Skip to Content

Alzheimer's Disease

The Genomics of Neurons. And Alzheimer’s. And Everything Else.

The brain is a complicated organ. Let’s start there. It’s complicated at every level that you care to examine, and if you get down to the genomic sequences of individual neurons, it’s worse than ever. The sheer variety of neurons and other cell types is quite extreme, and a lot of work over the years has gone into trying to figure out how this huge range of morphology and function is generated. As became clear not too many years ago, on a cellular level the brain is a mosaic tissue: the different cells in it can have somewhat different DNA sequences, distinct but (apparently) rather random. This arises (as far as we know) during development, and leads to single-nucleotide variations, copy number variations in individual genes, and other changes. And like so many odd things that you come across in cell and developmental biology, the odds are that after all these hundreds of millions of years it’s something with utility rather than just this stuff that happens out there in a void – the “Chesterton’s Gate” analogy is very strong in evolutionary biology.

One of the genes that shows a big range of copy number variation is APP, which codes for the (in)famous amyloid precursor protein. That’s a 751-amino acid beast, and the beta- and gamma-secretase enzymes carve out the far shorter beta-amyloid protein from the middle of it. Now, if you ask me why they do that, da bin Ich überfragt, as they say in German, “there I am overasked”, because that’s a major mystery. There’s a (controversial) theory that the beta-amyloid protein itself might be a defense mechanism against infectious disease in neuronal tissue, but it’s safe to say that a great deal of work on APP and amyloid have so far failed to resolve a lot of the “how come” questions.

Now it turns out that the APP gene shows the effects of genetic recombination as well. And how. This paper reports that there are thousands of cDNAs present in human neurons (“gencDNAs”) with variations of APP – all sorts of single-nucleotide variations and intra-exonic junctions, just a real mish-mosh. And these are actually inserted (via reverse transcription) into the genomes of post-mitotic neurons and can be transcribed and translated into new protein species. And they find what they describe as a “marked shift” in both the number and composition of these genomic cDNA species in the tissue of patients who were diagnosed with Alzheimer’s disease, including the sequences of several APP variants that had already been described as causal in the familial forms of the disease.

Those are mutations in APP, ones that make it easier to cleave the protein to give you beta-amyloid, that are associated with greater risk of early onset Alzheimer’s (or at least something that’s close enough to make little difference). There are various family lineages in Sweden, the Netherlands, Colombia, and other countries that have been discovered with such risk-factor mutations, a scary thought for several reasons. This, coupled with the obvious deposits of beta-amyloid itself in all  Alzheimer’s-affected brains (the very hallmark of the disease) are why so much time, effort, and money has been spent on the amyloid hypothesis for the disease. I’ve said a lot of disparaging things about it here over the years, but it’s a very compelling story.

And what this means is that any new theory of Alzheimer’s is going to have to encompass the amyloid one. It’s the same sort of thing as when Einstein’s relativity came along. He didn’t invalidate Newtonian physics – he subsumed it into a larger framework. Newtonian mechanics works perfectly well as long as you’re not (say) approaching the speed of light or under the effects of a strong gravitational field – under more “normal” conditions you have to take fanatically precise measurements to see that there’s anything off at all. And so with Alzheimer’s: any true theory of what the disease is and how it develops is going to have to explain all the amyloid connections, because they’re real – but it will also explain why all the efforts to target amyloid (through secretase inhibitors, antibodies, etc.) have not worked. Speed the day of such a theory’s arrival.

This new work might well be part of such an explanation. If the population of human neurons varies so much in its APP profile, developing a single drug to affect a single target might well be. . .well, inappropriate at best, futile at worst. These results could complicate human neuroscience immensely, which if I’m going to be honest is not what I would have asked for. But like any scientist, what I really ask for is truth, for the real details of what’s really happening, with the full knowledge that I might not like it when I get it. The implications, though, go far beyond Alzheimer’s disease:

Additional genes may be transcriptionally modified and genomically retro-inserted in response to selective activities in neuronal populations. Such a mechanism might enable preferential gene re-expression that bypasses splicing or further RNA modification. More broadly, gencDNAs could provide neurons with an activity-dependent mechanism for recording and retaining information over long periods of time, perhaps placing multiple forms of a gene under transcriptional control distinct from a wild-type locus, which could be produced through diverse genomic integration sites that remain to be determined. Such a process could have relevance to known neuronal functions that depend on transcriptional activity including Hebbian plasticity, synaptic wiring, learning and memory, and cognition. Thus, gencDNA production may represent both a ‘recording’ and a ‘playback’ mechanism for expressing a symphony of variants beyond wild-type gene forms. It would be surprising if APP were the only gene to undergo this form of recombination. . .

Oh yes, it would be. This is both exhilarating and terrifying. An entirely new feedback and regulatory universe could be opening up on us here if this pans out, with neurons modifying their own genomic sequences (and the rate and degree of that modification) in response to environmental stimuli. I would add that there may be no reason to suppose that such processes are confined to neurons (indeed, it’s already known that “standard” genetic recombination is how you get the huge variety of antibodies). The advent of numerous modern sequencing technologies on both DNA and RNA species, along with single-cell techniques for their use, has already been telling us (for example) that tumors are far more difficult and complex environments than we’d imagined. But what if our normal tissues are just as variable – indeed, with variability in the variations themselves? Hold on, hold on tight.

35 comments on “The Genomics of Neurons. And Alzheimer’s. And Everything Else.”

  1. Orvan Taurus says:

    Yow. The idea of ‘programming’ DNA/genes and then getting just exactly what is desired seemed difficult (at best!) before. And now it looks like it involve the thing programmers are warned against and is considered Insanely Dangerous: self-modifying code. This might be the best “proof of evolution” – life is a kludge.

  2. Derek Jones says:

    Great article.

    “mish-mosh” jumped out at me. Is this a technical term in brain genetics?
    Or did you mean “mish-mash”?

    My claimed level of expertise is once sharing digs with a first year medical student.

    1. Derek Lowe says:

      Hah! No, that’s just a phonetic rendering of how I say it. . .

      1. Margot Otway says:

        I like “mish-mosh.” Combination of “mish-mash” and “mosh pit”; very appropriate.

  3. Not a Neuroscientist says:

    Something that jumped out at me here: Is it too much of a leap to wonder if this variation in neuronal DNA is how the brain stores information?

    1. Derek Lowe says:

      No, apparently not. See that last quoted paragraph – the authors of the paper are going there, too.

      1. zero says:

        Superficially, that would be easier to conceptualize than some kind of weird holographic signal network, like a Bohr model of the atom is easier to understand than the terrifying reality of quantum mechanics.

        Practically, I suspect that makes ‘brain data storage’ far more difficult to predict, model, read out and modify as one would need to sequence (or measure the expression of) colossal numbers of individual neurons.

        If this checks out then the sci-fi vision of human memory hacking / mind transfer becomes virtually impossible.

        1. aairfccha says:

          It would also explain how something you know you knew but forgot can pop back with or without apparent reason.

        2. eub says:

          Why’s that? It would seem to me that the more the memory and the mind are encoded in DNA, the more the computer scientists can get their grubby little fingers onto them.

          1. Ed says:

            Genome sequencing is a completely destructive process. And if every neuron has to be individually sequenced, then every neuron has to be destroyed. In the process, the rest of the information that underlies brain function and memory (structure, connections between neurons and other cells, protein expression and localization, things we don’t even know about, etc.) will also be completely destroyed.

            In other words, if you can magically make a few thousand copies of yourself, it might be conceivably possible to destructively analyze all of them and make an approximate model of your brain.

  4. anon says:

    This is really fantastic work! Also, in reference to the use of cART therapy, it will be interesting to see how healthy people who are now on PrEP treatment for HIV prevention from a young age for presumably the rest of their lives fare to the general population for Alzheimer’s.

    1. Andrew says:

      They mention in the discussion that reverse transcriptase inhibitors have been shown to be protective in patients with HIV and they call for further investigation in light of this results.

  5. Project Osprey says:

    I remember reading Blood Music by Greg Bear at a time when DNA was still described as being a fixed set of instructions, like words in a book. Like a lot of sci-fi, it’s ideas seemed too far fetched – cells rewriting themselves in response to stimuli. Yet somehow, here we are.

    1. Diver Dude says:

      I just re-read it for the first time in 20 years. The parallels are thought-provoking. To put it mildly.

  6. truthortruth says:

    Beautiful work! This story reminded me of the amazing LAIR1 insertions observed in antibody genes of malaria patients (link in name).

  7. Peter Ellis says:

    While the evidence was clear that there are cDNA copies of the APP gene in neurons, I wasn’t convinced by all their claims.

    1) I’m not sure what their grounds are for claiming the cDNA is genomically integrated. RT activity (from ERVs) is well documented, and so one might expect neurons to contain low levels of ‘free’ cDNA. Having no promoter or means of further transcription, these likely have little effect on the cell but might be a bio marker for increased ERV activity.

    2) Their evidence for novel isoforms was mostly unconvincing, since it almost all came from PCR-amplified cDNA. The novel variants look exactly like PCR artifacts arising from partial strand extension, internal mispriming, etc.

    3) The CHO cell data supposedly exploring the biogenesis of these makes little sense. They transfected CHO cells with a full length cDNA, treated them with peroxide, and found a bunch of new variants with small internal deletions. Congratulations, you just discovered NHEJ / MMEJ.

    4) Their discussion talks about translatability of the cDNA copies. OK, some of them translate if you put them into an appropriate vector in a cell culture system. So does random DNA as long as there’s a good start codon. They have zero evidence that the cDNA copies are either transcribed or translated in neurons.

    Overall, the parts of it I’d say are rock solid are the fact that you have endogenous cDNA in neurons, and this varies according to disease status – one would guess due to ERV activity. Everything else is a bit shaky.

    1. Derek Lowe says:

      Very interesting! What do you think the killer experiment (experiments) would be for the more far-reaching claims?

      1. There are a number of methods using sequence enrichment feeding into long read (reads thousands to tens of thousands of bases long) which could be used to map out the diversity of integration sites, if these copies really are integrated into the genome. That could be very convincing.

    2. Not a neuroscientist says:

      These may reflect my ignorance but…

      1) Wouldn’t ERV-derived cDNAs (or otherwise free-floating cDNAs that are not genomically integrated) be single stranded and therefore not cleaved by the restriction enzymes (e.g. Fig 2j/k)? And is ERV-derived RT nuclear or cytoplasmic, because a lot of the isolation was from sorted nuclei, so cytoplasmic ss cDNAs should be excluded?

      2) Is the in situ hybridization and direct sequencing of enriched DNA (no amplification, therefore no PCR-derived errors) not sufficient for this?

      1. Peter Ellis says:

        For question (1): anyone’s guess! Reverse transcription and packaging of retroviruses is very complex, with lots of strand swapping, circularisation, etc driven by endogenous signals within the virus mRNA. Depending on the secondary structure of the APP mRNA, many things may be possible including formation of dsDNA, which could be linear or circular.

        Heck, given how evolution reuses these things, it’s not outside the realms of possiblities that certain genic mRNAs actually derive partially from the corpses of old HERVs and still retain some of their funky secondary structures. It might be interesting to scan these gcDNA copies for polypurine tracts (PPT), as this is one of the key aspects that prevents full degradation of the RNA part of the RNA/cDNA heteroduplex and allows second-strand priming to occur. Potentially one could predict which genes accumulate cDNA copies over time.

        I don’t know about localisation of ERV reverse transcriptase activity – but then again, the paper doesn’t say whether they looked for cDNA copies in the cytoplasm. Might be there’s quite a bit of it about, and some copies end up in the nucleus.

        Re (2). Yes, but the in situ hybridisation only looked for one canonical splice boundary and one non-canonical boundary, so it doesn’t necessarily prove the existence of the full zoo of novel isoforms claimed. I don’t think they have any sequencing data that doesn’t involve at least some PCR amplification. The do say they have sequence “without primary PCR amplification”, which was an Agilent capture process followed by next-gen sequencing. I’m not sure there isn’t a risk of creating occasional chimeric products during the next-gen sequencing step, but I’m very happy to be corrected.

        As to proving functionality, rather than these being dead-end side products of endogenous ERV activity… that’s a very tall order. Some thoughts:
        * Cloning an insertion site by inverse PCR (or some such) would help demonstrate that these can actually be integrated rather than being free-floating.
        * Looking for the mRNA corresponding to the abnormal splice forms would give some indication that the gcDNA copies are transcriptionally active, especially if you show you _don’t_ get the abnormal mRNAs in cells that don’t have gcDNA copies.
        * As for claiming translation…. you need to show the presence of the protein product (or at least a partial peptide) that comes from the novel variant. No way round it.

        1. Peter Ellis says:

          Further thoughts – one general principle is that virologists studying retroviruses must have methods that would be applicable here.

          For looking at integration:
          * PCR using divergent primers should show if you can amplify from terminal exons back round to the start. If so, then either they’re integrated as tandem repeat tracts, or they’re present as free, nonintegrated circles.

          * Careful size fractionation of a high molecular weight DNA prep, plus qPCR, might show whether the gcDNA copies are enriched in a HMW fraction (say >50kb) or a low weight fraction (say <20kb). This again touches on the question of chromosomal integration vs free linear or circular molecules.

          And for shits and giggles, since Derek asked for the "killer experiment": de-differentiate the affected neurons to get an IPS line that you can grow indefinitely, sequence to whatever depth you like, and look for the genome rearrangements…

  8. Luigi Facotti says:

    “Alzheimer’s-affected brains (the very hallmark of the disease)” – depends what you mean by :AD-affected brains: – the presence of amyloid, the presence of cognitive dysfunction/dementia or both.
    There is a lot of evidence – some dating back to 1912 that amyloid deposits can occur without dementia (30-40% of patients with amyloid do not have dementia) and other data that dementia can be present without amyloid. And the 2018 NIA/AA Framework recommends using their ATN/ATXN biomarker designations rather than AD to describe AD.

    1. Derek Lowe says:

      Amyloid can indeed be present without dementia, and dementia without amyloid (for example, vascular dementia and other such injuries). But my understanding is that if you don’t have amyloid plaques, then you do not have Alzheimer’s.

    2. dearieme says:

      “30-40% of patients with amyloid do not have dementia”: I didn’t know – another bit of science I’ve never seen in the newspapers.

  9. anon says:

    Looking at amyloid in AD or any neurodegenerative disease is like looking at the aftermath of tsunami and saying the debris caused the carnage.

    This blog keeps rehashing the same tired points on AD, plaques, as does the research in general.

    Where’s Lane Simonian? He made the most sense on this topic.

    1. Anonymous-lab-mouse says:

      If amyloid is an effect and not the cause, then a competing theory needs to explain (as Derek points out) why mutations that make it easier for amyloid to get made, increase risk of AD (or cause full blown early-onset dead-by-fifty, as in the case of the unlucky Dutch family). Just saying it’s an effect is a clear oversimplification and won’t cut it.

      1. loupgarous says:

        The husband of a good friend in South Louisiana died around that age from AD with very rapid onset and ruthless progression.

        That said, I can imagine one potential explanation for mutations that allow more rapid amyloid formation causing more rapid onset and progression of AD
        – if the mutations actually accelerate the underlying cause of AD (be it amyloid formation or not).

        What Big Pharma seems to have shown is that disrupting β-amyloid formation doesn’t by itself influence the clinical course of AD (anyone, please correct me if I’m wrong here). Not the same thing, is it?

  10. matt says:

    Any connection with previous papers pointing to viral activity in AD brains? Peter Ellis above points out ERVs, but what about non-endogenous RVs? Those previous papers suggested the up-regulation of APP in the presence of a certain family of virae, perhaps this paper is providing a mechanism by which that up-regulation might happen?

    1. not lane... says:

      that was my first thought, too…

  11. loupgarous says:

    Just a quibble, Derek. It’s “Chesterton’s fence”.

    The ‘paradox’ was actually grounded in a war of words between G.K. Chesterton and George Bernard Shaw and his pals which wouldn’t shed any more light if I were to explain it here.

  12. JimM says:

    “In the middle of the journey of our life I came to myself within a dark wood where the straight way was lost. Ah, how hard a thing it is to tell what a wild, and rough, and stubborn wood this was …”

    1. Derek Lowe says:

      Yeah, well, Dante eventually got out. . .

  13. The quoted text suggests that humans weren’t coded in binary & EMPs won’t mind-wipe us.

    That latter may be a “why” for encoding memory in genes. Geomagnetic storms^ from solar flares and CMEs (coronal mass ejections), and movement through Earth’s magnetic field, might otherwise negate memory storage.

    ^ – link to this perspective in my name. ( has more info.)

  14. MCS says:

    A number of thought occurred to me:

    Nowhere above does the word prion occur. My own knowledge doesn’t allow anything more then suspicion that there is some sort of connection.

    Is this phenomenon limited to humans, primates, mammals, organisms with brains?

    Does it extend to peripheral nervous tissue?

    Eating central nervous system tissue and especially cannibalism seems to be an increasingly bad idea.

  15. Johannes Høher-Larsen says:

    Derek, one also has to answer how amyloid relates to many unrelated risk factors; in that context it’s more likely that amyloid is just cellular debris. Is the dutch connection really the only thing that suggest a causal one?
    I don’t see anyone developing anti-drusen antibodies for AMD

Leave a Reply

Your email address will not be published. Required fields are marked *

Time limit is exhausted. Please reload CAPTCHA.