual bouts of wheel running and energy expenditure, and we can go ahead and determine what the relation is between them, and we generally get really, really good, tight correlations. The nice thing here is that you do not require a treadmill to get this kind of information. There are no shocks involved, there’s no cortisol, it’s entirely voluntary and you can obtain information from all of the animals in a whole system overnight.

Now let’s go to real behavior monitoring here. So, this is the actual graph that I had up on my screen when I suddenly realized that my goodness, in having designed the Promethion system, we’ve actually designed a really good behavior monitoring system as well. So, let’s look at this rather complex slide just briefly. The body mass habitat mass is in black, the food intake is in red, water intake is in blue, and the wheel running – bindery wheel running revolutions – is in gold. So, let’s start off over here. The animal is in the body mass habitat, it takes a drink of water, gets out of the habitat, takes a drink of water, gets onto the running wheel, gets back into the body mass habitat, eats. You can see the reduction in the hopper mass there in the red trace, really obvious. Takes a huge drink of water after eating, gets back into the habitat, gets restless, takes another drink of water, a small one, on the running wheel, back in the habitat, running wheel, habitat, running wheel. Now it does nothing. So, this is a short roam or a short lounge where it is basically just wandering around the cage and not interacting with any of the sensors. Then it takes a drink of water, goes to the body mass habitat, wheel, body mass, wheel, body mass, wheel, body mass, wheel. Again, doesn’t do anything for a short moment, takes a drink of water, body mass, wheel, drinks water, gets into the habitat just very briefly, and you can see how this whole thing goes.

So, I suddenly realized this actually is exactly what I remembered learning about when I was an undergraduate, when we dealt with Jean Altman’s work. Because what we have here is what she categorized as focal animal sampling, the only difference being that rather than an individual having to look at one particular animal at one time, here we are actually looking at all animals simultaneously. So, we have a good behavioral record using focal animal sampling. And from this we can build a behavior list. And you can see here, for example, for 11 seconds we have a short roam or short lounge behavior. For 12 seconds it’s drinking water, 55 microliters. Doing for 205 seconds it does 171 revolutions on the wheel, gets of the habitat, weighs 25.63 grams, and so it goes on. So, we have a complete then list of all of the behaviors that the animal actually sequentially did during each of these. And another nice thing is that we can actually optionally also add any parameter that you wish, such as, for example, energy expenditure, RER, or whatever you wish, synchronized with each behavior. So, you might say, for example, what is the energy expenditure when it’s in the habitat for more than a certain length of time or when it’s running on the wheel or whatever it may be. Okay, so from this then, we can actually generate time budgets and locomotion budgets. So, we can determine how much of the animal’s time is spent doing each of these behaviors and also what distance it travels during each of these behaviors.

So, this is a slide courtesy of Cathy Kotz at the University of Minneapolis, and she was using some of the behavior capabilities of her Promethion system, looking at the effects of high-fat diet here on time spent being inactive. And you can see the high-fat diet animals spend a lot more time inactive than the chow-fed animals – and eating, other way around. The high-fat diet animals spend much less time eating than the low-fat diet. So, you can get some very useful information from the time budgets, and with the time budgets then you can feed these into advanced statistical analyses such as hierarchical object clustering and you can then say, okay, I have a group of control animals and a group of experimental animals, do their time budgets differ? In this particular case, they do. Here is one group over here. This is the control, and this is the experimental group. So, it gives you a very nice objective way of separating out or looking at the effects of a treatment, a knockout, or whatever it may be.

Then, we move on to something even more interesting, which is the behavioral transition probability matrix in Ethoscan. I’ll lead you through this just briefly, so let’s look here. We have the eating food behavior. After eating food, the animal is 42% likely to drink water. It will not go on the wheel. So, let’s look then at what happens after drinking water. After drinking water, again, it will not go on the wheel. It’s about a quarter of the time it’ll go into its habitat, but about more than half of the time it will go into a short lounge or short roam behavior. Now let’s look at what happens after the short lounge. The most likely behavior is to get on the wheel. And so, this really gives us a way of looking into the mind of the mouse to see the ways in which it goes from one behavior to another to another. And it’s a very powerful tool.

Again, this is courtesy of Cathy Kotz, looking at the relation between, or basically the partitioning of overall daily energy expenditure between exercise, NEAT, which is non-exercise activity thermogenesis, and the thermic effect of meals, SDA or DIT, and teasing these apart. So here we have spontaneous physical activity. High and low animals will accumulate larger amounts of body fat if they are low SPA and smaller amounts if they are high SPA. And their orexin levels also of course differ as well. So, what about the effect of activity? So here we have eating, this is the time spent on chow or fat. You saw that a moment ago. Short bouts of inactivity, so short roams. The fat animals, the HFD-fed animals, tend to have more of the short roams, the long bouts of activity, but not significantly so. The long bouts of activity, you can see that they definitely have much larger numbers or larger proportion of their time is spent in long bouts of inactivity. Now we go to the probability of inactivity after eating, and the HFD animals spend significantly less time with short bouts of inactivity right after eating, and far more time with long bouts of inactivity. So, you can see, you begin to get a picture of the way in which their energy balance is likely to change.

So needless to say, you can also look at the behavioral transition matrix data in all kinds of different ways. This particular one is using Markov Chain Analysis, or Markov Chain Visualization, and you can see the ways in which the animal will transition from one behavior to another to another, and you’ll get more of this in Dan Lark’s talk. So here, for example, you can see after the short lounge, it will generate, will often transition to the wheel. After the wheel, it will transition back to a short lounge, et cetera. Now, all of this is made possible by the fact that we have a very, very high bandwidth. We get about 400 times more samples per unit time from our sensors in the system than the older technology systems. This is made possible by an all-digital data transfer system, which uses self-correcting, error-correcting technology. And so, it’s very high speed, and we can use very high-resolution numeric techniques to actually get the information from the system.

So, I was looking for a metaphor to describe this huge amount of information that you get from the system, which gives you a lot of versatility. And I came across the light field camera, which basically you point this camera at a scene, and it captures all of the raw light data. And then afterwards, you can go ahead and change the focus of the picture as you wish. And so, to continue the metaphor, let’s say, for example, you’ve just finished a study, you are about to submit it, and someone comes up with a new analytical technique, which is an improvement on the one that you were using. And so, you would like to go ahead and use that new technique. It may be with an older system you’d have to rebook the metabolic phenotyping center. With the Promethion data, you can simply change the focus of the analysis, and there you go. Now you submit the paper, one referee thinks it’s wonderful, another referee thinks, well, I really would like to see detailed meal pattern analysis from these creatures. You gave me total amounts; I really need to see the meal pattern. In my opinion, the paper really is not publishable without that. With an ordinary metabolic phenotyping system, you go back, order more mice, go back to the metabolic phenotyping center, book a new time, six months away. With the Promethion data, you just change the focus, and you have it immediately because everything that the experiment can yield, you have. So, the result is that you can start off with a focus that you think at the time is okay. It turns out, in light of later findings, that you need to refine your focus and you can easily do that with the system.

Daniel Lark: So, John finished his talk with an analogy of how you can take Sable Promethion data and go from maybe a broad scope to a more refined scope. And that’s really, I’m going to show you in practice how we’ve done this in the lab. And also I should note that you know my expertise is really a mitochondrial function, and so we started off, you know, looking at some of this data through the lens of mitochondrial function, and we really didn’t find anything of interest and so we were able to delve deeper into the information that we already had. Like John said, we took the data set that we had and really started peeling layers off it, and that’s really how we came to the conclusions that we’ve made, that I’m about to show you.

And so, the focus of the talk really is energy balance and regulation of body mass. And to remind everyone, the formula for that is energy supply minus energy demand. And so, in an energy balanced state, energy supply will match energy demand. And we can think of this pretty simply as basal metabolic rate and non-exercise activity being these components of energy demand, and then you can manipulate the scale and increase energy supply. And if energy supply exceeds energy demand, you have an increased risk of developing diabetes, obesity, components of the metabolic syndrome. Conversely, if you exercise and increase energy demand relative to supply, you elicit weight loss and you can improve metabolic health. And so really the question that we wanted to address in this study is when you add exercise, are these you know components of energy demand retained or is there some sort of compensation to demand that occurs. And so, we see some of this in epidemiological studies with humans where when we give people a diet or exercise program, they don’t lose as much weight as we would predict and it’s likely because of a number of factors. And the question that we had was what contribution does non-exercise activity, or in this case off-wheel activity, have on energy balance?

So, there’s some evidence that this is important in the literature. This is a pretty recent paper from Alexi Kravitz’s group at NIH, and what they did was they just took a camera and viewed a mouse’s movement around its cage without a wheel and then with the wheel, so this is at baseline without a wheel. And when you give mice access to a running wheel, obviously they use the running wheel a lot, but then they also move around their cage less, and so when you quantify this you see that you get a significant drop in off-wheel activity when mice have access to a wheel. And so, the question then becomes what impact does this have on energy expenditure? What impact does off-wheel activity in general have on energy expenditure and then what’s the impact of this drop in off-wheel activity on energy expenditure?

So, this has been surprisingly elusive to measure and so there have been, to date, no papers that have been able to report an independent contribution of off-wheel activity to energy expenditure. I’m going to show you one of the studies that attempted to and was unable or had concluded that there is no contribution. So, this is energy expenditure on the y-axis and off-wheel activity on the x-axis. And what they did is they took individual mice and plotted these data and looked at the activity rate and the energy expenditure and found that there was no correlation between how much a mouse moved around in its cage and its overall energy expenditure. And when they took individual data points from these mice and bin them based on their activity rate, they did find a correlation, but they found that it was most linear at these low rates of activity. And so, they used these low rates of activity to calculate active energy expenditure. And when they then took that active energy expenditure and plotted against activity, they found an R-squared of 0.24, but it was really dismissed because of these two outlying data points. And if you remove those, the correlation is essentially zero. And so, they concluded from these data that off-wheel activity is not an independent contributor to energy expenditure. And what I’m going to show you today is that by using these low rates of activity, they may have actually selected for a more complex behavior that masked the contribution of off-wheel activity.

And so, the study design that we used was we took our Sable Promethion multiplex system – and this is with five-minute sampling intervals – we used male chow fed C57 black 6J mice purchased from Jackson laboratory presumably at least some of these were littermates. They’re all bought at the same time at the same age. They’re 19 weeks of age when they went into the system. And the way we did this was we had them in a cage that had a running wheel the entire time. But for the first four days, we locked the running wheel, so they weren’t able to use it. And then for nine days we had the wheel unlocked. And over the entire two-week span that they were in the sample Promethion system we measured energy expenditure, energy intake based on the composition of the chow diet that they were consuming, off wheel activity, and voluntary wheel running when the wheel is unlocked. And just to show you the challenges I guess with using voluntary wheel running as an exercise stimulus, there is quite a lot of variability within this cohort of genetically identical mice that makes interpreting these sorts of data sometimes challenging. And so, the other kind of the second part of the talk is really going to be using some of these more precise behavioral outputs that John talked about to get a better understanding of what factors are actually determining these differences in wheel running behavior.

And so just to show you as a group, the mice’s overall energy expenditure increased about 25% when they were able to run on the wheel in the dark phase. Because these were all paired comparisons, we do see an increase in energy expenditure during the light phase, but this is a relatively small effect size compared to the dark phase. And the net effect of this on energy balance – which again is energy intake minus energy demand – energy balance became negative when they had access to a wheel as you might expect. Just like others have shown, a number of different groups have shown this in the last few years, that off-wheel activity decreases when mice have access to a wheel. That was true in our studies as well. And when we started looking at relationships between the change in off-wheel activity, so that delta, pre and post access to a wheel, and their metabolic variables, we didn’t see any relationship with energy expenditure or energy intake as individual variables. But when we calculated the energy balance, we found a correlation. And what this is suggesting to us is that the greater the drop in energy balance or the more negative the energy balance, the more likely that mouse is to lose weight or the more weight it might lose, the greater the drop in off-wheel activity. And so, this suggested that maybe there’s a metabolic link, a link between the metabolism of the animal, the change in metabolism of the animal, and the change in behavior that we observed just at the gross level here, just looking at distance traveled around the cage.

And so, in order to really quantify this, we needed to be able to measure the contribution of off-wheel activity to energy expenditure, and I mentioned that has not been shown before, and so we took a stab at it with the data that we had. And so we took our five minute sample data – and this was three consecutive dark phases of data that we used – and came out to be 435 individual data points per mouse that we used for this analysis and we separated the data out based on their off-wheel activity speed over that five minute period. And we stratified that data into 0.1 meter per minute bins and then extracted the corresponding energy expenditure over that same period of time. And then we just plotted that, and this is what the data looks like. And so, you can see at zero meters per minute traveled (and keep in mind this is all dark phase data so this is presumably there may be a little bit of sleep involved but this is presumably awake rest. This would be, you know, I think what you could probably say is you’re resting energy expenditure and then as the mouse starts moving you see an increase in energy expenditure that’s pretty clear. At one point you get about a 50% increase in energy expenditure. And so, the shape of this data looks very similar to what was shown previously, where they, again, tried to find a relationship between off-wheel activity and energy expenditure at these low rates of speed. What we wanted to do was see, do we find the same relationship at low rates of speed, so say above zero but below 0.3 meters per minute, versus maybe these high rates of speed. And so that’s what we did. And so, to calculate our energy expenditure for off-wheel activity, we had to calculate the active energy expenditure, which was the fraction of total energy expenditure above rest. And then we divided that by the percent of time that a mouse spent running at that speed. And what that gave us was an absolute kcal per hour measurement of the contribution of that rate of off-wheel activity to total energy expenditure. And then we just added that up for all the speed all the bins that we had for each individual mouse. And so, we added those up, and then we plotted those against the off-wheel activity distance of the of the animal. And all of this was before they had access to a wheel, so this is not with the wheel, this is just exclusively basal cage behavior. And what we found was when we looked at the total energy expenditure attributed to off-wheel activity there was no relationship with off-wheel activity distance. When we did the similar comparison to what had been previously published, we found essentially the same thing, which is that there was no relationship between low rates of off-wheel activity, the energy expenditure at low rates of off-wheel activity, and off-wheel activity distance. However, when we looked at the high rates of speed, we found a strong positive correlation. And this was the first indication that maybe this high rate of off-wheel activity was a truer indicator of the independent contribution of off-wheel activity to energy expenditure.

And so, then we wanted to dig deeper into this and see if there were other behaviors that could be masking the contribution at low rates of speed. And the obvious one was food intake, and so here what we did was we took those data points below 0.3 meter per minute but above zero and ask the question are mice eating food over that same five-minute period?  And if they are we put them in one bin, and if they’re not we put them in the other.  And so, what we found was if mice were eating food at these low rates of activity it was a pretty major contributor to their energy expenditure, about a 30% increase. By contrast if we do the same comparison when mice are moving above 0.3 meters per minute, there’s still a statistical increase because of the paired comparison that we’re using, but this is about a 3% increase in our energy expenditure. And so, what we can conclude from the data that we have so far, and this is, again, really pretty gross scope, a pretty big picture scope here, is that at low rates of off-wheel activity, there’s probably some contribution from the movement around the cage to energy expenditure, but it appears that food intake is maybe a more important contributor. And this is probably including water intake and maybe some other behaviors. This is just an example of one activity that is occurring together with the off-wheel activity. But certainly, I’m not meaning to imply that it’s this simple. By contrast, when mice are traveling at a higher rate of speed around their cage, it does appear that the movement around the cage is the primary contributor to energy expenditure, at least that food intake is not.

And so, we take this to be the true contribution of off-wheel activity energy expenditure. So, then we wanted to take that data now that we’ve shown an independent contribution of off-wheel activity to energy expenditure and see what effect the drop in off-wheel activity that we observed has on energy balance. So this is just to remind you that when we see a drop in off-wheel activity, when we take the energy expenditure attributed to off-wheel activity at these high rates of speed, comparing when the wheel is locked now to when the wheel is unlocked, we see about a 50% drop in that contribution. And that equates to about a 5% change in total energy expenditure, which may not seem like a lot. But when you calculate energy balance and you say, what is energy balance that we measured and then what would energy balance be predicted to be if there was no drop in off-wheel activity? If off-wheel activity were retained, then what would energy balance look like? And what it comes out to be is about a 45% difference in energy balance. So, to phrase it a little differently, the drop in off-wheel activity appears to be a pretty potent mitigator of negative energy balance. And the idea is that this is a physiological response of the animal to conserve body mass.

And so, to summarize this first part of the talk, with the introduction of exercise, you would predict that a significant amount of weight loss would occur. However, there is a drop in the contribution of non-exercise activity to energy expenditure, which results in only a modest, if any, weight loss versus what would be predicted to occur. And so, I showed you that there was a drop in off-wheel activity, I showed you that it’s important for energy balance, but now we wanted to dig deeper into what components of off-wheel activity change and whether these changes in specific off-wheel activity behaviors are related to wheel running distance and activity.

And so, in order to do that, we use the behavioral transition mapping and the behavioral list data that John showed at the end of his talk. And just to clarify some definitions really quick, this is a short roam or short lounge where mice aren’t triggering a mass sensor, they’re just moving around their cage or sitting still, and then the long roam is a behavior of greater than 60 seconds, and these are pretty intuitive; the wheel, eating food, drinking water, etc. And so, we first wanted to see as a group what happens to these different behaviors when you give mice access to a running wheel. And an important note to this is that we removed the time spent running on the wheel for these analyses. And the reason why is because of the variance in wheel running behavior, we wanted to pull that out and say, OK, what is just the contribution? What is the effect of just giving them a wheel, independent of how much they actually use it? And so, what we find is that there’s an increase in short roaming, time spent short roaming, a decrease in time spent long roaming, a decrease in their time at the water and drinking water, but no change in their feeding behavior. And if we were to push this out longer, we may see some of these effects in feeding behavior, but I think it’s probably just because we only used nine days.

And so, one of the things that jumped out to us was when we saw an increase in short roaming, and so we took those individual data points for each mouse and said, okay, what’s the relationship between the change in the short roam, so how much it increased, and their wheel running distance traveled. And what we found was a positive correlation between their change in short roaming but not their long roaming. And so, this got us thinking that maybe the short roaming behavior was somehow functionally linked to their wheel running activity. And so, we took the behavioral probability transition mapping and generated our Markov chain – this is just a visualization of the data. And to get to the point here, because John went through some of this, just to orient you to this graph, the thick lines are transitions that occurred greater than 20% of the time. The dashed lines occurred between five and 20% of the time, and below 5% aren’t shown, just for simplicity. And so, like John mentioned, the most likely transition from a short ROAM is to go on to the wheel and likewise the most likely transition from a long roam is to go on to the wheel. However, when the mouse is transitioning off of the wheel, 86% of the time they’re going into a short roam versus only 9% into a long roam and so this connection between short roaming and the wheel suggested that maybe these things are happening in series and may represent a circuit of behavior that could explain some of the variability in wheel running between mice. Maybe this is one of the behaviors that’s kind of tied to wheel running behavior. So just to give you a sense of the kind of scale of this, this is just the full plot. Obviously too much to go into right now.

And so, with this idea that short roaming and wheel running could function as a circuit, we wanted to look at how the frequency with which a mouse transition from a short room to a wheel, which we knew as a group was 39 percent, I want to see if the fraction of transitions to that, relative to the total number of short-running or short-roaming events predicted wheel running. And so, does the abundance of this transition predict how much they run on the wheel? And indeed, it does. It explains about 54% of the variance in this particular group of mice. When you do the converse and you look at the transition from the wheel to the short roam, there’s no relationship. So that was interesting to us. And so we wanted to see if this circuit really existed, and indeed it does, and I’ll show you this data in just a second, but then we wanted to see if the number of repeated bouts, so that is going from the wheel to a short roam and then back to a wheel in series, so three consecutive behaviors, whether that predicted wheel running distance. And surprisingly it didn’t. And so, what got us thinking is maybe all mice have these repeated bouts, but it’s not as simple as them getting on and off the wheel. Maybe some of these more conventional predictors, like bout duration or speed, are important for this as well. And so, we looked at average bout duration first and we found that that also was a positive predictor of wheel-running behavior. And then when we took the long running bouts, so we had a cutoff of 100 seconds of consecutive running, and we said, okay, if they’re engaging in a long running bout that is also repeated, so it starts with a long running bout, it goes to a short roam, and then it goes back to the wheel for a repeated bout, maybe that is a better predictor. And indeed, that was actually the best predictor that we found is the repeated long running bouts. And so, this really illustrates the link between an off-wheel activity, the short roam, and wheel running. And so, this integration between exercise activity and non-exercise activity is really what we think is probably the most important takeaway from this particular data set. We’re really just scratching the surface. John mentioned that you can measure energy expenditure during each of these individual behaviors and we really haven’t done any of that. So, there’s certainly a long way to go.

So, to summarize this data, we found that off-wheel activity is indeed an independent contributor to energy expenditure, that it certainly can be detected at high off-wheel activity speeds. The contribution of off-wheel activity energy expenditure decreases when mice engage in wheel running and this mitigates quite a lot of their potential weight loss, the effect on energy balance. And then finally wheel running drastically changes mouse behavior and not just their off-wheel activity distance and it reveals what we believe to be in a behavioral circuit that integrates exercise and non-exercise behaviors. And so, with that I do want to thank my mentor, David Wasserman, the Vanderbilt MMPC, which is where we house the Sable Promethion system. Mary Gay is the technician who runs the Sable system. Jamie is an undergrad in the lab who’s been instrumental in these studies, as well as John Lighton for helping with some of the analyses and NIH for funding.

Haley: So, John, this question is for you. Do you need to use the behavioral analysis system in conjunction with metabolic measurement?

John Lighton: Sorry, you asked if the behavioral system is in conjunction with metabolic measurement? Yeah, I mean the two of them occur simultaneously. So pretty much you have synchronized behavior and metabolic information. However, you can have the system for behavior without doing the metabolic analysis at all, if you wish. And so, if the metabolic information, the energy expenditure is not of interest for the research questions you’re asking, you can simply just get the system without the metabolic measurement, which means it’s a good deal simpler to operate and a lot less expensive.

Haley: Right. And another question, can you export the data from the system using other behavioral analysis programs?

John Lighton: You can certainly export data from the system to other behavior analysis programs. So, all of the information in the system, raw or processed, can be exported as CSV files. You can send given sections straight to Excel, if you wish, and so on. And so, any programs that can cope with CSV files and raw data should be able to be used with it. We also have R and Python utilities to read our data files directly into R or Python.

Haley: Okay. Dan, a couple of questions for you. Can you detect the acute contribution of voluntary wheel running to energy expenditure like you showed for off-wheel activity?

Daniel Lark: So, the answer is absolutely yes. That’s actually the easier one to measure and people have already shown previously that you can detect the contribution of wheel running to total daily energy expenditure. But yes, we can do the same thing that I showed with off-wheel activity, these five-minute increments. We actually see the same thing with wheel running. And in fact, some of the data that I didn’t show is that the efficiency of wheel running, So, the energy expenditure per meter traveled is actually a pretty strong predictor of overall distance. And so, not only can we do that, but you can also start digging into relationships between, again, the behavior and the energy expenditure during wheel running, just like we did for off-wheel activity.

Haley: Excellent. And does the change in energy expenditure correlate with the amount of voluntary wheel running performed?

John Lighton: Yes, it absolutely does. So, this is something that has been another elusive thing in the literature. And so there are a number of studies that have shown that, like I just mentioned, that wheel running can increase daily energy expenditure. But to my knowledge, the only studies that have been able to show a linear relationship between the distance traveled and the energy expenditure increase has been from Sable Promethion studies. So, I don’t know of any other instrument system that has been able to show that.

Haley: Okay, and a couple of questions from the audience have just come in. Claus asked, how high or what is the maximum flow rate of the system? John, could you answer that question?

John Lighton: Okay, if we’re dealing with mice, the default flow rate is two liters per minute, so that’s about five times higher than the conventional systems. And you can easily increase that for mice to three or four liters a minute, if you wish. There’ll be a slight reduction in terms of the delta. There’ll be, of course, corresponding reduction in the delta O2 delta CO2, but we have enough resolution to manage that without any particular problem. So, the maximum flow rate of the system is around, depending on the particular system, would be four or five liters a minute, and so obviously the higher flow rates are used for rats mostly. But if you really want to get a very fast time constant, you can certainly increase the flow rate above two liters per minute for mice.

Haley: Okay, and a question from Ben. Is there any detriment to an experiment if one of the cages for some reason becomes compromised?

John Lighton: I’m not quite sure what compromise means in this particular case.

Haley: For example, if it became broken.

John Lighton: Well, if the cage breaks you have some problems definitely. Now obviously the cage is not completely sealed, so in other words if the seal is slightly broken in general that’s not going to have a very significant effect because we are pulling from the bottom of the cage at a fairly high flow rate. But it depends again on the nature of the compromise. We have a lot of people who do things, for example, they will take a mouse out of the cage briefly, inject it, and put it back into the cage. And in our experience that has minimal effects in terms of downstream disruptions of metabolic information. Obviously, there’s a short-term one, but within three or four minutes basically you’re back to pretty much where you were before.

Haley: Okay.

Daniel Lark: So just to add on to that really quick, John, I think you can probably speak to this obviously better than I can, but the real-time monitoring of the system is really quite an advantage in that you can detect problems. So, say a mass sensor isn’t working, you can detect that. And I think you mentioned that you’re developing some online app type stuff to deal with that as well, right?

John Lighton: Yeah, exactly. In The latest version of the Promethion system is web-enabled, and you can use it on an intranet or even with suitable proportions on the internet. You can control the system. You can download full raw data or process data from the system. There’s a background real-time analysis kernel, which does a really, really good job of analyzing the raw data in real time as well. And you can get graphical output from that onto a desktop or laptop or phone or whatever you wish.

Haley: Okay, and a question from Ashley. Can you restrict food during certain times with this system?

John Lighton: Absolutely, yeah, no problem at all. You can restrict food based on times, you can restrict food based on the amount that the animal is allowed to eat, you can do paired feeding, you can do yoked feeding, all of the usual paradigms, and if you come up with a paradigm we don’t already support, just let us know and we’ll be sure that it becomes supported in the future.

Haley: Excellent. Javier has asked, is it possible to house several animals in a behavior measuring Promethion and use some type of RFID readers to detect activity from the different animals?

John Lighton: You can certainly do that. I mean obviously if you do the metabolic information, you will be getting the overall metabolic rate of all of the animals combined so you can’t really distinguish them. We are actively developing RFID at the moment, and we actually have a major RFID system which is going to be delivered fairly soon and you know there’ll be more information on RFID and multiple animal housing and behavior information within, I would say by the end of quarter one of next year we’ll have a lot to show.