Kimberly Gilmour is Director of Laboratory Medicine at Great Ormond Street Hospital for Children. She utilises next generation sequencing and whole exome and genome sequencing to diagnose patients with primary immunodeficiency. Kimberly discusses the challenges of validating variants of uncertain significance, the importance of publishing benign and null results and the joy of solving surprising cases.
Please note the transcript has been edited for brevity and clarity.
FLG: Hello everyone. Welcome to the latest ‘A Spotlight On’ interview. Today, I’m joined by Kimberly Gilmour, and we’re going to be talking about variants of uncertain significance. Kimberly, if you could please introduce yourself and tell everyone a little about what you do.
Kimberly Gilmour: I’m Kimberly Gilmour and I’m Director of Laboratory Medicine at Great Ormond Street Hospital. My personal focus, and focus of our laboratory, is investigating patients with genetic defects to their immune system. One of the things we found recently was that a lot of these patients have variants of unknown significance. Those variants of unknown significance are a change in someone’s DNA, and we don’t understand if they’re pathological and cause disease, or if they’re benign, and don’t cause disease. My whole career led up to setting up tests to functionally validate these and identify if they are benign or pathogenic.
I did a PhD at the State University of Stony Brook in Long Island. After I finished my PhD, I moved over to what was then the Imperial Cancer Research Fund, now known as Cancer Research UK, where I did a research fellowship. From there, I was hired by Great Ormond Street to take my PhD work and turn it into clinical diagnostics. For the last 22 years, I’ve been doing that; I was hired to set up what was initially diagnosing children with primary immune deficiencies. The way we used to do it before next generation sequencing was by screening for protein expression or function, and if it was missing, we would know to sequence that single gene. We didn’t have the ability to sequence the whole genome to look at lots of genes, so we tried to do very targeted genetics. As things moved along, we had the advent of next generation sequencing and whole exome and genome sequencing. We then almost flipped the paradigm – we would find lots of mutations or variants, not know if they were pathological or not, and then develop the functional tests to see if they were disease causing or not.
FLG: Are the definitions of variants of uncertain significance consistent? What is the prevalence of these variants?
Kimberly Gilmour: In both the United States and Europe, there is a classification of DNA changes. They go from ‘benign’, and we know they’re benign, because 20% of the population has them and obviously not 20% of the population has these very rare diseases. They are present in the population at a lower level, but still common enough to be unlikely to be disease causing. ’Variants of unknown significance’ means we don’t know if it causes disease or not. The end of the scale is ‘likely pathogenic’. They’re rare variants, they’re associated with disease, they prevent protein function, they’re likely to be causing disease, once they’ve been proven in multiple families, from multiple ethnicities or different backgrounds. And they cause disease. That is the range of changes.
Unfortunately, a lot of changes end up in that middle batch. It’s very specific on different genes, because some genes we have a lot of information on, they’ve been sequenced many, many times, we’ve seen them in lots of populations, and they’re very well classified. So, you don’t get very many variants in those particular disease genes.
For example, cystic fibrosis, we get fewer variants, because it’s a disease we’ve known for a long time. We’ve been sequencing the CF gene since the 1990s, we initially knew about hotspots. Now we have sequenced the whole thing and we have some good functional tests to go with it. The more you know about a gene, the fewer variants of unknown significance you get, because every time you get a change, you classify as disease causing or benign. You can only have so many changes, the genes are only so big. With rare diseases, you get many more variants, because we know much less about the disease. So, it is somewhat gene type specific in terms of variants. If you look at a human being and you have sequenced their whole genome, you can get as many as 30,000 changes. But, most them are going to be benign and they’re not going to cause disease. It’s filtering down to the ones that are likely, or potentially, pathogenic. A lot of times we only look for variants in the genes of interest. We don’t look for variants in genes influencing hair or eye colour, because we don’t care about those.
FLG: You have talked about our current ways of functionally validating these variants, from protein expression and protein function, what are some of the challenges of our current methods?
Kimberly Gilmour: Initially, almost all variants run through what’s called in silico, or computer-based analysis. That gives you predictions if they’re likely pathogenic or not. We generally think that a stop mutation, where the protein is not made to its full length, is likely to be pathogenic. For example, if there is a big deletion where a bunch of the protein is chopped out, that is likely to be pathogenic. If it’s an amino acid change, but it’s conserved from zebrafish to humans, and everything from zebrafish fish to humans has that same amino acid at that position, it’s likely to be pathogenic. If it’s very rare in the population, it’s likely to be pathogenic because, we’re not all wandering around with these rare diseases. All of those are big hints that they’re pathogenic, but it doesn’t tell you for sure. You have to do something else.
You can look at RNA, because if you don’t make RNA, you’re not going to make protein, and therefore, it’s likely pathogenic. That’s very useful for splicing defects or defects that might affect transcription of a gene and then the protein. You can then look for protein expression by immunoblot, flow cytometry, or by microscopy. If the protein is not there, then you have a very big hint that it’s pathogenic. But a lot of proteins are made but non-functional. You have a change in a single amino acid, it’s essential for a signal transduction pathway, or sending messages between cells, but the rest of the protein is there, it just can’t work correctly. In that case, it’s very difficult because you have to do a functional assay.
To do a functional assay, the biggest challenge is having the material to do it. You can only look at where that protein is expressed or where that message is made. I’m an immunologist, it’s relatively easy for me, I’m interested in white blood cells. Assuming someone hasn’t had a bone marrow transplant, we can get a sample and do that. But if you’re a neuroscientist, and you’re interested in brain signalling, that’s very hard. You’re not going to be able to take a little piece of someone’s brain to see if it’s a functional mutation or not. You’re much more dependent on animal, cell culture or tissue culture models to try and do it.
More and more, we’re using induced pluripotent stem cells. So, you can take a biopsy, make fibroblasts, turn them back into stem cells, and then re-differentiate them into what you want. But that’s complex and not very efficient. We’re now doing whole genome sequencing at scale within the NHS in the UK. We have lots of patients going through this. Those are really interesting for new genes or one-off assays, but they’re not useful when you have lots of patients coming through. If we can take tests we’re already running in clinical labs, or develop tests that can be run at scale and clinical labs, that’s much more useful for the patients.
FLG: What are the benefits and limitations of whole genome sequencing?
Kimberly Gilmour: The benefits and limitations are almost the same thing. The benefit is you get so much data. If you do whole genome sequencing, not only are you looking at the exons, the coding pieces of the gene, but you’re looking at all the other areas that help control gene expression. You’re looking at promoters, enhancers and splicing machinery. You have a much greater rate of picking up disease causing mutations, which is great for patients who may have been undiagnosed for many years, but you also pick up many more variants. These are even less likely for the computer programme to be able to model, because we’re much better at modelling amino acid changes than we are for something sitting at a promoter site. The functional validation becomes much more important – you don’t get those clues from the in silico. The benefits… We find many more mutations and we help diagnose patients who haven’t been diagnosed before, but these are also the drawbacks, because we get these extra variants to functionally validate.
FLG: What data do we need in order to identify variants more effectively?
Kimberly Gilmour: A lot of our algorithms identifying the variants are based on the fact that you have a child who might be very unwell. Parents who have the child, you assume that they’re fairly well, so you’re looking for things that the parents are heterozygous for, and the patients are homozygous for; they’ve inherited a bad copy of the gene from each parent. Or, things that are de novo, where neither parent has it, but the child does. You still miss a lot of mutations, such as things that might be partially penetrant. The parent may have it but doesn’t get really sick with that change, and the child does get really sick with it. Having trios or family groups is really helpful, but doesn’t necessarily solve the problem.
I think the bigger thing that we’re trying to work with our international colleagues for is we need much better data, data storage and data banks. If a lab in London says, “This is definitely pathological, here’s our evidence”, then a lab in France doesn’t have to redo that assay. You have that evidence, or you know exactly what assay was done and you could repeat that one assay rather than trying different things.
There are really good ways of recording your variants but, there are not really good ways of recording a functional data that may have been done. I think we’re really good about publishing and sharing positive findings where something is definitely disease-causing, no proteins are made, therefore you do a little case study, you write it up and you do something with it. When it comes to a completely normal protein, with a completely normal function, people tend not to share that information as much. But that’s equally important, because you don’t want to waste someone’s time trying to prove it’s pathological if it’s completely benign, and the cells do everything they’re supposed to do. We’re less good at sharing that, because that doesn’t get published and you’re less likely to get grant money for that.
FLG: Publish or perish strikes again! Earlier this year, we saw a biotech company announce their $100 genome, what technological advances have either already happened, or you think are around the corner that could advance your work?
Kimberly Gilmour: I think we’re really good at sequencing and sequencing has become cheaper and cheaper. You can get a sequence for $100. It’s the analysis and interpretation of that data that’s needed. I think as artificial intelligence gets better at it, and we get better at sharing that data, we’ll have a bigger data pool or databanks to draw from. I think it’s going to be the artificial intelligence analysing those genomes to pull out what’s likely to be pathogenic. But it does depend on the algorithms we use.
We had a case recently where we found one disease in a patient, but actually, it looked like something else was going on as well. But the algorithms used assumed it was autosomal recessive, you have a bad copy from mum, a bad copy from dad, baby gets two bad copies. That’s why they were sick and that identified the first disease in the baby. But, we missed the second disease. We went back and did some better family history, and we had excluded some genes because Mum was homozygous for it, when the disease was involved in height and mum was very tiny. Mum was affected. But the algorithms being used by the computers excluded that gene, because Mum was homozygous, so she couldn’t have it, even though she did have it. I think the problem with artificial intelligence is how those algorithms are written, and to understand how they’re written.
FLG: Empowering the patient often requires more involvement from them as well. Obviously, this is quite intense, family-focused patient-led medicine. What challenges do you see, as we shift our systems to patients with these complex diagnosis… What can we learn from the way that you are diagnosing and treating?
Kimberly Gilmour: I think it’s really important to have patient and parent involvement. Because they are the ones giving the samples, it’s their material, it’s their DNA, they’re getting benefit potentially from those diagnoses. But this diagnosis can be life changing, and some people don’t necessarily want to know that information. I think it’s really important to be really clear with families and patients, what you are and are not looking for. Are you going to do whole genome sequencing, but are you only going to look at a panel of genes? Are you going to report back on variants or changes in genes that aren’t related to the disease you’re looking for? What if you sequence for immunodeficiency, but find a BRACA1 mutation? I think there are a lot of ethics around that.
I think families really have to understand what you are looking for and what the limitations of the data are. We’ve had several families that have consented to re-analysis of that data, and things that we ran previously in 2012 or 2015, we didn’t know as much then, so we didn’t look for the things we’re looking for now. Re-analysing that data has led to diagnoses in some of these patients. As we move to gene therapy, this is really important because it’s essential that you know that mutation is disease causing so you put back the right gene. At Great Ormond Street, we’ve treated a lot of our patients with gene therapy already.
FLG: Great Ormond Street has been a huge chunk of your professional and personal history as well. Can you tell us about the history of the hospital and their cell and gene therapy history?
Kimberly Gilmour: Great Ormond Street is the oldest children’s only hospital in the United Kingdom. It was founded around 1852. It has always been a charitable children-only hospital. It started very early on doing bone marrow transplants, not only for haematology in paediatric leukaemias and things like that, but for these rare immunodeficiencies and has had a real interest in immunodeficiency along with our collaborators at UCL in child health.
Together, a gene therapy programme was developed, led primarily by Professor Adrian Thrasher, who’s still at Institute of Child Health (ICH). Our first patient was treated for X-SCID gene therapy at Great Ormond Street 22 years ago. He has done really well and is an adult. From that very rare disease, with severe combined immunodeficiency, often known as ‘the boys in the bubble’, children who have to be isolated until they either have a bone marrow transplant or gene therapy, that programme has expanded significantly. We treat a number of other primary immune deficiencies, and we’re beginning to treat some metabolic diseases as well.
That’s all personalised medicine. You identify the defective gene, we replace that defective gene. It’s all somatic gene therapy, we’re only replacing genes in blood cells or in tissue cells, not sperm and egg. The kids that have had it, have done really well. That’s beginning to expand into haematology, we’re beginning to treat things like beta thalassemia and other diseases as well.
FLG: Can you run us through a really impactful case study?
Kimberly Gilmour: One of our patients was on intensive care, and we were able to fairly rapidly diagnose them with X-SCID. They had no gamma chain protein and the signalling pathways below that weren’t working, but we couldn’t find a mutation. But they didn’t have a good donor for bone marrow transplant.
For gene therapy at the time, you were required to identify the defective gene. And we couldn’t identify the defective gene, although we knew they didn’t make RNA for common gamma chain, all the signalling was not working. We went back and did some very old genetics… We did what was called X inactivation. When mothers have this particular disease, their T cells are non-randomly X inactivated, they only develop T cells from the good X. We could show that mom had non-random X inactivation, we could show the child inherited the inactive X. But, we couldn’t find a mutation. The parents were very educated and chose to have gene therapy. They were counselled that there was a risk that we were putting in the wrong gene. But there were limited other options.
We did gene therapy and the child has done very well, they must be about 18 now. But it was about six or eight weeks after they started getting a new immune system, we knew we’d put the right gene in because this was an in vivo complementation assay, and then we found the mutation. And then, because it was the reverse of what we’re doing now, we had all the functional data. It was actually down in the polyadenylation site, which controls RNA stability. Although the coding region was okay, it was a mutation that controlled the stability of the RNA, and therefore, you didn’t make protein, and the cells couldn’t develop and function properly.
FLG: That’s a fascinating case study. Thank you so much for sharing. If we look to the future, what do you think is needed to streamline clinical implementation equitably?
Kimberly Gilmour: I think data sharing is important. I think patient education is really important, because patient and parents don’t always come forward, or they’re concerned about who’s going to control their data. I think there’s some misunderstanding in some communities, and I think there’s a huge piece of work about education across patients around the world so they understand what genetic sequencing can and can’t do. I think we need real clarity as to who owns the data. I don’t think you can patent a human genome. I think people should own their own genetic data then they can share it with their health care professionals. But, that’s not necessarily how it works in all the different countries. I think that needs real clarification.
I think outside of places like the UK, a lot of genetic sequencing is on a paid-for basis, so you have to be wealthy to do it. Even though I appreciate a $100 genome, the quality analysis and other things are tied in with that. Obviously 100 pounds or dollars is still a huge amount of money if you’re in a less developed country. I think the technology is going to continue to get cheaper and cheaper every few years. It’s the analysis we need to get more efficient to get it available to people outside the more developed countries. Even within more developed countries, there’s an equity issue about who’s willing to have it and who’s not, or who’s scared of having it: someone’s going to know about me, someone’s going to do something with my data. There’s a big educational discussion to be had with the whole community, not just patients and parents, but government regulators. How do we prevent being hacked? Your DNA is your ID, is that ever going to be used as such? There are a lot of questions, I think that need ethical discussion. Much more so maybe, than the scientific discussion?
FLG: What are your thoughts on the democratisation of genomic data?
Kimberly Gilmour: I think it’s really important and I think people should be able to access their data, and they should be able to understand their data. That data, if they’re happy in an anonymous way, should be shared, so other people can benefit. I think it does come back down to education. I think people still think “I’ve had my whole genome sequenced, nothing was found, therefore it’s not a genetic disease”, and people don’t understand the limits of the technology in the analysis.
We’re really good at sequencing, we can do the whole genome, that’s not a problem, we can say it’s A, T, T, C, C, G, A, T, C. That’s not the problem. The problem is understanding what it means and particularly what all the non-coding space means, remembering the vast majority 97 plus percent of the human genome is non-coding. It’s all regulatory, we don’t even understand what a lot of it does. I think having people understand the bits that we don’t understand is really important too.
FLG: Absolutely. And you mentioned biosecurity there as well. What do you think is a key thing that people should take away as we set up these biobanks, both national ones, as well as international consortiums, what’s the number one thing we need to consider?
Kimberly Gilmour: We need to make sure that that data is truly anonymised when it’s put in there. Depending on how it’s done, you may want it reversibly anonymised because you may want to be able to say, “Oh my gosh, this person has this disease, and we can treat it.” If we find out, we can make a difference and save their life. But you don’t want any researchers or pharmaceutical companies or anyone else having anything other than that generic minimal data. You can say they’re a 50-year old male, that’s it. The problem is you might be able to immediately say they’re from this ethnic group, and they’re predisposed to heart disease and diabetes.
In the UK, it’s less of an issue, because we don’t worry so much about insurance, and our insurance is not dependent on your genetic data. But in countries where they want genetic data to get health insurance or life insurance, I think you can’t do that. Because if you can’t get health insurance if they find something bad on your genetic sequence, would you ever have a genetic sequence done? If you can’t get life insurance, and you need life insurance, because you have small children, what if you can’t get it because you have a predisposition to diabetes? But if you knew you had a predisposition to diabetes, maybe you would lose weight, be more active, change your diet, and never develop the diabetes. You get into this very complex system, which I don’t think is helpful. I think there needs to be real clarity around how that data is used by commercial providers of insurance and things like that.
FLG: Very optimistically, what are your hopes for the future? As we learn more about these variants of uncertain significance?
Kimberly Gilmour: I think we’ll have fewer and fewer variants of unknown significance because we’ve classified them as benign or pathogenic… That would be huge. Parents and families won’t think, “Well, we found this and we don’t know what it means”. I think in five to ten years, there may be no variants in unknown significance… We might have identified them. At least in the coding space. In the non-coding space I think there will still be a lot to learn. I think for families, if you have a sick child, and you come into intensive care, we can get a sequence back in two weeks and a diagnosis, and you get optimal care, I think that would be amazing.
I do think people have to understand that genetics are not the whole thing, there’s still a huge environmental component. COVID being the classic example, it doesn’t really matter what your genes are, very few people have not gotten COVID. We haven’t really found any good genes that say you don’t get COVID, so your genotype didn’t really matter. For maybe a very, very small percentage, you were more susceptible to it, you got sicker with it. But there’s also some fatality in “Oh, I have a genetic predisposition for heart disease”, and then they have a heart attack. The flip side is actually, you have a genetic predisposition to heart disease, why don’t you lose some weight, exercise and take blood pressure medication, because there are other things you can do to prevent that from happening. There are some genes that say you have no T cells, you need a bone marrow transplant, or gene therapy, and that is an absolute. But a lot of genetics, isn’t that absolute, and it’s a combination of genes and environmental factors. It’s still remembering all those environmental factors and all the things we can do ourselves to take care of our own health.
FLG: What is the most common answer you get when you’re validating a variant of uncertain significance? Are you getting more ‘Yes, this is pathological’, ‘No, it’s benign’, or a lot of ‘Maybes’?
Kimberly Gilmour: We get a lot more ‘definites’ than ‘maybes’. A part of it is because we’re screening a lot in silica, and we’re taking the ones that have a higher probability. Most of those, maybe 70%, we are confirming is pathological. But we’re not looking at every variant the child has, we’re looking at the variants in the genes that are likely to contribute to their phenotype. There are certain genes that have a lot of variants that are benign. There are a couple of genes where you have to be careful that you just don’t go, “Oh, it’s just another variant in that gene”, because you do actually have to look enough to think would that fit? How common is it? Is it worth following up? I think at the moment, we’re choosing which ones to follow up, because we don’t have the resources to follow up all of them.
FLG: Are you left feeling like Sherlock Holmes or Miss Marple having solved the case?
Kimberly Gilmour: When you solve the case it is really exciting, particularly if you solve it in a way that’s really unexpected. A couple of years ago, a colleague of mine, Shamima Rahman, a mitochondrial physician, came to me and said they had this family with what was thought to be a mitochondrial disease, and we looked at their mitochondrial genes and didn’t find anything. So, we did whole exome sequencing. The only thing we found was this variant in a gene called STAT2, and STAT2 is an immune gene. She’s going “I don’t understand” – the only gene found where both siblings are homozygous for it and the parents are heterozygous. Genetically it fit, but why was it causing mitochondrial disease?
These patients had really long mitochondria. In a normal human being your mitochondria grow, then they split in half, and then they grow again, and they split in half. They’re either small, or they get to a little bit bigger, but then they split and they keep going through this cycle. These kids just had really long mitochondria, they were never going through the cycle. They were just getting bigger, so they definitely had the mitochondrial disease, but the only gene we found was an immune gene.
Then we did a lot of work. We did complementation assays. And if we put the correct gene back into those cells, their mitochondria then went through the cycle of growing and fissioning. We were then able to go a bit further and show that there was a phosphorylation defect in a particular protein in the mitochondria that this immune gene STAT2 was controlling. That was really exciting. And you did feel absolutely like Sherlock Holmes, because you take in this case, that didn’t make any sense at all. It was a mitochondrial disease with an immune gene. I think we’re seeing a lot more of that. I think there’s a lot more interaction of genes and we need to work much more together. That was a really exciting experience.
FLG: Thank you so much, Kimberly, I’ve learned so much and completely agree that the importance of null results being just as valid as everything else that we’ve talked about today. And I’m very jealous that you get to play Sherlock Holmes so often. And thank you for joining us today.
Kimberly Gilmour: Thank you very much, Poppy.