Why do other measurements of stress suck worse than cortisol?

After an overwhelming number of requests (2) for a sequel to my post Why cortisol sucks as a measurement of stress, I am obliging. The fact that I am in the middle of writing this particular section of my thesis and need some high-level perspective on it might also have something to do with it. So: why do other measurements of stress suck worse than cortisol?

When I left you, you were trying to design a study of stress in hospitalized dogs using cortisol as your marker of psychological distress. You were confounded by the fact that cortisol measures both psychological and physiological distress, and that it varies a lot between individuals. I haven’t been around to keep an eye on you lately, so you have started investigating other approaches to measuring stress other than cortisol.

Cortisol is a messenger used by the HPA (hypothalamic-pituitary-adrenal) axis, for the brain to send a message about stress levels out to the body, for the body to pass that message along to the organs that need to change their operations as a result, and for the body to then report back to the brain that the message has been received, so the brain can stop yelling about it. There are multiple levels in this axis; cortisol comes from the bottom-most level, the adrenals. Why not go up one level, to the pituitary? It is actually in the brain, so it is closer to the source of the message and might be less distorted by the game of telephone.

The hormone that the pituitary gland releases as part of the HPA axis is ACTH (adrenocorticotropic hormone, or “the hormone that makes the adrenal cortex change”). ACTH causes cortisol release. Why don’t you measure ACTH release directly? Unfortunately, ACTH can only be measured in the blood; it doesn’t get into the saliva. (Or urine, hair, or feces, three other places you can go to get an estimate of cortisol levels.) The owners of your hospitalized dogs aren’t going to be happy if you tell them you need to draw blood from their dogs for your study. And remember, you’d have to draw the blood pretty quickly in order to get it before the brain mounted a stress response as a result of having a needle stuck into the body. Cortisol levels change in under three minutes. I don’t actually know how long it takes ACTH levels to change, but I will hazard a guess that since they are farther up the telephone chain, they change faster.

What about farther down the chain? CBG (corticosteroid binding globulin, a.k.a. transcortin) is a protein that carries cortisol around in the blood. The body uses CBG as a way of regulating the stress response. When there is less CBG, cortisol is more able to jump inside cells and do its work. OK, no one actually uses CBG to measure stress levels, because we have no real idea how it works. But it is a very cool system that I’m really curious about. And stress researchers would do well to remember that it is there. If the dogs you are studying are very sick, they might not be able to make as much CBG as a healthy dog would, and that would affect their cortisol levels.

That pretty much exhausts using the HPA. Luckily there is an entire second axis for you to mine: the SAM (sympatho-adrenomedullary) axis. This is the series of chemicals that regulate the well-known “fight or flight” response. This particular game of telephone includes adrenaline (epinephrine), the effects of which which many people enjoy abusing when they go on roller coasters. This axis works much more quickly than the HPA. If you hear a sudden loud noise, you will get an adrenaline rush within a second. So you can try to measure adrenaline levels in the blood, but there is just no way you will be able to get the blood out fast enough to not have the stress of the needle (damn needle) affecting them. If you had a very controlled population of animals, with catheters already placed that they were used to, so that you could draw out blood without stressing them, that might work, assuming you could catch the animals without stress. (Catch a mouse without stressing it: difficult. Catch a dog without stressing it: actually, when I went into the runs with the hospitalized dogs I was studying, they definitely experienced eustress, or happy stress.)

You can also measure adrenaline levels in pee! This turns out not to be useful, though. Adrenaline levels go up and down, as we’ve said, very quickly, in response to individual stressors. Pee collects all those changes and averages them out over however many hours (say six). So this approach is definitely not good for measuring responses to specific stressors, like a sudden loud noise. It might be better at measuring something longer term (hey, like the response to being in a hospital!) but initial studies haven’t shown it to work very well at that, either. Adrenaline is just the most interesting when you can map it as it goes up and down, not when you have to look at an average and guess about what was smoothed out.

What about the other end of the SAM? When you get an adrenaline rush, you have some physical changes. Among many other things, your heart rate gets faster. Can you measure that? Well, again, good luck measuring that in a dog without having the excitement of interacting with a human confound your measurement! And heart rate is very sensitive to physical changes; you might be measuring whether the dog is standing up versus lying down, rather than its level of distress.

It turns out that what is a better way to measure physiologic changes from SAM activation is heart rate variability. Your heart rate normally speeds up a little when you breathe in, and slows down a little when you breathe out. (I actually did notice this in a dog once, in a lab where I was supposed to be learning how to find abnormal heart rhythms, and I had to call a vet over to ask if it was actually normal, because it sounded so weird once I noticed it.) When you are stressed (physically or psychologically), this variability goes away. This is not a bad way to measure stress, but you can’t measure it with a stethoscope; you have to hook up equipment to the dog in the form of a little vest with a monitor attached. This is expensive (too expensive for you to use, because your project is on a shoestring budget!). You would also have to get the dog used to the vest, so that you were sure you wouldn’t be measuring stress from having clothing on when the dog is used to being naked. It is therefore not a good measurement for hospitalized dogs on their first day in the hospital, but it is a good measurement for some studies. It’s best when used in conjunction with cortisol, so that the two measurements can catch each other’s mistakes.

That uses up the SAM, but there is a system that is the opposite of the SAM. When your body is not in “fight or flight” mode, it is in “rest and digest” mode. This mode is regulated by the parasympathetic branch of the ANS (autonomic nervous system). (The SAM is the sympathetic branch of the ANS.) Can you measure parasympathetic activity? It should increase when stress decreases, and vice versa. It turns out that when your body is thinking it’s time to rest and digest, it releases a digestive protein into your saliva, known as α-amylase. This protein is useful for pre-digesting carbohydrates. More α-amylase suggests less stress. And it’s even in the saliva, so it can be measured non-invasively! You are very excited until you find a paper from the 1950s (I am not kidding) which is the last time anyone bothered to look for α-amylase in dog saliva. Dogs don’t make it. Because they are not meant to eat lots of carbs? Oh wait, this isn’t a post about nutrition.

(For those of you who say “OK, but what about measuring stress via α-amylase in humans?” — I didn’t delve any deeper into this one after I learned it wasn’t useful in dogs. My guess is that it suffers from similar problems to measuring cortisol: it measures more than just [lack of] distress. It also has been less widely used than cortisol, so we understand its pitfalls less. This would be another good measurement to use as a complement to measuring cortisol. If you want to use it in humans, read lots studies that have used it before you commit.)

So much for the ANS. But you know that increases in stress cause decreases in parts of the immune system. In fact, that’s partly why we care about stress in hospitalized dogs — stressed dogs may not heal as quickly or as well. Can we measure the immune system?

We can. Your saliva normally contains a kind of antibody called IgA. This presumably provides a first line of defense against the bugs on your food. When you are stressed, you make less of it. (At a guess, this is because when you’re running from a lion, you’re not likely to be eating. You’re more likely to be getting bitten, so your immune system needs to focus on defenses against open wounds instead of microbes in food.) Salivary IgA is known as “sIgA.” Can you measure that in dogs? You can, and it is being fairly widely used in humans, in fact. Only some initial work has been done on it in dogs, though. It seems to be prey to some of the same issues cortisol is — varying regularly throughout the day, varying irregularly between individuals — so it’s not yet clear if it’s really a better option. It might be a good way to go for a long term project. For something short, though, it might be better to stick with what is well-understood.

Are there any other ways to measure immune system function as it relates to stress? As I said, your immune system reorients when it thinks you’re running from a lion, to protect against open wounds. It does this in part by packing the blood full of a kind of white blood cell called a neutrophil. Neuts are the first line of defense against microbes coming in through open wounds. You can measure their ratio to another kind of white blood cell, a lymphocyte, to measure stress. A greater N : L (neutrophil : lymphocyte) ratio implies greater stress levels. In some ways, this is a really great measure of stress, because it takes a little while — an hour or so — for the N : L ratio to change after a stressor. So when dogs first come in to the hospital, if you can get blood right away, you could actually measure their unstressed baseline. A later blood sample could provide a comparison. Then you could ignore all that annoying individual variability, because you would be measuring the difference pre- and post-stressor in the same individual. I would have loved to have use this measurement.

But, as always, good luck getting an owner to consent to not one but two unnecessary blood draws. I am not sure I would have felt good about adding that much stress to an already stressed dog’s hospital visit, either. For a different kind of study, this might be a really good option, though as always, it measures the effects of multiple systems, so there is going to be some extra variability to account for.

And that is why, though cortisol is a really appalling way to try to measure stress (looking at my salivary cortisol data right now, I keep saying “why does anyone use this hormone?!”), it is still the most widely used approach. As we learn more about how all these systems interact, it is possible that some day we will develop a method of taking multiple kinds of measurements and basically triangulating distress. Or maybe we’ll develop hand-held fMRI scanners and be able to directly measure activation of specific parts of the brain. For now, we are stuck with spit.

Rats, dogs, foxes, and the SHRP

Working on my Master’s degree has made me yen for more letters after my name, so I’ve been doing some spare-time reading on subjects that might yield PhD-type projects. My putative interest is in development of the stress system in young dogs. The idea is that if a dog’s stress system develops poorly, whether through bad genetics or a bad early environment, then that dog is more likely to bite people when it grows up. The more we know about how their stress system develops, the more we can know about how to grow healthy dogs with good bite inhibition.

For several months I thrashed around in the literature, reading about development of the stress system in rodents (about whom we know quite a bit, because we are more willing to do experiments on them than on dogs), and reading about socialization periods in dogs. It was hard to find good direction, and I wasn’t quite sure where to start. Recently I have had a breakthrough, however.

First, some orientation. You are walking through the woods. You see a shape on the ground. Your brain interprets the shape: long, thin. Your amygdala (part of the limbic system of your brain) yells SCARY SHAPE SCARY SHAPE and you get a blast of adrenaline in your system. Half a second later your cortex (the thinking, conscious part of your brain) catches up: hey, that looks like a snake. Your hypothalamus (which deals with a lot of hormone regulation) sends a message to your pituitary (which releases a lot of your hormones), and the pituitary releases a hormone which travels down to your adrenals, near your kidneys. Your adrenals release our old friend cortisol, which gets into your blood and tells your body that you are having a stressful experience. Cortisol, you of course remember, is what I like to extract from the saliva of dogs to tell if they are unhappy about being stuck in a noisy hospital run. This whole system is what I’ve been referring to as the “stress system,” more properly called the HPA (hypothalamic-pituitary-adrenal) axis.

If you were a rat or mouse, instead of releasing cortisol, your adrenals would release corticosterone. It is a very similar hormone with similar effects. Dogs actually release equal parts cortisol and corticosterone, but we just study their cortisol levels. I still haven’t figured out why we chose cortisol to focus on in them; there are a lot of tools available for studying cortisol, since humans make it primarily, but also a lot for studying corticosterone, since we study rodents quite a bit.

Now, to get back to my recent reading, very young animals don’t get as frightened by scary things as slightly more mature animals or adults. This phenomenon has been studied intensively in the rat: rats younger than two weeks of age don’t show this corticosterone spike when exposed to something upsetting. This is called the “stress hyporesponsive period,” or SHRP.[1] There has been work on what part of the HPA system is responsible for this blunted response: the amygdala? The hypothalamus? The pituitary? Or are the adrenals themselves not responsive yet?

A good way to stress out an infant rat is to expose it to the odor of an adult male rat. Left to their own devices, adult males will happily eat infants, so the young rats are quite right to fear them. An infant rat, upon smelling a strange adult male, will become immobile. However, a neonatal rat younger than 14 days (in other words, one still in the SHRP) will not become immobile: it hasn’t yet developed the machinery to feel, or possibly just to express, fear. If you remove the infant’s adrenals, so that it is unable to make corticosterone, then even when it matures to older than 14 days it will still not properly become immobile when exposed to the scary smell. Moreover, if you inject corticosterone into one of these pre-14 day rats, it will be able to develop the immobility behavior at age 14 days, just like a normal rat. [2] This suggests that corticosterone is responsible for the immobility behavior. However, if you remove the adrenals of a rat which has already developed the immobility behavior (one which is older than 14 days), it will continue to become immobile in the presence of the scary smell. [3] And if you inject extra corticosterone into a rat too young to have developed the immobility behavior, it will develop it early. [4] This suggests that corticosterone is responsible just for the development of the behavior, not for allowing it to actually happen at specific times once it has initially appeared.

What’s going on up in the brain while all this is happening? When infant rats are too young to express (or possibly feel) fear, are their amygdalas just failing to activate? When neurons in a particular brain region have been recently active, they contain a protein called c-fos. You can check a brain region for the prescence of extra c-fos to see if it has been doing anything in the recent past. This was done with young rats. Rats too young to have developed the fear response did not have amygdala activity (no extra amygdala c-fos) after exposure to the scary smell; if they were injected with corticosterone to cause them to develop the fear response early, then they did have amygdala activity; rats old enough to have developed the fear response did have amygdala activity; and rats whose adrenals were removed prior to developing the fear response did not have amygdala activity. [4] Unfortunately, this study does not appear to have looked at whether rats which were allowed to normally develop the fear response (intact adrenals), but then had their adrenals removed after initial development of the response, still showed amygdala activation. Perhaps that question has been answered elsewhere.

So what does all this mean for dogs? Do dogs have an SHRP? I found one unreferenced assertion that they do, but I have not yet found a study actually examining the canine SHRP. The SHRP does exist in various species, and it seems likely to me that it exists in the dog. Puppies start out fearless, and develop fear later. I suspect that a canine SHRP will prove to be an important part of socialization: the time that puppies don’t yet feel fear may be an important one for introducing them to lots of different kinds of people, so that they can learn that these people are a normal part of puppy life and are not to be feared later on.

The development of the HPA system has been studied in domesticated silver foxes — foxes selectively bred to not fear humans. (These foxes show surprising physical similarities to other domesticated animals in body shape and color, despite not having been bred for these features, leading to speculation that there is some general mechanism of domestication. That general mechanism of domestication is actually what I’d like to get at in a PhD project.) Researchers took two groups of foxes: domesticated foxes, and foxes bred for increased aggressiveness to humans. They tested them for behavioral reactions to humans and cortisol level increases after exposure to humans, at ages 30 days, 45 days, and 60 days. The aggressive foxes did not show aggressive behavior or cortisol spikes at 30 days, but they did show it at 45 and 60 days. The domesticated foxes, on the other hand, did not show aggressive behavior until 60 days, and their behavior at that time was described more as “defensive” than “aggressive.” They never showed the cortisol spike. [5]

Is this the same thing as a silver fox SHRP? I’m not sure that this study exactly gets at that, but it seems suggestive. Questions I’d like to ask about the SHRP in dogs are: Does the SHRP definitely exist in dogs? Is the SHRP length different in dogs and wolves? Does the length of the SHRP affect the socialization of the dog? Is the SHRP length different in different dog breeds? And, most important but most difficult to get at, does length of SHRP have anything to do with a dog’s fearfulness as an adult?


[1] Walker Claire-Dominique, Perrin Marilyn, Vale Wylie, Rivier Catherine. Ontogeny of the Stress Response in the Rat: Role of the Pituitary and the Hypothalamus. Endocrinology. 1986;118:1445-1451.

[2] Takahashi L. K., Rubin W. W. Corticosteroid induction of threat-induced behavioral inhibition in preweanling rats. Behavioral neuroscience. 1993;107:860-866.

[3] Takahashi L. Organizing action of corticosterone on the development of behavioral inhibition in the preweanling rat. Developmental Brain Research. 1994;81:121-127.

[4] Moriceau S. Corticosterone controls the developmental emergence of fear and amygdala function to predator odors in infant rat pups. International Journal of Developmental Neuroscience. 2004;22:415-422. [Free full text.]

[5] Plyusnina I., Oskina I., Trut L. An analysis of fear and aggression during early development of behaviour in silver foxes. Applied Animal Behaviour Science. 1991;32:253-268.

Epigenetics of stress

Last week in journal club I presented "Prenatal exposure to maternal depression, neonatal methylation of human glucocorticoid receptor gene (NR3C1) and infant cortisol stress responses," by Oberlander et al., published in Epigenetics in 2008. This paper tries to get at one small part of the mechanism for how the in utero experience can affect a fetus, possibly even affecting the baby's personality.

Oberlander's study builds on earlier work done in rats. Researchers found that the offspring of particular rat dams were less fearful than average. Specifically, these rat moms were spending extra time licking and grooming their babies, and performing "arched-back nursing." They were dubbed "LG-ABN" dams (licking grooming, arched-back nursing.) Their babies acted less fearful in stressful situations, and had a blunted stress response on the HPA axis.

The HPA axis... This is in large part what I'm studying for my Masters project, so it's hard to limit myself to a short explanation of what it is. In short, one way in which one's brain (or part of it: the hypothalamus, the H in HPA) responds to stress is to send a message to the pituitary (P), which in turns sends a message to the adrenals (A). The adrenals release cortisol, which is known as the stress hormone. Cortisol is what researchers look for in your blood if they want to quantify how stressed you are. Stress out two animals, check their cortisol levels, conclude that the one with higher levels reacted more strongly to the stimulus — that's the formula for any number of experiments, including this one. (It is, as always with physiology, more complicated than that, but the idea that there's a correlation between increased cortisol levels and increased stress is a good start.)

So these babies of LG-ABN moms had a smaller cortisol spike in response to stress. Further work elucidated part of why that happened: the receptors to which cortisol binds in the brain send a message back to the top of the HPA to tell it to stop releasing cortisol ("that's enough, there's plenty out here!") in a negative feedback loop. These receptors (glucocorticoid receptors, abbreviated "GR") were in excessive supply in the brains of these less fearful rats, so the negative feedback loop worked well, and the rats' brains responded to rising cortisol levels by releasing less cortisol — with the result that the spike of cortisol was smaller in response to a stressful stimulus.

Meanwhile, the researchers also found that this trait of decreased fearfulness was not genetic in the normal sense. If they fostered baby rats from non-LG-ABN dams on LG-ABN dams, these babies who were not genetically related to the LG-ABN mom grew up to be less fearful, presumably simply by being raised by her. And they passed the trait on to their offspring! It turned out that the trait was being passed along epigenetically. We all learned in elementary school about genetic traits — getting brown eyes because you got brown eye genes from mom and dad. And we all know that genes are coded on DNA. Epigenetic changes involve not different genes, but changes to the higher-level structure of the DNA. Instead of involving changing the building blocks of the DNA (the genes), epigenetics involve changing the shape of the building, or sometimes tacking something new on to the outside of it.

In this case, it turned out that in order for GR to be produced (remember that receptor for cortisol, necessary for negative feedback?), the machinery for reading genes had to have free access to the GR gene itself. However, an area of that gene had become methylated — in other words, another object was sitting on it, blocking access. The machinery for reading genes couldn't read that gene as well, so fewer GRs were made. Fewer GRs meant less negative feedback and a more easily stressed baby rat.

That's all background. Oberlander, who wrote the paper I presented, wondered whether the same mechanism applied to humans. He knew that human mothers who are depressed during pregnancy often give birth to babies with more reactive HPA axes. Could that be because those babies had fewer GRs, as a result of methylation of the GR gene? He also wondered about the effects of SRI medication, such as Prozac, on this system.

82 pregnant women were enrolled in this study. 33 were taking SRI medication. All were tested using a scale for depression, which resulted in a numeric score; higher scores implied greater depression. Blood was drawn from the moms in their second and third trimesters, and when they gave birth. Blood was taken from the babies' umbilical cords at birth. Then the babies were tested at three months of age for their response to a mild stressor.

The researchers found that the babies of depressed moms did tend to have increased methylation of the GR gene, exactly in the spot that they expected. That increased methylation correlated with an increased cortisol spike when the babies were mildly stressed. SRI exposure in utero didn't have any effect on the size of the spike, although babies whose moms were medicated did tend to have lower cortisol levels in general.

This paper spoke to me on two levels. I enjoy reading about mechanisms; I like imagining how all these little machines in our bodies interact to form our personality and affect how we experience the world. I also liked the study's methods, because I'm interested in finding ways of learning about living individuals. I want to study dogs, so I want to find ways of looking into their brains figuratively, not literally. Examining changes in DNA extracted from a blood draw is cool — it's something I could potentially do to someone's pet, perhaps as part of a study aimed at understanding why some dogs are more easily stressed to the point of biting than are others.

I think that the people who attended journal club found the paper interesting. Two professors who were in attendance work in this area of genetics and behavior, and had useful input for me. One pointed out that the list of variables that the paper's authors checked for in the pregnant women was very small. (It consisted of things like age, whether this was a first pregnancy, whether the pregnancy ended in C-section, whether the woman smoked or drank.) She listed some other things she would have checked for, such as body weight (fat can apparently produce cortisol). She also noted that the baby's blood sample came from umbilical cord blood, which is actually a mix of infant and maternal blood. Also, different parenting strategies weren't taken into account — did depressed mothers treat their babies differently in some way? She concluded that we'd all like to be able to see useful DNA changes just by taking blood samples (which is precisely one of the things that drew me to this paper), but it's actually very hard to do so, so this paper's results should be taken very cautiously.