antisense oligonucleotides: probably precluded by adverse reactions

A colleague recently pointed me to a recent paper on ASOs in prion disease that I had missed on my first scan of the literature [Nazor Friberg 2012].  Earlier studies had shown that chemically modified ASOs (phosphorothioate oligonucleotides, which are nuclease resistant and so more stable in cells [Bennett & Swayze 2010]) reduce PrPSc formation regardless of sequence in cell culture [Karpuj 2007] and in mice [Kocisko 2006].  That’s right – even if not designed against PrP mRNA, their chemical structure itself is somehow inhibitory of prion conversion.  Kocisko provided some evidence that this may be because these ASOs bind PrP^C and induce its endocytosis.  Kocisko also did therapeutic trials using random ASOs and found significant effects for prophylactic treatment or co-incubation of inoculum with the ASOs.

Nazor Friberg’s study appears to be the first to look at ASOs designed specifically to target the PrP mRNA sequence.  The authors screened 78 ASOs designed for complementarity to PrP mRNA’s 3′UTR; the most effective ASO they identified achieved a knockdown of 97% in cell culture.  Curiously, the maximum inhibition of protein (as opposed to mRNA) was 84% and was achieved only after two weeks of incubating cells with the ASOs – a timeframe much longer than the mRNA and protein half-life would lead us to expect – does this reflect slow uptake of ASOs?  In vivo, with intraventricular infusion, the best ASO was able to achieve a ~70% knockdown of PrP – both of mRNA and of protein – in brain regions immediately adjacent to the infusion site.

In a survival study, mice infused with ASO at 1 day post infection (dpi) appear to have lived to an average of something like 200 dpi (see Fig 6; exact average is never stated) compared to 136 dpi for controls, so a ~50% extension of survival.  The dose given was 100 times lower than used in the earlier studies of generic (non-sequence-specific) ASO interference with PrP^Sc formation, so the effect is presumably due to specific PrP mRNA degradation.  Troublingly, though, mice treated at 60 dpi (i.e. deep into prion infection but still before obvious symptoms) had a severe adverse reaction and had to be sacrificed.  The reasons for this reaction are not clear.

Intracerebral infusion of ASOs is invasive enough that it’s never going to be something that healthy volunteers – even PRNP mutation carriers – would sign up for.  As of now, it seems ASOs’ only realistic route to becoming a treatment is if they work in symptomatic prion disease patients, and the adverse reactions in prion-infected mice make it unlikely that ASOs could be used as treatments so late in the disease course.

Of course, since we don’t know exactly what triggered the reaction at the molecular level, it’s too early to say whether it’s an insurmountable problem.  Perhaps a different delivery method, or a different chemical modification to the ASO backbone, would avoid the adverse reactions.  The biggest news in gene therapy of late is that Roche and Isis have announced they’re investing $33M to bring anti-huntingtin ASOs to clinical trials for Huntington’s Disease [HDBuzz], following on last year’s landmark demonstration of feasibility and efficacy in animals [Kordasiewicz 2012].  Roche is talking up a ‘brain shuttle’ technology for ASO delivery across the blood-brain barrier without intrathecal or intraventricular infusion, though so far the web seems to be scant on details of what this ‘shuttle’ consists of.  In any event, there is sure to be a lot of exciting development on ASOs for the brain over the coming years, and it’s worth staying tuned.  But for now, at least, ASOs don’t seem to be the most promising path forward for prion disease.

maximizing the efficacy of RNAi

That leaves RNAi as the other method of antisense gene therapy.  The major study to date of RNAi in prion disease [White 2008] achieved an 18% extension of survival with one lentiviral shRNA injection late in disease course.  That’s pretty good – treatments are almost never effective late in prion disease – but 18% is also far from being a cure.  Does that mean RNAi is not so promising either?

Not so fast.  White’s study was an excellent demonstration of feasibility and a good piece of evidence for Mallucci’s ‘window for intervention’ hypothesis.  But that study wasn’t necessarily designed to all-out maximize therapeutic effect.  Here are a few reasons why:

• siRNA screening.  The siRNA used in this study, dubbed MW1, only achieved about an 80% knockdown of PrP mRNA in cell culture (N2a cells, see Fig S1) and in the mouse hippocampus (Fig 1).  80% mRNA knockdown is pretty good for an siRNA, on the high end of what you’d expect from an off-the-shelf siRNA that you order online, but by no means the best of the best.  It’s not clear how much time White and Mallucci spent screening siRNAs to find the best sequence – all that the paper says is that “siRNA sequences targeting PrP were screened in vitro for efficacy of knockdown (data not shown)”.  With extensive screening to identify the most potent siRNA sequences, knockdown levels of well over 90% are possible.  Alnylam, a startup in Cambridge, MA focused on delivering synthetic siRNAs (i.e. direct siRNA delivery, no viral vector) has claimed 99% knockdown of transthyretin mRNA and protein in the mouse liver for ALN-TTR01.  Arrowhead Research has made similar claims of 99% knockdown of target genes in monkeys [Wooddell 2013 (ft)].
• dosing. White’s study only examined one dose, delivered in a one-time injection into one brain region (the hippocampus).  Testing multiple doses would allow one to build out a dose-response curve and be able to test a dose with maximum efficacy.  Multiple timepoints and/or multiple injection sites could also help to achieve greater therapeutic effects, and none of that was explored in this study.  If intracerebral RNAi ever makes it to the clinic for prion diseases, it’s likely to involve multiple injection sites – see for instance the 12 injection sites used in a current Batten Disease clinical trial.
• delivery.  By far the biggest problem in RNAi for the CNS is delivery – distribution across the brain.  Indeed, the 18% effect seen in White’s study is surely an overestimate of what a similar delivery technique would achieve in humans, for the simple reason that mouse brains are small and human brains are huge.  Indeed, this is a problem not just for RNAi but for any intracerebral delivery of therapeutics.  Intraventricular infusion, for instance, may penetrate 2mm into tissue from the ventricles, which would cover a large fraction of the mouse brain and negligible fraction of the human brain.  This is one factor that may have contributed to the failure of pentosan polysulfate in humans with CJD.  Luckily, RNAi delivery is under heavy investigation for a variety of diseases, and efforts in prion diseases will be able to benefit from those discoveries.  One simple (and by no means new) technique is ‘convection-enhanced delivery’, i.e. slow, pressurized injection, which can increase diffusion across brain regions and has demonstrated in monkeys and in human patients [Bankiewicz 2000 (ft), Voges 2003].  There are also efforts to develop ways to deliver RNAi (and drugs) to the brain from the periphery, for instance with lipid nanoparticles [reviewed in Bondi 2012].  So far none of these methods have proven very effective, but keep your eye on the Roche/Isis anti-huntingtin ASO developments.

outlook: how promising is RNAi?

We know from studies of Tet-off PrP mice that even 10% of wild-type PrP levels is sufficient to sustain a lethal prion infection, albeit with dramatically lengthened incubation times [Safar 2005 (ft)], while Cre-mediated knockout (reducing PrP probably very close to 0%) cures prion infection [Mallucci 2003].  There is presumably a threshold somewhere in between 90% and 100% knockdown where symptoms would be reversed and a patient would not experience prion toxicity within their lifetime.  As of now we don’t know if that threshold is something like 95% or something more like 99.999%.

As mentioned above, the best siRNA sequences can achieve as much as 99% knockdown of targeted transcripts and their protein products.  But even with such an optimized anti-PrP siRNA and optimistically assuming that the magic threshold to reverse prion infection is only 95% knockdown, RNAi is probably not poised to be an outright cure for prion disease given today’s technology.  The human brain is large and not super permeable to viruses, ASOs or synthetic siRNAs; no one has yet been able to demonstrate a uniform high distribution of gene therapy molecules across the whole brain.  Optimistically, 99% knockdown might be achieved adjacent to an injection site, but prion infection would still proceed elsewhere.

But that’s just using today’s technology.  The point of research is to make things better.  And in the scheme of neurodegenerative diseases, prion disease is actually a pretty good candidate in which to move RNAi forward, for a few reasons.  Prion diseases are entirely untreatable at present, and very rapidly fatal.  An 18% increase in survival may not seem like much, but it’s an entirely reasonable endpoint for a clinical trial.  For instance, earlier this year the FDA approved the use of Kadcyla (Herceptin with a toxin conjugated to it) for breast cancer based on an average survival of 30.9 months compared to 25.1 months for patients treated with regular Herceptin and other drugs – that’s a 23% extension of survival.  Or for an example from neurodegenerative disease, the FDA approved riluzole for ALS in 1995 based on a ~3 month increase in survival, against a background of about a 4 year life expectancy from diagnosis – that’s something like a 6% extension of survival.  The ASO and RNAi treatments currently under investigation for Huntington’s, Parkinson’s, and Alzheimer’s may well prove to have similar effects, but because those diseases progress slowly, it will be years before we know.  In comparison, average survival with sCJD is 7 months [Pocchiari 2004 (ft)].  The variance on that number is fairly high and depends on codon 129 genotype and a few other things, but still, if RNAi were to hypothetically increase survival by just a few months, then a clinical trial would probably be able to determine that result in a year or so.

The rapid mortality of prion diseases may also mean that the inherent risks of RNAi are more tolerable than in a disease like Huntington’s where people can live with symptoms for many years.  The experience with pentosan polysulfate has shown that families are not just willing to accept the risks of intracerebral infusion, but in fact will actively campaign to be allowed to take on risky treatments for their loved ones.

As of today, RNAi is not positioned to be a cure, though that could change if delivery systems improve and if the ‘reversal threshold’ proves to be low enough.  Cure or not, it does have potential to be a treatment for a class of diseases that have so far proved untreatable.  And we don’t know yet, but it’s possible that even a moderate knockdown of, say, 50% or 80%, could prove highly effective if combined with other treatments such as the newly introduced anle138b.  As of today, we are just at the beginning of having treatments that look promising in mice, and researchers have not even begun to explore what could be achieved by combining different treatments.  The road to the end of prion diseases may involve a few different strategies, and RNAi could yet prove to be one of those.

defining the prion promoter

The first effort to characterize the promoter region of PrP concluded that a 273bp region was necessary for regulating PrP transcription [Mahal 2001].  Mahal undertook a series of deletion experiments, transfecting HeLa cells with plasmids that expressed luciferase under various deletion mutants of the sequence near the PrP transcription start site (TSS), in order to identify the crucial region without which luciferase was not expressed.   The resulting 273bp region was defined as “-148 to +125, relative to the cap site”.  ”Cap” appears to refer to the 5′ cap of the PRNP transcript, i.e. the transcription start site.  In human genome hg19 coordinates, PRNP RefSeq transcript variant 1 is transcribed from chr20:4666797-4682234.  (Note the RefSeq link gives length and sequence for the mRNA, while the UCSC link shows the entire pre-mRNA.)   If chr20:4666797 were indeed the  TSS to which Mahal referred, then the promoter would extend from chr20:4666649-4666922, but when I get DNA for that region it doesn’t quite match Mahal’s DNA, so Mahal must have been working from a different transcription start site.  Instead, the 273bp promoter identified by Mahal appears to be located at precisely chr20:4,666,877-4,667,149.  Here’s the sequence:

>hg19_dna range=chr20:4666877-4667150 5'pad=0 3'pad=0 strand=+ repeatMasking=none
CAAGCGAATCTCAACTCGTTTTTTCCGGTGACTCATTCCCGGCCCTGCTT
GGCAGCGCTGCACCCTTTAACTTAAACCTCGGCCGGCCGCCCGCCGGGGG
CACAGAGTGTGCGCCGGGCCGCGCGGCAATTGGTCCCCGCGCCGACCTCC
GCCCGCGAGCGCCGCCGCTTCCCTTCCCCGCCCCGCGTCCCTCCCCCTCG
GCCCCGCGCGTCGCCTGTCCTCCGAGCCAGTCGCTGACAGCCGCGGCGCC
GCGAGCTTCTCCTCTCCTCACGAC

As far as I can tell, the term ‘prion promoter’ is used loosely, and may refer to exactly this 273bp, or to a larger or smaller region.  Borchelt 1996 pioneered the use of a much larger swath of PrP regulatory elements as a promoter to drive expression of transgenes in mouse models of other human diseases (unrelated to PrP), for instance the N171-82Q Huntington’s Disease mice.   And Asante 2002 reported using a yet smaller, 214bp prion promoter to drive transgenes.  Transcriptional regulation is complex: regulatory elements in DNA can exist far upstream or downstream of the transcription start site, and we don’t always understand how they play their regulatory role, so it’s unlikely that Mahal’s critical region is the only region that matters.

Mahal’s bioinformatic analysis predicted binding sites in the prion promoter for Sp1, a transcription factor involved in embryonic development, along with AP-1 and AP-2.  But as far as I can tell, no one actually tried to determine experimentally which transcription factors bind, or how PrP transcription is regulated, until a few years later.

stress response

The earliest study I could find that addressed PrP transcription [Shyu 2002] found that PrP protein was upregulated in response to heat shock (cells placed at 42 C).  PrP mRNA was never measured.  Instead, a reporter construct with luciferase under the prion promoter was used to confirm transcription as the mechanism of upregulation.  A predicted binding site for heat shock transcription factor 1 (HSF1) was found in the promoter, and when this binding site was deleted, the luciferase response was greatly reduced.  Shyu then looked for HSF1 binding to the prion promoter using an electrophoretic mobility shift assay (EMSA) [reviewed in Hellman & Fried 2007], and confirmed binding.

Years later, Steele 2008 (ft) infected HSF1 knockout mice with prions and found that they succumbed to prion disease about 20% sooner than wild-type mice.  Curiously, they did not have elevated PrP^Sc production nor infectivity nor earlier onset of symptoms – only fatality was accelerated, implying perhaps a reduced ability to cope with the PrP^Sc.  In any event, Steele found no difference in baseline PrP levels in the brains of the HSF1 knockout mice vs. wild-type.  That doesn’t rule out a role for HSF1 in regulating PrP under heat shock conditions, but it does suggest HSF1 is not important for PrP transcription under normal conditions in vivo.

Another study [Wang 2005] found slightly elevated PrP protein and mRNA following exposure of N2a cells to oxidative stress agent nitric oxide (NO) and the inflammatory agent lipopolysaccharide (LPS).  NO was hypothesized to act through a guanlyl cyclase → MEK → p38 MAPK pathway, and pharmacological inhibitors of these proteins reduced PrP levels. The authors were not super rigorous about proving that this pathway regulates PrP: there was no knockdown and rescue, no confirmation of DNA binding, and many of the effects were of marginal statistical significance.  Another study reported that pharmacological MEK inhibitors abolished PrP-res, in cell culture [Nordstrom 2005 (ft)].  In Nordstrom’s Fig 3 you can see that both PrP-res and total PrP are reduced after treatment.   But there was no deeper investigation as to whether the MEK inhibitors acted through a transcriptional mechanism.

Another study hypothesized that the unfolded protein response (UPR) transcription factor XBP-1 might be involved in prion pathogenesis, but found no difference in disease course nor PrP levels in XBP-1 knockout mice [Hetz 2008].  A later study suggested that, at least in breast cancer, ER stress does induce PrP transcription via XBP-1 and BiP [Dery 2013 (ft)].  PrP levels are elevated in several cancers, and high PrP levels are associated with a poorer prognosis, reportedly because PrP promotes cell survival.  Dery found that ER stress-inducing agents caused an increase in XBP-1 mRNA and prion protein levels, and one experiment with siRNA against XBP-1 reduced PrP levels.  Mutation of predicted ER stress response element (ERSE) sequences in the prion promoter reduced the PrP transcriptional response.

copper response

It’s been known for a while now that PrP^C binds Cu²⁺ ions at its PHGGGWGQ octapeptide repeats [Brown 1997, Stockel 1998] and that Cu²⁺ binding causes PrP^C endocytosis [Pauly & Harris 1998].

It was later shown that exposing primary cultured neurons to copper causes an increase in PrP mRNA and protein [Varela-Nallar 2006 (ft)].  This led to a natural hypothesis as to what proteins might be regulating transcription.  A class of proteins called metallothioneins, which manage heavy metals, are regulated by metal transcription factor 1 (MTF-1) [Heuchel 1994 (ft)], which binds to a a sort of loosely defined DNA sequence motif called a metal response element (MRE) [Giedroc 2001].  Varela-Nallar describes the MRE motif as “a highly conserved 7-base pair functional core motif [5′-TGC(A/G)CNC] flanked by a less conserved 5-base pair GC-rich domain”.  Varela-Nallar identified a few sites that almost matched the MRE motif, located ~2000-2500 bases upstream of the PRNP TSS, and through deletion experiments (similar to Mahal’s method) found that one of these MREs (“−2,653 TGCGtCCCCTGC”; I was unable to find this sequence near PRNP in hg19), as well as some nearby non-MRE regions, appeared to be pretty important for copper-induced PrP expression.  Varela-Nallar then looked for MTF-1 binding to that particular 12bp sequence using an electrophoretic mobility shift assay (EMSA) [reviewed in Hellman & Fried 2007].  The EMSA did not reveal any MTF-1 binding to this particular 12bp sequence; also, zinc tends to activate MTF-1 yet had no effect on PrP levels.  For both of these reasons, Varela-Nallar concluded that MTF-1 was not involved in the transcriptional regulation of PrP.

A more recent study has disagreed [Bellingham 2009 (ft)].  Bellingham’s study relied on cultured fibroblasts from a patient with Menkes disease.  Menkes disease is an X-linked disease caused by loss-of-function mutations in the gene ATP7A, which encodes Menkes protein, a P-type ATPase which pumps copper out of the cytosol and into the secretory pathway [Barnes 2005 (ft)].  Patients with Menkes disease accumulate excess intracellular copper.  Bellingham compared the original unaltered patient fibroblasts, dubbed MNK(Del), to ones transfected with a vector overexpressing ATP7A, dubbed MNK(++), as well as an empty vector control, MNK(v/o).  The comparison of these lines was striking: PrP expression appeared to be completely abolished in the MNK(++) lines – neither the protein nor mRNA were detectable at all (Fig 3).  Bellingham hypothesized that Sp1 and MTF-1 were involved in copper-dependent regulation of PrP and sought to confirm this experimentally.  In the MNK(Del) cells, knocking down Sp1, MTF-1 or both reduced PrP levels, and transfecting the cells with additional Sp1, MTF-1 or both increased PrP levels.  The MNK(++) cells stubbornly refused to express any PrP at all under any conditions studied.

The most striking aspect of Bellingham’s study is the suggestion that copper is absolutely necessary for PrP transcription – at least in fibroblasts, under the conditions studied, etc.  If it’s true, it begs the question of what transcription factors mediate this copper dependence.  MTF-1 is well known to respond to copper [Heuchel 1994 (ft)], and interestingly, it appears to be on call for both copper surplus and copper deficit, which it manages through transcription of different genes, at least in Drosophila [Selvaraj 2005 (ft)].  Sp1 has also been reported to respond to copper [Song 2008 (ft)].  Bellingham didn’t do an EMSA or any other experiments to confirm that either of these proteins actually binds to any location in the prion promoter.  However, the finding that overexpression and knockdown both affected PrP levels is compelling functional evidence that they’re involved in some way, and seems to outweigh the earlier claim that MTF-1 is not involved, which was based on a lack of binding to just one predicted site [Varela-Nallar 2006 (ft)].

Another study reported a different copper-dependent transcriptional mechanism for PrP [Qin 2009 (ft)].  Qin examined the time course of PrP mRNA and protein elevation in response to Cu²⁺ in some detail and concluded, as these other authors have, that the PrP response to copper is transcriptional in nature.  Qin proposed a pathway whereby copper causes phosphorylation of ATM at S1981, and activated ATM then activates Sp1 through the MEK/ERK pathway, and activates p53 by phosphorylation at S15.  Qin provides some pretty rigorous evidence for this in Fig 4, by using siRNAs to knock down various elements of the pathway and show that this abolishes the response in all the downstream elements. Importantly, Qin also used an EMSA to show direct binding of both Sp1 and p53 to the prion promoter. Qin proposes that the upregulation of PrP in response to copper may constitute a stress response, since copper causes the creation of reactive oxygen species.

Qin’s finding that the MEK/ERK pathway is involved in PrP transcription has been subsequently validated and extended [Cisse 2011 (ft)].  Cisse found that depletion of ERK1 or dominant negative inhibition of ERK1 or MEK reduced PrP mRNA levels, and transfection of cells with ERK1 or MEK increased PrP transcription.  Cisse notes that ERK1 activation can promote gene transcription either by phosphorylation of Sp1 or by phosphorylation of c-Fos, part of the AP-1 complex.   Mutation of the predicted AP-1 binding sites also reduced PrP transcription, but mutation of the predicted Sp1 binding sites had no effect, suggesting AP-1 was responsible.  Interestingly, ERK1 also  controls PrP proteolytic cleavage by ADAM10 [reviewed in Checler 2012].

amyloid intracellular domain

Here’s a lightning-quick review of the metabolism of amyloid precursor protein (APP).  Beta-secretase (BACE1) cleaves it on the exoplasmic side, and gamma-secretase (including PSEN1 & PSEN2) cleaves it on the cytosolic side, producing an extracellular fragment amyloid beta (Aβ) with roles in Alzheimer’s disease, and an intracellular fragment called the amyloid intracellular domain (AICD) [Vassar 2009].

PrP and amyloid beta interact physically, leading to downstream signaling changes that are only now beginning to be understood – see my latest PrP/Aβ post.  PrP also regulates the production of Aβ through its inhibition of BACE1 [Parkin 2007, Rushworth 2013].  It turns out that APP metabolism may also regulate PrP: it has been reported that the AICD activates transcription of PrP [Vincent 2009 (ft)].

Vincent found that knockout of PSEN1 and PSEN2 in fibroblasts reduced PrP mRNA and protein by about half.  betaAPP overexpression increased PrP, betaAPP knockout reduced PrP, and direct expression of AICD fragments C50 or C59 increased PrP.  Vincent proposes that p53 mediates the AICD’s effects on PrP.  p53 knockout reduced PrP expression (rescued by p53 cDNA transfection) and abolished the stimulatory effects of C50 and C59; ChIP confirmed direct binding of p53 to the prion promoter.  Vincent’s group had previously reported that AICD activates transcription of p53 [Alves da Costa 2006], so this is postulated as the mechanism for the AICD → p53 → PrP connection.

no unbiased screens so far

All of these studies started from a particular hypothesis about what regulates PrP transcription and then sought to prove it.  Most started from completely different hypotheses and did not explore or directly seek to confirm one another’s findings.

No study has yet utilized an unbiased screen for proteins binding to the prion promoter.  One unbiased method for identifying regulators of a particular promoter is to prepare a biotinylated synthetic DNA oligomer of the desired promoter, dip it into a nuclear extract, and then perform mass spectrometry on the proteins that come out bound to the DNA [Nordhoff 1999].  This method has its pitfalls – it by no means captures the range of physiological conditions under which transcription might be stimulated – but at least it is unbiased and has the potential to catch things that no one has thought to hypothesize.

Something big and unbiased has happened since 2009, though: the ENCODE project.

ENCODE data

ENCODE (Encyclopedia of DNA regulatory elements) aims to create a genome-wide map of everything that regulates gene expression: histone modifications, DNA methylation, and transcription factor binding among other things.  Most of the data comes from hundreds of genome-wide ChIP-seq experiments.  DNA-binding proteins are cross-linked to DNA with formaldehyde, one protein of interest is pulled down with antibodies, de-cross-linked, and then the DNA that had been attached to it is sequenced.  This gives you a map of where in the genome the protein had been bound.

Though ENCODE is unbiased in that it is genome-wide, it is limited to the transcription factors that were chosen to be studied.  The full list of ChIP-seq experiments includes Sp1 but neither p53 nor MTF-1.  ENCODE currently has transcription factor binding data from 690 experiments on 161 transcription factors, but finding the data for all of them is confusing.  If you simply turn on the ENCODE Txn factor track in the UCSC genome browser, you’ll see only 26 of the 690.  If you instead visit the Uniform TFBS Track page you’ll be confronted with the entire list of 690 and no way to display data from all of them except by checking each and every box.

The easiest way I’ve found of getting the data for all 161 transcription factors is through the UCSC table browser, settings as shown below. I used chr20:4665648-4668377, so ~2700bp, a 10x zoom-out from Mahal’s promoter region.

Results are here: [TXT].  The table contains 48 entries, apparently only the transcription factors with scores > 0.  Three transcription factors got the maximum possible score of 1000: JunD, FOSL2 and c-Fos.  All three of these are components of the AP-1 transcription factor complex, thus confirming Mahal’s original bioinformatic prediction of AP-1 binding [Mahal 2001].  The thickStart and thickEnd columns tell you where exactly the peaks are; FOSL2 and c-Fos both overlap Mahal’s 273bp promoter region of chr20:4666877-4667149.  Besides the AP-1 proteins, STAT3 also comes close to the maximum, with a score of 950.  Sp1 is not on the list, implying it had a score of 0.  Some AP-2 transcription factors are present lower on the list.

summary

This table attempts to summarize the findings from the above studies.

If we very generously (i.e. unskeptically) believe everything that’s been reported, the picture of prion protein transcription that has been developed so far looks as follows:

If we’re more skeptical and only consider those things which (A) are supported by more than one study and (B) for which both physical and functional evidence are available, the picture thins considerably:

In either case, what we have at this point is certainly a very incomplete picture of how PrP transcription is regulated.  Most of these studies focused on transcriptional upregulation in response to a particular environmental stress, rather than characterizing what drives baseline transcription levels.  Only recently [Cisse 2011 (ft)] was AP-1 activity implicated in the literature, even though according to ENCODE it is the most active transcription factor in the prion promoter.

The most important piece that is missing is what drives neuronal transcription of PrP.  PrP  mRNA is expressed ~10 times higher in the CNS than in most peripheral tissues [Novartis BioGPS microarray data], implying there must be some neuronal-specific transcription factors, upstream regulatory mechanisms or epigenetic marks at work.  Of the above-cited studies, just one used primary cultured neurons [Varela-Nallar 2006 (ft)] and two used CNS-derived cancer cell lines (N2a and NT-2) [Qin 2009 (ft), Shyu 2002 respectively].  The rest used fibroblasts or peripheral cancer cell lines.  Browsing the ENCODE human cell experiment list, I also don’t see any CNS cell types there.

assessment of therapeutic potential

It’s unclear if any of these pathways lend themselves to therapeutic intervention.  The most potentially interesting finding is that cytosolic copper depletion by overexpression of ATP7A reduces PrP expression to undetectable levels [Bellingham 2009 (ft)].  If this result (originally from fibroblasts) could be replicated in neurons, and could then be associated with the inactivation of one particular transcription factor or one particular protein upstream thereof, that could potentially constitute a drug target.  If Bellingham is right that MTF-1 is involved in some way, it is interesting to note that MTF-1 knockout is embryonic lethal, perhaps due to liver decay [Gunes 1998 (ft)] but that Cre-mediated adult knockout in the liver is survivable [Wang 2004 (ft)].  As far as I can tell from the literature no one has done a Cre knockout of MTF-1 in neurons, and no one has checked whether PrP is expressed in MTF-1 knockout cells.

Transcription factors are not the ideal drug targets – they tend to promote hundreds of genes, so inhibiting them to turn off one substrate is truly razing a village to catch one man.   But blocking a transcription factor is not entirely unheard of – for example disulfiram inhibits NF-KB [Schreck 1992 (ft)] and has been explored for treatment of HIV and cancer, though its relevant mechanism of action in those diseases may be something different [Lovborg 2006 (ft), Doyon 2013].  Also, disrupting a particular DNA binding event or a recruitment interaction between two transcription factors can be more specific than just inhibiting a transcription factor altogether.

what is glycosylation?

Glycosylation is the attachment of sugar chains to proteins.  Proteins with sugar chains attached are called glycoproteins, and the sugar chains are called glycans.  There are a few different ways that a sugar chain can be attached to a protein; of relevance here is N-linked glycosylation, where the glycan is covalently bound to a nitrogen (N) atom in an asparagine (N) amino acid in the protein.

Glycans attached to proteins can be quite diverse.  This image conveys just a few of the major groups of glycans.  (adapted from Berninsone, P.M.):

Glycans can differ on several dimensions: the length, branching structure, and the particular sugar molecules at each position.

brief review of PrP cell biology

To understand PrP’s glycosylation, it’s useful to have a quick background on PrP’s cell biology.  For more details, see my secretory pathway notes or [Harris 2003 (ft)].  Here we go in brief.

Human PrP is 253 amino acids long, but the first 22 are a signal peptide which causes the mRNA/ribosome complex to get moved, mid-translation, to the endoplasmic reticulum (this is ‘cotranslational translocation’).   Those 22 amino acids (MANLGCWMLVLFVATWSDLGLC) get cleaved off by signal peptidase and the protein continues to be translated directly into the ER.  While it’s still being translated, oligosaccharyl transferase (OST) adds glycan chains at 0, 1 or 2 of 2 possible sites.  In general, OST recognizes two amino acid motifs – NXS and NXT – as its sites for attaching glycans.  In human PrP these occur at codon 181-183 (NIT) and 197-199 (NFT).  But OST’s hit rate is not 100%.  Any given molecule of PrP might end up with glycan chains attached at codon 181, codon 197, both, or neither.

The C-terminus of PrP is a 23 amino acid sequence (SMVLFSSPPVILLISFLIFLIVG) which signals for the addition of a GPI anchor.  In this process, called glypiation, these 23 amino acids are cleaved off and replaced with GPI, which is itself another kind of sugar chain, which embeds in the membrane, thus anchoring PrP to the ER membrane.  Depending who you ask, glypiation may also be called ’glycosylation’, but is not the subject of this post.  This post is about the N-linked glycosylation at codons 181 and 197.  Once the signal peptide and GPI signal have both been cleaved, the initial 253 amino acid protein is just 208 amino acids.

From its beginnings in the ER, PrP transits through the secretory pathway, acquiring a disulfide bond between its two cysteines (C180 and C214) and undergoing several modifications to its N-linked glycan chains.  Secretory pathway proteins with a DXE signal, called a ‘di-acidic’ signal since D is aspartic acid and E is glutamic acid, get exocytosed.  PrP has a DYE at codons 144-146, and thus travels from the Golgi to the cell surface in an exocytic vesicle.  Once there it remains GPI-anchored to the exoplasmic (outside) surface of the plasma membrane.

how people study PrP glycosylation

Any two molecules of PrP could differ from one another in glycosylation state, in two different ways.

First, as mentioned above, PrP can be glycosylated at N181, N197, both, or neither.  That’s four different states.  But the way that these can be studied in the lab is by separating them by molecular mass in gel electrophoresis.  When this is done, you only see three bands: diglycosylated, monoglycosylated, and unglycosylated.  Among the monoglycosylated species of PrP, it’s not possible to distinguish the N181- from the N197-glycosylated molecules.

Below: Gel electrophoresis from Levavasseur 2008, Fig 1. PrP (in this case, PrP-res) separates as three bands corresponding to di-, mono- and unglycosylated PrP.

Second, even if two molecules of PrP are both, say, diglycosylated, they might still differ in the types of glycan chains attached (again: length, branching structure, types of sugar molecules included).  This is hard to study too.  The enzyme PNGase F can be used to cleave the glycan chains off of PrP, and then you can identify the exact glycan chains themselves using mass spectrometry.  But you won’t know whether each glycan chain you see came from N181 or N197.  Or you can do a two-dimensional gel electrophoresis, where you leave the glycans atatched to PrP and then separate PrP molecules by size along one axis (thus separating di-, mono- and unglycosylated, and probably also separating out different cleavage products) and by isoelectric point on the other axis.   Isoelectric point is the pH at which a molecule is neutrally charged, and this differs for many glycan chains, though unlike mass spec, you can’t precisely identify every glycan chain.

Below: 2D gel electrophoresis from Hopf 2011, Fig 4.  Proteins (in this image, a bacterial protein, not PrP) are separated by molecular weight on the y axis and isoelectric point on the x axis.

From my reading of the literature, the term ‘glycoform’ seems to be used loosely to refer to either the di/mono/un-glycosylated distinction or to the which-glycan-chain distinction, or to both.  I’ll use it loosely here too.

is glycosylation how prion strain information is encoded? (answer: no)

In humans and in other animals, there are different strains of prions, distinguishable pathologically by the specific disease symptoms, incubation times and the brain regions most heavily affected, and distinguishable biochemically by the size of cleavage products and, as we’ll see shortly, the ratio of di-, mono- and un-glycosylated species.  These properties are maintained over serial transmission between animals, in the absence of any amino acid sequence difference.

Ever since the notion of infectious proteins was proposed [Griffith 1967, Prusiner 1982], a central mystery has been how proteins could encode different strains.  Before the advent of the prion hypothesis, all of the infectious agents that people were used to seeing had nucleic acids, and so differences in DNA or RNA sequence could code for different strains.  You might ask: could different strains simply be different glycoforms of PrP?

A considerable body of evidence says the answer is no.  The first study to show that glycosylation is not necessary for strain information considered two strains of hamster prions, named hyper (HY) and drowsy (HY) for the different behavioral phenotypes observed in infected hamsters [Bessen 1995].  Bessen discovered that even when PrP^Sc from each strain was first amplified in a cell-free conversion assay [Kocisko 1994] using purely unglycosylated PrP as a substrate, strain properties were still preserved on transmission to a new set of hamsters.  The finding that unglycosylated PrP can preserve strain information has since been replicated in mice with the RML and 301C prion strains [Piro 2009].  No one has yet been able to induce spontaneous PrP^C > PrP^Sc conversion in a cell-free system with wild-type PrP – though several efforts have come very close [Legname 2004, Castilla 2005, Edgeworth 2010, reviewed in Benetti & Legname 2009].  But we do now know from studies of yeast prions that conformation alone can be sufficient to encode strain information [Tanaka 2004, Tanaka 2006].

All that is pretty convincing evidence that glycosylation is not necessary for the existence of prion strains.  That doesn’t by definition rule out the possibility that glycosylation might still contribute to certain strains, and one recent study has sought to show that passage of prion strains through mice expressing glycosylation mutants does indeed change the strain properties [Cancellotti 2013].  A challenge in interpreting the results of this (and many other in vivo studies, as we’ll see below) is that in order to abolish one or both of PrP’s glycosylation sites, the investigators had to mutate the NXT sequences, thus confounding the effects of glycosylation changes with the effects of amino acid sequence changes.

As an interesting aside, it has recently been shown strain properties can be changed depending upon the cofactors available for prion conversion.  Using a version of the protein misfolding cyclic amplification (PMCA) assay [Saborio 2001, Castilla 2005, Deleault 2010], it was shown that the presence or absence of different cofactors can affect the faithfulness of prion strain propagation – specifically, when phosphatidylethanolamine was the only cofactor available, three different prion strains converged into one [Deleault 2012].  This is still consistent with the idea that PrP^Sc strain information is (or at least can be) encoded in conformation alone, but that cofactors are necessary for the templating of that strain information onto new PrP^C molecules.  We may eventually learn that glycosylation, similarly, is necessary for the propagation of certain strains, but this hasn’t been shown convincingly yet.  But whether or not it’s strictly necessary, there is good evidence that glycosylation does affect the efficiency of prion conversion for various strains, as we’ll see later in this post.

different strains have different glycosylation patterns

Although glycosylation does not appear to be necessary to encode strain information, strains do have clearly different glycosylation patterns.

In humans, different strains of sporadic and acquired Creutzfeldt-Jakob Disease exhibit different relative proportions of un-, mono-, and di-glycosylated species of PrP, and these differences are stable over passage through mice [Collinge 1996].  Similarly reproducible differences in the extent of glycosylation were also observed for several rodent strains of PrP^Sc [Somerville 1997].  In that latter study, a limited amount of within-strain variation in glycosylation was still observed, and available experimental techniques were not sufficiently precise to uniquely identify all strains based solely on glycosylation state [Somerville 1997].   Still, it was clear that strains have characteristic glycoform ratios.  Indeed, the identical glycoform ratio observed in human variant Creutzfeldt-Jakob Disease (vCJD) and bovine spongiform encephalopathy (BSE) upon transmission of each to mice was used as one of several lines of evidence demonstrating that these two strains are one and the same, and that the vCJD epidemic in the U.K. arose from the transmission of BSE or ‘mad cow disease’ to humans [Hill 1997].

In an even more extreme example of characteristic glycoform patterns, it has recently been shown that two human prion diseases – familial Creutzfeldt-Jakob Disease caused by the V180I mutation, and a newly characterized sporadic prion disease dubbed variably protease-sensitive prionopathy (VPSPr) [Zou 2010] – contain no PrP^Sc molecules glycosylated at N181, and therefore no diglycosylated PrP^Sc molecules at all [Xiao 2013].  This feature had previously been observed in a familial form of Creutzfeldt-Jakob Disease caused by the T183A mutation, but in that case, the amino acid substitution abolishes the N181 glycosylation site, which requires an NXS or NXT motif.  In contrast, VPSPr is a sporadic disease not associated with any amino acid substitution, and cells expressing V180I PrP do produce PrP^C glycosylated at N181.  Therefore it appears that PrP glycosylated at N181 does exist in both of these diseases, but simply cannot be converted to the particular strain of PrP^Sc that is present.

PrP^Sc of different strains may differ not only on the di/mono/unglycosylated dimension, but also in the nature of glycan chains attached.  One study has shown a difference in glycan chains between PrP^Sc in two different human prion diseases: sporadic fatal insomnia (sFI) and the M/M2 subtype of sporadic Creutzfeldt-Jakob Disease (sCJD) [Pan 2001].  These diseases exhibit different phenotypes in the absence of any amino acid substitutions in PrP, and all subjects examined (n=2 for each disease, i.e. n=4 total) were homozygous MM at codon 129.  The relative abundance of di-, mono- and un-glycosylated forms was indistinguishable between the two diseases, but sCJD appeared to contain a greater diversity of glycan chains than sFI.  2D gel electrophoresis revealed several dots (corresponding to different isoelectric points) present only in sCJD and not in sFI.  These disappeared when the PrP^Sc was deglycosylated prior to electrophoresis, confirming that it was the glycan chains that caused the different isoelectric points.  And no such differences were observed in PrP^C isolates, suggesting the observed differences were the product of prion infection and not just differences between the patients’ original sets of glycoforms.

the preferential conversion hypothesis or ‘selection model’

The finding that glycosylation states differ between strains led Collinge to propose two hypotheses: (1) that each PrP^Sc conformation may have an easier time converting certain glycoforms of PrP^C rather than others, and (2) that the selective vulnerability of different brain regions to different prion strains may therefore owe in whole or part to the local abundance of compatible glycosylation species [Collinge 1996].

Hypothesis #1, which I will call the preferential conversion hypothesis and which some authors have called the ‘selection model’, would imply that a given strain (defined by its conformation alone) might efficiently convert, say, monoglycosylated PrP but inefficiently convert diglycosylated PrP.  (Or the efficiency might vary by the type of glycan chains attached).  This might happen because the glycan chains stabilize a particular structure, or get in the way of conversion, or get in the way of a particular folding conformation, thus raising the energy barrier that has to be crossed to convert from PrP^C to a particular conformation of PrP^Sc.

If this is true, then PrP^Sc glycoforms should differ from PrP^C glycoforms in the same brain. Pan 2001‘s and Xiao 2013‘s results both seem to support this notion, for glycan chain types and for di/mono/unglycosylated PrP molecules respectively.  But not all investigators have agreed.  Mass spectrometry of glycans isolated from uninfected and infected Syrian hamster brains revealed 52 different glycan chains attached to PrP, every one of which was present in both PrP^C and PrP^Sc [Rudd 1999].  The relative abundance of some glycans differed, with PrP^Sc enriched for tri- and tetra-antennary and depleted for bi-antennary structures relative to PrP^C, but this was speculatively attributed to a global disruption of GnTIII enzyme activity in the course of prion-mediated neurodegeneration and thus considered a consequence of PrP^Sc infection rather than an intrinsic property of PrP^Sc.

Perhaps a more testable implication of the preferential conversion model is that prion strains should have trouble infecting mice expressing incompatible glycoforms of PrP.  For instance, a strain with a ‘preference’ for diglycosylated PrP should cause mild or no disease in mice that don’t express diglycosylated PrP.  As mentioned above, the troubling confounder here is that in order to abolish glycosylation at one of PrP’s N-linked glycosylation sites, you have to mutate either the N or the T in the NXT signal.

That amino acid change matters a lot, because prion propagation efficiency depends on compatibility between the PrP^Sc amino acid sequence and the target PrP^C amino acid sequence [reviewed in Collinge & Clarke 2007].  Prion strains have different ‘species barriers’ which tend to go away after the strain is passaged in the new recipient species or when the recipient animal expresses a transgene of the original species’ PrP sequence [Prusiner 1990] and humans with homozygous codon 129 genotypes are much more vulnerable to sporadic CJD [Palmer 1991] as well as vCJD and have earlier onset in some genetic prion diseases [reviewed in Lukic & Mead 2011].  So it’s to be expected that if you mutate an N or a T in mice to some other amino acid, that might affect vulnerability to prion strains regardless of the resulting glycosylation changes.

Keep that in mind while interpreting the results from mouse studies.  Tuzi 2008 created mice with the Ns mutated to Ts in the first, second or both glycosylation sites of PrP (called G1, G2 and G3 mice respectively).  The resulting mice were then infected with 79A prions (a low glycosylation strain) or ME7 prions (a medium glycosylation strain).  The incubation times of the different strains differed dramatically among the different mouse mutants and versus wild-type mice.  Perhaps most convincingly, the G3 mice (with entirely unglycosylated PrP) were capable of sustaining a 79A infection, but they got sick very late in life if ever (only 4 of 21 mice died of prion disease), and they appeared wholly incapable of sustaining an ME7 infection.  The G1 mice also could not support ME7 infection, suggesting the first glycosylation site was indispensible for ME7 infection.

Tuzi’s results could all be explained by amino acid sequence changes, or could be explained by glycosylation.  Indeed, the story is quite plausible if you think of it in terms of glycosylation: 79A, the ‘low glycosylation’ strain, could infect any of the mutant mice (though with extended incubation times), while ME7, the ‘medium glycosylation’ strain, seemed to require glycosylation at the first site but could do without glycosylation at the second.

The same problem – is it amino acid sequence or glcyosylation that matters – affects interpretation of similar work in cell culture [Salamat 2011] and of the later work showing changes in strain properties after passage through Tuzi’s mice [Cancellotti 2013].  Of note, Salamat discovered that totally unglycosylated mutants of PrP are not properly trafficked to the cell surface, which could account for some of the extended incubation times that Tuzi observed, though it still cannot account for the differences in those effects between the 79A and ME7 prions.

To my knowledge, no study has yet really convincingly separated the effects of glycosylation from amino acid sequence in these mutant PrP models.  Therefore these studies cannot be taken as proof of the preferential conversion hypothesis, though overall they do seem to support it.

Rather, the most convincing evidence for the preferential conversion model comes from the fact that prion strains maintain their own characteristic glycoform ratios even in cell-free conversion assays, provided that all the cofactors and a diversity of PrP^C glycoforms are available as substrates [Castilla 2008].  In Castilla’s Fig 1 you can see that the respective glycoform ratios of four different prion strains are virtually identical before and after PMCA – and consistently different from each other.

Following Collinge’s original hypotheses, Somerville had proposed two other models to explain the glycoform differences of prion strains, suggesting that strains might differentially affect the glycosylation of PrP at the point of synthesis (similar, but not identical, to Rudd 1999‘s suggestion that enzyme disruption in the disease process led to differences in PrP^Sc vs. PrP^C glycosylation) or that different glycoforms might be differentially deglycosylated by other enzymes during infection by different strains [Somerville 1999].  Neither synthesis nor degradation is occuring in the PMCA assay, so preferential conversion seems to be the only model to explain the maintenance of glycoform ratios.  In interpreting his own data, Castilla concludes that the selection model must be correct [Castilla 2008].

selective vulnerability of different brain regions

If strains do exhibit a ‘preference’ for converting different glycoforms of PrP, that could go a long ways towards explaining why strains preferentially affect different brain regions, a phenomenon called strain-specific neurotropism.

Total PrP^C levels differ across the brain, being highest in the thalamus [DeArmond 1999].  But that’s true regardless of what strain an animal is infected with, so differences in host PrP expression could explain at most one strain’s neurotropism (brain region preference).  Some other brain region difference is needed to explain the neurotropic differences between strains.  And indeed, PrP^C glycosylation does differ across brain regions.  2D gel electrophoresis has been used to separate PrP^C molecules both by size and by isoelectric point, showing differences in the glycan chains of diglycosylated PrP^C across several different mouse brain regions [DeArmond 1997].   Later studies using gel electrophoresis or antibodies with differential affinity for di-, mono- or un-glycosylated species of PrP were also able to show differences in the prevalence of these species across different brain tissues both in mice [Somerville 1999 (ft), Beringue 2003] and in humans [Kuczius 2007]. All of these findings for PrP specifically are consistent with the well-established knowledge that, for glycoproteins generally, glycosylation patterns are tissue-specific [Rademacher 1988].

Together these results suggest sufficient heterogeneity in PrP glycoforms across different brain regions to potentially account, at least in part, for selective vulnerability.

But to my knowledge no one has really been able to show this experimentally.  To do so would seem to require somehow changing the glycoform ratios (and not changing anything else) in different brain regions and then showing that this alters which brain region is affected by which strain.  That’s a tall order, especially when we don’t have the tools to even convincingly separate glycosylation changes from amino acid changes in the whole organism.

On the other hand, in a quick thought experiment, I find it hard to talk myself into the idea that glycosylation doesn’t matter for neurotropism.   As discussed above, it’s clear that the relative ratios of glycoforms of PrP^Sc collected from brains of humans or other animals do differ between strains, sometimes dramatically so.  That’s consistent with the idea that strains have preferences for certain glycoforms, and that those preferences drive them to accumulate in different brain regions where their preferred glycoforms are abundant.  One alternate interpretation of this data would be that different strains affect different brain regions for some other, unobserved, non-glycosylation-related reason, and that then because they are active in different brain regions, they end up converting the glycoforms that are locally available.  That alternate interpretation is hard to reconcile with the existence of glycosylation preferences in vitro [Castilla 2008].

But that’s not proof, and in any event, glycosylation could contribute to strain-specific neurotropism without explaining all of it.  For instance, since it’s clear that cofactors matter too [Deleault 2012], the availability of different cofactors across brain regions could also contribute to explaining it.

All this is assuming an implicit causal link: that the amount of PrP^Sc produced in each brain region determines the extent of pathology there.  The couple of studies I was able to find do suggest that PrP^Sc levels and pathology are pretty strongly correlated across brain regions, both in CJD [Parchi 1996] and FFI [Cortelli 1997].  Interpretation of this is complicated by the fact that when we say “PrP^Sc“, what is usually being measured is actually proteinase K resistant PrP (PrP-res), or PrP plaque staining, both of which appear to be just proxies for disease and not direct measures of what is toxic, and FFI doesn’t produce much PrP-res anyway [Jackson 2009].

summary of the literature

Based on my above read-through of the literature, the following points seem very well-supported by available evidence:

1. Glycosylation is not necessary for encoding strain information in prions.
2. Prion strains exhibit reproducible glycoform ratios which are faithfully maintained during propagation both in vitro and in vivo.
3. Prion strains have different ‘preferences’ for converting different PrP glycoforms.
4. Ratios of PrP^C glycoforms differ across brain regions.

All this suggests a role for glycobiology in explaining strain-specific neurotropism, but this has never been experimentally demonstrated.

is glycosylation a potential therapeutic target in prion disease?

At first glance, based on the above evidence, it would seem that targeting glycosylation could be a therapeutic strategy, but not an ideal one.  For one, since glycosylation ‘preferences’ differ between prion strains, glycosylation-based strategies seem likely to have strain-specific efficacy, which would limit their usefulness.  Second, with a few exceptions (VPSPr, V180I CJD [Xiao 2013] and ME7 in G3 mice [Tuzi 2008]) the glycoform ‘preferences’ of prion strains do not appear to be absolute.  That is, a strain may ‘prefer’ to convert diglycosylated PrP but still can convert un- or monoglycosylated PrP.  That suggests that changing the glycosylation state of PrP^C might delay prion disease a bit, but not forever.

Two studies that examined prion replication in the presence of compounds that alter glycosylation seem to support these conjectures [Browning 2011, Oelschlegel & Weissmann 2013].  These studies were not exactly therapeutic-oriented, but rather, used glycosylation-altering compounds as tools to probe the hypothesis that each prion strain is actually a quasi-species of several substrains, which can be selected for or against by different drugs.  In addition to their biochemical and phenotypic properties, prion strains can now be distinguished by their relative ability to infect different cell lines, and their relative resistance to different strain-specific drugs such as curcumin and cpd-B, in a ‘cell panel assay’ [Mahal 2007].  Scrapie-infected cells were treated with a variety of glycosylation-modifying drugs and other small molecules, including kifunensine (kifu), castanospermine (CST) and swainsonine (swa).  In support of the ‘quasi-species’ or ‘many substrains’ hypothesis, some strains of prions were initially sensitive to some drugs, but could develop drug resistance or drug ‘dependence’ (increased efficiency of propagation in presence of the drug) after being cultured in the presence of the drug [Browning 2011, Oelschlegel & Weissmann 2013].

The strain specificity of these compounds, and the related fact that they may merely select for resistant substrains like quinacrine does [Ghaemmagami & Ahn 2009], indeed seems to suggest they wouldn’t be ideal therapeutics, though these studies did not examine in vivo efficacy.

But another property of glycosylation-altering drugs is their ability to in some cases increase the degradation of glycoproteins, reducing the total amount of the protein produced.  That could fit into the therapeutic strategy of depleting PrP, a possibility which I’ll consider briefly here.

My interest in learning about PrP glycosylation and writing this post was triggered by a conversation I had with glycobiologist Nikolay Kukushkin.  He pointed out that alpha glucosidase inhibitors, a class which includes several approved drugs (and the experimental compound CST), can not only alter glycosylation but also reduce the total amount of some glycoproteins produced.  As one example we discussed a drug called n-butyl-deoxynorijimycin (NB-DNJ; commonly known as Miglustat and commercially known as Zavesca; shown below), which is prescribed for Gaucher disease and Niemann-Pick Type C.

This drug’s mechanism of action is fascinating.  Its therapeutic value in the storage diseases Gaucher and Niemann-Pick is due to its inhibition of ceramide glucosyltranferase, thus reducing the amount of certain glycolipids produced and lessening the overall burden of storage material.  This approach is called ‘substrate reduction therapy’.  However, many of the glycan-processing enzymes have similar enough binding sites that one drug may inhibit several of them.  It so happens that NB-DNJ also inhibits alpha glucosidases [see for instance Fischer 1995, Tian 2004], enzymes which normally trim the glycan chains of a freshly synthesized, immature glycoprotein in the ER, allowing that glycoprotein to interact with folding chaperones calnexin/calreticulin.  The protein will fold in the ER with their help, and then later in the Golgi, it will have its glycan chains further extended and modified by other enzymes.  When glycans are not trimmed (due to NB-DNJ), the glycoprotein cannot interact with calnexin/calreticulin, and remains in its initial unfolded state with immature glycan chains attached.  That state doesn’t last forever: our cells have redundant mechanisms for trimming the glycan chains later in the Golgi, and so eventually the glycan chains will be extended and modified like always.  But this redundancy isn’t perfect, so NB-DNJ does result in changes to the final glycosylation state of some proteins, and more importantly, the added time that the glycoprotein spends in an immature and unfolded state in the ER gives it more chances to be degraded (via ERAD) before it ever passages through the Golgi.  This means that the total amount of some glycoproteins is reduced by NB-DNJ.

NB-DNJ’s inhibition of alpha glucosidases leads to totally non-specific downstream effects – it affects all glycoproteins – and raises the interesting question of how it would affect PrP.  I did PubMed and Google Scholar searches for ‘prion n-butyl-deoxynorijimycin’ and similar search terms and could not find any studies that have examined this question. NB-DNJ/Miglustat is also not included in MSDI’s Spectrum Collection recently used in the Ghaemmagami lab’s ELISA assay for PrP-res inhibitors [Poncet-Montagne 2011] nor in MSDI’s US Drug Collection used in the Lasmezas lab’s FRET assay for PrP reducers [Karapetyan & Sferrazza 2013], so there doesn’t seem to be any available high-throughput screening evidence on its effects either.

We can make a reasonable conjecture that any glycosylation-altering effects of NB-DNJ would result in strain-specific inhibitory effects, as seen for CST [Browning 2011].  But the experiments in these studies were run on cell lines expressing wild-type PrP.  It is interesting to consider the possibility that in mutant cell lines, alpha glucosidase inhibitors might increase the degradation of the mutant protein.

That bit of speculation is based on a study looked at the metabolism and secretion of PrP in cell lines transfected with DNA expressing either of the D178N mutants: D178N cis 129M FFI and D178N cis 129V CJD [Petersen 1996].  In both cases, Petersen found that the unglycosylated form of the mutant (but not wild-type) PrP was almost completely degraded before reaching the cell surface.  At the time of synthesis in the ER, unglycosylated PrP appeared to account for ~20% of total PrP in both the mutant and wild-type varieties, but by the time it reached the cell surface, unglycosylated D178N 129M PrP was undetectable, and unglycosylated D178N 129V PrP had dropped to 0.6% of total PrP. Petersen then looked at postmortem brains of FFI patients, which confirmed that the same phenomenon is present in vivo, albeit to a somewhat lesser extent than in cell culture.  Unglycosylated mutant PrP was under-represented compared to unglycosylated PrP from the patient’s wild-type allele, suggesting that it might be preferentially degraded.

Petersen’s results can be taken to suggest that in the absence of glycan chains, the mutant PrP misfolds in some way that triggers a cellular response to degrade it.  Could a similar effect be achieved through alpha glucosidase inhibitors, promoting early degradation in the ER?  It’s a risky proposition: by promoting misfolding of some sort, the drugs might accelerate prion disease.  Indeed, tunicamycin, a toxic, non-drug experimental compound which blocks glycosylation altogether is used as a positive control for ER stress [for instance in Moreno 2012] because it causes such a dramatic accumulation of misfolded proteins.  It’s also not clear if ER-associated degradation could be achieved: Petersen found evidence that the degradation of unglycosylated mutant PrP in some late endosomal/lysosomal compartment, rather than ERAD.  More recent work has agreed that mutant PrP degradation appears to take place in acidic compartments past the Golgi, rather than in the ER [Ashok & Hedge 2009].

To my knowledge, no research has yet examined whether alpha glucosidase inhibitors can promote early degradation of mutant PrP before it reaches the cell surface.  A first step would be to test the effects of alpha glucosidase inhibitors in cell lines expressing mutant PrP and check for reduced expression of the mutant allele.  If expression indeed appeared to be reduced, effects in vivo could be determined in knock-in mice.  If it happened to have therapeutic value, that would be exciting -NB-DNJ has good blood-brain barrier penetrance [Patterson 2007] and is an approved drug with an apparently decent safety record [Ficicioglu 2008].

conclusions

N-linked glycosylation turns out to be a fascinating part of prion biology.  Different strains of prions exhibit different characteristic glycoform patterns, a phenomenon which has now been shown, fairly conclusively, to be due to the ‘preference’ of different strains for converting certain glycoforms of PrP^C.  This ‘preference’ might even contribute to explaining why strains of prions affect different brain regions more dramatically.

Glycosylation is not an ideal therapeutic target in prion diseases, because its effects would be likely to be strain-dependent, and because for the most part the ‘preferences’ of different strains are not absolute, so even at best, altering glycosylation could only slow, and not stop, prion disease.  Still, it would be interesting to know whether alpha glucosidase inhibitors, a class of drugs which can alter glycosylation and/or increase degradation of some proteins, could reduce expression of mutant PrP.  These have apparently never been tested for their effects on genetic mutants of PrP.

In this recent post I reviewed the list of drugs that have been attempted as therapeutics in symptomatic prion disease patients.  The short answer is that none of them worked.  Most of these drugs are covered in a published review [Stewart 2008], who summarizes by saying that the only positive result reported for any drug was that flupirtine appeared to help preserve cognitive function in sporadic CJD patients, but did not extend survival at all [Otto 2004].  The results of some more formal clinical trials for quinacrine [Collinge 2009] and pentosan polysulfate [Bone 2008, Tsuboi 2009] have come out since Stewart’s review was published, so here are quick summaries of those drugs.

Quinacrine reduces prion formation in cell culture models [Doh-Ura 2000, Korth 2001], but no study has ever shown quinacrine to reduce prion formation, extend survival or delay onset of prion disease in any animal model [Collins 2002, Barret 2003, Doh-Ura 2004].  All reports on human use declared no therapeutic effect [Furukawa 2002, Benito-Leon 2004, Haik 2004, Collinge 2009].  It was later shown that even in cell culture, quinacrine’s effects are only transient, apparently because it merely selects for certain strains of prions [Ghaemmagami & Ahn 2009].

Pentosan polysulfate is highly effective in mice, at least against some prion strains and if administered early in the disease course [Doh-Ura 2004]. It cannot cross the blood-brain barrier, and so in both mouse and human trials it was infused directly into the brain with an implanted device, a risky procedure which resulted in complications in several patients.  It did not demonstrate any therapeutic effect in clinical trials in sporadic and variant CJD patients in the U.K. and Japan [Bone 2008, Tsuboi 2009].

The only clinical trial that is ongoing as of May 2013 is for doxycycline.  Doxycycline and the related compound tetracycline delayed disease in certain animal models [Forloni 2002, Luigi 2008] and doxycycline was used in symptomatic sCJD patients, where it showed a small but possibly statistically significant effect.  The results of those human trials have never been formally published as a study, but some results have been presented at conferences or mentioned in other articles [Luigi 2008, Stewart 2008, Zerr 2009 (ft), see also tetracyclines post].  Based on the animal results and the marginal results from symptomatic patients, doxycycline moved on to a clinical trial in 11 asymptomatic FFI carriers.  This clinical trial is not expected to report results until 2022.

That date – 2022 – reflects the extreme difficulty of running clinical trials for genetic prion diseases.  The onset of these diseases is late and highly variable, and the number of patients is small.  Prion diseases are rare even as a whole – ~250 deaths/year in the U.S. for instance [Holman 2010] – and then just ~15% of prion disease patients have genetic mutations [Appleby & Lyketsos 2011].  And then they’re not all the same mutations: you have E200K and D178N and P102L and so on – so any study will need to either control for this difference or only enlist people with a particular mutation (D178N in the doxycycline study).  Finally, because patients are perfectly healthy up until onset and will be taking the drug for years or decades, it’s only realistic to consider approved drugs with excellent safety profiles. For drugs with nastier side effects (rapamycin) or unapproved experimental compounds (anle138b), the risks would probably outweigh any benefits in genetic carriers.

The gold standard of biomedical research is the double-blind, placebo-controlled clinical study in a large number of people.  That’s probably never going to happen in genetic prion diseases.  There are too few patients, and too few would agree to be randomized and maybe get a drug and maybe get a placebo.  And even for the tiny (11 people), non-double-blind, non-placebo-controlled clinical trial that is underway for doxycycline, it’s a tall order to ask anyone to wait until 2022 to get answers.

This leads naturally to the possibility of off-label use - of doctors prescribing to patients a drug which is already approved for other uses, even though clinical evidence is not yet in place to demonstrate the drug’s efficacy in prion diseases.

Informally, the National Prion Disease Pathology Surveillance Center in Cleveland tells patients to take doxycycline just in case it turns out to work.  When my wife Sonia and I received her genetic test results from the Surveillance Center in December 2011, indicating she carries the D178N mutation, the manager of the Center at that time (who was not an MD) recommended she take doxycycline.

I have asked around a bit, and though no one agreed to be quoted for this blog post, physicians appear to be divided on the issue of whether to prescribe doxycycline to genetic prion disease carriers.  On one hand, doxycycline has a rather low rate of adverse events [Smith & Leyden 2005] and an overall strong safety record, even in long-term use for rosacea [Valentin 2009].  Other physicians, however, feel that drugs shouldn’t be prescribed for experimental uses outside of a clinical trial setting.

From a research standpoint, the evidence for doxycycline’s efficacy is not as strong as the evidence for many other drugs.  Doxycycline was able to delay onset of prion disease in infected mice [Luigi 2008], but there are a few caveats to this study.  First, the mice were either infected peripherally (“peripherally” meaning “not in the brain”) and treated peripherally, or infected in the brain and treated with direct injections into the brain.  So Luigi’s results don’t establish that enough doxycycline crosses the blood-brain barrier to be effective if taken orally for a disease in the brain.  Second, Luigi treated the mice starting on the day they were infected.  Lots of drugs have shown efficacy if administered on the same day of inoculation, when the prion infection has yet to take hold.  It is much harder for drugs to be effective once infection has taken hold.  Third, the closely related compound minocycline was shown to have no effect when administered to mice late in prion infection [Riemer 2008 (ft)].  As for the human results from doxycycline, the results have never been published but I have heard it argued that the extension of survival in symptomatic patients treated with doxycycline was small enough that it could potentially be explained by the increased medical attention paid to study participants, or the antibiotic effects of doxycycline helping to prevent chest infections.

But the caveats relating to efficacy (or lack thereof) late in disease may not apply here.  Carriers who know their genetic status may actually be able to start taking a drug years or decades before any prion infection takes place, so we cannot rule out the possibility that drugs effective early in disease course in mice (which might include doxycycline) could be helpful.

Still, it is informative to note that several other drugs have been effective in more stringent mouse models – mice infected intracerebrally and treated orally or peripherally late in the disease course.  Notably, statins – which also have an excellent long-term safety record [Law & Rudnicka 2006] – have consistently been able to extend survival or delay onset by ~10% in mouse models [Mok 2006, Kempster 2007, Haviv 2008 (ft), Vetrugno 2009 (ft)].  Consistent with that, there is some unpublished evidence that high LDLs (‘bad cholesterol’) are correlated with shorter survival in sporadic CJD.

The fact that the Surveillance Center recommends doxycycline, and not statins, to genetic carriers may owe to a few reasons.  First, a mechanism of action has been suggested – tetracyclines such as doxycycline are suspected to bind PrP and inhibit its misfolding and/or aggregation [Tagliavini 2000]  – whereas the mechanism of action of statins in prion disease is still mysterious.  Second, though results are unpublished and effects are marginal at best, the fact that doxycycline may have extended survival slightly in symptomatic patients provides a layer of human evidence where statins only have mouse evidence.  Third, the fact that doxycycline is in a current clinical trial for FFI may make its off-label use seem more justified.

In any event, neither doxycycline nor any other drug has been shown to delay onset of prion disease in patients with genetic mutations, so any off-label use at this point is highly speculative.  Indeed, doxycycline (and statins) have never even been tested in genetic mouse models of prion disease.  Most drug studies in animals are done in mice infected with prions.  But mouse models are now available for several genetic prion disease mutations including D178N [Jackson 2009], E200K [Friedman-Levi 2011] and A117V [Yang 2009].

That’s why one of our foremost research goals at Prion Alliance is to take several of the compounds that have shown some efficacy in prion-infected mice and test them in genetic mouse models of prion disease.  We’re currently working on gathering all the data on previous mouse studies in order to objectively pick the best candidates.

In choosing the best candidates for further animal studies, we’ll need to consider not only which compounds are most potentially effective, but also which are most realistic to become treatments for genetic carriers.  To date, the only drug that has extended survival in a genetic mouse model of prion disease is rapamycin [Cortes 2012], but rapamycin is an immunosuppressant with very serious complications including a long-term increase in cancer, which makes it a very poor candidate for anyone to take for years or decades in the hopes of delaying prion disease.  Experimental new molecules like anle138b [Wagner 2013 (ft)] are not available commercially and their side effects are as yet unknown, which means that their safety and efficacy will need to be shown first in clinical trials in symptomatic patients before any doctor will consider prescribing them to genetic carriers – though it would certainly still be of interest to know whether they are effective in genetic prion diseases.  Meanwhile, even delays in onset of a few percent could correspond to meaningful years of a person’s life, so compounds of small or arguable effect – curcumin [Riemer 2008 (ft)] or astemizole [Karapetyan & Sferrazza 2013] for example – shouldn’t be ruled out automatically, but rather evaluated on the basis of efficacy as well as safety and availability.

Sometime in the next few months, expect a blog post reviewing the different studies to date and evaluating which compounds might be most worthy of further study.  For now, unfortunately, little effort has gone into studying potential treatments in genetic mouse models, so we don’t have much animal data to go on.  And no treatment has been shown to delay the onset of genetic prion disease in humans.

As mentioned above, I’m not a physician and nothing in this post is medical advice.  As a researcher and as someone affected personally by genetic prion disease, I wrote this post to gather the available information into one quick reference.  If readers would like to discuss their questions with a physician, Dr. Brian Appleby of the Cleveland Clinic has agreed to be contacted by any genetic prion disease carriers seeking information on this subject ().  Still, note that only your own physician will be able to provide you with medical advice.

Relative to my previous post, the update is that recent research (2012-2013) has shifted towards identifying the specific molecular changes that are triggered downstream of PrP binding to Aβo.  The results have been fascinating, and may represent the first signs of consensus in this controversial area.  Two papers released last year [Um 2012, Larson 2012] agree that Aβo binding to PrP triggers the activation of Fyn, a Src family kinase.  PrP’s ability to activate Fyn was one of the first native functions of PrP to be identified [Mouillet-Richard 2000], but no one knew Aβ was involved, and no one had really begun to uncover the significance of it until now.  Um showed that the activation of Fyn by PrP/Aβo leads to phosphorylation of NR2B, a subunit of the NMDA receptor, leading to calcium signaling changes which might explain the excitotoxicity observed in Alzheimer’s.  Larson showed that this activation of Fyn also leads to phosphorylation of Tau, providing a potential mechanistic link between the two major pathologies observed in the Alzheimer’s brain: Aβ accumulation and Tau hyperphosphorylation.

PrP is GPI-anchored to the outside of the cell’s membrane, and Fyn is cytosolic, so they would seem to need some intermediary in order to communicate with each other.  Mouillet-Richard 2000 had originally identified caveolin-1 as this intermediary, and Larson 2012 co-immunoprecipitates PrP, caveolin-1 and Fyn all in one go, thus supporting this model.  Now, the membrane topology of caveolin-1 is itself a topic of study and may not be fully understood yet [Razani 2002 (ft), Williams & Lisanti 2004, Aoki 2010] but it appears to be embedded in the membrane from the cytosolic side without fully penetrating through to the outside of the cell.  So there may be yet a fourth intermediary here (integrins?) but no one has investigated this in too much detail yet with regards to PrP.

I’ve tried to capture the current best understanding in this graphic:

Here is the review as submitted for class: [PDF].  And the full text is also below for direct blog reading.

Abstract

In 2009 it was discovered that prion protein (PrP) binds amyloid beta oligomers (Aβo), potentially acting as a receptor on the surface of neurons.  PrP was additionally suggested to mediate at least some of the toxic effects of Aβo in Alzheimer’s disease (AD), a proposition which has proven enormously controversial.  In the past three years, several studies have produced apparently strong evidence both for and against the notion that PrP mediates Aβo toxicity and AD pathology.  Though no consensus has been reached on the overall importance of PrP in AD pathology, recent efforts to elucidate the signaling pathways that are activated by Aβo binding to PrP  have provided consistent evidence that this binding event causes activation of the Src family kinase Fyn, leading to downstream phosphorylation events with potential relevance to AD.

Introduction

Background on PrP

Prion protein (PrP) was discovered, and gained notoriety, for its capacity to convert into an infectious agent that can propagate and cause disease in the absence of nucleic acids [Prusiner 1982].  Human PrP is a 208 amino acid GPI-anchored glycoprotein of unknown native function encoded by the gene PRNP, ubiquitously expressed and abundant in the central nervous system.  It is ordinarily found in a healthy ‘cellular’ conformation (PrP^C) but its conversion to an infectious ‘scrapie’ conformation (PrP^Sc) is responsible for the class of rapidly fatal, untreatable diseases known as transmissible spongiform encephalopathies (TSEs) or prion diseases.  These include Creutzfeldt-Jakob Disease, Fatal Familial Insomnia, Gerstmann-Straussler-Scheinker syndrome, and kuru in humans, and scrapie, bovine spongiform encephalopathy (‘mad cow’) and chronic wasting disease in other mammals.  The PrP amino acid sequence is highly conserved among mammals, yet knockout mice, cows and goats are healthy, fertile and viable [Bueler 1992, Richt 2007, Benestad 2012].  Certain knockout phenotypes have been reported under stress or late in life [Steele 2007, Bremer 2010].

Although PrP’s native function is unknown, its participation in certain interactions is well established.  Over a decade ago it was shown that cross-linking of PrP on mature neurons leads to activation of the tyrosine kinase Fyn [Mouillet-Richard 2000], but the significance of this signal transduction pathway has not been understood.  More recently, it has been shown that PrP inhibits the activity of BACE1, reducing the formation of Aβ [Parkin 2007]. The importance of these two roles will be explored further below.

Background on Aβ

Alzheimer’s disease (AD) is the most common cause of dementia.  It is characterized by two main neuropathological features: the accumulation of amyloid beta (Aβ) plaques and the formation of neurofibrillary tangles (NFTs) of hyperphosphorylated Tau protein.   95 – 99% of AD cases are late onset and idiopathic (of unknown molecular origin); the remainder are early onset familial forms caused by mutations in amyloid precursor protein (APP) or presenilins 1 or 2 (PSEN1 and PSEN2) [Bekris 2010].   APP is a Type I transmembrane protein which undergoes sequential cleavage by β-secretase (BACE1) and γ-secretase (a complex which includes both PSEN1 and PSEN2) to produce 40- or 42-residue N-terminal fragments known as Aβ₄₀ and Aβ₄₂ [Vassar 2009].  Mutations in APP, PSEN1 and PSEN2 that cause familial AD have been reported to increase the ratio of Aβ₄₂ to Aβ₄₀.  These peptides are present in the AD brain as monomers, as oligomers and as larger plaques.  This review will use the term “Aβ” to refer generally to these peptides, regardless of aggregation state, and “Aβo” to refer specifically to oligomers.  Oligomers of Aβ₄₂ are suspected to comprise the toxic species in AD [Bekris 2010].   The relationship between Aβ accumulation and Tau pathology in AD is not yet clear.  Mutations in Tau (MAPT) in humans cause frontotemporal dementia (FTD), not AD, and some transgenic mouse models with only APP or PSEN mutations fail to form NFTs, leading some researchers to employ mice with both APP and MAPT mutations in order to recapitulate a more full set of human neuropathological changes associated with AD [Bryan 2009].

Though the toxic role of Aβ in the Alzheimer’s brain is still poorly understood, certain species of Aβ are known to acutely impair neurons.  Injected Aβ inhibits long-term potentiation (LTP) in the rat hippocampus [Walsh 2002] and cause memory and behavioral changes in rats [Lesne 2006].  However, the exact size (monomer, dimer, trimer, etc.) and structure (globular, spherical, fibrillar, etc.) of the toxic Aβ species is highly controversial [Benilova 2012].

Interest in the mechanism of this acute neuronal impairment by Aβ led to a cDNA library screen to identify proteins that physically interact with synthetic Aβo, which revealed a single strong hit: PrP [Lauren 2009].  PrP bound to synthetic Aβo with an order of magnitude stronger affinity than any other Aβo binding partners and appeared to account for about 50% of total and Aβo binding to cells.   Moreover, Aβo was found to inhibit LTP in wild-type mouse hippocampal slices but not Prnp^-/-slices, implicating PrP as a receptor necessary for Aβo-induced neuronal impairment.  This phenotype could be rescued by 6D11, a monoclonal antibody against mouse PrP residues 93-109, approximately covering the putative Aβo binding site at residues 95-110.  For clarity, note that all mice and cultures used in these experiments were uninfected with scrapie and therefore all findings are assumed to refer to PrP^C and therefore to describe a native function of PrP.  Aβo binding to PrP^Sc has not been shown.

The finding that PrP binds Aβo has been replicated several times and never disputed [Biasini 2012].  However, the finding that PrP is required for Aβo impairment of neurons and, by implication, for Alzheimer’s pathogenesis, has proven enormously controversial.  Though no clear answers have emerged yet, this paper will review recent findings with a view to understanding the current status of three questions:

1. Does PrP mediate acute neuronal impairment by Aβo?
2. Does PrP mediate Alzheimer’s pathology in transgenic animal models?
3. What are the molecular consequences of PrP/Aβo interaction?

1. Does PrP mediate acute neuronal impairment by Aβo?

The claim that PrP is required for Aβo-induced neuronal impairment was quickly challenged in a battery of experiments [Kessels 2010].  Most importantly, Kessels reproduced the exact same system that Lauren had studied: long-term potentiation in the CA1 region of the mouse hippocampus using brain slices from wild-type and Prnp^-/- mice.  Kessels’ synthetic Aβ₄₂ oligomers impaired neuronal LTP equally in both genotypes, implying that PrP expression is not necessary for the Aβo blockade of LTP.  Kessels further examined two other established neuronal phenotypes of Aβo exposure: dendritic spine loss in neurons overexpressing APP or exposed to Aβ₄₂ oligomers, and synaptic depression in neurons expressing APPct100, a truncated form of APP that results in high Aβ production.  Both of these phenotypes were equally observed in wild-type and Prnp^-/- brain slices, confirming the dispensibility of PrP for Aβo’s effects.

In agreement with Kessels’ findings in brain slices, another group reported that intraventricular injections of synthetic Aβ₄₂ oligomers caused behavioral changes equally in wild-type and Prnp^-/-  mice [Balducci 2010].  The behavioral measurements involved exposing mice to a collection of novel and familiar objects and measuring the time that the mice spent with each.  By default, mice prefer novel objects over familiar ones, and Balducci found that this held true for wild-type mice after injection with vehicle, injection with synthetic Aβ in its ‘initial state’ (presumably monomers), or injection with Aβ fibrils, confirming previous findings that Aβ monomers and fibrils are not acutely toxic.  The preference for novel objects was diminished, however, following injection with synthetic Aβ₄₂ oligomers.  The authors interpreted this as an inability to discriminate between objects, thus implying memory impairment.  This reduction in discrimination was found in wild-type and Prnp^-/- mice, thus supporting the conclusion that Aβo’s effects on memory are independent of PrP.  This finding has been questioned: Balducci’s data show Aβ₄₂ oligomer-injected mice spending more time with familiar objects than novel ones, which other authors have taken to imply a change in preference, not discrimination [Gimbel & Nygaard 2010].  Indeed, Balducci’s data suggest that this change in preference is actually stronger in Prnp^-/- mice than in wild-type mice, which could suggest some PrP-dependent mechanism, albeit with an unexpected direction of effect.

While the best interpretation of the mouse behavioral data can be debated, Balducci’s and Kessel’s results uniformly failed to support the notion that the acutely deleterious effects of Aβ₄₂ oligomers on the brain are mediated by binding to PrP.  However, it is increasingly recognized that preparations of Aβo and/or of Aβ aggregates can differ dramatically between laboratories [No Authors Listed 2011].  It is therefore of great importance that a different team not only replicated Lauren’s original result with synthetic Aβ₄₂ oligomers but also extended it to authentic materials, using Aβ species derived from human AD brain tissue [Freir 2011].

Noting that the size, structure, and concentration of Aβo could explain the discrepancies between Kessel’s and Lauren’s results, Freir was careful to provide an exact description of the procedures used to synthesize biotinylated and unbiotinylated Aβ₄₂ oligomers in vitro.  Freir’s synthetic oligomers proved to inhibit LTP in hippocampal slices from wild-type mice but not from Prnp^-/- mice.  Further noting that synthetic Aβo may not represent a toxic species found in vivo, Freir also exposed the hippocampal slices to soluble extracts isolated from postmortem brain tissue from AD patients and from control, non-demented subjects.  As expected, the control extracts had no effect.  The AD brain extracts were found to inhibit LTP in wild-type brain slices but not in Prnp^-/- brain slices.

Freir next considered whether the PrP/Aβo interaction could represent a therapeutic target in AD.  In the wild-type brain slices, Freir discovered that LTP impairment by Aβo could be blocked by either of two well-characterized monoclonal antibodies against PrP.  One of these antibodies, ICSM-35, binds to PrP residues 95-110, identical to the putative binding site of Aβo.  The other, ICSM-18, binds at PrP alpha-helix 1 – residues 145-154 [Riek 1996] – and thus its mechanism of action with regards to Aβo is not immediately obvious.   Based on the known epitope of ISCM-18 and a crystal structure of the antibody bound to PrP^C [Antonyuk 2009] Freir dismisses the possibility of steric hindrance as a mechanism, instead proposing that the antibody may block hypothetical PrP:PrP interactions which could be required for Aβo binding.  ISCM-18 and ICSM-35 are slated for a clinical trial as therapeutic agents in prion diseases [MRC Prion Unit, accessed April 22, 2013], raising the possibility that if successful they may also be investigated as AD therapeutics.  Another team also investigated the feasibility of blocking the Aβo-PrP interaction using antibodies against PrP, and found that Fab D13, which targets PrP residues 96-104, prevented Aβo-induced LTP impairment in rat hippocampal slices, while Fab R1, which binds to PrP at residues 225-231, the most extreme C-terminus of the protein, had no effect [Barry 2011].  This is in agreement with Freir’s finding that the PrP/Aβo interaction can be disrupted immunologically, at least through binding at certain PrP epitopes if not others.

By using extracts from AD postmortem brain tissue, Freir and Barry both confirmed that authentically derived species of Aβo induce LTP impairment in a PrP-dependent manner.  This may be reconciled with Kessel’s and Balducci’s findings by assuming that the latter two groups’ synthetic Aβo preparations differed from the synthetic preparations of both Lauren and Freir in containing an oligomeric Aβ species that (1) induces neuronal impairment by a PrP-independent mechanism, and (2) is not found in vivo in the AD brain.  Alternately, point (1) may be accepted but point (2) may be substituted for an assumption that said species does exist in the AD brain but was not successfully extracted by Freir’s or Barry’s protocols.

In either case, though, hippocampal LTP impairment is merely one system in which to model the deleterious effects of Aβo in the brain.  At the same time as the above studies were being carried out, experiments were underway to assess the importance of PrP for the appearance of AD pathology in vivo in transgenic mouse models of AD.

2.  Does PrP mediate Alzheimer’s pathology in transgenic animal models?

Many transgenic mouse models of AD use familial early onset AD-causing mutations in either APP, PSEN1, or both [Bryan 2009]. Two separate groups crossed mice with both APP and PSEN1 mutations (hereafter “AD” mice) with Prnp^-/- mice in order to assess the relevance of PrP expression to AD pathology in vivo, with completely conflicting results [Gimbel & Nygaard 2010, Calella 2010].

The first group used already-characterized APPswe/PSen1ΔE9 mice, which form Aβ plaques but have not been reported to form NFTs [Jankowsky 2004].  These AD mice have also been reported to show significantly reduced lifespans compared to wild-type mice, though the investigators reporting this offered several caveats to their study, including the use of non-littermates, mixed genetic background and possible viral infections [Halford & Russell 2009].  This first group reported that AD/Prnp^-/- mice were indistinguishable from controls in Morris water maze performance (a test of spatial learning and memory), synapse loss, and survival, even while AD/Prnp^+/+  mice deteriorated severely on all of these measures.  The AD/Prnp^-/- mice produced APP and Aβ, and deposited Aβ plaques, in equal quantities as AD/Prnp^+/+  mice, yet did not appear to be impaired by the presence of this Aβ [Gimbel & Nygaard 2010].

The second group used a separate, but similar, already-characterized AD mouse model known as APPPS1⁺ [Radde 2006].  Compared to APPswe/PSen1ΔE9, the APPPS1 model carries the same ‘Swedish’ APP mutation but a different PSEN1 mutation (L166P instead of ΔE9).  These mice were crossed to Prnp^-/- mice and evaluated for a set of phenotypes different than used in the above-described study [Calella 2010].  Calella evaluated the mice primarily by electrophysiological measurements of LTP at four months of age, and reported no difference between AD/Prnp^+/+ , AD/ Prnp^+/- , and AD/Prnp^-/-  mice.  Calella reported no reduced survival in any genotype, precluding any comparison with Gimbel & Nygaard’s results.  A potential confounder was impure genetic background: the mouse crosses performed to generate these mice were insufficient to recombine mouse chromosome Mmu2, location of both Prnp and the AD transgenes.  To rule out these confounding effects, Calella also crossed the AD mice to transgenics overexpressing PrP or overexpressing a mutant form of PrP lacking the GPI anchor.  These transgenes did not exacerbate the LTP impairment phenotype in AD mice; to the contrary, the overexpression of PrP trended toward rescue, and the overexpression of unanchored PrP significantly rescued the phenotype.

Because these studies used similar but not identical AD mouse models and measured different phenotypes, a direct comparison is nearly impossible.  Gimbel & Nygaard’s behavioral phenotypes are more variable and difficult to measure than Calella’s electrophysiological measurement; whether either represent an informative model of human AD pathology is debatable.   Calella notes that Gimbel & Nygaard fail to address the problem of residual genetic linkage on Mmu2.   Further, Calella’s finding that overexpression of PrP does not exacerbate LTP impairment does appear to rule out a dose-dependent effect of PrP, at least on LTP.

A third mouse study, using yet a third mouse model, successfully used passive immunization with a monoclonal antibody against PrP to rescue AD phenotypes  [Chung 2010].  This study used APP/PS1 transgenic mice [Holcomb 1998], which express the same APP ‘Swedish’ mutation as the two mouse models mentioned above, along with yet a third PSEN1 mutation, M146L.  Treated mice received peripheral injections of the 6D11 antibody, which recognizes PrP amino acids 93-109 (covering the critical region for Aβo binding), while controls received IgG or vehicle.   After two weeks of treatment starting at 8 months of age, the mice were subjected to a maze test and a blinded observer counted the number of ‘errors’ (entries into an arm of the maze where the reward had already been consumed) the mice made.   The APP/PS1 mice treated with vehicle or IgG consistently made 8-10 errors before successfully consuming all 8 rewards, while the mice treated with 6D11 made only ~4 errors, similar to wild-type controls.  Aβ and PrP levels in the 6D11-treated mice were indistinguishable from vehicle- or IgG-treated, implying a mechanism whereby 6D11 occludes PrP, blocking binding to Aβo, rather than accelerating degradation of either protein.

The results of this study are fairly surprising for several reasons.  First, this is the earliest evidence of peripheral antibody injections significantly affecting PrP activity in the CNS.  Peripheral injections of anti-PrP antibodies have been shown to abolish peripheral prion infections but have so far proven unable to occlude a large enough fraction of PrP in the brain to detectably slow the course of established CNS prion infections [White 2003]. To achieve this, the authors report having used ‘large’ (1 mg/day) doses of antibodies, and cite evidence that ~0.1% of passively administered antibodies enter the CNS [Bard 2000].  Second, APP/PS1 mice are reported to have significant Aβ plaque burden by eight months of age, and so an apparently complete reversal of one behavioral symptom after only 2 weeks of treatment is contrary to the growing body of evidence that AD immunotherapy is most effective if administered early in the disease course [Lemere & Masilah 2010].  Third, if cognitive ability as measured in the radial arm maze is presumed to reflect LTP-based memory and learning, then this result is contrary to that of Calella.  For all of these reasons, Chung’s study represents a potentially very high-impact finding, limited by the authors’ decision to rely on only one behavioral test and no other phenotypic measurements.  Although Chung’s study was published after Calella’s, it does not address possible reasons for the discrepancy between their results.

In sum, the three AD mouse model studies provide no more clear agreement than the brain slice and Aβo-injected mouse studies discussed earlier.  All authors agreed, however, that Aβo does bind to PrP.  Calella’s conclusions are phrased mildly, arguing that PrP mediation of Aβo toxicity is not ‘universal’ and allowing the possibility that further research may clarify whether PrP is a potential therapeutic target in AD.  Accordingly, the molecular consequences of the PrP/Aβo interaction have been intensively studied over the past two years.  The next section of this review will summarize current hypotheses.

3. What are the molecular consequences of PrP/Aβo interaction?

Because PrP is known to transduce signals through Fyn [Mouillet-Richard 2000], and overexpression of Fyn has been reported to exacerbate AD phenotypes [Chin 2005], Fyn is a natural candidate for mediating Aβo/PrP signaling. Recently, antibodies specific to pY416-Fyn detected increased Fyn activation after incubation with human AD brain-derived Aβo in wild-type neurons but not in Prnp^-/- neurons, demonstrating that PrP acts as an Aβo receptor leading to Fyn activation [Um 2012].  Fyn signaling through other receptors was preserved in Prnp^-/- neurons but no Aβo-induced Fyn activation was detected, implicating PrP as an indispensible intermediate in all Aβo/Fyn signaling.  In PrP-expressing neurons, exposure to Aβo caused a five-fold increase in the amount of phosphorylated NR2B, a subunit of the NMDA receptor.  The effect was absent in cells lacking PrP or Fyn, and reduced proportionally in heterozygous knockouts of either gene, suggesting a dose-dependent mechanism.  Phosphorylated NMDARs were shown to endocytose less frequently, with a peak threefold increase in the ratio of cell surface to internalized NMDARs.  The resultant changes in calcium signaling caused excitotoxicity detectable by cellular release of LDH, and also caused dendritic spine loss.  This dendritic spine loss could be prevented by PrP ablation or by antibodies against NMDARs, and was proposed to possibly explain the seizures observed in AD/WT mice but not in AD/Prnp^-/ mice.  Taken together, Um’s findings imply that the Aβo-, PrP- and Fyn-dependent phosphorylation of the NMDA receptor has dramatic consequences, potentially explaining the epileptic activity, excitotoxicity and dendritic spine loss associated with AD.

Shortly thereafter, another group provided evidence for an equally compelling piece of the AD puzzle, demonstrating that Aβo/ PrP-induced Fyn activation leads to Tau phosphorylation, potentially tying together the two major histopathological phenotypes of AD [Larson 2012].  First, Larson answered a basic question which Um had left untouched: because PrP is anchored to the exoplasmic leaflet of the plasma membrane while Fyn is intracellular, the two are likely to need an additional interacting partner.  Co-immunoprecipiation of PrP and Fyn from AD brain extracts showed caveolin-1 as the third party, confirming earlier findings from cell culture [Mouillet-Richard 2000].  Larson next used a battery of antibodies to ask which species of Aβ was co-precipitating with PrP, and identified Aβ₄₂ dimers as the sole species binding PrP.  Cortical neurons derived from primary mouse tissues were shown to exhibit elevated Fyn activation after incubation with human AD brain-derived Aβo, and this effect could be abrogated by antibodies against certain epitopes of PrP.  Moreover, this activation of Fyn more than doubled the amount of Tau protein phosphorylated at Y18, and changed the intracellular distribution of Tau; this effect could be abolished by anti-PrP antibodies.  In vivo experiments on AD mice showed Fyn activation and Tau pathology to be somewhat reduced in heterozygous PrP knockout mice, dramatically reduced in homozygous knockout mice, and exacerbated in PrP overexpressers.

Together, Um’s and Larson’s work tell an important story about how PrP may mediate AD pathology.  The two studies agree that Aβo binds PrP, activating Fyn and causing downstream pathology.  Although the studies point to two different effects, on NMDA receptors versus Tau, the findings are non-conflicting.  Importantly, Um’s and Larson’s colleagues were on opposite sides of the original debate, suggesting that a pathological activation of Fyn may be, for now at least, the first non-controversial result of Aβo/PrP binding.

Current work seems to support Um’s and Larson’s conclusions.  Another study recently confirmed Fyn activation upon Aβo binding to PrP, and detected membrane disruption as a cytotoxic marker when this binding event occurred [Rushworth 2013].  Interestingly, that study also found that the binding event induced endocytosis and prevented PrP’s inhibition of BACE1, leading to increased production of Aβ, pointing to a possible vicious cycle mechanism.

Discussion

After a few years of controversy, the field may finally be heading toward consensus on at least a subset of conclusions about Aβo and PrP.   It is widely agreed that Aβo binds PrP with high affinity and it now appears clear that this binding event induces activation of Fyn via caveolin-1, and that Fyn’s phosphorylation of downstream targets could explain at least some of AD pathology.  The downstream targets proposed here – NR2B and Tau – have the potential to finally explain the role of excitotoxicity in AD and the molecular pathway connecting Aβ accumulation with Tau tangles in the AD brain.

Meanwhile, although the reasons for discrepant results in earlier experiments have still not been fully elucidated, investigators seem to have accepted the importance of using authentically derived Aβ from human brains and, after some nudging [No Authors Listed 2011], recent work shows a trend toward explaining oligomer preparations and methodologies in greater detail, which may help to make results more reproducible in the future.

Today I needed to do the same thing in CRAN R, and a few quick Google searches for “R equivalent of SQL coalesce” didn’t turn up anything.  So I wrote this quick function which accepts any number of R vectors of the same length and returns the first non-NA value in each position.

# accepts a list of vectors of identical length and returns one vector with the first non-NA value
coalesce = function(...) {
    # convert input arguments into a list of vectors
    input_list = list(...)
    # check that all input vectors are of same length
    vectorlength = length(input_list[[1]])
    for (j in 1:length(input_list)) {
        if(length(input_list[[j]]) != vectorlength) {
            stop(paste("Not all vectors are of same length. First vector length: ",vectorlength,". Vector #",j,"'s length: ",length(input_list[[j]]),sep=""))
        }
    }
    # create a result vector to fill with first non-NA values
    result = rep(NA,vectorlength)
    # fill with first non-NA value
    for (i in 1:length(result)) {
        for (j in 1:length(input_list)) {
            if(!is.na(input_list[[j]][i])) {
                result[i] = input_list[[j]][i]
                break
            }
        }
    }
    return(result)
}

# examples to show how it works
most_preferred_measurement = seq(1,10,1)
most_preferred_measurement[c(1,4,5,6,7,8)] = NA
backup_measurement = seq(11,20,1)
backup_measurement[c(2,3,4,5)] = NA
least_preferred_measurement = seq(21,30,1)
least_preferred_measurement[c(5,9)] = NA
other_vector_of_different_length = seq(31,35,1)
coalesce(most_preferred_measurement,backup_measurement,least_preferred_measurement) # this works
coalesce(most_preferred_measurement,backup_measurement,least_preferred_measurement,other_vector_of_different_length) # this will throw an error

If you run this code, you’ll see the expected result:

> coalesce(most_preferred_measurement,backup_measurement,least_preferred_measurement) # this works
 [1] 11 2 3 24 NA 16 17 18 9 10
> coalesce(most_preferred_measurement,backup_measurement,least_preferred_measurement,other_vector_of_different_length) # this will throw an error
Error in coalesce(most_preferred_measurement, backup_measurement, least_preferred_measurement, : 
 Not all vectors are of same length. First vector length: 10. Vector #4's length: 5
>

Enjoy!

what are stem cells

A stem cell is a cell with two key properties:

• self-renewal
• differentiation

In other words, they can divide to form more stem cells or to form more committed lineages of cells.  Stem cell populations are maintained either by asymmetric mitosis, where one daughter cell becomes differentiated and the other remains as a stem cell, or by stochastic differentiation, such that some mitosis events in a population will result in two stem cells while others will lead to two differentiated cells.

The use of the word “stem” here is believed to have arisen from analogy to the stem of a plant, which can branch out into terminal leaves.

Here is a lengthy introductory video:

how pluripotent is pluripotent?

We can think of a hierarchy of stem cells, in terms of how diverse of lineages they can give rise to:

• Totipotent – can make all possible cell types of an organism. (zygote, morula)
• Pluripotent – can make cells of any germ layer (endoderm, mesoderm, ectoderm), but not extraembryonic tissue, ex. the placenta.  ESCs and iPSCs in the lab cannot become placental tissue (which requires CDX2 expression).
• Multipotent – can make different cell types within a tissue. For instance hematopoetic stem cells (HSCs) can make a variety of blood cell types.
• Unipotent – can only make one cell type.  These do not have the property of differentiation, but do have the property of self-renewal.

As cells specialize, they lose their pluripotency.  This is hypothesized to have evolved as a defense against diseases.

Tissue-specific stem cells can be slow cycling or quiescent at steady state, but in response to stimuli can rapidly ramp up production of themselves.  An example is hematopoietic stem cells.  Differentiated blood cells have high turnover – neutrophils last ~12h, red blood cells last ~120d.  These are the best-characterized adult stem cells, well-studied since the 1960s.  They are used clinically in bone marrow transplantation.   By the way, “cord blood” is largely hematopoietic stem cells.

Stem cells live in a stem cell ‘niche’: a supportive environment made of cells and extracellular material that regulate the stem cells’ fate decisions.  Cellular signals may be secreted or transmitted through direct cell-cell contact.

stem cells in the lab

Embryonic stem cells (ESCs) used in laboratory settings are derived from culturing the inner cell mass of a blastocyst.  Because the inner mass does not include the Cdx2-positive outer layer destined for placental cell fate, ESCs are pluripotent, not totipotent.

Ways to reprogram cells:

• Nuclear transfer.  Take the nucleus from a differentiated cell into an enucleated egg. [Gurdon 1958, reviewed in Gurdon 2003, Nobel Prize 2012]
• Cell fusion.  Fuse a differentiated cell with pre-existing ESCs – this will generate 4N cells.
• Induced pluripotency.  Reprogram differentiated cells with Oct4, Sox2, c-Myc and Klf4 [Takahashi & Yamanaka 2006 (annotated full text here), Nobel Prize 2012].

All of these properties result in epigenetic changes to DNA that give rise to pluripotency.

experimental methods

To determine whether an isolated cell population is multipotent, label them (e.g. with GFP), then transplant them into a non-labeled irradiated recipient animal.  Irradiation prevents the host animal from rejecting the transplanted tissue and clears the stem cell niche.  Then watch to see if the GFP-expressing cells populate the host animal, and what cell types they end up being.  For example, if you see erythroid, myeloid and T, B, and NK cells, then that means you got HSCs to start with.  If you only get erythroid and myeloid cells, you had probably isolated the CMPs.

Cell fate mapping is well-studied in C. elegans because the origin and fate of every cell are known.

random grab-bag of facts about cancers

Many cancers display stem-like characteristics, but without the ‘checks and balances’: apoptosis is inhibited, proliferation is largely unchecked, and differentiation may be blocked.  (By the way, warts, moles and skin tags are all benign tumors).

Here are six changes that often (but not necessarily) occur in cancers:

• Proliferation – tendency to divide
• “Enabling replicative immortality” – (how is this any different from the previous bullet?)
• Evasion of growth suppressors – ways of ignoring anti-growth signals from nearby cells
• Angiogenesis – ability to induce vascularization so that the tumor continues to have access to nutrients.
• Resisting death – ways of ignoring pro-apoptotic signals from within or without
• Invasion & metastasis – ways of escaping from own tissue and invading other tissues

Two major classifications of cancer:

• Carcinoma – arise from epithelia (endoderm) or ectoderm.  Gut, skin, nervous system.
• Sacroma – arise from mesoderm. Muscle, blood, and their precursors.

Glioblastoma is not quite either of these, arising from glial cells in the brain.

Tumors can be characterized in vitro as having undergone “transformation” if they can grow indefinitely in the absence of growth factors on a variety of media.  Human tumors are often studied by xenograft into model organisms, to study angiogenesis, metastasis and other processes in real-time.

Sometimes inherited germline mutations contribute to cancer, even if not oncogenic on their own.  Other genetic lesions contributing to cancer are somatic mutations, arising spontaneously or in response to mutagens such as cigarette smoke, asbestos, etc.

How do proto-oncogenes convert to full-on oncogenes?  Four types of mutations often contribute:

1. SNPs that created a constitutively active protein product.  Examples include mutations in RTK that mimic ligand binding by introducing an amino acid substitution that mimics the phosphorylated state of the protein, and mutations that make Ras less able to hydrolyze GTP, thus making it ‘activated’ by GTP binding all the time.
2. Gene fusion by chromosomal translocations, creating a novel fused protein product. For instance, a chr9-chr22 translocation creates a BCR-ABL fusion protein under the BCR promoter, resulting in overexpression of a constitutively active ABL mutant which promotes cell proliferation, causing chronic myelogenous leukemia.  Imatinib targets the BCR-ABL fusion protein very specifically (it can’t bind to wild-type ABL) and is a very effective treatment.
3. Chromosomal translocation moving a gene under a new regulatory region, resulting in overexpression or ectopic expression.
4. Duplication of a gene leading to overexpression.

Tumor suppressor genes are usually haplosufficient, so both copies must be lost in order to give rise to a cancer.  Most tumor suppressors fall into one of the following categories:

1. Cell cycle regulators (ex. p16, Rb)
2. Receptors or signal transducers (ex. TGF-beta).  Examples include a V>Q HER2 mutation which allows kinase activity in the absence of a ligand, and a deletion of the Erb2 extracellular ligand-binding domain, allowing the intracellular domains to dimerize and activate without ligand.  Receptors may also be activated by viral proteins that mimic signaling molecules.
3. Checkpoint controllers (ex. p53)
4. Pro-apoptotic.  For instance, loss of Fas and Fas ligand. Mutations in p53 also lead to a loss of Bax expression.  (Gain of function of anti-apoptotic genes will also do: overexpression of Bcl-2 can happen in cancers).
5. Caretaker (ex. DNA repair enzymes)

Here’s a detailed example involving TGF-beta.  Mutations in TGF-beta receptors can cause a loss of incoming pro-apoptotic signal, shutting off the Smad signaling pathway and causing a loss of p15 and Pai-1 exepression.  Loss of p15 deregulates the cell cycle, and loss of Pai-1 allows invasion of the extracellular matrix and escape from the tissue for metastasis.   In order to metastasize, cancer cells must get through the extracellular matrix.  They harness cofilin and WASP in order to form an “invadopodium”.  You can see hair-like structures on some tumor cells in micrographs, which are precursors of invadopodium.

Most cancers are at least ‘two-hit’, involving both activation of oncogenes and inactivation of tumor suppressors.  One interesting line of evidence for a ‘multi-hit’ model of tumor formation is the age-related incidence of cancers.  For a variety of cancers, incidence rises exponentially with age from < 1/100,000 individuals in their 20s – 40s up to ~100-500/100,000 by people’s 70s and 80s. This suggests it is the accumulation of multiple somatic mutations in the same cell which finally allow tumor formation.  For instance, in colon cancer, loss of APC alone will cause small benign polyps. Only when combined with loss of function of K-ras will it cause larger (but still benign) polyps.  Finally when p53 is lost, then the tumor is capable of metastasizing.  Current estimates for the most lethal cancers suggest it takes 5-6 ‘hits’ i.e. 5-6 different genes being mutated, before a malignant tumor can form.  Mouse models also support the multi-hit model.  Mice overexpressing myc or expressing mutant ras^D will both get tumors, but double mutant mice with both of these genes get tumors much earlier in life.

One consequence of the multi-hit nature of cancers is that highly specific cancer drugs such as imatinib, which targets BCR-ABL fusion protein, may not be potent enough to stop the whole tumor.  This is why specific drugs are still often combined with global therapies (“controlled poisons”).  Radiation and chemotherapy just target all rapidly proliferating cells, which is why people lose their hair.

However until now I’ve been confused about exactly what number is meant by the “correlation” of their phenotypes.  I learned about parent-offspring regression from a recent review [Visscher 2008] which says that the slope of the regression line is what is important.  For one parent – one offspring regression, you take 2x the slope, and for midparent – offspring regression, you take 1x the slope.

After I read this I wondered, why the slope and not the variance explained?  Suppose I do a single parent-offspring linear regression and I get a slope of 0.4 (implying h² ≤ 80%) but an R² of only .12.  Couldn’t you argue that if the parent’s phenotype only explains 12% of the variance in the child’s phenotype, then the trait can’t be more than 24% heritable?

To convince myself of which interpretation was correct, I had to do some experiments in R, which I’ll share in a moment.

But first, I also had to refresh myself on how the Pearson’s correlation coefficient (denoted r, rho or ρ) relates to linear regression.  tl;dr: the Pearson’s correlation r is the square root of the unadjusted R² (variance explained) from a linear regression.  In other words, r and R are the same thing.  If you need convincing, just make up any old dataset in R and check that the two are equal:

setseed(9876) # set a random seed so you get the same answer I do
series1 = seq(1,100,1) # make up some data
series2 = series1 + rnorm(n=100,m=0,sd=10) # make up some correlated data
m = lm(series2~series1) # linear regression
unadj.r.sq = summary(m)$r.squared # extract unadj R^2 from a linear model
correl = cor.test(series1,series2) # Pearson's correlation
r = as.numeric(correl$estimate) # extract r from cor.test result
sqrt(unadj.r.sq)
#[1] 0.9385556
r
#[1] 0.9385556

Moving on, my goal is to figure out whether the slope or the variance explained is the figure that relates to heritability.  To do this, I simulated phenotypes for a series of 1000 made-up siblings:

# make up a bunch of random vectors
set.seed(9876)
a = rnorm(n=1000,m=0,sd=1)
b = rnorm(n=1000,m=0,sd=1)
c = rnorm(n=1000,m=0,sd=1)
# create a series of sib pairs for which a common factor (a) ought to explain ~50% of variance in "phenotype"
sib1 = a+b
sib2 = a+c
correl = cor.test(sib1,sib2)
r = round(as.numeric(correl$estimate),2)
m = lm(sib2~sib1)
slope = round(summary(m)$coefficients[2,1],2)
adj.rsq = round(summary(m)$adj.r.squared,2)
plot(sib1,sib2,pch=19,main="made up sib correlation",sub=paste("slope = ",slope," r = ",r," adj R^2=",adj.rsq,sep=""))
abline(m,col="red")

The idea is, there are 1000 pairs of siblings, so every index in the sib1 and sib2 vectors represents a pair, i.e. for all i, sib1[i] is the sibling of sib2[i].  These vectors hold the phenotypes of the two siblings.  A common factor a contributes about half of the phenotype for each of them.  Because the sibs share half their phenotype and share on average half their genome, this corresponds to a trait that could be up to 100% heritable (or it could all be shared environment effect).  When you run the above code, you get this:

Since I made up the data, I know that the effect of a (representing half the additive genetic effect plus all the childhood shared environmental effect) should explain ~50% of the phenotype.  The slope reflects this (.48 ≈ .50) and the variance explained (.24) does not.  If you double the slope 2*.48 = .96 you get a reasonable estimate of the “true” heritability (~100%).

Or let’s do another example, in which the heritability should be only ~50%.  Here I set the standard deviation of e and f (representing stochastic or non-shared environmental factors), to 1.7 (≈√3) so that d (the shared genetic factor), should only account for 25% of overall phenotypic variance.

# an example where the trait should be only ~50% heritable
set.seed(2222)
d = rnorm(n=1000,m=0,sd=1)
e = rnorm(n=1000,m=0,sd=1.7) # 1.7 ~ sqrt(3)
f = rnorm(n=1000,m=0,sd=1.7)
var(d)
#[1] 1.006501
var(e)
#[1] 3.038408
var(d)/var(c(sib1,sib2))
#[1] 0.2578496 # this is 1/2 of the "true" heritability
# now try to measure heritability by regression
sib1 = d+e
sib2 = d+f
correl = cor.test(sib1,sib2)
r = round(as.numeric(correl$estimate),2)
m = lm(sib2~sib1)
slope = round(summary(m)$coefficients[2,1],2)
adj.rsq = round(summary(m)$adj.r.squared,2)
plot(sib1,sib2,pch=19,main="another made up sib correlation",sub=paste("slope = ",slope," r = ",r," adj R^2=",adj.rsq,sep=""))
abline(m,col="red")

Again, 2*slope = .48 is pretty close to a correct heritability estimate (actually, according to 2*var(d)/var(c(sib1,sib2)) the true heritability ended up being ~51%).  The R² is miniscule, at .06.

So it appears to be 2*slope, not 2*(variance explained), that estimates the upper limit on heritability.  Again, it’s an upper limit because the “shared” factors I spiked in (a and then d) could represent genetic or shared environmental factors.  Indeed, as the collective genius of Wikipedia has put it:

sibling phenotypic correlation is an index of familiarity – the sum of half the additive genetic variance plus full effect of the common environment

But the use of the term “correlation” in this quote alludes to a fact you’ve probably noticed in both of the above examples: slope ≈ r.

Of course, there’s no rule that says r has to equal slope.  By definition, Pearson’s correlation does not tell you anything about the steepness of slope (though its sign tells you direction of slope), as is beautifully illustrated in this Wikimedia Commons graphic:

Or if that doesn’t convince you,  just make up some data in R to convince yourself:

x = seq(1,100,1)
y = x^2
r = as.numeric(cor.test(x,y)$estimate)
r
# 0.9688545 
slope = summary(lm(y~x))$coefficients[2,1]
slope
# [1] 101

That said, it seems that r and slope are likely to be pretty close in any plausible model of a real genetic relationship.  After all, for what earthly biological reason would one sibling’s phenotype ever be the square of the other’s?  Especially when you consider there is usually no logical ordering to sib pairs (it’s random which member of any pair ends up plotted on which axis), you’re just unlikely to get a really steep slope with a really small r, or a really tight fit (large r) for a slight slope.  They can vary a bit, but I can’t think of a good reason why the two quantities should be hugely different, so if I ever find that slope and r are wildly different, I’ll consider that a cue to do double check my data and my analysis.

The observation that r and slope are likely to be pretty similar may resolve a confusion I’ve had.  The literature on twin studies [ex. Deary 2006 (ft); see also the above quote and Falconer's formula] and sib-sib correlation tends to refer to the “correlation” (presumably r) of the two siblings, while the literature on parent-offspring regression refers to the “slope” [Visscher 2008].  Perhaps this discrepancy owes simply to the fact that these quantities are very likely to be quite close in any real genetic dataset.  If anyone can think of a reason why r is more correct for sibs and slope is more correct for parent-offspring, leave me a comment.

This lecture will cover two different ways cells can die: apoptosis (programmed cell death) and necrosis (unplanned cell death).  It is easy to tell these two apart morphologically under the microscope, as shown in this Wikimedia Commons image:

necrosis

Necrosis is when cells die accidentally due to, say, trauma (ex. a poisonous spider bite), or lack of nutrients (ex. lack of blood supply).  Necrosis begins with cell swelling, the chromatin gets digested, the plasma and organelle membranes are disrupted, the ER vacuolizes, the organelles break down completely and finally the cell lyses, spewing its intracellular content and eliciting an immune response (inflammation).

apoptosis

Apoptosis can constitute cell suicide or cell murder.  Cells will commit suicide when they lack any incoming survival signal in the form of trophic factors, or when they detect extensive DNA damage in their own nucleus.  Cells will murder other cells to clear out unneeded cells or to eliminate potentially self-attacking immune cells.

Either of these processes constitutes programmed cell death.  During embryonic development, people have webbed hands and feet and tails; the cells that constitute those parts later apoptize.  Apoptosis also goes on constantly in many tissues including the intestines.

Here’s a stunning Wikimedia Commons image of apoptosis (read left to right, top to bottom) thanks to Egelberg:

Major steps of apoptosis:

• Cell shrinks
• Cell fragments
• Cytoskeleton collapses
• Nuclear envelope disassembles
• Cells release apoptotic bodies

Notably absent from this list is ‘send out a signal.’  Apoptotic cells do not send out any signal, with one exception: they release apoptotic bodies and ‘engulfment proteins’ to induce other cells (‘phagocytic’ cells) to engulf the apoptotic bodies and and break them down in their lysosomes, but this is not much of an immune response.

Proteins important in apoptosis:

• ‘killer proteins’: the caspases (discussed in detail below).
• ‘destruction proteins’ that digest DNA, fragment the cell and break down the cytoskeleton
• ‘engulfment proteins’ that elicit and promote phagocytosis by other cells

C. elegans has been the major model organism for understanding apoptosis, both by forward and reverse genetics.  Forward genetics is observing a phenotype and then determining which gene gives rise to it; reverse genetics is introducing a mutation into a known gene in order to see what phenotype results.

The key pathway in C. elegans apoptosis is shown in this Google Drawing I created:

Here’s an explanation of how each of these proteins does its job, from bottom up:

are:

• CED-3 pulls the trigger, activating apoptotic proteins that destroy the cell.  (In the mammalian equivalent, CED-3 is Caspase 9, which cleaves-thereby-activating Caspase 3, which in turn destroys the cell.)
• CED-4 activates CED-3.
• CED-9 binds to CED-4, preventing its activation
• EGL-1 is transcriptionally activated in response to death signals and catalyzes the release of CED-4 from CED-9.

Note that there is no robustness in this system – it is single points of failure all the way through.  If CED-3 is knocked out, no apoptosis can occur.  If CED-4 is knocked out, no apoptosis can occur.  If CED-9 is knocked out, every cell in the worm will apoptose.  If EGL-1 is knocked out, no apoptosis can occur.  Note that the order the arrows point in the above diagram reflects the flow of information in the system.  For instance, if EGL-1 and CED-9 are both knocked out, it’s the same as if CED-9 alone was knocked out: every cell will apoptose.

In mammals, apoptosis is governed chiefly by caspases (cysteine-aspartic proteases).  The entire caspase pathway is post-translationally regulated: the caspases are always present in inactive form (called procaspases, containing a prodomain, which contains a caspase recruitment domain (CARD)) and can be activated by cleavage. This allows a very quick response if cell suicide is needed.  In order for apoptosis to occur, the initiator caspases  must be cleaved and dimerize.  Thus activated, they must then cleave the effector caspases (aka pro-caspases), triggering a ‘caspase cascade’.  This amplifies the number of activated caspases in the cell.  The effector caspase have many targets including the nuclear lamina and cytoskeleton.

There are both pro-survival and pro-apoptotic caspases, and they share many common domains.  Pro-survival caspases have BH1, 2, 3 and 4; pro-apoptosis caspases have either BH1, 2 and 3 or just BH3.

Inhibitor of apoptosis proteins (IAPs) restrain both the initiator and effector caspases.  They each have a zinc binding domain that binds directly to caspases, inhibiting their activity.

However, there are also mitochondrial proteins called SMAC and DIABLO which inhibit the inhibitors.  Upon mitochondrial injury they are released and will bind IAPs, freeing the caspases to go cause apoptosis.  Another collection of mitochondrial proteins called Htra2/Omi, apoptosis-inducing factor (AIF) and endonuclease G can also be released and will cleave IAPs.  AIF also causes chromosome condensation and DNA fragmentation independent of caspases.

Indeed, the mitochondria are central regulators of apoptosis.  Outer mitochondrial membrane proteins Bcl-2, the BH3-only proteins and Bax are involved: Bax can form a pore in the membrane to allow cytochrome c, normally located in the intermembrane space, out into the cytosol.  Bax monomers move from the cytoplasm to the outer mitochondrial membrane, where they oligomerize and permit the influx of ions through the membrane.  This has also been shown in in vitro experiments where you can show that vesicles made of outer mitochondrial memrbanes are permeabilized in the presence of Bax.  It is not currently known why this influx of ions leads to cytochrome c release.

Bcl-2 prevents release of cytochrome c, thus blocking apoptosis.  Bcl-2 was the first mammalian apoptosis gene to be cloned.  In some lymphomas, it gets translocated to a position under a stronger promoter, causing overexpression that prevents the cancer cell from apoptosing. See also bad & bid.

Once cytochrome c is released, it binds to Apaf-1 (apoptotic protease activating factor), causing the latter to hydrolyze the ATP to which it is usually bound, thus causing a conformational change that activates Apaf-1 and triggers the caspase cascade. Apaf-1 forms a disc-shaped heptamer called the ‘wheel of death’ or apoptosome which activates caspases (Wikimedia Commons image by Org1012):

When  a trophic factor is present, the receptor activates PI3K, which activates PKB/Akt, which phosphorylates Bad.  p-Bad is then retained in the cytosol by 14-3-3, preventing p-Bad from inhibiting Bcl-2. Thus apoptosis is prevented.

Trophic factors are an example of a cell extrinsic signal that promotes survival.  There are also extrinsic signals that promote death (this is cell murder).  Tumor necrosis factor (TNF-alpha) is released by macrophages to trigger cell death by binding to ‘death receptors’.  Death receptors have a single transmembrane domain.  They must trimerize in order to activate FADD (Fas-associated death domain).  These serve as adapters for caspase-8 and -10 and form a death-inducing signaling complex (DISC) which can initiate the caspase cascade.  Though this whole process originates independent of mitochondria, it can also activate (?) t-Bid, leading to a mitochondrial apoptosis signal as well.

Cells can become murder-resistant by expressing decoy receptors which have only the ‘death ligand’ binding domain and no active cytosolic domain.  This occurs sometimes normally in animal cells but is also a trick that some viruses use – they encode decoy receptor proteins to keep their host cells safe from immune attack.

TNF-alpha usually promotes death, but can also promote survival in certain cell types by activating NF-κB.  Sometimes cells use decoy receptors to promote an inflammatory response instead of death.

p53 is a key regulator of DNA damage response and can promote DNA repair, apoptosis or cell cycle arrest.  It does this by binding to promoters of target genes.  It is still not clear what determines when p53 will induce cell cycle arrest versus apoptosis.

experimental methods

Apoptotic cells exhibit a particular chemical signature.  One of these is that an endonuclease cleaves DNA into fragments in the linker regions between nucleosomes and the resulting fragments form a ladder when run on a gel.  Another is TUNEL (Terminal deoxynucleotide transferase dUTP Nick End Labeling) staining.  This involves adding a Tdt enzyme and a BrdU which Tdt will add to the ends of cleaved DNA.  After giving it a chance to do this you wash away excess BrdU and then use an antibody against BrdU.  Yet another method is that phosphatidylserine (PS) is normally located in the cytosolic leaflet of the plasma membrane; during apoptosis, it flips to the exoplasmic leaflet, where it serves as a signal to request other cells to phagocytose the dying cell.  A fluorescently labeled annexin V protein can label PS on the outside of apoptotic cells.

Double-stranded DNA cannot get through the plasma membrane of intact cells – and that means healthy cells and apoptotic cells.  If it does get out, that is a sign of necrosis.  So you can stain with annexin V for exoplasmic PS and with 7-AAD for dsDNA; apoptotic cells are those which are positive for annexin V but negative for 7-AAD.

concluding video

In sum, here is a disturbing video about apoptosis:

relevance to PrP

In general, in nature, cells either die by apoptosis, necrosis or by autophagy (meaning, in this case, getting engulfed whole by other cells).  There aren’t really any other ways to go.  One of the many mysterious things about prion diseases is how neurons die in the prion-infected brain – they do not obviously appear to follow any of these paths.  Here’s a quote from an excellent recent paper on toxic mechanisms of prion disease [Moreno 2012]:

Caspase 12 cleavage occurred at 10wpi, following rising CHOP expression… coincident with onset of neuronal loss… however the exact effector mechanism of neuronal death is unclear: we found neither apoptosis, nor autophagy, nor necrosis on examination of hippocampal slices… and neither Bax deletion, nor Bcl-2 overexpression, nor caspase 12 deficiency are neuroprotective in prion disease.

In addition to her own evidence, Moreno cites the study of prion infection in mouse models with Bax (a pro-apoptotic protein) deleted or Bcl-2 (an anti-apoptotic protein) overexpressed – two different ways of blocking apoptosis.  Neither of these mouse models had any delay or amelioration of prion disease [Steele 2007a].  Another apoptotic protein, Caspase-12, undergoes proteolytic processing during prion infection, but deletion of Caspase-12 also did not change the course of prion disease [Steele 2007b].

That word – ‘late’ – is very important here.  Hosts of compounds have shown an ability to delay symptoms of prion disease if administered early in the incubation period, long before symptoms.  But 85% of prion disease cases in humans are sporadic CJD patients [Appleby & Lyketsos 2011], who will never know they have a disease until they are already sick.  A compound that is going to work for these people has to work in mice that are already showing symptoms of disease.

But that turns out to be a tricky thing to quantify, because whether mice are showing symptoms depends in part on how hard you look.  Mallucci’s first paper on Cre-mediated PrP knockout in mice [Mallucci 2003] didn’t report any symptoms prior to 10 weeks post infection (wpi), but a few years later, through more meticulous behavioral testing, she was able to find certain symptoms starting at 8 wpi, in the exact same mouse model [Mallucci 2007].

Wagner 2013 makes the claim that the mice treated with anle138b at 80 dpi and 120 dpi were treated “after onset of disease”.  That is certainly true for 120 dpi – Wagner provides his own data for weight loss beginning by that time – but for 80 dpi he cites Mallucci 2007.  That’s where comparisons get tricky.  Mallucci found behavioral symptoms at 8 wpi = 56 dpi, in a model where mice die at ~12 wpi = 84 dpi.  That’s a much faster mouse model than Wagner’s model, in which controls live to ~180 dpi, so symptoms may well appear earlier.

That prompts me to re-iterate my statement from the anle138b post: different prion mouse models get sick at different times.  Incubation time of prions in mice is exquisitely correlated with both the PrP expression level of the mouse and the infectious titer of the inoculum.  It’s also affected by prion strain and mouse background.  So comparisons between models need to adjust accordingly.

Whether or not Wagner’s mice had symptoms at 80 dpi, it looks like they definitely did by 120 dpi.  Yet other investigators have treated similar mouse models at similar timepoints – say, treatment at 100 dpi in a model where controls lived to 180 dpi in [Riemer 2008 (ft)] – and not made the claim of treating already-sick mice.  That’s probably just because, again, finding early symptoms depends on how hard you look.  Importantly, Wagner’s mice were not at the stage that researchers call ‘clinical signs’ or ‘clinical onset’ (impaired gait, etc) at the time when they began treatment.

Terms like ‘late’ and ‘after onset’ are a bit subjective, and to anchor this conversation in real past experience, I think a comparison to pentosan polysulfate (PPS) is merited. Wagner et al largely omit PPS from their summary of other proposed anti-prion compounds (Fig S20 on p. 31 of supp1), because they only consider compounds that have been administered peripherally to animals infected intracerebrally.  I completely agree on the importance of only considering such stringent mouse models going forward, but by omitting the original PPS experiments in mice [Doh-Ura 2004], Wagner et al have left out an important piece of prion history.

Doh-Ura used a very rapid mouse model in which controls die at 51 dpi.  Doh-Ura didn’t look hard for early symptoms; he reported that the animals exhibit “ambiguous signs of reduced activity about 2 days prior to death”.  Because PPS does not cross the blood-brain barrier, it was administered by continuous intraventricular infusion with a surgically implanted cannula.  It extended survival by a huge amount in mice infected with 263K prions (the primary model used), but its effects were time-dependent.  It more than doubled incubation time if administered starting at 7 dpi, but had no significant effect when given at 42 dpi.  Because the raw numbers in terms of days post-infection are so different between Doh-Ura’s study and Wagner’s study, I find it useful to convert dpi numbers into a new figure as follows:

%diseasecourse =  treatment started dpi / control survival

And I also convert therapeutic effects into a percentage delay:

%delay = treatment survival / control survival – 1

When I do this and compare PPS with anle138b, I get the following table (PPS numbers are eyeballed from Doh-Ura Fig 2B since he doesn’t give the raw data):

You can see that Doh-Ura’s trial of PPS at 42 dpi was ~80% of the way through the disease course, later than any timepoint tested by Wagner.  The fact that Doh-Ura reported no symptoms until ‘ambiguous signs of reduced activity’ at 49 dpi while Wagner reported that 80 and 120 dpi (respectively 48% and 70% of the way through the control survival time) already constituted ‘treatment after onset of disease’ might reflect real differences in mouse models, but more likely just reflects how hard they looked for symptoms.

When I plot these data in R:

ppsx=c(14,41,69,82)
ppsy=c(145,86,47,0)
anlex=c(0,48,70)
anley=c(97,44,30)
main = "extension of survival of PPS vs. anle138b"
sub = "PPS [Doh-Ura 2004], anle138b [Wagner 2013]"
xlab = "% of disease course that had passed before treatment began"
ylab = "% extension of survival in treatment group vs. control"
plot(ppsx,ppsy,col="purple",pch=19,xlim=c(-.01,100),xlab=xlab,ylab=ylab,main=main,sub=sub)
points(anlex,anley,col="orange",pch=19)
points(anlex,anley,col="orange",type="l")
points(ppsx,ppsy,col="purple",type="l")
legend("topright",legend=c("PPS","anle138b"),col=c("purple","orange"),pch=19)

Then you can see the difference in survival between the two compounds:

By these measures, PPS was actually more effective at all tested timepoints in terms of extending survival than anle138b, and yet still it had no effect when it was administered at 42 dpi, ~80% of the way through the disease course, a later timepoint than any tested by Wagner.

In spite of the evidence that PPS did not work late in disease, it went on to clinical trials in Japan and the U.K., which were roundly unsuccessful.  To be sure, PPS was a rather problematic drug which had to be infused intraventricularly via cannula in humans just as it had been in mice.  But PPS didn’t fail clinical trials due to cannula-related complications (though there were some of those too), it failed on efficacy, showing no effect at all on survival of treated patients.

Maybe that was due to strain specificity or other problems in translating drugs from mouse to humans, but one can make a good argument that the timepoint of treatment is sufficient to explain PPS’s failure.  If you read carefully the official report on the U.K. clinical trial [Bone 2008], you’ll see that most of the patients were already unable to walk by the time they began pentosan polysulfate treatment. That’s a lot worse than Doh-Ura’s “ambiguous signs of reduced activity” which the mice reached only in their last day or two of life, and probably represents a timepoint far later (perhaps at ~170-180 dpi) than any timepoint where Wagner began treating mice with anle-138b.

True, you can notice symptoms in mice very early if you look hard enough. And one would certainly hope that we humans are better at recognizing symptoms in our fellow humans than we are at recognizing symptoms in mice.  It could be speculated that some even earlier timepoint in Wagner’s mice – 30, 40, 50 dpi? – corresponds to when we would notice symptoms in humans.

However, the path from first symptoms to a diagnosis can be long, at least when there is no family history of prion disease.  From the time when patients first show symptoms, family members need to notice, become worried, see a doctor, get referred to a specialist, and then the specialist has to run a battery of tests and puzzle over what the problem is.  sCJD is notorious for appearing to neurologists as lyme disease, a viral infection, an adverse drug reaction, or a host of other possible conditions [Murray 2011].  Once a diagnosis finally is made, then also allow time for a referral to a specialized center such as Cleveland Clinic or MRC Prion Unit.  No wonder the patients treated with PPS were in such bad shape by the time they finally entered the clinical trial.

So treating mice at a time when some subtle behavioral symptoms or weight loss are present may not represent a timepoint at which humans will be treated.  On the other hand, if we were to simply accept that patients will not be available for treatment until they are already in such a debilitated state, then we are basically accepting that no treatment is possible.  By the time a patient is bedridden, a large fraction of neurons in the most-affected brain regions have probably already been lost, and though a total knockout of PrP (were it possible) would still stop the spread of prions, it would not bring back cognitive function, because neurogenesis in adults is fairly limited.

Therefore we simply need to find a way to act earlier.  Mallucci has made an excellent case for the existence of a window for intervention in prion disease where neurons are impaired but not yet lost [White & Mallucci 2009, Moreno 2012].  Mallucci’s stage is one where some behavioral symptoms are present in mice and so it may correspond to a stage in humans where we could diagnose the disease, even if we aren’t doing so consistently yet.

And having a treatment available would itself drive earlier diagnosis.  Murray’s diagnostic paper, oriented towards clinical neurologists, is very enlightening in this regard.  The abstract declares that the paper “highlights the relatively rare treatable mimics which must not be missed” (my underlining).  Table 2 lists about 15 other conditions (some as broad as ‘viral encephalitis’) that can be difficult to distinguish from sCJD, and – this is crucial – the caption on the table reads “Those in bold represent potentially treatable conditions which must not be missed” (again, my underlining). The paper discusses the various diagnostic tools for sCJD including the test for 14-3-3 protein but finally, in the Practice Points, it summarizes that, for sCJD, “Disease duration is the most reliable distinguishing feature” [Murray 2011].

This is the practice of neurology today: there are a lot of conditions that can cause overlapping sets of symptoms, and it is difficult to wade through them all to find the right diagnosis, so doctors justifiably prioritize looking at the diagnoses that would be treatable.  So simply having a treatment available will make a difference in diagnosis – it will get CJD to be one of the bold items in Murray’s list.

Despite the critical view I’ve taken here of how ‘late’ 80 dpi and 120 dpi really are, I actually do not disagree at all with Wagner’s general claim that anle138b is the most promising treatment yet proposed for prion diseases.  Its BBB permeability, oral bioavailability, low toxicity, large and (by in vitro PMCA at least) non-strain-specific inhibitory effects are a compelling package of attributes that no previous drug has combined.  I am hopeful that it will make its way to clinical trials soon.  When it does, it will take an all-out effort to get patients treated as early as humanly possible.

And, when it does, we will also need to consider the ambiguous questions about what ‘success’ will look like.  From what we see in mice, anle138b does not reverse disease, it just extends survival.  Until we have a clinical trial, we are unlikley to know whether or not that will translate into any improvement in quality of life for patients who are already debilitated by prion diseases.  If a drug extends life without improvement in quality of life, can we call that a success?  Clinical trials follow pre-defined clinical endpoints, of which time to death is often one, so yes, a drug could potentially be approved even if it didn’t improve quality of life.  If such a drug got approved, would that be a success?

I tend to think yes, because the existence of an approved drug would further raise public and physician awareness and the priority placed on diagnoses of prion disease, making it possible for patients to be treated earlier and earlier in the disease course when it might start to make a real quality of life difference too.  And it would also be our first step at fighting these diseases, to be combined with future treatments yet to be discovered.