Do Synapses Really Store Memories?

By Neuroskeptic | December 4, 2016 2:34 pm

Most neuroscientists will tell you that long-term memories are stored in the brain in the form of synapses, the connections between neurons. On this view, memory formation occurs when synaptic connections are strengthened, or entirely new synapses are formed.


However, in a new piece in Frontiers in Systems Neuroscience, Austrian researcher Patrick C. Trettenbrein critiques the synapse-memory theory: The Demise of the Synapse As the Locus of Memory.

Trettenbrein acknowledges that “the idea that learning is essentially the modification of synapses in an ever-changing plastic brain has become one of the dogmas of modern neuroscience”. However, he notes that some within the field of cognitive science have long been skeptical of this idea, or at least of its cousin, associationism.

He goes on to discuss theoretical problems with the synapse-memory theory. For instance, Trettenbrein says that “the brain must adhere to the abstract architectural properties of a universal Turing machine. A defining feature of a Turing machine is its ability to read and write a long-term memory bank. Trettenbrein argues, following a book by Gallistel & King that given that memory is so fundamental to a Turing machine, memory ought to be a function of individual neurons, not of the connections between them.

I haven’t read the Gallistel and King book, but to me this argument seems overly abstract and not very convincing. I’m not sure what it would mean for a single cell to store information.

Trettenbrein next discusses a number of pieces of experimental evidence against the synapse-memory theory. For example, he cites a 2014 paper that reported non-synaptic long-term storage of information in neurons from the sea-slug Aplysia. However, as I said at the time in my blog post about this paper, it’s not clear how this mechanism would work in the case of the much more complex human brain.

Overall, this is an interesting and thought-provoking paper. However, as Trettenbrein acknowledges, it’s hard to see how we could even begin to explain memory if we reject the role of synapses:

To sum up, it can be said that when it comes to answering the question of how information is carried forward in time in the brain we remain largely clueless… the case against synaptic plasticity is convincing, but it should be emphasized that we are currently also still lacking a coherent alternative.

ResearchBlogging.orgTrettenbrein, P. (2016). The Demise of the Synapse As the Locus of Memory: A Looming Paradigm Shift? Frontiers in Systems Neuroscience, 10 DOI: 10.3389/fnsys.2016.00088

CATEGORIZED UNDER: animals, papers, select, Top Posts
  • Gabriel Finkelstein

    According to Ehregott Andreas Wasianski, Immanuel Kant left himself reminders to forget (in this case, his servant’s name):

    “Other people record what they wish to remember; but Kant had here recorded what he was to forget. ‘Mem.: February, 1802, the name of Lampe must now be remembered no more.'”

  • David Littleboy

    One thing that’s interesting here is that if he’s right, then the whole “neural network” hype fest that infects AI is, at best, completely irrelevant. Oops.
    (Personally, I’m more than happy with that. YMMV, of course.)

    • Neuroskeptic

      Well, neural networks might be a useful approach for AI even if they don’t actually model the human brain. But yes, if the brain operates on non-connectionist principles it would certainly take away a lot of the allure of neural networks.

  • derrick

    What a charming anecdote.
    Kant … Quirky
    It would never have occurred to me.
    Thanks for the smile.

  • M Mahin

    Thanks very much, Neuroskeptic, for drawing attention to Trettenbrein’s paper, which is very worthy of a read. But you didn’t mention the heart of his case, which is in section 3.3. of his paper, “Rate of Synaptic Turnover.” The molecules that make up synaptic weights have very short lifetimes of only a few weeks or less. Stettler’s research indicates a lifetime of synapses of only about 3 months. Dendritic spines last only about a month in the hippocampus, and less than two years in the cortex. So it seems that synapses are an unstable “shifting sands” kind of platform, rather than the stable platform one needs for memories that can reliably last for 50 years. See below for links to lots of relevant research:

    • Neuroskeptic

      Thanks. I agree that section is very interesting!

      • David Littleboy

        Hmm. The synapse could still be the locus of memory. All you need is a mechanism that checks and replicates the data synapses hold as necessary. Some of the recent work on recollection of memories being necessary (I.e. memory is way more complicated than just data readout) indicates some sort of mechanism like this. Whatever, speculation here is hard, especially since we don’t have a clue as to what the data being stored looks like…

        • Neuroskeptic

          Indeed. In fact, each synapse could just promote the formation of new synapses nearby, at a rate sufficient to counterbalance the ‘death rate’ of synapses.

          This would ensure that despite each synapse having a finite lifespan, the total number and distribution of synapses would remain stable over time. In much the same way that the populations of cities and towns are fairly stable even though individual humans are constantly being born and dying.

          • Levan Bokeria

            In fact, within the synapse itself AMPA receptors get recycled about every 15 minutes! So “the synapse” is not a stable entity during its lifetime either.

        • Maia

          Was just going to post this point which is very similar to yours: like the replicaton of DNA, you don’t have to hold onto the same molecules over a lifetime, you retrain the pattern by passing it on. All of reality seems to be in flux, not stable over a lifetime in material form, but can be stable over a lifetime and longer, if copies and distributed. Hericlitus’ “everything changes” is not a problem, but a clue.

          • Albert Slump

            And, in that way, we have the answer to why the memories “change” or are distorted over the time… like mutations in the dna…

          • pabajo

            Great discussion. I imagine that for fluent memories to work a kind of code should be necessary. Hard to imagine a nerve tissue code with all that mess…

  • William Holz

    Out of curiosity, what’s unusually weird about the cell itself containing ‘data’ vs. everything happening at the synaptic level?

    As somebody who’s more an educated amateur, both of them seem strange, but the idea of the synapse being the primary mover actually seems less intuitive, especially given how sophisticated cells can be.

    I also got the impression from Michael Dickinson during his ‘How the Fly Flies’ talk that arthropod neurons were already an example of the synapse being more than an on/off switch (or perhaps there was some sort of hybrid chemical modulation/design variance thing going on, I never really delved into it).

    Just curious, it’s a fascinating topic and I’ve only recently started re-delving into the nuts and bolts of brains. :)

    • C. Ed Felt

      “On/Off” switches are how we have implemented the digital age. This is probably why we mapped our understanding of the brain to this idea. It’s great we are figuring out we are wrong.

      • William Holz

        Cool timing! This has been a fun convo in a couple of circles.

        We already know that a number of single celled organisms have very rich behaviors without nerves and the information is stored inside a single cell.

        It also kind of makes sense that nerves would be a HUGE value-add for communication (hey feet parts! Go where the mouth parts want to go!)

        So it’s not even radical to think that the bulk of computation/calculation/thinking would be inside individual cells, and the neural network sitting on top of synapses would be way less granular and sophisticated. We don’t really know the give and take IMHO, but we know it’s not ‘all synapse’ or ‘all cell’, and nature digs analog.

        This one is starting to feel ‘obvious in retrospect, just struggling with a little inertia’, isn’t it? :)

  • Roy_Lofquist

    The number of neurons in the human nervous system is estimated to be about 86 billion. This is equivalent to about 10 gigabytes of random access memory in a computer. But the computer circuitry is more than a million times faster than the brain.

    Modern computer role playing games, such a Fallout or Skyrim require 15-30 gigabytes of memory to represent worlds that are but a small fraction of the size of human memory. The average human can drive a car (real time 4 dimensional calculations) while seemingly instantly recognizing faces, smells, voices music etc. and carry on a conversation in real time.

    The brain as computer model is quite obviously inadequate.

    • Stephan Zielinski

      You assume one bit per neuron. There is no particular reason to believe this is the case.

      • Roy_Lofquist

        Assume as many as you want. The argument still holds. The apparent capabilities of human consciousness are orders of magnitude greater than computer equivalents.

        • Stephan Zielinski

          We don’t know that one way or the other, as we don’t have a computing system that is capable of human-equivalent performance.

          There’s an guesstimate from 2009 of the raw computing power involved in a brain in Computers have a lot to learn from the human brain, engineers say: “A computer comparable to the human brain, he added, would need to be able to perform more than 38 thousand trillion operations per second and hold about 3,584 terabytes of memory.”

          • Roy_Lofquist

            Yup, many orders of magnitude. The logical inference? Supernatural, in the sense of impossible within our current understanding of the laws of nature. The same analysis applies to the notion that the origin of life was spontaneous (abiogenesis).

          • Stephan Zielinski

            Well, no. You still don’t have anything to compare it to. You’re starting from beliefs in dualism and creationism and arguing in a circle.

          • Roy_Lofquist

            Nope, the other way around. If you start with a rejection of dualism and creationism there is, to date, no way to explain the world in which we live. It is the null hypothesis. Nothing to compare it to? Life and its capabilities are sui generis, unique, we will never have a comparable human constructed equivalent.

        • Дмитро postman31 Булах

          So why so capable humans’ toddlers can’t manage a spoon, aside of a car?

          • Roy_Lofquist

            I have no idea. Life and consciousness have no analogs in nature. If life is the result of design then viable offspring are tailored to optimize survivability. Humans are the only animal capable of extended education of their young so it is not necessary for them to be born with complete physical survival traits.

        • C. Ed Felt

          I love this conversation. The possibilities are awesome.

    • C. Ed Felt

      I have a degree in and have been working in the Computer Science (including data science) field for almost 20 years. I’m no neuroscience expert, but from many comments here, and based on this and other articles, in my opinion, the brain does not store memories. The brain is an interface to something else that is not in this plane of existence. Most new theories (and I admit, I’m a novice at understanding it all) seem to indicate the brain is more like a CPU than memory, and the nervous system like buses to interfaces (in Computer Science terms).

      • Roy_Lofquist

        Well, Ed, I don’t have a degree in Computer Science. Mainly because they didn’t exist when I started – early 1960s. Back in the days of vacuum tube and discrete component machines the architecture was much more transparent – the CPU was in one of those big cabinets, the memory in 1 to 8 of them, the IO processors (IOPs) had another and each device (disk, tape, card reader/punch, analog to digital (ADC) and digital to analog (DAC) converters) had their own cabinets with their own specific device controllers. Unsurprisingly 90%+ of all failures were due to faulty cable connections.

        I view the brain as a big I/O system/device controller with both the CPU and memory in another plane.

  • Stephan Zielinski

    Assuming monism, a Turing machine that simulated the interactions of all the atoms and subatomic particles in a brain would serve as an emulation of all the computation taking place in the brain. This would work whether the brain stores data in the cells, the synapses, the pattern of firing neurons, or in any combination thereof. That the Turing machine emulation requires a reliable form of memory does not mean that everything it emulates must also have a reliable memory.

    • Roy_Lofquist

      Stephan, I refer you to the works of Roger Penrose:

      “Penrose believes that such deterministic yet non-algorithmic processes may come into play in the quantum mechanical wave function reduction, and may be harnessed by the brain. He argues that the present computer is unable to have intelligence because it is an algorithmically deterministic system. He argues against the viewpoint that the rational processes of the mind are completely algorithmic and can thus be duplicated by a sufficiently complex computer.”

  • Pingback: Indistinguishable from Magic 12/5 – Disruptive Paradigm()

  • stargene

    Before we lose focus here, there is ample reason to suspect that sub-cellular processes, aside from, or in addition to, synaptic activity,
    may underlie storage and processing of memory. See:
    … which suggests the possibility of ‘good’ prion-like structures in neurons acting as the ground of memory…

    Additionally, as an example, see “Cytoskeletal Signaling: Is Memory Encoded in Microtubule Lattices by CaMKII Phosphorylation?” at
    I don’t post these as future ’truth’ but as wider, richer possibilities
    which ought to be testable in the near future.

  • Kamran Rowshandel

    This doesn’t seem very looming at all. There is so much basis to synaptic memory that it’s difficult to even begin to describe the evidence against that crappy article. I’m going to take the position that entirely epigenetic memory is impossible mathematically.

  • Franck Ramus

    A cell can certainly store a lot of information in the form of epigenetic marks. Now, how exactly this would support storage of our memories is not so clear.

  • Jonathan Edwards

    I was always dubious about Randy Gallistel’s suggestion of memory in nucleic acid. I also doubt that one should start from the idea that we are looking at a Turing machine using symbols. My guess is that the semantic atoms in brain are whole propositions, not symbols. Nevertheless I am interested in the idea that cell structure might store memory not so much through synapse ‘strength’ as through some more complex pattern-based system. That might not so much involve memory at a molecular level as at polymer level – in the lengths of chains of monomers. Prion type proteins and microtubules are obvious possibilities. The result might be more like tuning the sympathetic strings of a sitar than sliding contacts along rheostats in an amplifier mixer system.

  • Pingback: A lire et classer | Pearltrees()

  • Pingback: A lire et classer | Pearltrees()

  • Pingback: Doing science backwards | Neuro-patch()

  • PingThing

    What if physical memory doesn’t exist?, What if what is actually happening is that the brain is simply remote viewing actual events as they happened/ are happening on a quantum level. Who says that the brain is not some sort of quantum biological antenna. And all events in the brains timeline are actually happening in the present. In fact where is the proof that the brain has any processing power at all. I can’t help thinking that if there was a physical mechanism for storing memory that some clever scientist would have worked it out by now. Time to think left field.

  • Trip

    “the brain must adhere to the abstract architectural properties of a universal Turing machine” — What an utterly absurd, unsupportable statement. There are numerous ways that the brain clearly does NOT act anything like a Turing machine. A classical Turing machine is not even capable of interaction – all information must be provided at the start! Our brains are fault-tolerant, in a way that the linear layout of a Turing machine’s memory couldn’t remotely account for (clip out a cancerous area of someone’s brain, and they may work around it; try clipping out a segment of a Turing machine’s memory and watch the entire program fail), and our brains certainly don’t work by moving a pointer that targets one cell in a row at a time. (An infinite row, in a true Turing machine!) Even if you accepted the absurd proposition that the brain somehow works like a Turing machine, the claim that “that given that memory is so fundamental to a Turing machine, memory ought to be a function of individual neurons, not of the connections between them” is a further nonsensical pronouncement. There’s nothing obviously magical about a neuron’s interior that makes it the only possible residence of some sort of symbol.

  • Pingback: Scientists Transferred Memories From One Snail to Another. Someday, Humans Could, Too.()

  • Pingback: | Scientists Transferred Memories From One Snail to Another. Someday, They Could Do The Same in Humans. –

  • Pingback: Scientists Transferred Memories From One Snail to Another. Someday, Humans Could, Too. – Technology and Longevity Feed()

  • Pingback: Scientists Transferred Memories From One Snail to Another. Someday, Humans Could, Too. | Elexonic()

  • Pingback: Scientists Transferred Memories From One Snail to Another. Someday, They Could Do The Same in Humans.()

  • Pingback: Scientists Transferred Memories From One Snail to Another. Someday, They Could Do The Same In Humans. | No Nonsense Only News | Paperdabba()

  • Pingback: Scientists Transferred Memories from One Snail to Another | Splice()

  • Pingback: Optogenetics, Neurohacking, Brain Reprogramming and Mind Control – HUMANITY+()

  • Pingback: Researchers Were Able To Transfer Memories From One Snail to Another. What Does It Mean For Humans. » LiveScience()



No brain. No gain.

About Neuroskeptic

Neuroskeptic is a British neuroscientist who takes a skeptical look at his own field, and beyond. His blog offers a look at the latest developments in neuroscience, psychiatry and psychology through a critical lens.


See More

@Neuro_Skeptic on Twitter


Discover's Newsletter

Sign up to get the latest science news delivered weekly right to your inbox!

Collapse bottom bar