Cassava Sciences has been under attack since 8/21 reaching a crescendo recently — https://www.science.org/content/article/co-developer-cassava-s-potential-alzheimer-s-drug-cited-egregious-misconduct and https://www.science.org/content/blog-post/saga-cassava. Both contain demands that their current double blind placebo controlled studies on Simufilam, their Alzheimer drug be stopped.
Full disclosure. I do not own any Cassava stock and have not received anything from them for writing about them (aside from a free meal from Lindsay Burns 11/21 at the CTAD (Clinical Trials in Alzheimer’s Disease) in Boston. There was and is no quid pro quo about anything I’ve written.
I do know Derek Lowe and have spent several pleasant afternoons discussing organic chemistry and drug development with him at my cousin’s New Year’s Day bash, before COVID put a stop to the affair.
I’ve known Lindsay since she was a teenager, when I was practicing neurology in Montana. This was primarily through her parents who were sheep ranchers and good friends of my wife and me.
The basic position of the two articles above, is that some of the protein electrophoreses backing up Cassava’s model of Simufilam were either fudged or missing.
For details about Cassava’s model and the science behind it please see — https://luysii.wordpress.com/2023/04/23/the-science-behind-cassava-sciences-sava-the-latest-as-of-19-april-23/
But now it’s time to talk about models of disease causation and how important and reliable they are. I’ll start with an impeccable model of Alzheimer’s disease and the exquisite chemistry, biochemistry and genetics backing it up. It’s technical, but so are the protein electrophoreses which have been criticized.
Basically the model says that the accumulation of Abeta peptides, the major component of the senile plaque, causes Alzheimer’s disease, and removing them should help the disease.
The following is pretty long and if you want to skip the details for the rest of the argument — fast forward to *****
First (and probably the best evidence) is the mutation that protects against Alzheimer’s disease. As most of you know, the aBeta peptide (39 to 42 amino acids) is part of a much larger protein the Amyloid Precursor Protein (APP) which contains 639 to 770 amino acids. This means that enzymes must cut it out. Such enzymes (called proteases) are finicky, cutting only between certain amino acids. In what follows A673T stands for the 673rd position which normally has amino acid Alanine (A) there. Instead there is amino acid Threonine (T). The enzyme cleaving at 673 is Beta Secretase 1 (BACE1).
[ Nature vol. 487 pp. 153 ’12 ] A mutation in APP protects against Alzheimer’s disease. First the genome sequence APP of 1,795 Icelanders were studied to look for low frequency variants. They found a mutation A673T adjacent to the site that is cleaved by beta secretase 1 (BACE1) which doesn’t vary — it’s gamma secretase which cleaves at variable sites leading to Abeta40, Abeta42 formation. The mutation is at position 2 in Abeta. The mutation results in a 40% reduction in the formation of amyloidogenic peptides in vitro (293T cells transfected with variant and normal APP). Amazingly, a different variant at 673 (A673V — V stands for the amino acid Valine) — increases Abeta formation. Because BACE1 can’t cleave APP containing the A673T mutation, alternative processing of APP at another site the alpha site (which is within aBeta preventing formation of the full 39 – 43 amino acid peptide).
So if you can’t make the full aBeta peptide you don’t get Alzheimer’s (or have less chance of getting it).
Then there are the mutations in the part of APP which code for the aBeta peptide which increase the risk of Alzheimer’s. They cause the different forms of familial Alzheimer’s disease. Now that we know the actual structure of the aBeta amyloid fiber, we can understand how they do this. This is further evidence that the aBeta peptide is involved in the causation of Alzheimer’s disease.
You’ll need some protein chemistry chops to understand the following
Recall that in amyloid fibrils the peptide backbone is flat as a flounder (well in a box 4.8 Angstroms high) with the amino acid side chains confined to this plane. The backbone winds around in this plane like a snake. The area in the leftmost loop is particularly crowded with bulky side chains of glutamic acid (single letter E) at position 22 and aspartic acid (single letter D) at position 23 crowding each other. If that wasn’t enough, at the physiologic pH of 7 both acids are ionized, hence negatively charged. Putting two negative charges next to each other costs energy and makes the sheet making up the fibril less stable.
To make an amyloid fiber just stack 1000 or more of these flounder aBeta peptides on top of each other.
The marvelous paper (the source for much of this) Cell vol. 184 pp. 4857 – 4873 ’21 notes that there are 3 types of amyloid — pathological, artificial, and functional, and that the pathological amyloids are the most stable.
In 2007 there were 7 mutations associated with familial Alzheimer’s disease (10 years later there were 11). Here are 5 of them.
Glutamic Acid at 22 to Glycine (Arctic)
Glutamic Acid at 22 to Glutamine (Dutch)
Glutamic Acid at 22 to Lysine (Italian)
Aspartic Acid at 23 to Asparagine (Iowa)
Alanine at 21 to Glycine (Flemish)
All of them lower the energy of the amyloid fiber making them more stable
Here’s why
Glutamic Acid at 22 to Glycine (Arctic) — glycine is the smallest amino acid (side chain hydrogen) so this relieves crowding. It also removes a negatively charged amino acid next to the aspartic acid (which is also negatively charged). Putting two negative charges next to each other costs energy — because like charges repel. Both effects lower the energy of the amyloid fiber
Glutamic Acid at 22 to Glutamine (Dutch) — really no change in crowding, but it removes a negative charge next to the negatively charged Aspartic acid
Glutamic Acid at 22 to Lysine (Italian)– no change in crowding, but the lysine is positively charged at physiologic pH, so we have a positive charge next to the negatively charged Aspartic acid, lowering the energy because like charges attract.
Aspartic Acid at 23 to Asparagine (Iowa) –really no change in crowding, but it removes a negative charge next to the negatively charged Glutamic acid next door
Alanine at 21 to Glycine (Flemish) — no change in charge, but a reduction in crowding as alanine has a methyl group and glycine a hydrogen.
As a chemist, I find this immensely satisfying. The structure explains why the mutations in the 42 amino acid aBeta peptide are where they are, and the chemistry explains why the mutations are what they are.
****
It doesn’t get any better than this, yet therapy based on the model (monoclonal antibodies to remove the aBeta peptide) have produced minimal benefit (a less than 30% decrease in the rate of decline) and serious side effects namely brain hemorrhage, because the aBeta peptides don’t confine themselves to the senile plaque but can be found in the walls of blood vessels. This is called cerebral amyloid angiopathy (CAA).
What about wildly successful therapies which were found without ‘benefit’ of any theory, with theoreticians scrambling to explain them post hoc.
Here are 3:
l. Lithium for mania– the story is particularly fascinating — https://www.nature.com/articles/d41586-019-02480-0. We still don’t know how it works but theories abound, despite the original discovery reaching the ripe old age of 70.
2. Ketamine analogs for depression. For decades we were taught that antidepressants take a few weeks to work, and this was true of the drugs we had (tricyclics, selective serotonin reuptake inhibitors). Hell, I taught it myself back in the day at Montana State University. This was until ketamine analogs (esketamine) lifted depression within hours. We are still trying to figure out how this works, and what it tells us about depression (which is almost certainly a lot).
3. Wegovy, Ozempic for weight loss. I’ve been reading about glucagon like peptide (GLP) for years, because it is a peptide neurotransmitter. People have been studying it for years. Only when GLP agonists used to treat diabetes produced weight loss did theorists scramble to figure out why.
As an old med school friend, warped by his experiences at the University of Chicago used to say. “That’s how it works in practice, but how does it work in theory?”
So here we have a superb model for Alzheimer’s disease which hasn’t produced a terribly useful therapy, and three incredibly useful therapies for which we presently have no satisfying model.
So dumping on Cassava and wishing to stop an ongoing study because the model underneath it is based on flawed data just doesn’t make sense.
In particular, it doesn’t make sense given the data Cassava has already released, but that’s for part II.