Yesterday a new study, [Moreno 2013, paywalled] made the rounds as a big news story.  In just a few hours, several different people including my mom, my former boss and a close friend have emailed me the BBC article calling this study a “turning point” in Alzheimer’s research.  It seems safe to say this is a greater amount of attention than I’ve seen for any other news story in the two years I have been following prion disease.

The underlying study is well-executed research of substantial importance.  However, the press hype is out of all proportion to the impact of this research.  This post walks through what this study does and does not show.


Last year, the same group of researchers described a signaling pathway which appears to be involved in neurotoxicity in prion disease [Moreno 2012].  Prion diseases such as Cretuzfeldt-Jakob disease and fatal familial insomnia are caused by misfolded prion protein, dubbed PrPSc.  PrPSc accumulates over the course of the disease and Moreno showed in mice that it eventually triggers the unfolded protein response, leading to phosphorylation (and therefore activation) of PERK,  which then phosphorylates eIF2α, dramatically reducing the overall rate at which new proteins are translated in the cell.  This draconian mechanism might have evolved as a way to give the cell a break from degrading misfolded proteins, but it backfires in the case of prion disease because PrP mRNA has multiple untranslated open reading frames (uORFs) in its 5′UTR, a property which allows it to escape translational repression.  In fact, Moreno showed that PrP paradoxically is produced in even greater-than-usual quantities when the unfolded protein response is activated.

In that original study, the authors tried two different ways of intervening in this process [Fig 4].  Injecting the mice with lentiviral vectors to overexpress GADD34, a phosphatase which dephosphorylates eIF2α, (thereby inhibiting the unfolded protein response), extended survival of the mice by 10% (90 days vs. 82 days control).  Meanwhile, treating the mice with salubrinal, a small molecule which inhibits GADD34 (thus promoting the unfolded protein response) accelerated disease, leading to death a few days earlier (the exact numbers are never given but it’s probably a < 5% difference).

The interpretation of these results was that the unfolded protein response is bad in prion disease: it accelerates neuronal death by reducing protein production, while simultaneously failing to halt the production of the protein at the root of the problem: PrP.

The lentiviral overexpression of GADD34 provided one way to intervene against the unfolded protein response, but it wasn’t a practical therapy.  After all, if and when we perfect the viral delivery of gene therapy, we’ll probably just use it to knock down PrP itself – which Mallucci has also tried [White 2008] – rather than looking for other drug targets such as the unfolded protein response.  But in the same year, GlaxoSmithKline published its discovery of a drug-like small molecule called GSK2606414 which can penetrate the blood-brain barrier and specifically inhibits PERK [Axten 2012].  This opened up an opportunity to attack the same pathway with a more therapeutically relevant approach: a potent compound that could be delivered orally and reach all areas of the brain.

this study

The new study, published yesterday, uses the new GlaxoSmithKline molecule, GSK2606414, to inhibit the unfolded protein response and delay the clinical signs of disease in prion-infected mice [Moreno 2013].

The study used Mallucci’s preferred mouse model, tg37 mice which overexpress PrP and have short prion incubation times, with terminal illness occurring about 85 days post infection (dpi).  For any readers who aren’t familiar with prion research: prion diseases have a long silent incubation period in which PrPSc replicates but there are no symptoms – no behavioral abnormalities and no pathological changes visible in the brain.  Mallucci has been a pioneer of looking for the very earliest indicator signs of prion disease and the earliest phenotypes she has found occur about 2/3 of the way through the disease course – at 56 dpi in mice that succumbed to disease at 84 dpi [Mallucci 2007].  Because different mouse models have different incubation times, I standardize by referring to the day of infection as timepoint 0 and the day of terminal illness as timepoint 1, so the earliest symptoms in Mallucci’s model are found at timepoint ~0.66

In this new study, Mallucci treated the mice with GSK2606414 at times 0.58 and 0.75.  The mice in the early group were free of symptoms, while some mice in the late group already had some early symptoms.  While all of the untreated control mice had reached late-stage, “clinical” illness by 84 dpi, none of the treated mice had done so.  Behavioral tests and examination of the brains of treated mice at 84 dpi revealed early signs of prion disease in some mice (~25% of the early-treated mice and ~50% of the late-treated mice) but no signs of terminal or “clinical” disease in any of them.

However, the mice were not monitored for longer to see how long they would survive or how long they would remain disease-free.  GSK2606414 doesn’t affect protein translation only in the brain, it acts throughout the whole body, and its effects in the pancreas appeared to cause some pre-diabetic changes, with mice having increased blood glucose and weight loss of ~20%.  According to the animal welfare rules of Mallucci’s institution, the weight loss meant that the mice had to be sacrificed, and so they weren’t monitored past 84 dpi to see when disease would set in.

Therefore we are left to guess just how effective this treatment was.  Since all control mice had reached clinical illness by 84 days but none of the treated mice had, the treatment must surely have delayed clinical illness by at least a couple of standard deviations, say ~10 days, which would be a delay of at least 10/84 = 12%.  That much we can be fairly certain of.  Meanwhile, control mice were showing early signs of prion disease by 63 dpi, while not all of the treated mice had these signs even by 84 dpi, suggesting the delay might have been more like 84/63-1 = 33%.  Of course, it could have been even more – but we can’t conclude that from this study.

In general, it is always hard to say how percentages like these will project onto the human disease course.  It’s especially hard in this case because the PERK inhibitor approach does not target the underlying cause of the disease – PrPSc – but rather seeks to allow neurons to tolerate a greater accumulation of PrPSc before dying.  That’s exciting because it provides a proof of principle that targeting this more general pathway can be therapeutically valuable.  But it probably also limits the efficacy of this approach for prion disease.  It is unlikely that the unfolded protein response is the only thing toxic about PrPSc – there are several other proposed toxic mechanisms (partial review at the beginning of this post) which may kill neurons if the unfolded protein response doesn’t get them first.

All in all, it’s a great study, very interesting and I’ll be excited to see the followup.  But it does not deserve the extreme hype it is receiving right now.  Here are a few reasons for a more moderate view.

1. there is no evidence this compound prevented neurodegeneration

In a statement to BBC, Mallucci is quoted as saying:

What’s really exciting is a compound has completely prevented neurodegeneration and that’s a first.

In order to know that the compound completely prevented, and did not merely delay, neurodegeneration, it would be necessary to follow the mice for the entirety of their natural lifespan (> 2 years), and observe that they never develop signs of neurological disease.  That was not done in this study – the mice were only followed to 84 days after prion infection, at which point they had to be sacrificed due to pre-diabetic metabolic changes associated with chronic use of the drug.

For some perspective, consider that several compounds including anle138b, cpd-B, and 2-aminothiazoles have given delays in onset of at least the same magnitude as this study when mice were treated at similar timepoints [Wagner 2013, Lu & Giles 2013 (ft), Berry 2013 unpublished].  In any of those studies, if the mice had been examined just a few weeks after the time of disease onset in the control mice and then monitored no further, it probably would have appeared that these treatments too had “completely prevented” neurodegeneration.  However, by following the mice for longer those authors were able to observe that the treatments merely delayed neurodegeneration.

2. the adverse effects may be unavoidable

Commenting on the adverse effects that led to the premature termination of the study, the BBC writes:

Side effects are an issue. The compound also acted on the pancreas, meaning the mice developed a mild form of diabetes and lost weight.

In fact, this is probably not a side effect.  It’s more likely a main effect.  In the paper, the authors cite evidence that hemizygous PERK knockout mice suffer a similar weight loss and hyperglycemia phenotype [Harding 2001], suggesting that this could be directly due to GSK2606414′s intended effect – PERK inhibition – rather than due to an off-target interaction with some other protein.  If so, then avoiding this adverse effect while trying to develop a drug for human use will be challenging indeed.

Addressing this issue, BBC responds that “Any human drug would need to act only on the brain.”  Is there any precedent for such a thing?  Can anyone point me to a drug taken orally that localizes only in the brain while sparing the rest of the body’s tissues?  I am not aware of any drug that is able to do this.  update – see comment below.

3. the relevance to Alzheimer’s remains to be shown

Time, CBSBBC and The Independent all structured their articles mostly around Alzheimer’s disease. No doubt, there are intimate links between prion disease and Alzheimer’s disease.  But there is not a ton of evidence to say that the very specific pathway targeted in this study is shared between the two diseases.

Activation of the unfolded protein response reduces the translation rate of most proteins, which intuitively seems helpful for neurons trying to survive an accumulation of misfolded proteins.  But this backfires in prion disease partly because PrP itself is among the few proteins spared from this translational repression [Moreno 2012].

There isn’t much literature out there discussing whether Alzheimer’s shares this peculiarity.  Mallucci’s study does not cite, nor did I find online, any evidence that APP – the protein involved in Alzheimer’s – is one of the lucky few proteins to escape translational repression when the unfolded protein response is activated.  I did find one study showing that BACE1, the protein that cleaves APP to create amyloid beta (Aβ), is upregulated during translational repression [O'Connor 2008] and therefore that amyloid beta production does increase when the unfolded protein response is activated. This would indeed suggest that PERK inhibition could be helpful for Alzheimer’s.  But there are also papers on the opposing side.  One article suggests that blocking the unfolded protein response actually increases the sensitivity of neurons to amyloid beta toxicity [Lewerenz & Maher 2009] and another study using salubrinal – a compound which has the exact opposite effect of GSK2606414 – to try to prevent amyloid beta toxicity in cell culture [Huang 2012].  In other words, the unfolded protein response might actually be helpful in Alzheimer’s, and so blocking that response with GSK2606414 might actually make Alzheimer’s worse.  All in all, there hasn’t been a ton of research on this specific question, and it is far too early to claim that Mallucci’s results in a prion disease model could be extended to Alzheimer’s.