How do geneticists determine if a gene mutation is pathogenic?

Question

I am analysing information about patients with neurodevelopmental disorders using the DECIPHER genomics database. I am looking for patients who have only a specific gene deleted and no other mutations and have intellectual disability.

One thing that I have noticed is that some these patients have deletions of just one gene and have intellectual disability. However the pathogenicity of the gene deletion is reported as benign.

Why is this the case? Since the patient has deletion of only one gene and no other mutations, it's most likely that the gene deletion is responsible for the patient's intellectual disability.

In general, I was wondering how do geneticists who study patients determine whether a given gene mutation is pathogenic or benign? Are there any established standards/conventions for determining this?

Any insights are appreciated.

Answer 1

The presence of a single variant relative to reference in a single patient is usually not considered adequate evidence to say that the variant causes a patient phenotype. It might be considered an interesting anecdotal observation, but there is plenty of room for error in genotyping, phenotyping, or the possibility that the disorder is not genetic.

Another important note is that you are looking at a specific database (DECIPHER) which has their own bespoke workflow for inferring pathogenicity. Their practice relies upon and integrates information from other data sources, but it may not be identical to any other process for determining variant pathogenicity or disease causality.

For this variant specifically in DECIPHER, it is marked as "Benign" not because of this patient, but because of prior evidence that variants in the gene don't have strong phenotypes. You can click the "?" symbol and it gives you some info:

So you can see that the benign classification is the result of a statistical model looking across patients. (Though as you can also see in the blue box, it does incorporate information from this patient; for more information about this data's lifecycle, you can see the doc here). I should be clear that I don't see documentation of what models exactly DECIPHER is using, or how they are applied, though I haven't searched exhaustively.

You can further click through the OMIM link for the gene in question, which gives more info about what is known about the gene- in this case, that it doesn't seem to have strong effects.

Also, this variant is heterozygous, which is to say that the patient has a functional copy of the gene. There is not evidence for haploinsufficiency of this gene (e.g. that one copy knocked out is bad enough to cause a phenotype). You can see this under the "ClinGen DS" entry for the gene.

If many more patients with this deletion and also similar phenotypes were observed, and there were no control individuals with this deletion, then that might change the weight of evidence towards it being pathogenic. But at present there simply isn't enough evidence to indicate that this variant is pathogenic.

Alternately, if there were evidence that the disorder were genetic from a pedigree of a family, and that this variant were co-inherited with the disorder, that would also be strong evidence.

Yet another form of evidence would be if someone made the gene deletion in e.g. a mouse or a human cell line, and showed that there were a related phenotype.

CAVEAT

Having written all that, I have to say that there are probably situations in which most people would probably accept a one-patient anecdote variant as causal. Examples might be large previously unobserved chromosomal abnormalities coupled with strong phenotypes. But that would be a gut-feeling kind of decision, not really based on rigorous scientific reasoning.

Answer 2

This is a bit of an off-the cuff/high-level answer, but I will leave you with a few things to think about.

1. The human genome, while one of the highest-quality and most thoroughly analyzed ones we have, is still "poorly" understood and "under construction".

There are large stretches of genomic material called scaffolds that we know exist but don't know where to place in the rest of the genome. Regions around chromosome centromeres are also poorly understood as centromeres are notoriously difficult to sequence though. There are also alternate loci and other genomic phenomenon that are less understood and often overlooked when people analyze genetic data.

Why does this matter? Just because you have sequenced someone's genome doesn't mean we have a complete set of information about it. Maybe they have a SNP on a scaffold, maybe they have a SNP or indel in an unannotated cis- or trans-regulatory element, ect. You are only looking at variation in "known" things but there are also a lot of known "unknowns" that you are not going to have the means to quantify.

2. The quality and depth of the sequencing will greatly impact your ability to call SNPs and indels (read: genetic variation).

Often in clinical settings and wide-scale studies people go for quantity over quality since the goal is often to gain statistical power through aggregating at the population level. If your sample falls into this category you likely don't have either the depth or read length (ie sequencing quality) to call some of the more subtle genetic variation that might be present.

3. Everyone has genetic polymorphisms (SNPs, indels, ect.)

Estimates vary on exactly how many the average person has, but no one is ever truly genetically "normal". The real question is, "are any of that person's variants detrimental?" and that is often a very hard question to answer. Even if polymorphisms in a gene are usually benign, you can never completely rule out the possibility that the particular polymorphism you are seeing just happened to hit a sweet spot that perturbs the gene's function enough to manifest in disease.

So it is almost certain your given patient has genetic variation in locations other than just the one gene you mentioned. It is possible they are in unknown/unannotated regions (1) or not detectable given your sequence quality (2), but they almost certainly exist.

4. Phenotype = Genetics + Environment

With any phenotype there is always the possibility that some combination of genetics and environment play a role in its manifestation. Maybe its purely genetic, maybe its purely environmental, often its a mix of the two. You always need to keep these two factors in mind when studying any disease.

Putting it all together

If after sequencing you only detect a significant polymorphism in one gene, why can't you conclude that must be the root cause? The answer is a combination of 1-3 as well as a litany of other reasons. Essentially genetic sequencing (in terms of genome-wide or other forms of high-throughput sequencing) has a lot more slop and squishyness to it than most people make it out to have.

You should never believe you have seen all there is to see, and often should be fairly critical of even what you do see.

Another point I want to make relates to point 4. The intellectual disability you see in your patient may be entirely due to environmental causes. For example, in Autism Spectrum Disorder there is a theory that severe viral infection of the mother during pregnancy may cause the maternal immune system to attack the developing fetal brain, leading to developmental deficits and eventually ASD. It doesn't matter if this theory is actually accurate (although there is a good amount of data that seems to suggest this), but it does suggest one of many ways that a recurring phenotype in the population may be caused entirely by environmental rather than genetic factors. This possibility is always important to keep in mind.

So how do we actually figure out if genetic variation is causing disease? The very short answer is through many complimentary methods and that it's still quite complicated.

The slightly longer answer is that we usually start with correlational data over a large population. We gather a bunch of subjects, measure some trait (in this case presence or absence of disease), sequence their genomes, and see if we can find any recurring associations between genetic polymorphisms and the trait in question. One form of these types of studies is called a Genome Wide Association Study (GWAS) which you might want to read up on.

Another method is by looking at conserved sequences. By looking at similar genes across species, or identical genes across many samples of the same species, we can identify areas of genes that have identical or near-identical sequence (ie that are well conserved). The theory goes that these sequences tend to get conserved because they are vital for a given gene's proper functioning, whereas less conserved sequences may be able to tolerate some change without causing significant issues. So, if a given patient has a significant polymorphism in a highly conserved region, that's usually good evidence something might be going on and often warrants further investigation.

These types of approaches are just the start, and often lead to a list of "candidate genes" for a given disease/phenotype. At this point biochemical or other forms of experimental methods take over to physically investigate the functions of these genes. Often this involves manipulating the gene in some sort of model system (cell lines, model organisms, ect.) with the end goal usually being to a) determine the gene's functional role in disease and b)the mechanism by which the gene carries out said function.

Often the overall path looks something like this: association with gene -> deleterious variants within gene identified -> experimental modeling of variants-> biochemical investigation of models -> mechanism. The standard can vary by field, but almost no one will say a given mutation/gene is "pathogenic" without at least some experimental evidence to back that claim up.

So, as a final thought, why is this still so complicated/why doesn't this approach work for all diseases? Rather than answer directly, I'll leave you with a (hopefully) thought-provoking example.

Let's go back to my ASD example and point 4 (P=G+E). Let's say that the theory is true and severe viral infection during pregnancy causes a significant proportion of ASD cases. If you were to do a GWAS study on ASD, you would likely turn up polymorphisms in immune system related genes (if we assume that genetic variation in certain genes could make you more susceptible to viral infection). Now those genes aren't really the "cause" of the disease, they might just make you more likely to get infected (or more likely to become severely infected, ect.), but the viral infection itself is probably what most people would think of as the "cause". Even more subtle still, while it is these genes in the mother that confer vulnerability to disease, they may still show up as a signal in the affected children since a decent proportion of them will inherit these genes from their mother. So while what might show up from your study is variation in immune-related genes in the affected individuals, this actually tells you very little about what actually causes the disease. This is just one of many ways that trying to nail down a disease using genetic sequencing alone can be significantly challenging.