There have been a number of press releases and feel-good articles circulating among my 22q13 deletion syndrome Facebook friends celebrating the advancement of mouse models of “Phelan McDermid syndrome”. I am all for enthusiasm! However, cheering is more fun if you know how the game is played and what to cheer for. With that in mind, I would like to look carefully at models. What is a good model and what might it tell us?
I started out in engineering, where modeling is very important. Let’s imagine modeling an airplane. The first issue of modeling is, what aspects of the plane are motivating us to build a model? In this case we might want to learn about how the airplane will perform if the tail is damaged in flight. We can construct a scaled-down miniature with wings, tail, etc. and omit the inside furnishing (cockpit, storage compartments, etc.). If the miniature has the exact same shape and moving wing/tail parts as the real airplane, we would say our model has construct validity. That is, it is constructed in a way that reflects the original plane. The next step is to put our model into a wind tunnel and see if it flies. The model is held on a wire. We can adjust the wing flaps and other control surfaces. If the model tries to rise at the same wind speed as the real plane, and tries to bank left or right with the same amount of wing flap adjustment as the real plane, we can say this model has face validity. That is, with regard to what we are testing, the model behaves much like the real plane. We are finally ready to benefit from all our work. Let’s take the model for a test flight and then poke a hole in the tail. What settings of the flaps and other control surfaces allow us to keep control of the plane? We may struggle with this for weeks or months hoping we can learn to control our crippled model plane. If we find a solution, maybe we have found a way for pilots to rescue their plane in the event of a similar emergency. If this scheme works, our model has predictive validity. Thus, we measure a model’s worth by:
- does it model what we want?
- is it constructed in a way that tests what is important to us? (construct validity)
- does it perform in a way that mimics what we already know? (face validity)
- will model manipulations tell us how the real thing will respond? (predictive validity)
I hope you are getting a picture of what an animal model should do. Let’s look at a mouse model of “Phelan McDermid syndrome” or “PMS”. I use quotes because different scientists have different definitions of “PMS”. See Introduction. For this blog, I will omit the quotes, but remember that there are numerous definitions floating around.
The definition of PMS is important for modeling. The definition of PMS tells us what people claim to be modeling. Some scientists define PMS as a deletion or mutation on chromosome 22 that involves SHANK3. That is fine with me, but that omits the rest of 22q13 deletion syndrome, since there are numerous cases of “interstitial deletions” that don’t affect SHANK3. So, SHANK3 might not be a good choice of model for many families. It depends on the deletion and it depends on what is causing your child the greatest difficulties. There are currently mouse models for 11 different PMS genes for deletions of 1 Mbase or larger. Every gene on this list is relevant to 95% of patients (See Understanding deletion size.) The 11 genes are BRD1, CHKB, CPT1B, MAPK8IP2, MAPK11, MAPK12, NCAPH2, PANX2, PIM3, SHANK3, TYMP. Most of these mouse models are very well studied. If you go further up the chromosome you find other well studied genes with mouse models, like ATXN10. So, the choice of gene is all about what aspect of a disease or syndrome you wish to study.
SHANK3 is popular not because of PMS. It is popular because is has been associated with autism. There are under 1,000 identified people in the USA with PMS, but there are an estimated 36,500 children born each year in the USA with autism. Parents of 22q13 deletion syndrome children should appreciate that researchers who study Shank3 mice are tapping into the national (and world) autism crisis. Our children are a convenient source of subjects, which is why the big national study officially excludes PMS families with interstitial deletions that do not affect SHANK3. Children with SHANK3 mutations are of greatest interest even though they technically do not have 22q13 deletion syndrome (that is why the name PMS was created). Note that only 1/3 of children with PMS have autism, so patients with SHANK3 mutation and autism are the most valuable research subjects.
Although SHANK3 is of great current interest, autism is caused by hundreds of genes. Most parents don’t realize that many 22q13 genes are autism-related or suspected to contribute to autism. Some of the autism genes on chromosome 22 are BRD1, CELSR1, CHKB, MAPK12, PANX2, BRD1. Further up the chromosome (associated with somewhat larger deletions) you can find CELSR1, WNT7B, TCF20, EP300 and others.
Now we understand that Shank3 mice need to be models of something. From the above lists of genes, it is pretty clear that a mouse missing only one gene is not a universal models of either 22q13 deletion syndrome or autism. Both conditions involve a large number of genes. The Shank3 mice are single gene models. The mice are fashioned after super-rare cases of people with specific SHANK3 mutations. Let’s see how these models stack up in terms of construct, face and predictive validities.
All of the published papers so far describe Shank3 mutations and microdeletions, not deletion. I am not going to cite the specific papers here. I have written brief reviews on most of these papers. Contact me if you would like specifics. Some of the mouse Shank3 models use deletions that reflect mutations found in actual patients. For those specific patients (often just one or two), the genetic manipulations have construct validity. That is, the mouse gene has been changed in a way very similar to the human gene. As for the rest of us PMS parents, 95% of our children are missing the SHANK3 gene altogether, along with 30 to 100 more genes. So, these mice do not have very strong construct validity for our children. It is well known that mutations can be very different from deletions. (see Gene deletion versus mutation.) Of course, mutating a single gene may not be very helpful to understanding your child when so many other important genes may be involved. (see How do we know which genes are important?) There is also the problem of using a mouse to model a human. The gene is mutated in an animal that lacks brain areas that are crucial to human behavior, like the granular prefrontal cortex. If these brain areas are important to autism, then the construct validity is weaker.
To be fair, a lot of work goes into to creating a knockout mouse. I don’t have first-hand experience, but I work down the hall from a colleague who is an expert. He works hard and I can read the frustration in his brow on tough days. Just making a mouse is not good enough, you have to prove you have modified the right gene in the right place without messing up the rest of the genome. Then you cross-breed, back breed and then do more validations. My hat is off to those people who make a living this way. When done, the mouse has construct validity in that the targeted gene has been modified.
Face validity is a huge problem with mouse models. Our model airplane ascends and descends, banks and aerodynamically behaves much like its real-life big brother. We know the flight behaviors that are important and we can directly (although not perfectly) compare our model plane to the real thing. How do we compare our model disease to the real disease? Generally, the first step is to compare normal mice to our genetically modified mouse. We note what is different between the two. Then, we compare normal human subjects (“developmentally typical”) to those with the syndrome or disorder. The question then becomes, do the mouse differences seem to reflect the human differences? Mouse models of kidney function and cancer have been very successful with face validity. Urine output and tumor size are easier to measure than social behavior and eye contact in mice. Biopsies of human kidney and many tumors are also much easier than brain biopsies. It is no surprise, then, that mouse models of neuropsychiatric disorders are hard to validate.
The differences between normal and mutated mice can be observed in brain structure, chemical signatures, cellular changes and gene expression. For the most part, there is very little human data for comparison. There are studies with human post mortem tissue that can be helpful, but most of that tissue is from normal human brain. As you might guess, screening for gene expression and other changes in a donated brain from a PMS patient will have issues. Most patients have too many genes involved. One helpful approach is induced pluripotential stem cells (iPScs), grown from skin samples or hair follicles. One might even be fortunate enough to find a donor who’s genetic mutation matches the mouse (or vise versa). However, iPScs do not undergo normal development, so developmental studies are impossible at the moment. They also don’t produce circuits like an intact brain. These are all limitations that impact face validity.
The one common feature you can examine between mice and humans is behavior. There are two extreme viewpoints in this regard. At one extreme, some scientists argue that any behavioral difference between the normal mouse and the genetically modified mouse is “autism-associated” because of the construct validity. At the other extreme, a few scientists argue that the mouse behavioral difference should reflect the standard manual that defines neuropsychiatric disease (DSM-5). The remaining scientists are in the middle until it comes to publishing a paper. Then, they make the argument that their animals’ behavior somehow reflects the behavior of a person with autism. It is my observation that the more popular or prestigious the journal, the more their mouse sounds like a miniature human.
A major problem with Shank3 mice is that, almost universally, they don’t have a significant aberrant behavior. How can you study behavior when it seems normal? There are two solutions, but both have problems. With one approach, researchers find a mutation spot on the gene that is extra good at producing a pathological behavior (e.g., ignoring other mice, not learning, rubbing its fur until the skin is damaged). Another approach is to mutate both copies of the gene. Each of these solutions creates a new problem. In the first case, we are no longer modeling PMS or autism, because this behavior only shows up in very specific mutations. In the second case, we have lost construct validity. We were never trying to model mutation or loss of both copies of SHANK3. That would likely be a different syndrome (just as mutation is different from deletion).
In the end, current models of SHANK3 mutation (autism and SHANK3 mutation aspect of PMS) provide information about molecules and neurons, but are very limited models of the human diseases. They don’t necessarily say much about 22q13 deletion syndrome in general. It is interesting to note that mouse models of other genes of 22q13 deletion syndrome may be more promising. Several mouse models have clear consequences with only one gene mutated or deleted. For example, the Cpt1b mouse model is used to study cold tolerance and insulin sensitivity, and the autism and schizophrenia-associated gene, BRD1, is under intense research right now by a group in Denmark. In both cases the behavioral effects are clear when only one copy of the gene is affected.
We are not done yet, because the most important and most difficult type of validity is predictive validity. That is, does your mouse model tell you what will happen in your human subjects? Given all the problems with making a mouse model of a neuropsychiatric disorder, it is no surprise that the models have not lead to translational (treatment) gains. (See Hope for autism treatment dims as more drug trials fail).
I have heard many people in the research field try to sound hopeful to parents. I believe that hope, when anchored in reality, is very helpful and important. But, too many people have alternative reasons for fanning the flames of hope by knowingly or unknowingly misrepresenting the size of strides in research. My recent blog (How can the same deletion have such different consequences?) touches on why so many parents are mislead by researchers, writers and support organizations. The subsequent blog (22q13 deletion syndrome and science leadership) explains that it does not have to be this way if we bite the bullet and hire qualified leaders.
As a parent, my heartstrings feel the tug of hope every day. The urge to follow those feelings is managed by having a deeper understanding of the research world, and remembering that too many people are eager to give those strings an extra tug.
arm22q13
Previous posts:
How can the same deletion have such different consequences?
22q13 and the hope of precision medicine
22q13 Deletion Syndrome: hypotonia
Understanding gene size
Gene deletions versus mutations: sometimes missing a gene is better.
Understanding translocations in 22q13 deletion syndrome: genetics and evolution
Understanding deletion size
22q13 deletion syndrome – an introduction
arm22q13 