Finding Answers for Undiagnosed Patients with Rare Genetic Diseases

Daniel Levine: Matthew. Thanks for joining us.

Matthew Bainbridge: Thank you for having me. I really appreciate the opportunity to talk to you.

Daniel Levine: Well, we’re going to talk about Rady Children’s Institute for Genomic Medicine, the use of whole genome sequencing as a diagnostic tool, and a recent research collaboration you’ve entered into with Pacific Biosciences. Perhaps we can start with the challenge of diagnosing for rare disease. We hear a lot about the diagnostic odyssey, but what does it take today for a parent with a child with a rare disease to get an accurate diagnosis?

Daniel Levine: That’s a great question. First and foremost, it takes luck. So even with the best diagnostic tools we have, the genetic diagnostic rate is still in the 30 to 40 percent range. Most parents will start out getting a microarray, then they might get a panel, and then they might move on to an exome, and then they might move on to whole genome sequencing, although that’s still pretty rare. And if all of those fail, then often they’ll end up in a research study, possibly in the undiagnosed diseases network or something similar. It usually takes a lot of tenacity on the parent’s part or a very, very good and dedicated clinician to push all those diagnostic tests.

Daniel Levine: What is the consequence of a delayed diagnosis?

Matthew Bainbridge: I mean, at its very worst it’s death; at its very, very worst, it’s a preventable death, right? Where there was a treatment for what the kid had and it was some extremely rare disease and the MDs just didn’t get to it in time. I will say that I think that’s very rare. A lot of our physicians are very committed and very strong, and so, they eventually get around to the right treatment. But that’s the absolute very worst thing. Equally quite terrible is you could get permanent damage. You can have a kid who was seizing and not responding and they seized for so long that they suffered permanent brain damage. You could get kids that had unnecessary procedures, right? And some of them are benign—a blood draw’s pretty benign, urinalysis is pretty benign, but some kids could end up getting liver transplants that actually aren’t beneficial for them, so a really expensive and really traumatic procedure. Even when you don’t have those sorts of very physical and expensive problems, you have these psychosocial issues. You know, parents, I always picture them sort of they’re lost at sea, right? They’re in a raft and they don’t know what’s wrong with their child and they don’t know what the future is going to hold. And they don’t know if they’re ever going to find a treatment and they don’t have anyone they can relate to. I remember early on in my career, I was shadowing an MD and he was meeting some parents who had a child with an undiagnosed, rare disease. We would go to one group and they’d say, oh, you’re so lucky, your kid can sit up. That’s amazing. None of our kids can sit up and then they’d go to a different group and they’d say, oh, we feel so bad for you because your kid isn’t vocal. All our kids are vocal. They can talk and they can respond to things. And so, they never really felt that they belonged in either of the support groups if they found. I think we really underappreciate and underestimate how hard that is for parents having a kid with an undiagnosed disease and going through this often very lengthy odyssey.

Daniel Levine: You mentioned a moment ago, even with the best diagnostic tools like whole genome sequencing, diagnosticians are often left without an answer to provide for why a child has a condition. Why is that? Why is there this gap even with today’s tools and being able to say, definitively, this is what’s wrong with the child.

Matthew Bainbridge: That’s a great question and I’ve thought about it a lot. In fact, that’s what I’ve dedicated the last 10 years of my life to figuring out why we can’t diagnose these kids. And on one level there’s probably a certain number of kids where it’s not genetic. And as someone who studies genetics, it pains me to say that because I’d like to think everything is genetic, but there’s probably some kids where their disease isn’t genetic, it’s something else, something environmental or something that happened to them that is causing their disease. Broadly speaking, there are things where we can’t detect it. That could be some variants like structural variants. That could be, you know, methylation signals where when we sequence, we aren’t looking for these things, we can’t detect the genetic variant that is causing the disease.

I think there’s a large portion of kids where we can detect the variant. We can detect the thing that’s causing the damage, but we can’t interpret it as being damaging. And so these are cases where—there’s parts of the genome we understand very well: the genes, the exons, the coding portion of the genome. We understand these parts very well, but we don’t understand other parts of the genome. So there’s non-coding portions of the genome. If you think of it as a recipe book, it’s the parts of the genomes that aren’t recipes and we just are not good at understanding what a typographical error or a deletion of a sentence of those things does. And then of course there’s novel disease genes, right? And so we’ve seen a really strong push in the last 10 or 15 years in identifying new disease genes and a kid who has a mutation in a gene that isn’t currently associated with disease, we can’t diagnose that child until we’ve done the research and can establish the association between mutations in the gene and the disease that the kid has.

Daniel Levine: In the past we’ve talking to Stephen Kingsmore about Project Baby Bear. Listeners may be familiar with that, but while I have you, this is work that Rady has done to really pioneer the use of whole genome sequencing into the NICU, the, the newborn ICU, to get rapid answers for kids who appear to have a genetic disease. I know a study ran recently in the American Journal of Human Genetics. What did the studies show?

Matthew Bainbridge: I think broadly speaking, it showed that if you have young—I mean, we concentrated on kids less than a year of age, acutely ill children that were admitted to an ICU, so a NICU or a PICU, pediatric ICU—if we don’t know what’s wrong with them, if it isn’t obvious, if they weren’t in a car accident or something like that, they should get rapid whole genome sequencing because it will save money and just on a purely monetary level, it will save money, not to mention any of the psychosocial benefits. And I think the thing that Project Baby Bear really showed was that genetics needs to be moved more to a first tier test rather than a second or third tier test; that people should be thinking, you know, genetics earlier on in the diagnostic odyssey for these kids.

Daniel Levine: What’s the lasting impact of the work? Is it now being adopted as a standard practice in newborn ICU?

Matthew Bainbridge: I will defer a little bit to grace here. I think we’re going to see California is going to start reimbursing whole genome sequencing for kids increasingly now.

Daniel Levine: Do you expect that at some point we’ll move to the use of whole genome sequencing becoming a universal newborn screening tool?

Matthew Bainbridge: You know that’s the dream, right? That is absolutely the dream. And I’ll tell you a little story about why this really speaks to my heart. I joined the Institute almost five years ago now and the very first kid they sequenced when I was here, they also diagnosed, and this was actually a 16-year-old girl who was just going out with her friends on a Friday night. And she just dropped and her friends didn’t know what to do, and they resuscitated her and they brought her in and we sequenced her and we found out she had a mutation in a gene called RYR2, and RYR2 helps coordinate your heartbeat. Right? So, she basically had a cardiac arrest event, and it was great that we got a diagnosis and there’s actually a drug you can give called flecainide. And the doctor was happy with getting a rapid diagnosis because he could put her on flecainide and she wouldn’t have another event and it could delay the surgery that she was going to get for a couple of days so that she could recover a bit more. So, the doctor was really appreciative. And because I was going to do a talk about it six months later, I looked her up and she had suffered brain damage, a huge harm for the amount of time that she was down. And it was great that we’d made this diagnosis. It was great that we had some treatments that we could give her, but wouldn’t it have been so much better if we could detect this mutation a week before she had the event, but better yet, the moment she was born, and could have done something about it to prevent it and be able to do sequencing, broad sequencing, whether that’s whole genome or whole exome or a couple thousand genes—it would just be tremendous. And I can think that there’ll be so many kids that we could benefit.

You think if it’s a screen you’ll get some kids early on, but you can also imagine… the thing that we do at Rady’s, and we do so well, is this really ultra-rapid sequencing, right? The kid comes in and the kid’s sick. We sequence them in 13 hours. But there’s an even faster genome, and it’s the genome you got the moment you were born before you actually need it. And I can’t imagine the benefits of living in a world where your kid could show up in the NICU or PICU and the doctors can just call up their genome and say, okay, given these symptoms, is there anything in the genome that could contribute to this? And that is the thing that could take five minutes, and to either find a diagnosis, and say, oh, it’s probably this. We should be doing this, or to not find anything and say, okay, we can probably eliminate a genetic cause—that’s amazing. That would just be hugely amazing.

Daniel Levine: The problem: cost—is it inability to translate the data into actionable information? Is it something else? What, what will it take to get to the point where this becomes a universal newborn screening tool?

Matthew Bainbridge: I mean, I think cost is a big component of it. I think newborn screening right now costs about a hundred dollars. I think sequencing right now would probably be on the same order of magnitude, so you’d be talking about doubling the cost, maybe even a bit more. Costs are not insurmountable though. Right? What we’ve seen is that costs have come down in sequencing tremendously. Twenty years ago, the first human genome, it cost about 500 million or a billion dollars, depending on who you ask. And now it’s a couple of grand. Costs are not a thing that I worry about too much. I think there’s a lot of political social issues around it. You know, do parents want the government or some other agency having the full genome sequence of their kid? A lot of these agencies feel that they’re already overburdened with even the limited number of newborn screening tests they do. And to say, oh, we want to screen for 2000 conditions, I think would make their heads explode. And they would see that there’s just going to be this huge burden. There are additional issues that right now, the logic under newborn screening is that those [screened for] has to be a specific essentially curative treatment, right? That they don’t look for all seizure genes, even though you could give kids phenobarbital or Keppra or something for most seizures, they just don’t look for it. They wait for the kid to have the seizure first. And so that would require a real paradigm shift for the people who do newborn sequencing now to be willing to screen for more things, even if there wasn’t necessarily a treatment for it, and I think that that’s going to be a major issue.

Daniel Levine: At the end of June, Rady Children’s announced a research collaboration with Pacific Biosciences. The study is focused on using long read whole genome sequencing in cases where short read whole genome or exome sequencing fails to yield an answer. Can you explain the difference between long-read whole genome sequencing and short-read whole genome sequencing?

Matthew Bainbridge: Absolutely, so imagine your genome as a book, and it’s actually in a way kind of a boring book, because lots of it is repetitive. So, imagine on page 50, chapter 3, and then page 100, chapter 8, we actually have a page that’s exactly the same. It reads exactly the same. As it turns out, the one on page 50 actually codes for a gene. So, it’s really important, say to the plot of the book, but the page on 100 doesn’t code for a gene, it codes for a pseudo gene. So, it’s not important for the book. In fact, you could skip all of chapter 8 and you wouldn’t miss out anything. So, when you get a mutation or a variant, which we had in this case, we’d think of it as like a typographical error. We have a typo. How do we know it came from page 50 versus page 100?

When you use short-read sequencing, you’re basically reading a sentence at a time or a couple of words at a time. And that means you can detect the typo, but you can’t tell whether it came from page 50 or page 100. Long-read sequencing is like reading whole pages or multiple pages at a time. And so, because we know that this typo is also associated with page 49 and page 51, we actually know that that typo must’ve come from page 50 and not from page 100. That’s really one of the major benefits of long read sequencing—that you can get into these regions of the genome that are highly repetitive and therefore can’t really be read one sentence at a time very effectively.

Daniel Levine: Are you at a point where you can make the case that long-read sequencing can identify disease causing genetic variants that short-read sequencing fails to identify?

Matthew Bainbridge: Absolutely. We can a hundred percent say that. In our case with our 25, 5 of which were controls, in the 20 we didn’t make a new diagnosis. So, in that case, it wasn’t beneficial. But in one of our controls, there was a mutation that we knew we could not pick up on short-read sequencing. And it was exactly this case. It was a gene that has a pseudo gene and we just could not place the reads over the correct. We couldn’t tell that the mutation was in the actual functional gene. And then when we did it with long-read sequencing, clear as day, no problem. When we took the 20 cases where we didn’t make a diagnosis and we said, well, did we still find mutations that look bad? So, they are highly deleterious, we know that they would stop gene function, that we didn’t pick up with short read. And we found a handful about a dozen of those sorts of mutations, in this case they weren’t diagnostic because various reasons, usually the gene is also recessive, so you need a second hit in the other copy of the gene, but we’re definitely picking up variants we weren’t able to pick up with the short reads.

Daniel Levine: How are you going about the study and how long will it run?

Matthew Bainbridge: The initial push was to just get cases that were very close to some of our clinicians’ hearts. There are cases where they think, absolutely this must be genetic. There’s something going on in this kiddo. And yet the short read sequencing did not yield a result. So, they said, I really want you to look at these kids because we really need to get a diagnosis. And we’re very convinced it’s genetic. Then we did a second cohort, which we’ve just gotten results back on yesterday and I haven’t had a chance to look at it at all. We concentrated on diseases where we think long-reads are going to be more beneficial because the genes are harder to sequence with short-reads. And as it turns out, that’s a lot of immune diseases and kids often don’t get a diagnosis in immune diseases. So, we picked out a cohort primarily of immune kiddos and we’re hoping we’re going to run this study for at least about a year. In the end, if we had a diagnostic rate of like 30 percent, long-reads could push us up to 33 or 35 percent, I would think that was tremendous. But in order to be able to make that case, we’re going to have to do 40, 60, 80 cases.

Daniel Levine: And from a process, time, cost point of view, how does long-read sequencing compare to short-read sequencing?

Matthew Bainbridge: The process is very similar. You know, you draw blood, you extract the DNA, you put it on the sequencer. You do some bioinformatics to figure out where the reads came from and look for variants. In that case, that’s very similar. The time is not as short. The long-read sequencers take longer to run, but it’s not an order of magnitude longer. Our fastest short reads are about 13 hours, most places do 48 72. And you can do long reads in 24, 48, 72, that sort of length, right? It’s not much longer, but it is definitely longer. Cost is a factor. It’s two to three times the cost of short-read sequencing, and I think that’s has slowed down the uptake, because generally if you told someone, you know, you can do 10 kids very thoroughly with long-reads, or you could do 30 kids almost as thoroughly with short-reads, they’d usually say it’s better to go after the 30 kids.

Daniel Levine: Tools tend to be good for specific purposes. You know, it’s much better to use a screwdriver for a screw than a hammer. Is long-read sequencing better for diagnosing certain types of genetic mutations than others?

Matthew Bainbridge: Yeah, like with my initial analogy, genes that have pseudo genes or repetitive elements, it’s really good at getting in there and finding the typos and pinning them down. But we also think that long reads are much better at picking up what we call structural variants. So, if a standard variant is like a typographical error in a book, a structural variant might be you rip out three chapters, you turn them upside down, you stick them back in. The reason why is that these kinds of big structural variants are often mediated by these repetitive elements. So, these two pages that are very identical, the structural variant actually relies on that identicalness to cause the inversion or the ripping out and the replacing. And it’s not always easy to actually tell that’s happened when you’re only reading a sentence at a time; when you’re reading multiple pages, it’s much easier.

And in fact, one of the cases that we did with PacBio was a case where we really thought we had a translocation and that’s where a piece of a chromosome goes and attaches to another piece of a chromosome. And it looked very convincing, really convincing off a short read. And then when we did it with PacBio, we actually saw that it was an insertion of a tiny little bit of process pseudogene, sorry for all the jargon, and that it wasn’t really anything that drastic as a translocation. And in that sense, it’s great that we could very quickly eliminate this as something that is disease causing and not waste a lot of time pursuing it. Although that’s less exciting than getting a diagnosis, it actually saves a lot of resources on the backend trying to validate things that aren’t actually there.

Daniel Levine: Are there usually clues as to when it might be more appropriate to use long-read sequencing or is it just a matter of progressing when you fail on an earlier test?

Matthew Bainbridge: I will say, I don’t really know the answer to that. I think there are some diseases where we’ll find that the causative genes are harder to sequence with short reads. Like I said with immune disorders, probably long reads are going to get you more diagnoses in immune disorders than in other kinds of disorders. But I don’t think we have a really clear of “we should immediately go to long read on this kiddo.” And it’s much more, let’s try short read first, if that fails, let’s move on to long read.

Daniel Levine: Are there other things you found or things you’re hoping to find from this work?

Matthew Bainbridge: Good question. Obviously, we’re hoping to make more diagnoses. I would really love to find out that in our immune cohort, we make a couple of diagnoses in there. And then I can just recommend that every immune kiddo we have, we can just send straight over to long read. I am hoping that we can find weird stuff. I like the weird stuff. Once you eliminate all the sort of common stuff and normal things that everyone looks for, all you’re left with is the weird stuff. And I’m hoping we make some diagnoses with some weird things as well, because it’s great for the kid to get a diagnosis. And it’s also good for us to figure out the kinds of other things that we should be looking for.

Daniel Levine: Matthew Bainbridge, principal investigator, and associate director of clinical genomics at Rady Children’s Institute for Genomic Medicine. Matthew, thanks so much for your time today.

Matthew Bainbridge: Thank you so much. It’s been a pleasure.

This interview has been edited for clarity and readability.