Second dispatch from the London cohort
A new pre-print from Simon Mead and T.H. Mok reports the latest updates from a cohort of pre-symptomatic people at risk for prion disease in London and provides some valuable new insights into the evolution of fluid biomarkers as people approach disease onset [Mok 2022].
Sonia and I have long said we’re racing against a clock we can’t see. People at risk for genetic prion disease seem to be totally healthy for some variable number of decades, then fall off a cliff: a totally unpredictable age of onset followed by a usually rapid disease course [Minikel 2019]. What if there was some way to predict when you only had 1 year left, or 5 years left? While many mutation carriers would find this information terrifying, it would also be really useful. While we think that the best opportunity for therapeutic intervention is before anything goes wrong, and so we don’t necessarily want to wait until you’ve got only a year left, one can also imagine a drug with good efficacy but a bad side effect profile where you wouldn’t want to, or couldn’t, take it for decades. Meanwhile, while you probably couldn’t design a clinical trial solely around people close to onset, because there are too few of them at any given time, seeing the effect of a drug in just a few such people could be powerful in convincing yourself whether a drug works. All this is to say: we would love to find fluid biomarkers — molecular readouts in blood and spinal fluid — that tell us if someone is close to disease onset.
Until a few years ago, very little was known about any changes before onset in presymptomatic people at risk for genetic prion disease. Efforts to find any prodromal brain imaging changes by MRI or FDG-PET found putative changes only months before onset, and this was based on a subjective, retrospective reading of the images, making it hard to set any quantitative cutoff for what imaging features did or didn’t count as a bad omen. More recent efforts to find subtle neurophysiological changes — things like reflexes, heat threshold, etc. — found changes only right at the onset of symptoms [Rudge 2019]. Several fluid biomarkers are good at discriminating symptomatic prion disease patients from normal controls or even from other neurodegenerative diseases, these markers often increase years or decades before onset in other neurodegenerative diseases, and in mouse models, we can see fluid biomarker changes before any obvious symptoms [Minikel 2020]. All these lines of evidence suggested that fluid biomarkers might be a valuable place to look for prodromal changes. In 2017, we launched a study, ongoing today, where we fly at-risk mutation carriers to Boston from around the U.S. to donate blood and cerebrospinal fluid (CSF) for research. In our first dispatch from that study, we found that cross-sectionally, most mutation carriers are normal. Most do not have elevated markers of neuronal damage in their blood or CSF, though we did find 1 person (out of 23) who had prion seeding activity in their CSF by RT-QuIC [Vallabh 2020]. Around the same time, the scientists behind the London study published their first batch of results, showing that in a group of people with slowly progressive mutations (such as P102L, OPRI), there was some signal of increased neurofilament light (NfL) a marker of neuronal damage, appearing in blood maybe as early as 2 years before disease onset [Thompson 2021]. In people with rapidly progressive mutations (E200K, D178N), however, there was at most a whiff of NfL increase only a few months before onset.
While we know more about pre-symptomatic biomarkers today than we did five years ago, our knowledge is very limited, and we need to expand on three dimensions. Cross-sectionally, we need more people, and we’re eagerly awaiting expansion of the Boston and London cohorts as well as news from other pre-symptomatic cohorts such as those in Tel Aviv, Barcelona, and San Francisco. Longitudinally, we need longer follow-up on the existing participants, and for that we can only sit and wait and try to keep people interested in coming back. Finally, breadth-wise, we need to look at additional candidate biomarkers, and see if anything can provide an earlier signal. On all of these dimensions — N, time, and markers — the new report from London provides some valuable progress.
The new report follows a total of 69 at-risk individuals. This includes people at 50/50 or 25% risk who were untested, and so there are some undisclosed number of individuals who proved mutation negative. These 69 people donated a total of 217 blood samples (average ~3 each) and 61 CSF samples (average ~1 each). Some people had just 1 study visit, but it appears from Figure 1 that others had as many as 10 serial samples over as long as 10 years. Out of this cohort, either 15 or 16 people developed prion disease. (The text at top of p. 18 says 16 but then breaks it down as “P102L = 10, D178N-FFI = 2, E200K = 1, 5-OPRI= 1, 6-OPRI = 1”, which adds up to 15). It is particularly out of this cohort of people who converted to active disease that we have some valuable insights.
For plasma NfL, the message is basically the same as last time, but the larger N gives us greater confidence in the results, which break down on the dimension of slow/rapid mutations.
For slow mutations (mostly P102L), there appears to be a prodromal NfL rise up to 2 years before onset, albeit with caveats. Most of the action occurs under the 90th percentile of healthy controls, meaning that an individual measurement would not be enough to determine that someone was prodromal — it’s really the serial sampling that allows you to make out someone’s approach towards onset. Even then, it’s noisy — many people have spikes and dips, at least one person who hasn’t developed disease had a huge spike (see Figure 5C), and at least one OPRI carrier actually looks like NfL went down leading up to onset (Figure 5B).
For rapid mutations (E200K, D178N), the prodromal rise, if any at all, is “explosive” right at the time of onset. Just eyeballing Figure 5B, one rapid mutation carrier had a large sudden spike in NfL maybe 2 months before onset, but another remained at their low baseline just maybe 3 months before onset and had a spike observed only after onset. Maybe if that latter person’s plasma had been sampled every 2 weeks it would have been possible to catch something before onset, but wow — it’s really the blink of an eye.
GFAP is a structural protein found in astrocytes, and because prion disease results in intense astrocytosis, increased GFAP in the brain is among the earliest pathological changes in prion disease in animals. Some studies have found a GFAP rise as early as 55 days post-inoculation in mice [Tamguney 2009]; we saw it just slightly later, at 70 and 73 in our two studies, still well before symptom onset [Vallabh 2022, Minikel 2020]. Could prodromal astrocytosis be a early warning sign in humans too? Initially my hopes were not high, because in the one study I’m aware of up to now GFAP in CSF didn’t seem too sensitive/specific even at the symptomatic stage. That study [Abu-Rumeileh 2020] found that CSF GFAP was about 50% higher in symptomatic prion disease compared to controls (1.028 vs. 0.665 ng/mL), with distributions widely overlapping [Abu-Rumeileh 2020]. Consider that NfL is sky-high in symptomatic patients, and as we saw just above and in 2020, even for NfL, the prodromal signal is subtle. What hope is left, then, for wringing a prodromal signal out of a marker that only goes up 50% even when patients are dying? Well, possibly more than I thought. Again, the findings from Mok et al break down by mutation.
For slow mutations, there appears to be an upward trend in plasma GFAP in the two years leading up to disease onset, and maybe even earlier than that. I see two, maybe three, P102L individuals who had a reading at 2.5 or 3 years before onset that looks higher than their baseline. The authors say four years, which I’m a little less convinced of. In either case, though, it looks a lot more like a signal than I had expected.
For rapid mutations, there’s really no appreciable signal even a few months before onset. Indeed, even in the handful of months these patients survived after onset, plasma GFAP spiked pretty high on one but only inched up slightly in the other two.
The difference between this study and my low expectations for GFAP appears to lie in the measurement of GFAP in CSF vs. in plasma. You can see this pretty clearly in the comparison of the leftmost panel of Figure 4A vs. 4B. In CSF, just as Abu-Rumeileh found, there’s really no GFAP signal even for sporadic CJD. In plasma, by contrast, there is a stepwise increase smaller in magnitude but similar in kind to that seen for NfL: sporadic CJD is higher than symptomatic genetic is higher than last two years before onset genetic is higher than healthy people (mutation carrier or not).
RT-QuIC is fantastic at diagnosing sporadic CJD, and pretty good for E200K genetic cases too. But historically it has had pretty limited sensitivity for detecting prion seeds in the CSF of symptomatic genetic cases with P102L, D178N, and other mutations. (More background and references in my fluid biomarker survey). Low sensitivity in symptomatic cases sounds like a pretty bad starting point for trying to find prodromal prion seeds in pre-symptomatic patients with those mutations. In 2020, I wrote an R01 grant proposal in which I planned to develop RT-QuIC conditions, using various mutant recombinant PrP constructs, optimized for detecting CSF prion seeds associated with different mutations. Reviewers panned the proposal (its score was “Not Discussed” in NIH parlance), and I set the idea aside for the past couple years. Upon opening this pre-print, I breathed a huge sigh of relief: perhaps I didn’t need to do any of that work after all — these folks went and did it!
At a minimum, it looks like the conditions the authors explored included: classic “IQ-CSF” conditions using truncated Syrian hamster PrP, bank vole PrP (which we also tried on CSF in our MGH cohort but couldn’t get to work well [Vallabh 2020]), full-length human PrP, and P102L PrP. The Results implies maybe they tried other mutant recombinants as well: “Optimum RT-QuIC conditions for D178N-129M, Y163X and classical 6-OPRI were not found despite extensive exploration.” I would love to hear more details on that extensive exploration — it could save others a lot of time to know what else didn’t work. In any case, one thing that did work decently is P102L: that recombinant construct, with specific buffer conditions, gave a positive RT-QuIC in 4/9 symptomatic P102L cases, 1 pre-symptomatic P102L individual, and 1 out of 57 negative controls. While not perfect, that looks like better sensitivity and specificity than we had for P102L up to now.
In terms of prospects for detecting potential prodromes, again, the results appear to break down by mutation, but in the opposite direction as before.
For slow mutations, there’s not yet any evidence of ability to detect prodromal seeding activity in CSF. That one pre-symptomatic P102L carrier who was positive was apparently not one of the individuals who converted (yet). According to Figure 1, only one P102L individual who converted had a pre-symptomatic CSF sample available, and that sample, taken 0.9 years before onset, was negative, even though the person’s RT-QuIC turned positive 0.6 years after onset.
For rapid mutations, specifically E200K, there’s a hint here that CSF RT-QuIC may provide that longer prodromal window for detection that we’ve been missing with NfL and GFAP. One pre-symptomatic E200K individual had a positive RT-QuIC and then became symptomatic 0.2 years later. In terms of people who actually developed active disease, that’s all we know, and of course, that’s not a long window. But interestingly, two other pre-symptomatic E200K people were RT-QuIC positive and remained healthy as of last follow-up 2-3 years later. While we can’t say with any certainty that this has any prognostic implications for those people’s future disease onset, it seems highly plausible, and if so, this might suggest that the window in which to detect RT-QuIC seeding activity before onset in E200K carriers could be a few years. In our MGH study, we had reported one E200K carrier with a positive RT-QuIC who remained asymptomatic 1 year later [Vallabh 2020]; the new findings from London add to the number of such individuals, hint at the potential duration of RT-QuIC positivity, and provide one example where a person who was positive did indeed convert to active disease.
caveat: actual onset vs. predicted onset vs. years of follow-up
One point I worry could confuse some readers. The authors use a total of 6 different terms for onset: onset, clinical onset, symptom onset, disease onset, actual onset, predicted onset, and estimated onset. Actual refers to a hard fact: someone got sick. Predicted and estimated onset both refer to someone who was still healthy the last time the authors saw them, and is therefore probably still healthy today, but who based on their mutation and age is predicted to develop disease X years from now. Age of onset is highly variable in genetic prion disease, so such predictions are very loose. The terms onset, clinical onset, symptom onset, and disease onset all seem to be used to refer to either actual or predicted, depening on context. To me, what’s really important to know about any at-risk sample is for how many years of follow-up they remained healthy afterwards. This number is not always easy to discern. Consider the following sentence from the abstract:
RT-QuIC… detected seeding activity in four CSF samples from three PRNP E200K carriers in the presymptomatic phase, one of whom converted shortly after but the other two remain asymptomatic after two and three years of follow up
So three E200K carriers were RT-QuIC positive at a pre-symptomatic stage. Trying to find the underlying data for these three patients in the text is difficult. The one who became sick very soon must be the one who developed disease after 0.2 years according to p. 15 right above Figure 3. But the others? Figure 3 describes a pair of samples from a single E200K carrier taken “7.1 and 5.1 years prior to estimated onset”. I think, but I’m not sure, that this refers to the person who remained healthy after two years of follow-up. The person who remained healthy after three years of follow-up is not mentioned anywhere in the Results text nor Figure 3 and must therefore be the person from Figure 2A, an E200K carrier whose sample the figure legend says was “drawn at an estimated 8.3 years from disease onset”. This person resurfaces in the discussion:
The E200K presymptomatic seeding period (as early as 8.3 years before predicted onset) appears unexpectedly long for an illness with such an explosive onset and short duration.
This sentence illustrates why the distinction between years of follow-up and predicted onset is absolutely critical. It would be so easy for a reader to become confused and think that RT-QuIC was positive 8.3 years before onset. In fact (if my reading is correct) all we know is that RT-QuIC was positive a minimum of 3 years before onset. I agree that even 3 years is surprising, for a disease with explosive onset and short duration. But 8.3 years would be yet more surprising.
In Figure 1, the times of sample collection are described for all individuals, with a single dashed vertical line indicating either expected onset or actual onset depending on whether the horizontal line representing that person’s life crosses said vertical line. In the grouped analysis of Figure 4, they group together “at-risk individuals less than two years to predicted/actual clinical onset”. What readers need to keep their eye on is the fact that these things being grouped together have very different meanings: actual onset being a person who actually got sick, and predicted onset being a very loose prediction. Indeed, it’s not clear to me how one ever gets a predicted onset <2 years away, because your expected onset is a function of what age you are today. In the E200K life table we constructed a few years ago, for instance, an E200K carrier who is 30 has an expected onset of 62, while an E200K carrier who is 60 has an expected onset of 68, and even an E200K carrier who is 80 has an expected onset of 83. Age of onset is so variable that it’s very rare for any individual’s expected number of years left to be less than 2. And, indeed, squinting at Figure 1, it appears that almost all of the “<2 years” group are actual onsets; I see only 2 or 3 dots that might be pre-symptomatic people with “predicted” onsets <2 years away. So why not just group together people <2 years from actual onset, and leave the “predicted” onsets out of it?
implications for trials and clinical care
This report is an important step forward. The patient longitudinal follow-up of pre-symptomatic individuals, meticulous sample collection, and broad array of biomarkers tested all contribute to giving us important new insights about potential prodromal changes in prion disease.
In terms of designing clinical trials, a lot of what I said back in 2020 still holds. Cross-sectionally, the number of mutation carriers in a prodromal state at any given time is pretty small. Here, they looked at 217 samples from individuals at risk for genetic prion disease collected over 14 years, and they saw 4 individuals with positive RT-QuIC (3 E200K and 1 P102L) and maybe 10 or 12 with “true positive” suspicious-looking NfL or GFAP trajectories. That suggests to me that if you looked in any one year, in this size cohort, you might find 2 or 3 individuals you would suspect were prodromal. You’re never going to power a whole trial that way. But, the information could still be really useful. I can imagine many people (pharma types, regulators) who might be swayed by evidence that you modulated a biomarker in just a handful of prodromal people, as an adjunct to data from non-prodromal individuals in a larger, better-powered trial. Moreover, a merciful trial designer might envisage an “off-ramp” whereby most pre-symptopmatic people screened into a trial are randomized (for a study brief duration, just to observe safety and target engagement, not to follow people to onset), but those who appear to be prodromal are siphoned out of the randomized cohort and into a compassionate or open label arm. Just an idea.
To be clear, in terms of clinical care, we’re still a much longer ways away (I think the authors agree). While the prior seems high, we still don’t have enough data to directly say that pre-symptomatic RT-QuIC predicts future onset, much less to say how many years out it predicts onset. While NfL and GFAP seem to have some predictive value in slow mutation carriers, provided you have baseline samples to compare to, they also shoot up and down enough that I think many clinicians would be reluctant to tell any individual patient that their onset is close at hand based on this information alone.
What remains to be done then? For one, we all need to follow a greater number of mutation carriers, for longer into the future, and collaborate more closely with one another to look at more markers in the samples collected. My wish is that everyone following pre-symptomatic PRNP mutation carriers can get plenty of funding to increase, expand, intensify. If grant reviewers say “but it’s been done”, you send them to talk to me. Meanwhile, even if, as this report suggests, RT-QuIC does prove sensitive to prodromal E200K and NfL/GFAP sensitive to prodromal P102L, there’s still a big gap we need to fill. For D178N and probably a handful of other mutations, we still really have no plausible prodromal markers to turn to. Even at the symptomatic stage, D178N people’s CSF is only occasionally positive by RT-QuIC [Sano 2013, Cramm 2015, Franceschini 2017, Foutz 2017, Rhoads 2020] and their plasma often has only a modest increase in NfL [Zerr 2018, Hermann 2022]. For all you methods development / biomarker discovery people out there, this is your call to action.