Cassie Parks at the 2023 Cedars-Sinai Intracranial Hypotension Conference
Cassie Parks, an MD/PhD student at the genetic lab of Hal Dietz at The Johns Hopkins University School of Medicine, presented this talk titled “Heritable Disorders of Connective Tissue and Spontaneous CSF Leaks” at the 2023 Cedars-Sinai Intracranial Hypotension Conference on July 8, 2023. The conference was hosted by Cedars-Sinai with generous support from the Spinal CSF Leak Foundation in Kohala Coast, Hawaii.
Slides from the talk
View a PDF of Parks’ slides here
Transcript
[00:00:09] All right, thank you so much for that introduction. It’s a pleasure to be here today. So today I’ll be talking about what our work has shown so far in determining whether there is a connection between spontaneous spinal CSF leaks and heritable connective tissue disorders. So first I wanted to talk a little bit about why we’re interested in this connection, although I think it’s pretty obvious from the morning section why that’s true.
[00:00:41] But we do know that patients with spontaneous spinal CSF leaks tend to have a preponderance of connective tissue disease characteristics where about 20 percent have nonspecific connective tissue findings such as tall stature, arachnodactyly, or long fingers, and stretchy skin, and among other things. There are also about 5 percent of spontaneous spinal CSF leak patients have defined diagnosable connective tissue disorders.
[00:01:08] Conversely, we know that patients with connective tissue diseases have higher incidences of spinal CSF leaks, which include patients with Marfan syndrome, Ehlers Danlos syndrome and Loeys-Dietz syndrome. Further, there are additional dural pathologies that are quite common in heritable connective tissue diseases.
[00:01:27] So for example, dural ectasia, in which there is dural outpouching of the dural sac, is one of the major criteria for diagnosis of Marfan syndrome, and it occurs in about 30 percent of Loeys-Dietz syndrome patients. Tarlov cysts, which we’ve spoken a bit about, um, are innervated arachnoid nerve root cysts and also occur commonly in patients with Marfan syndrome.
[00:01:49] And then of course, as we’ve said, CSF leaks are more common in connective tissue diseases. Because dural pathologies occur with a high frequency in patients with known connective tissue diseases, we hypothesize that genetic variants exist in a gene that could predispose patients to spontaneous CSF leaks.
[00:02:08] So what would be some of our expectations about such variants in a gene? We would expect such variants to have low effect sizes, meaning that any variant we find wouldn’t cause an extremely dramatic phenotype. It would cause a slight change in the tissues of patients such that over a lifetime and possibly with the correct environmental perturbation, the CSF leak phenotype would be elicited.
[00:02:31] And this is in contrast to syndromes like Marfan syndrome or Loeys-Dietz syndrome, where the patients have obvious dramatic characteristics that one can pick out across a room. Additionally, the gene would have low penetrance, that is, everyone with the mutation would not be expected to have CSF leaks necessarily.
[00:02:50] That is, there are potentially many people walking around with mutations in the gene that predispose them to the CSF leaks without the CSF leaks to a degree where they would seek out a doctor or obtain a diagnosis. So overall, we’re looking for a fairly common variant, as genetic diseases go, but with enrichment in CSF leak patients.
[00:03:13] In order to find this gene, we’ve conducted a case control study where we’re looking for a signal for enrichment of a rare deleterious variant in the CSF leak population compared to a matched control group. So to do this, we conducted a whole exome sequencing of 49 probands who have type 1b spontaneous spinal CSF leaks and collectively over 5000 controls who did not.
[00:03:38] We then filtered for variants that were rare, that is, they have a minor allele frequency less than 0.01, meaning that less than 1 percent of the patients has a variant. Then we filtered for variants that could have deleterious effects on the structure or expression of the protein, that is, missense, stop, gain, indel, or splice variants.
[00:04:00] We kept genes where at least 10 percent of the probands had rare variants, filtered for genes that could possibly be associated with connective tissue phenotypes, and ended up having a variant list. And when we assess this variant list, we found that 20 percent of patients with type 1b CSF leaks have rare missense variants in the gene fibrillin-2.
[00:04:21] And interestingly, one of the mutations, which is I2394T here, occurs in 1 out of 2350 people in the general population, but in 8 percent or about 1 in 12 of our patients with type 1b CSF leaks. With this in mind, we conducted burden tests to determine whether these rare fibrillin-2 variants are more common in the leak patients than in a variety of control patient cohorts.
[00:04:49] When we compared the leak patients to patients from the Mendel Initiative, which is a cohort from the U. S. and has about 2,000 patients, to patients from the Belgian Whole-exome cohort, which has about 1400 patients, and patients from the thoracic aortic aneurysm and dissection cohort, which has about 18,000 patients, we saw a consistent signal for enrichment of the rare fibrillin-2 variants in CSF leak patients. Interestingly, the fibrillin-2 variants that we saw in the patients did not exhibit a random distribution. So, within fibrillin-2, about 70 percent of the protein is comprised of calcium-binding EGF domains, which are indicated here by the green rectangles.
[00:05:33] However, about 90 percent of our patients’ rare variants fell outside of those domains. 60 percent of the variants lie within 8-cysteine, or TB, domains, which are the blue diamonds. And 20 percent of them lie in or around the only two RGD domains of the protein which are indicated by the red rectangles and I’ll explain why that’s notable in a moment.
[00:05:56] It’s kind of a negative control. We can look at the distribution of rare fibrillin-2 variants in a completely unrelated disease and a disease that we would not expect to be related to fibrillin-2 at all. So here again, 70 percent of the proteins comprised of calcium-binding EGF domains. And for this unrelated disease, 75 percent of the rare variants localize to calcium-binding EGF domains also, whereas 12 percent of this unrelated disease variants localized to TB domains, and none of them localized near RGD domains.
[00:06:30] So this is more what you would expect based on random chance. We were especially interested in the fact that two out of the seven unique variants lay in or around RGD domains, particularly because one of the variants directly disrupts one of the RGD residues, which we’d predict to lead to an obligate loss of integrin binding.
[00:06:51] The arginine-glycine-aspartate motif is the canonical binding motif of integrins, which are one of the major receptors that allows cells to bind to the extracellular matrix, and is responsible for a myriad of cell signaling events which affect the state of the cell. So we thought it was reasonable to hypothesize that mutations near or within these sequences could have effects on the ability of cells to bind to the extracellular matrix and on the signaling events that occur within them.
[00:07:20] Though the arginine-glycine-aspartate motif is the canonical integrin binding motif, there are also non-canonical binding sites that are being appreciated over the years but can be a bit more difficult to identify. One prediction you could make is that the non-canonical binding sites would share features with other integrin binding sites, and one such feature is the presentation of those sites on flexible loops so that they’re easily accessible by integrins.
[00:07:47] So the TB4 domain, which is shown in the middle, has been shown previously to be used for integrin binding. And as you can see, AI prediction of the structure of this domain asserts that the RGD domain, which is highlighted in green at the top, exists at the end of a flexible loop. When looking at the location of the other two mutations in TB domains by the same methods, we see that the mutations are also located in flexible loops.
[00:08:12] And this gives further support to the hypothesis that these mutations may be disrupting binding sites. To see whether the hypothesis that fibrillin-2 mutations could predispose patients to CSF leaks has a basis in human genetics, we sought to answer the question of whether there’s a precedent for RGD mutations within proteins that have been previously associated with a different disease, causing an unrelated and unexpected condition.
[00:08:40] First, we can turn to what we know about the human genetics of the other major functional fibrillin, which is fibrillin-1. We know that mutations that cause Marfan syndrome occur in calcium-binding EGF domains, where people are tall, they’re quite flexible, and then they have a high penetrance of eye lens dislocation and aortic root aneurysm.
[00:09:02] In contrast, mutations in or immediately adjacent to the only RGD domain in fibrillin-1 cause a distinct phenotype called stiff skin syndrome, where patients are short, they’re somewhat muscular, and they have no cardiovascular or eye disease, where their defining characteristic is skin fibrosis. So, different mutations in the same protein cause extremely discordant phenotypes that have no clinical overlap.
[00:09:28] And you could not have predicted the phenotype of stiff skin syndrome based on the phenotype of a patient with Marfan syndrome. When we look at fibrillin-2, we know that mutations in calcium-binding EGF domains cause congenital contractual arachnodactyly, where patients have contractures of the hands and feet, crumpled ears and scoliosis.
[00:09:49] This begs the question of what kind of phenotype would result from disruption of integrin binding to fibrillin-2. We think it’s plausible to hypothesize that mutations in and around the RGD domains of fibrillin-2 could have important clinical consequences that you couldn’t predict based on the phenotype of someone with CCA, and that consequence could potentially include spontaneous CSF leaks. So our central hypothesis is that mutations in functionally important domains of fibrillin-2 lead to altered tissue homeostasis and predispose patients to dural tears and spontaneous CSF leaks. So to interrogate this hypothesis, we’ve used CRISPR to create three mouse lines that have mutations corresponding to three of the most interesting patient mutations.
[00:10:35] The two mutations that are in and around RGD domains, and the mutation in TB7 that’s occurring in 8 percent of the patients. However, we would expect undertaking the phenotyping of this mouse to be quite different from the phenotyping of a mouse with Marfan syndrome or stiff skin syndrome, which have dramatic human phenotypes and mice which obviously reflect those phenotypes.
[00:10:56] And we’ve spent the better part of the morning discussing the difficulties in diagnosing humans with a headache that they can tell you they have. So it stands to reason that the mouse would also have a phenotype that’s at least as difficult to delineate. And then, so we’ve begun undertaking multiple approaches to assess the biomechanical integrity of the dura in these mice.
[00:11:17] And admittedly, we’re finding this quite challenging. So I’d be very receptive to any ideas anyone has afterwards, if you can find me. But as one example of mechanical testing avenue we’ve undertaken, we designed a device that can test the force required to puncture the dura overlying the parietal lobe of the brain.
[00:11:35] So to do this, I make bilateral craniotomies of a consistent size and location based on a 3D printed guide, and then use a blunt lacrimal probe to puncture the dura at a consistent speed. Because I can measure the force acting on the probe, I can determine how much force it takes to puncture the dura.
[00:11:53] As a positive control for these tests, we test mice with Marfan syndrome, and we’d expect them to have a weaker dura due to the Marfan patients’ increased incidence of dural pathology. However, we did not see a difference in the cranial dural strength between the Marfan mice and the litter-mates, uh, controls, no, nor between the fibrillin-2 mutant mice and their litter-mate controls.
[00:12:16] So we’re developing a test more sensitive than this to use, but that development is ongoing. Obviously, ideally we’d want to target the spinal dura, but that is also extremely challenging. So next we turn to light microscopy to determine whether mutations in fibrillin-2 cause differences in neural integrity that we can appreciate histologically.
[00:12:36] So we performed a variety of analyses and didn’t see any differences in the H&E stain between wild type and fibrillin-2 mutants, nor the VVG stain, which stains elastin black, nor the trichrome stain, where collagen is stained blue. And of note, a prior study that looked at human dura from patients with documented spontaneous CSF leaks noted that the dura of these patients didn’t appear different by microscopy.
[00:13:01] An additional avenue we’ve undertaken is using transmission electron microscopy that allows us to look at the ultrastructure of the dura at a very high magnification. One example of a phenotype we begin to notice is the structure of the elastin in the dura, which here stains very darkly and you can appreciate it with the red arrows.
[00:13:21] And you’ll notice in the top panels, which are the wild type mice, the borders of the elastin appear smooth, and the center of the elastin is completely filled in with tropoelastin. Where in contrast, in the bottom three panels, the elastin has more of a moth-eaten appearance, where the edges are not smooth and the tropoelastin does not fill the inside completely.
[00:13:43] And here’s just a higher magnification of that so you can appreciate those differences. But the significance of this finding remains to be elucidated. And then to interrogate whether there are differences in the state of signaling between dural fibroblasts in the fibrillin-2 mutant mice and the wild type mice, we’ve begun to undertake single nuclear RNA sequencing, which allows us to take a snapshot of the mRNA messages that exist in each nucleus, identify a specific cell type, and compare the expression in that cell type between groups. So excitingly, this is the first experience anyone’s had doing single nuclear or single cell RNAseq from mouse spinal dura. And through this sequencing, we’ve been able to reliably identify a multitude of cell types, which include the dural fibroblasts, cells of the pia mater and arachnoid mater, Schwann cells and oligodendrocytes, and neurons and astrocytes, among others.
[00:14:39] When we pull out only the dural fibroblasts, we can conduct differential expression to see what the signaling differences are between the fibrillin-2 mutants and the wild types, only in that cell type. So in this graph, the larger circles mean that more dural fibroblasts express a certain gene, and the darker circles indicate higher expression of the gene, where the lighter circles indicate lower expression of that gene.
[00:15:03] So you can see that the fibrillin-2 mutant has decreased expression of key extracellular matrix proteins, such as two of the fibrillar collagens, which are collagen 1 and collagen 3. Among other proteins. So it is possible that protein turnover differences didn’t allow us to appreciate these expression differences using microscopy.
[00:15:22] So we are going to undertake biochemical assays using the entire spinal dura to validate this. Lastly, we’d like to test our mechanistic hypothesis that fibrillin-2 RGD domain disruption prevents dural fibroblasts from binding to fibrillin-2 and creating a strong dural layer. To do this, we’re using in vitro assays that assess the binding of human dural fibroblasts to recombinant fragments of fibrillin-2 in the regions that correspond to the patient mutations, both in the wild type form and in the mutated form.
[00:15:55] This will allow us to both determine whether there are novel binding sites in fibrillin-2 that are as of yet unknown, and whether the patient mutations disrupt these binding sites. We’ve generated recombinant protein fragments corresponding to the regions of fibrillin-2 where the patient variants exist, and have used them to determine whether cells can bind to the fragments.
[00:16:17] The three fragments we’ve generated are TB3, which has an RGD domain; TB4, which also has an RGD domain; and TB7, which does not have an RGD domain. And what we’ve done is we coat Petri dishes with these recombinant fragments and see whether dural fibroblasts will adhere. Here you can see we were able to achieve robust expression of peptides in both their native non-reduced state and in their reduced state, both for the wild type and mutated genotypes.
[00:16:48] The change in migration upon reduction attests to the formation of secondary structures in vitro. So what we found is that TB3 didn’t show evidence of interacting with dural fibroblasts in either the wild type or mutant form. So in these graphs, the x axis is increasing fibrillin-2 fragment concentration, and the y axis is increasing cell binding.
[00:17:10] In contrast, TB4, which contains an RGD binding sequence, and TB7, which does not, showed clear evidence of interaction with dural fibroblasts, in blue, and that was greatly attenuated by the patient mutations, which are in red. Thus, we’ve used patient genetics to discover a novel, non-canonical fibrillin-2 binding site, and mutations which abrogate them, which are associated with CSF leaks.
[00:17:36] A key characteristic of dural fibroblasts that must be true for them to bind to RGD domains is the expression of integrins that bind them. Using qPCR, I’ve determined that human dural fibroblasts express alphaV beta1, alphaV beta5, and alpha5 beta1, which are all RGD binding integrins.
[00:17:57] They also express alpha3 beta1, which is not an RGD binding integrin, but has previously been shown to bind to an unknown part of fibrillin-2, which might be our novel TB7 binding site. So in conclusion, We know that there’s enrichment for rare and functionally consequential DNA sequence variants in patients with type 1b spinal CSF leaks, and our data taken together suggest that loss of cell-fibrillin-2 contacts impairs matrix synthesis and integrity, but obviously, there are future studies we have to do that could inform therapeutic strategies.
[00:18:32] A lot of people were involved in this work, so I just wanted to thank all of them. Thank you very much for your time.