A Theoretical Aside
So today’s post was inspired by a tweet I saw routing me to a story in ABC news pointing out a study done in one of the mouse models of Down Syndrome. In this study, the researchers made a single injection into Ts65Dn mice that resulted in increased cerebellar growth and some normalization of the behavioral phenotype, at least pertaining to hippocampus function.
Well I sent out a snarky tweet describing my feelings that the use of the water maze as a primary outcome variable in that study was at best uninterpetable, and at worst, invalid. I couched this assertion as a plea for someone to please hurry up and develop a better battery of tests for the Down Syndrome mouse models since I fail to see the clinical relevance of the ones currently in use (e.g., water maze, fear conditioning).
This post refers to “Hedgehog Agonist Therapy Corrects Structural and Cognitive Deficits in a Down Syndrome Mouse Model” from Roger Reeves and colleagues in Science Translational Medicine. I am not going to refer to their study at all except as an example of the behavioral status quo. More to the point, I am going to cover my own work and my ideas on how we can make research better in mouse models, particularly by making the tasks used in mice parallel those given to human populations Link,Link. This is an ongoing theoretical project of mine, so there will be a more posts on this topic whenever similar problems arise.
The paper from Reeves et al. Focused on hippocampus-based learning and memory and thus used the water maze. This is fine on face, but it has been demonstrated on a very, very large number of occasions that more than the hippocampus is involved for water maze performance. Personally, I find the water maze to be rather limiting in scope, hard if not impossible to clearly interpret, and confounded by anxiety levels-particularly in mice. as such, I tend not to accept water maze data unless they are part of a much wider behavioral assay that tests more specific memory functions (in practice with these criteria, I almost never accept the water maze data as the water maze is usually the sole learning and memory task presented).
So how do we overcome this problem? I like to complain about the water maze but appear to offer no solutions, isn’t that just being contrarian? Yes, yes it is. By that is not my goal, my solutions start now.
As stated in my theoretical review papers linked above, the first step we need to take in studying mouse models is to actually talk to the clinicians that test the population being modeled. In the case of Down Syndrome, it is important to find a research institution or local researcher that can be asked questions face to face or by phone-email is too impersonal for this kind of exchange. I consider this step imperative because, in my experience in the fragile X world, the information that is most useful for the mouse behaviorist is not the data that can be found in published papers. Rather, the information pertaining to the idiosyncrasies of the population, their subclinical deficits, and the general impression of the clinician concerning what should be tested in mice comprise essential guidance to developing a useful and clearly interpretable battery of behavioral tasks.
I think far too often we think of mice as dumb little animals so we underestimate their abilities and thus do not even try to find clinically valid tasks for mice. This is a problematic assumption when trying to model a human population in mice. In fact, such a predisposition toward mouse research essentially guarantees that the tasks selected by the experimenter will not have clear human analogues and worse, that any effects observed are unique to mice and thus clinically useless.
In the case of hippocampus function in Down Syndrome, it has been clearly demonstrated that individuals with Down Syndrome have deficits in paired associate learning on the CANTAB PALs. So, why do we not see paired associate learning test in the Down Syndrome model mice? Object-location paired associate tasks exist in spades for rats and have been modified for mice on numerous occasions. At present, a virtual reality based water maze is used to test spatial navigation individuals with Down Syndrome, but the data are still unclear as to the deficits. I think the big difference between the virtual and real water maze is the fact that the human populations do not have to escape deep water onto a platform, so there may be dramatic effects of anxiety confounding the behavior of the mice compared to the human performance. As such, I propose a dry land water maze as a more optimal task for the Down Syndrome mouse model-a task that has been in use since the late 1980s in rats and early 2000s in mice. Moving beyond the human data, it has been suggested that there is excessive inhibition in the dentate gyrus in Down Syndrome, so why not use tasks that require pattern separation: both in the mouse as well as human Down Syndrome populations?
Similarly, there are a few cerebellar tasks in use for Down Syndrome, mainly focusing on finger usage and sequence learning (sequence of finger taps, etc). Well, in mice we can do fairly well mimicking these tasks using very etymologically valid tests in mice. There is a test wherein the mouse is required to simply eat a small piece of dry capellini or else to shell sunflower or other seeds. Simply looking at a high speed camera recording of the animal’s behavior is sufficient to see if there are abnormal digit use or movements. In fact, simply recording grooming can provide a trove of useful cerebellar function related data. Additionally, one can use a skilled forelimb reaching task of some type to further assay cerebellar function (and cerebellar-cerebral connectivity) in the mouse.
For executive functional deficits in Down Syndrome, researchers use a rule switching task (intra and extra dimensional set shifting to be more precise). These do exist for mice, and at differing levels of difficulty and complexity. In fact, a collaborator of mine has been using a set shifting task with differing reward contingencies (80% reward /20% unrewarded choices instead of 100% rewarded) to phenocopy the data reported for autism spectrum disorders in the BTBR mouse model.
As you can see, it is clear that there are tasks available for use in mouse models that can provide meaningful data to inform research in the human population. In my opinion, it is often the stubbornness of us animal researchers that we do not reach out to the clinic for research ideas and rather just try to creatively interpret the tasks we always use anyway differently for each disease model we test. I do not know if the results of the study that brought us here are valid and will serve as a treatment option in the future, what I do know is that I will have much more confidence in the data if the behavioral tasks are chosen with the clinical population in mind.
TL;DR. Moving forward, we are going to have to do a better job of modeling human behavior in our mouse models. But to do this, we need to swallow our pride and talk to clinicians and take their advice as to what we should test seriously. Doing so will make our results more easily applicable to human disease. It will also allow research on animal models to actually inform, rather than distract from, human disease research.