The “Genetics” of Schizophrenia

 I have several posts on here related to schizophrenia and the claimed genetics of this psychiatric disorder and I thought it would be useful to combine twin studies, genetic studies, and polygenic scores  in one post to give some perspective. Let me start with twin studies:

During medical school and my psychiatry residency training, the genetic nature of schizophrenia was often emphasized. The strongest evidence, we were informed, was that if an identical twin was diagnosed with schizophrenia, there was a 50% chance that the other identical twin would be diagnosed with schizophrenia. While this might leave some wondering why the other 50% do not get schizophrenia with an identical genetic profile, 50% is hard to just ignore.  Well, as it turns out, it isn’t really accurate. Most of the studies that claimed such high concordance rates are from well over a half century ago. At that time, there was far more institutionalization and the diagnostic criteria were not really the same. As twin study critic Jay Joseph points out, if you take more modern studies the concordance rates are far lower, with an overall concordance rate for such studies after 1963 of 23% (some might recall a figure of 28%, but this does not include the Finnish study noted below, which seems to have been “disappeared”). 

Depending on which of these studies you examine, that 23% figure might even be a bit inflated. Take, for example, a couple of the Scandinavian studies and although I may be accused of cherry-picking, these countries tend to have better national records to draw from, which include all twin pairs and all twin pairs that are concordant for schizophrenia, so I think they arguably are more accurate. I have recently posted on these studies. One is a Finnish study from 1984 and one is a more recent Danish study from 2018. I discussed them both at more length in previous blog posts, here and here. These studies both had a full twin registry available to them and were thus able to identify all individuals diagnosed with schizophrenia, as well as any which had a monozygotic or dizygotic twin and whether they were also diagnosed with schizophrenia. The Finnish study found an 11% concordance and the Danish study found a 14.8% concordance if you look at the actual numbers. These are shocking numbers for those of us who were led to believe that it was closer to 50%. 

I hadn’t seen this Danish study until recently, and even if I had, due to what I think is some statistical sophistry, I might not have noticed the poor result, due to these claims in the abstract.

 The probandwise concordance rate of SZ is 33% in monozygotic twins and 7% in dizygotic twins. We estimated the heritability of SZ to be 79%. 

This is rather deceptive, in my view, as I discussed in the blog post linked above. So if you skim this abstract, you see a 33% concordance rate and think it’s close of enough to 50% when that is far from the truth. Moreover, if you aren’t familiar with heritability, the 79% figure might also seem like a “case-closed” number. How do you come up with a heritability figure like that when the concordance was only 14.8%? Well, they are comparing the higher rate of concordance for monozygotic twins vs. dizygotic twins and extrapolating from that difference. So even though the concordance is low, they claim high heritability because of the disparity. 

A couple of important points, here. First, what is the explanation for such a low concordance for a disorder that they claim has a strong genetic component? Why would one twin get schizophrenia and not the other? Since they generally grow up in the same household, and are genetically identical, what environmental factor could explain this? There is no adequate answer, really. If you look at a genetically driven neurological disorder like Huntington’s Chorea, if one identical twin gets this disorder, the chances that the other gets the disorder are 100%. For dizygotic twins, it is 50%. There is nothing really ambiguous about it. So when you talk about Schizophrenia having a genetic component and you can’t explain why one MZ twin would get it and not the other, you are already getting into fuzzy territory. The explanations for this are, in fact, circular, since the assumption is that there is a genetic component.

Secondly, is whether identical twins are more likely to be given the same diagnosis than fraternal twins with the same symptoms. This is no doubt true for the psychiatrists making the diagnosis, since if they know that an identical twin brother is diagnosed with schizophrenia, they are going to be very likely to give the same diagnosis. We are trained to take such a family history and put it into consideration when making a diagnosis. So it becomes circular. It’s genetic, because it’s genetic.  Moreover, is there something inherently different about being an identical twin than being a fraternal twin that might lead to such a diagnostic disparity. For example, in the Finnish study, there was this interesting detail. 

Of the MZ pairs concordant for psychiatric hospitalization, 47% had lived together for their whole life time; of those discordant, 16% lived together. The corresponding figures for DZ pairs were 18% and 15%.

Thus, the concordant Monozygotic twins were three times more likely to have lived their whole life together than the dizygotic twins, even when both were concordant for schizophrenia. This certainly makes it more likely that they would be diagnosed with the same disorder (and perhaps even had the same doctor) than if they had been living apart. In fact, I will extrapolate from this that it is likely monozygotic twins are more likely to live together and share their lives together than dizygotic twins, in general. In short, being a monozygotic twin is a lot different than being a dizygotic twin.

Where I’m going with this is making heritability estimates based on a diagnosis disparity between monozygotic twins and dizygotic twins suggests that their differences in concordance rates is related to genetics, and this is why these low concordance rates are important. A 50% concordance rate is a bit hard to write off, but it is entirely plausible to dismiss a 10 to 15% concordance rate for identical twins when you take into consideration the inherent biases of those diagnosing. 

If you consider for example, that religious affiliation is a very highly heritable trait, despite the fact that this is unrelated to genes, you can see where descriptions of high heritability are subject to many factors that are unrelated to which genetic variants an individual has. Thus, I would argue that twin studies have not convincingly demonstrated that schizophrenia has a significant genetic component. This in fact, might explain the lack of success for genetic studies, which have not confirmed this claimed heritability. In short, perhaps the “missing heritability” is evidence of a lack of genetic heritability. With that, I’ll move on to genetic studies.

The two primary types of genetic studies that have been used in an attempt to identify genetic variants associated with schizophrenia are “candidate gene” studies and “genome-wide association studies” (GWAS). First, I’ll discuss candidate gene studies. Some might think that I am being unfair, since most behavioral geneticists dismiss such studies out of hand these days and they have been widely refuted, as this study notes. However, that’s precisely my point. These studies served their purpose, which was to buffer the idea that schizophrenia is a genetic disorder, and were conveniently discarded when GWAS became popular.

The idea behind candidate gene studies is to pick genes that you think likely might contribute to the trait in question and then assess whether different variants of these genes are smaller or larger contributors. I’ll give a couple of examples here: Catechol-O-methyl transferase (COMT)  involved in the breakdown of dopamine, and the dopamine receptor genes (DRD2 and DRD3). 

If you are not familiar with why there might be a focus on dopamine-related genes, it is because many of the medications used for Schizophrenia, such as Haldol and Thorazine or newer drugs like Olanzapine and Risperidone act in part to block dopamine receptors. So it would seem like good candidate genes would involve dopamine. And when these were looked at, they frequently found correlations between different variants of these genes and Schizophrenia.  These, however, did not hold up (as the link above suggests). Generally, it is clear that candidate gene studies are very prone to false positives. Since behavioral geneticists no longer defend these studies, though, I won’t spend more time refuting them.

I will say this, however. I have been a critic of genetic claims of schizophrenia (and other mental disorders) for 30 years. For the first 10 or 15 of those years, I was dismissed and even had my sanity questioned because of the “overwhelming” evidence for a genetic component for schizophrenia. That “evidence” was candidate gene studies, which are now widely accepted to be false positives and twin studies, which were markedly exaggerated as I note above. I would suggest to anyone reading this, that you consider the fact that the field has done nothing but crank out false positives for decades is a good reason not to embrace whatever the latest study claims to demonstrate. With that, let’s move on to GWAS.

If you are unfamiliar with GWAS, they are genetic studies which identify the genetic variations between individuals and determine whether these variations occur more or less frequently in individuals possessing a particular trait; in this case, schizophrenia. Implied in this type of study, it should be noted, is the idea that the trait is complex, or dependent on many (presumably hundreds or thousands) different genetic variants. This assumption in itself is based on the fact that we are not able to find one or a very few number of genes involved for the trait, as we would find again with something like Huntington’s Chorea. Thus, it is an hypothesis by exclusion. The only evidence for it being complex is the fact that they couldn’t find causal genes.

The fact that GWAS were not likely to find much related to schizophrenia has been understood almost from the outset, as a study from 2003 pointed out, and for which I wrote this letter to the American Journal of Psychiatry. What has happened is that each GWAS finds several genetic variations that occur significantly more frequently for schizophrenia than in the general population. However, which genetic variants are identified are different in each new GWAS. They have, to date, never consistently matched a genetic variant from one study to the next. This indicates that the variants found were either false positive results, or pointed to some skewing of the data being used, such that certain genetic variants are more frequent among those with schizophrenia for reasons related to the group being studied (where some groups are more likely to be given that diagnosis) rather than actual genes related to the disorder. This is a phenomenon called population stratification.

In either case, we have what is called a “replication crisis.” This crisis pervades the field of behavioral genetics and is true of ALL psychiatric disorders. Behavioral geneticists have addressed this in a couple of ways. The first is by promising us that once their studies have large enough numbers of people, they will be able to get consistent results. This has not been the case, however. They now have access to hundreds of thousands of individuals due to large genetic databases like the UK Biobank and 23andMe and yet, while they are able to find more and more genetic variants associated with schizophrenia and other psychiatric disorders, they still do not match independently from one study to the next.

This brings us to the second way in which they address the replication crisis, which is to avoid trying to replicate anything. So, what they now do, is what they refer to as a meta-analysis and simply add new data to old, do a GWAS and find more new genetic variants without addressing whether the variants they found in previous studies independently replicate. Instead, they might try to show that the variants are better than false positives and leave it at that, but that doesn’t eliminate population stratification issues and feels like a dodge. There is no reason not look at the new data independently of the old data and see whether you are finding the same variants.

Case in point is the largest schizophrenia GWAS, the “PGC consortium.” I discuss this study in a bit more length at a blog post here. This study was just expanded in the way I described, with a previous version of the study having 36,000 cases and new cases being added to get it to 69,000. Thus, they do not (to my knowledge) take the new 33,000 cases and do an independent GWAS, which they could then compare to the old one. Instead they just add it on to the old one and this gives them a much larger number of genetic variants “associated” with schizophrenia. Although they attempt assess these variants and determine whether they might be causal in some way for schizophrenia, we don’t have any idea whether any of them are anything more than false positives that will likely never be replicated.  This study also does what is referred to as a polygenic risk score.

As I mentioned previously, these studies are based on the presumption that the trait (schizophrenia) is based on a combination of hundreds or thousands of genes. When genetic variants in a GWAS are found to be significantly associated (or even not so significantly) with the trait, it is assumed that they contribute in some way to the trait. If you take all of the variants you find in the GWAS and calculate how much each contributes to the schizophrenia you can give a percentage of the variance explained. Ideally, if you found all the variants involved in the trait it would explain 100% of the variance. 

These numbers have been notoriously low in GWAS to date, usually only accounting for 2 or 3 percent of the variance and this is considered part of the “missing heritability” problem, since twin studies (taken at face value) would seem to suggest that you should find a large number of genes explaining much of the presumed genetic heritability for schizophrenia and other mental disorders and they have not had much success in that regard. The company line has been that as their sample sizes increase, they will find more and more of the missing heritability.

So if we look at the previous PCG consortium from 2014, with 36,000 cases, they came up with 3.4% of the variance. This is a less than exciting figure. However, when the new cases were added in the most recent GWAS, that figure was 2.6% of the variance. Thus, adding more cases did not improve and actually worsened the situation. These numbers are close to a null finding. One can’t imagine that adding more cases is likely to change the situation in the future, and, if they have a more diverse group (this one is all white European), this tends to water down these calculations further, probably because they relied on population stratification.

The other purpose of the polygenic risk score is to predict the likelihood, based on someone’s genetic code, that they will inherit the trait in question. In other words, can we use the information to predict whether someone will become schizophrenic? I’ve commented previously on an attempt to test such predictions, I believe generating the score from the previous PGC consortium data in a sample of individuals from the Netherlands. This gave a dismal performance, with a 0.5% prediction success. To put that in perspective, if I was told that the person was a young male, and based on that said that he was schizophrenic, that would be twice as good a predictor. If I was told a person had gone to the doctor at some point in their life complaining of pain, and diagnosed them with schizophrenia based on that fact alone, that would be 4 times more accurate than the polygenic score. Clearly, this is not clinically useful and the updated PGC Consortium is no better as they acknowledge:

 ,,,the liability captured by PRS is insufficient for predicting diagnosis in the general population

So to conclude:

Twin studies of schizophrenia have been grossly exaggerated and arguably even their lower than billed results have more to do with differences between identical vs. fraternal twins, rather than actual genetics and the “missing heritability” may not even exist.

Genetic studies have consistently failed to identify genes for schizophrenia and do not independently replicate.

Polygenic scores appear to have peaked at a level far below clinical utility and shouldn’t be expected to improve with the addition of more cases.

After three decades of genetic studies, there is no real conclusive evidence to date that schizophrenia is a genetic disorder, in whole or in part.

Addendum (11/2023): I have had a few people make the case that individuals with certain genetic disorders, such as 22q11.2 Deletion Syndrome and Fragile X syndrome, have a much higher frequency of being diagnosed with schizophrenia, and that this suggests a genetic component for the disorder.  On the surface, this is true. However, this requires some understanding of how clinical psychiatry works. Most of the individuals with the disorders above have mental disabilities and behavioral problems. In the old days, we would give a diagnosis of mental retardation. However, you can’t simply diagnose someone with that and put them on medications. At an early age, they are essentially trained to describe their impulse outbursts as being in response to “voices,” and their temper tantrums are couched as severe “mood swings.” They might also be described as paranoid, believing people are talking about them or making fun of them. Many of these children likely grew up very much being made fun of or mocked. To give them medications, you then need to have a diagnosis. Hearing voices and paranoia can then be diagnosed as schizophrenia, justifying the use of sedating antipsychotic medications. Severe mood swings can give you a diagnosis of “bipolar disorder,” and then put on mood stabilizers like lithium and depakote. The underlying purpose is behavioral control. In fact, they are most often given the diagnosis of ‘Schizoaffective Disorder,” which is effectively a combination of schizophrenia and bipolar disorder. 

All of this is a far cry from someone who has classic schizophrenia, hears actual auditory hallucinations, specific paranoid delusions (CIA, Masons, etc.) and are grossly disorganized. It is also a far cry from classic bipolar manic episodes, where an individual goes days with no sleep, believing things like that they are millionaires, secretly married to a celebrity, etc. In short, people with mental disabilities do not have schizophrenia (or bipolar disorder) in the classic sense. There is no reason to believe that just because they get that diagnosis for practical (and unfortunate) reasons, and have a genetic disorder, that someone diagnosed with classic schizophrenia has it due to genetics, despite not having mental disabilities or other concomitant problems (seizures, heart problems, etc.). It is not that someone with a genetic disorder like 22q11.2 Deletion syndrome has schizophrenia in addition to a mental disability. The mental disability is exactly the reason they get the diagnosis.