In my last post I discussed an exicting new development in the Huntington disease field: study has shown that it is possible to measure the levels of mutant huntingtin in patient cerebrospinal fluid (CSF) and that this level correlates with the patient’s clinical disease progression [Wild & Boggio 2015]. This has made me wonder whether it would be possible to do something similar for PrP in prion disease, so this post is a roundup of all the literature I could find on measuring PrP levels in CSF.

In principle, measuring PrP levels in CSF should be much easier than measuring huntingtin levels. Huntingtin is a mostly cytosolic protein, meaning it will be scarce in the extracellular environment, whereas PrP is usually anchored to outside surface of the cell. PrP is known to undergo at least three different proteolytic events: alpha cleavage, beta cleavage, and shedding. Each of these releases a different length segment of the protein into the extracellular environment, so if you’re going to measure PrP anywhere outside the cell - say, in the CSF - the measurement you get will depend on which antibody you use. The diagram below shows PrP bound to the cell surface by its GPI anchor, and shows the proteolytic events (top) and the epitopes of antibodies used in the studies reviewed in this post (where information was available; bottom).

Before setting out on reading this literature, it was not at all clear a priori which direction of effect to expect in prion disease - should PrP levels be lower or higher in symptomatic prion disease patients compared to healthy controls? There are a few countervailing factors here. On one hand, as reviewed here, figures from several studies indicate that total PrP levels in the brain increase continuously over the course of prion disease due to accumulation of PrPSc [Bueler 1994, Schulz-Schaeffer 2000, Safar 2005, Moreno 2012]. On the other hand, even as PrPSc accumulates, PrPC appears to be down-regulated [Mays 2014]. And even there, this down-regulation is post-translational, with the exact mechanism as yet unclear. So if the down-regulation is due to total degradation of PrP, then one would expect less PrPC in the extracellular environment, whereas if it is due simply to cleavage or shedding of PrP, one might expect more PrP (or at least N-terminal fragments thereof) in the extracellular environment.

An additional confounder that I originally wondered about, but which it looks like is not actually be an issue, is the centrifugation of CSF. CSF samples are often (but not always) centrifuged to pellet out debris prior to freezing and subsequent analysis of supernatant. For instance, in these studies, centrifugaton is performed in [Picard-Hagen 2006, Meyne 2009, Torres 2012] but not in [Schmitz 2010]. There are two reasons why I wondered whether this affects PrP recovery. First, PrP is abundant on exosomes [Fevrier 2004, and see this post] and PrP-positive exosomes are found in CSF [Vella 2008], so I wondered whether centrifugation would affect the recovery of exosome-bound PrP. But from reading a bit it seems that you really need ultracentrifugation (like 100,000g) to pellet out exosomes [Coleman 2012], whereas CSF is usually centrifuged at something more like 2,000g [Wild & Boggio 2015]. The second thing I wondered about was pelleting out of PrPSc, but after checking on this it looks like protocols to pellet out PrPSc require detergent and usually involve more on the order of 50,000g - for instance, the original Prusiner protocol [Prusiner 1983] had a step with 19,000 rpm on a Ti19 rotor, which according to the manual translates into 53,900g at the maximum radius.

Summary of literature

Here is a quick summary of the results of each study I found. I have included studies in humans and other animals, and studies in people with prion disease or other diseases. If I’ve missed anything in the literature, please leave me a comment below.

Prion protein in the cerebrospinal fluid of healthy and naturally scrapie-affected sheep [Picard-Hagen 2006]

This was the only non-human study I found [Picard-Hagen 2006] compared nine healthy vs. seven scrapie-infected sheep. They used an ELISA with antibodies 12F10-AchE (epitope not stated) and SAF34 (N-terminal octarepeats). No difference was found in CSF PrP levels (10.9±1.3 ng/mL scrapie-infected vs 9.6±0.81 ng/mL healthy). I didn’t see any note in the paper about how advanced the disease was in the infected sheep.

Total prion protein levels in the cerebrospinal fluid are reduced in patients with various neurological disorders [Meyne 2009]

Here, Inga Zerr’s group used an ELISA based on Eppendorf/Bio-Rad’s Platelia® BSE Detection Kit. The instructions for that kit originally include a proteinase K digestion step, which was omitted in order to measure total PrP. Frustratingly, neither this paper [Meyne 2009] nor Eppendorf’s very limited documentation state which antibodies are used in this kit, or even which part of PrP is targeted, and as far as I can tell, the kit is no longer commercially available. Since it is designed for detecting PK-resistant PrPSc fragments, and PrPSc’s N terminus is digested by proteinase K, the kit presumably must use antibodies against the C terminus of PrP, meaning that the fragments one would be quantifying in CSF would have to represent full-length shed PrP and not the N1 or N2 fragments released by alpha and beta cleavage respectively. In any event, the study found that PrP levels in people with four different neurodegenerative diseases - CJD, Alzheimer disease, Parkinson disease, and dementia with Lewy bodies - were slightly lower than in controls. This signal, though statistically significant at a population level (p < .001, one-way ANOVA), was both small in magnitude and quite noisy: the people with the various diseases had CSF PrP levels of 164±~80 ng/mL (mean±sd; the overall sd is not stated so I am guessing based on the within-group sd values listed in Table 1), and the controls had CSF PrP levels of 226±79 ng/mL. Thus, the standard deviation is 34% of the mean, and the difference in people with disease versus controls is only -27% - not quite the level of specificity you’d hope for in a biomarker. The paper discusses a possible cutoff of 200 ng/mL and says this cutoff would give diagnostic specificity in the range of 63% to 76% for the various diseases studied. There was also the frustrating confounder that PrP levels were positively correlated with age among female controls.

As an aside, I found it interesting to calculate the molarity of PrP in the CSF using the values reported here. If the concentration is say 200 ng/mL, that is 200e-6 g/L, divided by a molar mass of say 35 kDa = 35,000 g/mol gives 5.7e-9 = 5.7 nM PrP in the CSF. Multiply by 6.022e23 and divide by 1000 mL/L and you get about 5.1e12 = 5 trillion molecules of PrP in one mL of CSF.

Codon 129 polymorphism and the E200K mutation do not affect the cellular prion protein isoform composition in the cerebrospinal fluid from patients with Creutzfeldt-Jakob disease [Schmitz 2010]

This paper from Inga Zerr’s group used a different commercial ELISA kit, the BetaPrion® BSE EIA Test Kit. This one has more extensive documentation but sadly, again, no indication of which antibodies are used or even which epitopes are targeted. As with the Platelia kit above, I reason that if its purpose is to detect PK-resistant PrPSc then the kit must be using C-terminal antibodies, and thus cannot quantify N-terminal fragments released into CSF by alpha or beta cleavage. The ELISA kit was not a major part of this study - it was only used on CSF from one patient to get a ballpark estimate of CSF PrP concentration, which was estimated at 55 ng/mL - a fair bit lower than the range found in the earlier paper [Meyne 2009]. Most of the study is devoted to measuring the relative intensity of Western blot bands between people with different prion disease subtypes - the study compares a total of 41 patients with various forms of prion disease, and 17 controls. I won’t get into those results in detail in this post, other than a few points of interest. The blots probed with the N-terminal antibodies 8B4, 3B5, SAF32 and 8G8 are all cut off just below 25 kDa. Hopefully no one would cut off a band, so presumably this means that there was no reactivity at lower molecular weights, meaning that no N1 or N2 fragments were detected, only full-length shed PrP. This paper also cites what is apparently the first paper ever to report detection of PrP in human CSF [Tagliavini 1992], which found only a ~35 kDa band corresponding to full-length PrP. In contrast to the N-terminal antibodies, some of the central and C-terminal antibodies - 12F10, 1E2, 8H4, and particularly SAF70, detected lower molecular weight fragments that might correspond to the C2 fragment and other cleavage products. As for comparison of prion disease patients versus controls, there is a control in every Western blot, and the intensity of PrP banding in the controls generally appears (to my eye) comparable to that of patients overall, but the authors don’t make any quantitative comparisons, since a Western is a hard way to do this.

Altered Prion protein expression pattern in CSF as a biomarker for Creutzfeldt-Jakob disease [Torres 2012]

This study compared 46 prion disease patients (35 sporadic and 11 genetic) and 16 controls from Chile. The study used Western blots with the 3F4 antibody, evaluated by quantitative densitometry. Although it is generally difficult to use Western blots to make any true quantitative comparison, the differences between cases versus controls in this study are extremely large and visually obvious from just glancing at the blots (Fig 2 & Fig 3). The intensity of the PrP bands in cases appears to have been on average about 2 or 3 times lower in cases than in controls (Fig 2), though the ranges overlapped considerably.

CSF prion protein concentration and cognition in patients with Alzheimer disease [Schmidt 2013]

This paper from the Zerr group compared PrP levels in CSF of 114 Alzheimer disease patients. Again, they used the BetaPrion EIA ELISA kit from [Schmitz 2010]. Instead of comparing to controls, the goal of this study was to see if PrP levels were correlated with any cognitive measures among the AD patients. No correlation was found.

PrP mRNA and protein expression in brain and PrP(c) in CSF in Creutzfeldt-Jakob disease MM1 and VV2 [Llorens 2013]

One study looked at levels of PrP and its mRNA in various brain regions, and levels of PrP in CSF, in a series of sCJD cases (15 MM1, 10 VV2) and controls (n=15) [Llorens 2013]. It found that mRNA levels were just slightly decreased in the cerebellum in VV2 and in the frontal cortex in MM1 (Fig 1). This is contrary to the long-accepted view that PrP mRNA levels are unchanged during prion infection [Kretzschmar 1986] and one can speculate that either the small differences found in the present study are just random variation, or that the lack of difference found back in the 1980s was just because our ability to measure mRNA levels was too crude back then. As for PrP levels, these were quantified in brain using quantitative densitometry on Westerns with the 3F4 antibody, and in the CSF using that Platelia kit from [Meyne 2009]. Total PrP was reported to be slightly reduced in the cortex in MM1 and about doubled in the cerebellum in VV2, and reduced by about a third to a half in all sCJD (Fig 2).

Association of Cerebrospinal Fluid Prion Protein Levels and the Distinction Between Alzheimer Disease and Creutzfeldt-Jakob Disease [Dorey 2015]

This study quantified CSF PrP using, again, the BetaPrion EIA kit discussed earlier. Values are reported in μg/L, but since this is the same as ng/mL which the other studies used, I’ll use that unit for consistency. They compared a large cohort of controls (n=23), Alzheimer patients (n=131) and sporadic CJD patients (n=79). They found that PrP CSF levels in controls had a median of 300 ng/mL, which is slightly higher than [Meyne 2009]. In Alzheimer disease patients the average was slightly higher (median 319 to 501 ng/mL in various subgroups) but not significantly different. In prion disease patients the level was only about half what it was in controls or AD patients, with a median of 154 ng/mL. These differences were highly significant, but the ranges still overlapped considerably, and the best discrimination they could get in using this as a diagnostic criterion was 82% sensitivity and 91% specificity. This is more power than some of the above studies, but still less than reported for other diagnostic criteria diffusion-weighted MRI [Vitali 2011] or RT-QuIC [McGuire 2012, Orru 2014]. The authors also developed a formula (the “Creutzfeldt-Jakob factor”) combining the total PrP level with CSF tau measurements, and this formula gives a higher area under the curve than any single measurement. A commentary in JAMA Neurology [Zetterberg 2015] speculates that the reduction in CSF PrP levels in prion disease “may reflect sequestration of the protein in PrP aggregates in the brain with lower amounts being able to diffuse via the brain interstitial fluid to reach the lumbar CSF where it can be sampled and measured”, similar to how Aβ is reduced in the CSF during Alzheimer disease.

Note that this study is in agreement with [Meyne 2009] that CSF PrP is reduced in prion disease patients, but finds unaltered or higher levels in Alzheimer disease patients, in contrast to [Meyne 2009] who found that PrP was also reduced on those patients.

Preparing for therapeutic trials in human prion disease: clinical and laboratory approaches to improve early diagnosis and monitoring of disease progression [Thompson 2014]

This is the PhD thesis of Andrew Thompson, a student at MRC Prion Unit who worked with Simon Mead on the development of the MRC Prion Disease Rating Scale [Thompson 2013] and several other efforts to prepare for clinical trials in the U.K. The work in this thesis shows that MRC Prion Unit has taken a pretty different approach to detecting PrP in CSF compared to the other studies above. For a while now, the folks at MRC Prion Unit have been interested in the fact that prions seem to bind to steel wires [Edgeworth 2009, Edgeworth 2010]. In Thompson’s thesis, he uses a “Direct Detection Assay” (DDA) in which steel powder is added to tissue samples, incubated in a thermomixer, and then subsequently removed using a magnet. This is thought to enrich for PrPSc since apparently PrPC has little or no affinity for steel. They then use a biotinylated ICSM18 to bind the PrP, and an HRP-conjugated avidin to detect the biotin in an ELISA format.

Thompson applied the DDA to a variety of tissues including blood, brain, and CSF. In brain, it looks like the assay gave about 50% higher signal in sCJD brain homogenate than in normal brain homogenate (Figure 23) which might reflect the higher accumulation of total PrP in brain during prion disease (discussed at top) or some specificity of steel for PrPSc. When brain homogenates were spiked into normal CSF, the difference between sCJD and normal brain homogenate diminished somewhat. When the assay was actually run on diagnostic CSF samples without any spike-ins, the assay detected an almost categorically lower signal in prion disease patients (n=4 sCJD and 1 vCJD) than in controls (other neurodegenerative diseases, n=11) for a p value of .0087 - see Figure 25. The fact that prion disease CSF was lower in signal than control CSF is perhaps unsurprising in light of the studies above, which have tended to find lower total PrP in CSF during prion disease, but it is surprising in light of the view that steel preferentially binds PrPSc.


Taken together, these studies indicate that PrP levels in CSF are on average lower in individuals with prion disease than in controls. The one sheep study I read found no difference [Picard-Hagen 2006], but the human studies are in general agreement that total CSF PrP is about 25-50% lower in symptomatic prion disease patients than in controls [Meyne 2009, Torres 2012, Llorens 2013, Dorey 2015]. The studies do not all agree as to whether this reduction in PrP level is specific to prion disease [Dorey 2015] or occurs in other neurodegenerative diseases as well [Meyne 2009]. In general, the reduction in PrP level associated with prion disease, though statistically significant at a population level, is quite noisy and at least on its own, does not appear to give strong specificity as a diagnostic criterion.

It is not clear how much the noisiness of the signal results from genuine variation between people or samples, versus noisiness of the ELISAs or Western blots used to measure PrP. But even if (optimistically) the noise is mostly a technical issue that could be minimized with further tweaking, the direction of effect observed - lower PrP in people with prion disease - means that PrP level in CSF is not an ideal biomarker of prion disease for any therapeutic clinical trials. Knocking down PrPC expression is an appealing mechanism for a potential drug, because PrP is required for prion disease pathogenesis [Bueler 1993] and reducing or eliminating PrPC in animals can slow or reverse disease [Safar 2005, Mallucci 2003]. That’s why so many groups have tried to knock down PrP using RNAi [Pfeifer 2006, White 2008, Pulford 2010, Lehmann 2014, Ahn 2014] or ASOs [Nazor Friberg 2012] or have screened for small molecules that reduce PrPC levels [Karapetyan & Sferrazza 2013, Silber 2014]. If such a drug were ever to be effective, on one hand one would expect it to push PrP levels in prion disease patients’ CSF even lower than they already are, but on the other hand, if this then caused the patient’s condition to improve, their CSF PrP levels would be expected to increase back towards the level seen in controls. Because of these two countervailing forces, it would be very difficult to tell whether a PrPC-lowering drug was working by measuring total PrP in CSF.

Perhaps the more promising biomarker for a therapeutic trial, then, would be to use RT-QuIC (or PMCA, if that’s your style) to quantify prions in patient CSF. There are ample data out there showing that CSF samples from prion disease patients are positive for RT-QuIC seeding activity, while CSF samples from healthy controls or other neurodegenerative diseases are negative [Atarashi 2011, McGuire 2012, Orru 2015]. What has not yet been explored, to my knowledge, is the status of CSF from asymptomatic carriers of pathogenic PRNP variants - do people develop seeding activity in their CSF before presenting with clinical disease? If so, how long beforehand? And would the number of prion seeds be predictive of time to disease onset? Meanwhile, even for symptomatic patients, it is not yet clear whether CSF seeding activity could ever be a therapeutic biomarker. It is well-documented that, in the mouse brain at least, prion infectivity plateaus at clinical onset [Prusiner 1982, Sandberg 2011, Sandberg 2014, and see this post]. Seeding activity is not identical to prion infectivity, but certainly the two are related. The one time series study of RT-QuIC that I have read, on hamsters, was somewhat ambiguous as to whether CSF or brain seeding activity continues to rise after clinical onset [Orru 2012]. It remains to be seen whether successful therapeutic intervention would actually reduce CSF seeding activity in animals or humans that are already symptomatic.

All of these unknowns will be important things to test over the coming years to establish whether RT-QuIC or another amplification technology could be used as a biomarker for prion disease in clinical trials. As for total PrP levels in CSF, my reading has convinced me that it is unlikely to be the biomarker we are looking for to enable therapeutic trials, though I welcome any opinions to the contrary - just leave a comment below.