Barking up the wrong therapeutic tree in Alzheimer’s disease

Billions have been spent by big pharma (and lost) trying to get rid of the senile plaque of Alzheimer’s disease.  The assumption has always been that the plaque is the smoking gun killing neurons.  But it’s just an assumption which a recent paper has turned on its ear [ Proc. Natl. Acad. Sci. vol. 116 23040 – 23049 ’19 ]

It involves a protein, likely to be a new face even to Alzheimer’s cognoscenti.  The protein is called SERF1A (in man) and MOAG-4 in yeast. It enhances amyloid formation, the major component of the senile plaque.  SERF1A is clearly doing something important as it has changed little from the humble single yeast cell to man.

The major component of the senile plaque is the aBeta peptide of 40 and/or 42 amino acids.  It polymerizes to form the amyloid of the plaque.  The initial step of amyloid formation is the hardest, getting a bunch of Abeta peptides into the right conformation (e.g. the nucleus) so others can latch on to it and form the amyloid fiber.   It is likely that the monomers and oligomers of Abeta are what is killing neurons, not the plaques, otherwise why would natural selection/evolution keep SERF1A around?

So, billions of dollars later, getting rid of the senile plaque turns out to be a bad idea. What we want to do is increase SERF1A activity, to sop up the monomers and oligomers. It is a 180 degree shift in our thinking. That’s the executive summary, now for the fascinating chemistry involved.

First the structure of SERF1A — that is to say its amino acid sequence.  (For the nonChemists — proteins are linear string of amino acids, just as a word is a linear string of characters — the order is quite important — just as united and untied mean two very different things). There are only 68 amino acids in SERF1A of which 14 are lysine 9 are arginine 5 Glutamic acid and 5 Aspartic acid.  That’s interesting in itself, as we have 20 different amino acids, and if they occurred randomly you’d expect about 3 -4 of each.  The mathematicians among you should enjoy figuring out just how improbable this compared to random assortment. So just four amino acids account for 33 of the 68 in SERF1A  Even more interesting is the fact that all 4 are charged at body pH — lysine and arginine are positively charged because their nitrogen picks up protons, and glutamic and aspartic acid are negatively charged  giving up the proton.

This means that positive and negative can bind to each other (something energetically quite favorable).  How many ways are there for the 10 acids to bind to the 23 bases?  Just 23 x 22 x 21 X 20 X 19 X 18 x 17 x 16 x 15 x 14 or roughly 20^10 ways.  This means that SERF1A doesn’t have a single structure, but many of them.  It is basically a disordered protein.

The paper shows exactly this, that several conformations of SERF1 are seen in solution, and that it binds to Abeta forming a ‘fuzzy complex’, in which the number of Abetas and SERF1s are not fixed — e.g. there is no fixed stoichiometry — something chemists are going to have to learn to deal with — see — https://luysii.wordpress.com/2018/12/16/bye-bye-stoichiometry/.  Also different conformations of SERF1A are present in the fuzzy complex, explaining why it has such a peculiar amino acid composition.  Clever no?  Let’s hear it for the blind watchmaker or whatever you want to call it.

The paper shows that SERF1 increases the rate at which Abeta forms the nucleus of the amyloid fiber.  It does not help the fiber grow.  This means that the fiber is good and the monomers and oligomers are bad.  Not only that but SERF1 has exactly the same effect with alpha-synuclein, the main protein of the Lewy body of Parkinsonism.

So the paper represents a huge paradigm shift in our understanding of the cause of at least 2 bad neurological diseases.   Even more importantly, the paper suggests a completely new way to attack them.

Technology marches on — or does it?

Technology marches on — perhaps.  But it certainly did in the following Alzheimer’s research [ Neuron vol. 104 pp. 256 – 270 ’19 ] .  The work used (1) CRISPR (2) iPSCs (3) transcriptomics (4) translatomics to study Alzheimer’s.  Almost none of this would have been possible 10 years ago.

Presently over 200 mutations are known in (1) the amyloid precursor protein — APP (2) presenilin1 (3) presenilin2.  The presenilins are components of the gamma secretase complex which cleaves APP on the way to the way to the major components of the senile plaque, Abeta40 and Abeta42.

There’s a lot of nomenclature, so here’s a brief review.  The amyloid precursor protein (APP) comes in 3 isoforms containing 770, 751 and 695 amino acids.  APP is embedded in the plasma membrane with most of the amino acids extracellular.  The crucial enzyme for breaking APP down is gamma secretase, which cleaves APP inside the membrane.  Gamma secretase is made of 4 proteins, 2 of which are the presenilins.  Cleavage results in a small carboxy terminal fragment (which the paper calls beta-CTF) and a large amino terminal fragment. If beta secretase (another enzyme) cleaves the amino terminal fragment Abeta40 and Abeta42 are formed.  If alpha secretase (a third enzyme) cleaves the amino terminal fragment — Abeta42 is not formed.   Got all that?

Where do CRISPR and iPSCs come in?  iPSC stands for induced pluripotent stem cells, which can be made from cells in your skin (but not easily).  Subsequently adding the appropriate witches brew can cause them to differentiate into a variety of cells — cortical neurons in this case.

CRISPR was then used to introduce mutations characteristic of familial Alzheimer’s disease into either APP or presenilin1.  Some 16 cell lines each containing a different familial Alzheimer disease mutation were formed.

Then the iPSCs were differentiated into cortical neurons, and the mRNAs (transcriptomics) and proteins made from them (translatomics) were studied.

Certainly a technological tour de force.

What did they find?  Well for the APP and the presenilin1 mutations had effects on Abeta peptide production (but they differered).  Both however increased the accumulation of beta-CTF.  This could be ‘rescued’ by inhibition of beta-secretase — but unfortunately clinical trials have not shown beta-secretase inhibitors to be helpful.

What did increased beta-CTF actually do — there was enlargement of early endosomes in all the cell lines.   How this produces Alzheimer’s disease is anyone’s guess.

Also quite interesting, is the fact that translatomics and transcriptomics of all 16 cell lines showed ‘dysregulation’ of genes which have been associated with Alzheimer’s disease risk — these include APOE, CLU and SORL1.

Certainly a masterpiece of technological virtuosity.

So technology gives us bigger and better results

Or does it?

There was a very interesting paper on the effect of sleep on cerebrospinal fluid and blood flow in the brain [ Science vol. 366 pp. 372 – 373 ’19 ] It contained the following statement –”

During slow wave sleep, the cerebral blood flow is reduced by 25%, which lowers cerebral blood volume  by ~10%.  The reference for this statement was work done in 1991.

I thought this was a bit outre, so I wrote one of the authors.

Dr. X “Isn’t there something more current (and presumably more accurate) than reference #3 on cerebral blood flow in sleep?  If there isn’t, the work should be repeated”

I got the following back “The old studies are very precise, more precise than current studies.”

Go figure.

We don’t understand amyloid very well

I must admit I was feeling pretty snarky about our understanding of amyloid and Alzheimer’s after the structure of Abeta42 was published.  In particular the structure explained why the alanine 42–> threonine 42 mutation was protective against Alzheimer’s disease while the alanine 42 –> valine 42 mutation increases the risk.  That’s all explained in the last post — https://luysii.wordpress.com/2017/10/12/abeta42-at-last/ — but a copy will appear at the end.

In that post I breathlessly hoped for the structure of aBeta40 which is known to be less toxic to neurons.  Well it’s here and it shows how little we understand about what does and what doesn’t form amyloid.  The structure appears in a paper about the amyloid formed by another protein (FUS) to be described later — Cell 171, 615–627, October 19, 2017 — figure 7 p. 624.

Now all Abeta40 lacks are the last 2 amino acids of Abeta42 — isoleucine at 41 and alanine at 42.  So solve the Schrodinger equation for it, and stack it up so it forms amyloid, or use your favorite molecular dynamics or other modeling tool.  Take a guess what it looks like.

Abeta42 is a dimer, a beta40 is a trimer, even though the first 40 amino acids of both are identical.

It gets worse. FUS (FUsed in Sarcoma) is a 526 amino acid protein which binds to RNA and is mostly found in the nucleus.  Neurologists are interested in it because over 50 mutations in have been found in amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD).   FUS contains a low complexity domain (LCD) of 214 amino acids, 80% of which are one of 4 amino acids (glycine, serine, glutamine and tyrosine).  At high protein concentrations this domain of FUS forms long unbundled fibrils with the characteristic crossBeta structure of amyloid.  Only 57/214 of the LCD amino acids are part of the structured core of the amyloid — the rest are disordered.

Even worse the amino acids forming the amyloid core (#39 -#95) are NOT predicted by a variety of computational methods predicting amyloid formation (Agrescan, FISH, FOLDamyloid, Metamyl, PASTA 2.0).  The percentages of gly, ser, gln and tyr in the core forming region are pretty much the same as in the whole protein.  The core forming region has no repeats longer than 4 amino acids.

The same figure 7 has the structure of the amyloid formed by alpha-synuclein, which accumulates in the Lewy bodies of Parkinson’s disease.  It just has one peptide per layer of amyloid.

When you really understand something you can predict things, not just describe them as they are revealed.

 

Abeta42 at last

It’s easy to see why cryoEM got the latest chemistry Nobel.  It is telling us so much.  Particularly fascinating to me as a retired neurologist is the structure of the Abeta42 fibril reported in last Friday’s Science (vol. 358 pp. 116 – 119 ’17).

Caveats first.  The materials were prepared using an aqueous solution at low pH containing an organic cosolvent — so how physiologic could the structure actually be?  It probably is physiologic as the neurotoxicity of the fibrils to neurons in culture was the same as fibrils grown at neutral pH.  This still isn’t the same as fibrils grown in the messy concentrated chemical soup known as the cytoplasm.  Tending to confirm their findings is the fact that NMR and Xray diffraction on the crystals produced the same result.

The fibrils were unbranched and microns long (implying at least 2,000 layers of the beta sheets to be described).  The beta sheets stack in parallel and in register giving the classic crossBeta sheet structure.  They were made of two protofilaments winding around each other.  Each protofilament contains all 42 amino acids of Abeta42 and all of them form a completely flat beta sheet structure.

Feast your eyes on figure 2 p. 117.  In addition to showing the two beta sheets of the two protofilaments, it shows how they bind to each other.  Aspartic acid #1 of one sheet binds to lysine #28 of the other.  Otherwise the interface is quite hydrophobic.  Alanine2 of one sheet binds to alanine42 of the other, valine39 of one sheet binds to valine 39 of the other.  Most importantly isoLeucine 41 of one sheet binds to glycine38 of the other.

This is important since the difference between the less toxic Abeta40 and the toxic Abeta 42 are two hydrophobic amino acids Isoleucine 41 and Alanine 42.  This makes for a tighter, longer, more hydrophobic interface between the protofilaments stabilizing them.

That’s just a guess.  I can’t wait for work on Abeta40 to be reported at this resolution.

A few other points.  The beta sheet of each protomer is quite planar, but the planes of the two protomers are tilted by 10 degrees accounting for the helicity of the fibril. The fibril is a rhombus whose longest edge is about 70 Angstroms.

Even better the structure explains a mutation which is protective against Alzheimer’s.  This remains the strongest evidence (to me at least) that Abeta peptides are significantly involved in Alzheimer’s disease, therapeutic failures based on this idea notwithstanding.  The mutation is a change of alanine2 to threonine which can’t possibly snuggle up hydrophobically to isoleucine nearly as well as alanine did. This should significantly weaken the link between the two protofilaments and make fibril formation more difficult.

The Abeta structure of the paper also explains another mutation. This one increases the risk of Alzheimer’s disease (like many others which have been discovered).  It involves the same amino acid (alanine2) but this time it is changed to the morehydrophobic valine, probably resulting in a stronger hydrophobic interaction with isoLeucine41 (assuming that valine’s greater bulk doesn’t get in the way sterically).

Wonderful stuff to think and speculate about, now that we actually have some solid data to chew on.

Abeta42 at last

It’s easy to see why cryoEM got the latest chemistry Nobel.  It is telling us so much.  Particularly fascinating to me as a retired neurologist is the structure of the Abeta42 fibril reported in last Friday’s Science (vol. 358 pp. 116 – 119 ’17).  

Caveats first.  The materials were prepared using an aqueous solution at low pH containing an organic cosolvent — so how physiologic could the structure actually be?  It probably is physiologic as the neurotoxicity of the fibrils to neurons in culture was the same as fibrils grown at neutral pH.  This still isn’t the same as fibrils grown in the messy concentrated chemical soup known as the cytoplasm.  Tending to confirm their findings is the fact that NMR and Xray diffraction on the crystals produced the same result.

The fibrils were unbranched and microns long (implying at least 2,000 layers of the beta sheets to be described).  The beta sheets stack in parallel and in register giving the classic crossBeta sheet structure.  They were made of two protofilaments winding around each other.  Each protofilament contains all 42 amino acids of Abeta42 and all of them form a completely flat beta sheet structure.

Feast your eyes on figure 2 p. 117.  In addition to showing the two beta sheets of the two protofilaments, it shows how they bind to each other.  Aspartic acid #1 of one sheet binds to lysine #28 of the other.  Otherwise the interface is quite hydrophobic.  Alanine2 of one sheet binds to alanine42 of the other, valine39 of one sheet binds to valine 39 of the other.  Most importantly isoLeucine 41 of one sheet binds to glycine38 of the other.

This is important since the difference between the less toxic Abeta40 and the toxic Abeta 42 are two hydrophobic amino acids Isoleucine 41 and Alanine 42.  This makes for a tighter, longer, more hydrophobic interface between the protofilaments stabilizing them.

That’s just a guess.  I can’t wait for work on Abeta40 to be reported at this resolution.

A few other points.  The beta sheet of each protomer is quite planar, but the planes of the two protomers are tilted by 10 degrees accounting for the helicity of the fibril. The fibril is a rhombus whose longest edge is about 70 Angstroms.

Even better the structure explains a mutation which is protective against Alzheimer’s.  This remains the strongest evidence (to me at least) that Abeta peptides are significantly involved in Alzheimer’s disease, therapeutic failures based on this idea notwithstanding.  The mutation is a change of alanine2 to threonine which can’t possibly snuggle up hydrophobically to isoleucine nearly as well as alanine did. This should significantly weaken the link between the two protofilaments and make fibril formation more difficult.

The Abeta structure of the paper also explains another mutation. This one increases the risk of Alzheimer’s disease (like many others which have been discovered).  It involves the same amino acid (alanine2) but this time it is changed to the more hydrophobic valine, probably resulting in a stronger hydrophobic interaction with isoLeucine41 (assuming that valine’s greater bulk doesn’t get in the way sterically).

Wonderful stuff to think and speculate about, now that we actually have some solid data to chew on.