Neuroscientists discover molecular mechanism that allows memories to form (2020)
news.mit.edu
news.mit.edu
Less click baity.
But that is practically untrue (I'm not just saying there are corner cases, but that the main topic is still very poorly understood) and a cop-out.
And yet I can speak concretely about a hydrogen atom. Or even photosynthesis. Or DNA.
In DNA, there are 4 base pairs. We can see how that maps to RNA and how that in turn codes to amino acids. (and exceptions exist to these things, but we have a pretty strong grasp on it.) We can even try to fold the resulting Amino Acids into shapes, although it's computationally very difficult to do so. So data storage within a cell has some concrete understanding. Empirically validated and useful (this is how we were able to engineer the mRNA vaccines).
Our knowledge of memory, our knowledge of data storage in the brain is far less concrete. And not just because it's not thought to be precisely molecular like DNA or RNA (although perhaps it could be!) but we have a poor grasp of even where, precisely, information is stored in the connections. Like, sure, you can say it's not stored in a particular location in 3D space, but there must be some transformation, some (non-euclidian) representation of the connections where memories CAN be located. Because our brains are capable of recalling memories on demand. But we have only a vague understanding of how that all occurs. It's very unlike DNA or RNA or whathaveyou in that our understanding is still primitive and vague.
You can talk about some abstractions about how a hydrogen atom works, but we don't fully understand everything about atomic structure. We don't know if there is something that makes up the structures of subatomic particles becuase we don't have particle accelerators that have enough energy for us to peer deeper into what quarks are made up of. We don't even know if we've found all the types of particles.
Corner cases in several layers lower than chemistry vs not even really knowing where a memory is stored!
Not sure where that hubris comes from. There are physical phenomena we have not been able to test, so there will always be unknowns and assumptions based on (abstract) modeling. eg the metallic liquid hydrogen described here https://edu.rsc.org/soundbite/hydrogen-falls-apart-under-pre...
And I don’t mean to say that neuroscientists are doing shoddy work. Far from it! The brain is a far more complex entity than RNA or a hydrogen atom. The task is MUCH harder! But I am showing that high level of specific, concrete knowledge IS possible in the physical sciences. Memory in the Brain is a physical process as well, but we have only a relatively vague understanding of the specifics of it. We can sequence DNA or RNA accurately with relative ease. We cannot do the same with memories in the brain.
You're missing the point. The idea that the predictions are the same as knowledge illustrates the misunderstanding. We do not know because we cannot prove it. I shouldn't need to get into the reason we produce experimental evidence eg https://www.forbes.com/sites/startswithabang/2019/07/06/ask-... - via the LIGO and Virgo detectors
I'm not addressing the hand wavy nature of the article, which is self-evident. This is irrelevant to the point being discussed.
You might want to reread the thread for context. You don't seem to be discussing the same point as the people you're replying to.
Memory shares a property with holograms or with lenses/mirrors. If you draw a spot with a marker on one and looks at/through it, you don’t see the spot rather you see a slight loss of image quality across the image.
If you wipe out a bit of brain your memories degrade a bit. Somehow the encoding of memory is holographic.
We just need a different set of mathematical skills/tools to understand this property. From the point of view of traditional scientists it’s not something they know how to tackle.
Personally I think we should look at the brain (and DNA) with a signal processing lens. How data is encoded in the brain is something engineers could answer. Neural networks too are analog computers and spike signals are encoded with a variety of modulation schemes. If we understand those we don’t need to model neutrons the way we do (which is very inefficient, using floating point etc.) but by custom single-transistor analog circuits. That might help advance computation in general...once we understand how the brain is doing it.
Surely there must be a better way to conduct this experiment than to terrorize innocent mice.
Why can't they
Open loop systems are simple...the effects of one block can be understood in isolation. Closed loop control systems with feedback have nonlinear responses to stimuli that can cause ripple effects across the system.
By analogy, on a modern cpu with good power management, and shared memory caches, if a program heavily writes to computer memory, the effects will be on the whole computer. Not only will memory power increase, but cache use will too (causing threads on other cpus to slow down) and the chip will heat up, causing every part of the chip to throttle.
It is still basic research, but if I understand the significance of this correctly having a confirmation there are specific physical changes neurons undergo when forming some kinds of memories allows at least theoretically extraction of those memories from pieces of inactive tissue. As a layman in the subject I've been always very much interested how much of our personality and consciousness is "encoded" in the electro-chemical activity (software) vs how things are connected and formed together(hardware). The more of "us" is in the hardware the more likely someone in future will come up with a way to "scan" features of dead tissue to recover (in a simulation perhaps) the person lost to death.
We're currently probably as far from that as a medieval battlefield "surgeon" from modern heart transplant, but at least we're moving in the right direction.
To form a memory is something completely different than to store data in a db, as it involves lots and lots of encoding factors. If anything is actually stored, it's not the memory itself, but the perception at time T in relation to the expectations for T, i.e. a delta (think of the times when you listen to a song from the old days and it hurts, because the emotional connotation from back then doesn't match the current situation). Retrieving this delta would be meaningless unless you have the matching interpreter, the entire status machine.
You transition to a difference in encoding which is a significant but separate consideration. I think the thrust of your statement agrees with the science but quickly nice into speculation and undue specificity that to me reads as though you are declaring fact. Language and good communication are hard.
A very important distinction is that a computer has clean separation between physical data representation (magnetic bits) and physical device implementation (transistors & etc). As an analogy to a biological system I'd suggest an FPGA running a self modifying program that's also reconfiguring the FPGA on the fly. No sane engineer would design such a system (I hope).
For ion flow through dendrites and axons, the separation of information transmission from the protein assemblies of neurons is similarly separated.
For neurotransmitters the particles are literally packed into bundles and separated from the ionic transmission that causes their release and cell boundaries at the synapse until re-up take.
For electric transmission, the electricity is definitely separate from the axon and junction.
I think the distinction is a little more in how a single computer's hardware is constrained to a (relatively) small set of finite states outside of outside modification. The brain can expand and contract the set of states that it's physical parts can enter on top of being, for better and worse, less discrete. If our computers could grow their silicon wafers and lay new circuits then we would have something similar.
> If our computers could grow their silicon wafers and lay new circuits then we would have something similar.
Yes, that's why my analogy included reconfiguring the FPGA (ie hardware) that's running the program on the fly. You highlight a number of examples where a reasonably clear line between the biological "hardware" and "software" can be drawn at a given point in time. However, you fail to explicitly mention any of the biological mechanisms involved in reconfiguration, some of which I would consider to blur those lines!
The addition and removal of synapses over time is an obvious but fairly slow and boring example. The software is very slowly modifying the hardware by small amounts, but the two are still clearly separate.
Epigenetics is more interesting. Any number of convoluted signalling pathways feeding back on the machinery that controls gene expression, with (in some cases) heritable effects. The distinction here remains fairly clear to me at any given point in time. The proteins are the hardware, the DNA the storage, and the program is manipulating the storage so as to modify the synthesis of future hardware components. Easy enough.
... or is it? Often the quantity of some component that gets synthesized is itself used as a signal. Does synthesis being used in this manner mean that the hardware has become part of the software? How far do things have to go before the hardware can be considered to be part of the software? But things get even weirder!
> the separation of information transmission from the protein assemblies of neurons is similarly separated
Not always! Ever come across GPCR heteromerization? This happens at your synapses (among other places), more or less in real time, in response to various convoluted signalling pathways. It can change how the receptor responds to a given ligand, or even allow it to respond to entirely different ligands. So in some cases, an important part of the "logic" for the neurotransmitter response is being played out by the physical configuration of the receptors. From my perspective, it looks decidedly as though the receptor (ie hardware) has become an integral part of the software.
Other examples abound, but this comment has gotten quite lengthy so I digress.
Gene expression, signaling proteins, and protein synthesis or epigenetics in general are excellent examples both of the duality but some circumstances where the distinction gets fuzzier, particularly depending on the context and perspective you take.
I had not heard of GPCR heteromerization and will be reading more. I ended up going down the complexity and theory path more heavily and assigning my brain into business software so my knowledge of the concrete mechanisms is shallower than I'd like.
I definitely enjoyed thinking about about where to more specifically pin the difference in this context so thank you for being gracious with my making that attempt to offer a formulation. I think there's another interesting question to ask whether the duality divide matters but perhaps another day.
Thank you for the conversation.
The target audience is really science journalists on EurekaWire. They're hoping this press release will get picked up by a science writer for the New York Times or Scientific American, etc.
A parallel goal is to demonstrate that the university is doing cutting edge work, which helps with student enrollment, fundraising, and recruitment. This also applies to the funders: they want to demonstrate that donations (or tax dollars) are being put to good use.
If it directly catches the attention of science aficionados, that's mostly a bonus.
This would be comparable to winning some scientific prizes, since they are accomplishments that are hard and competitive to achieve. Of course the general public does not care about which scientists win which prizes.
This is all part of marketing to get broad exposure to the scientific community first, general public second.
http://phdcomics.com/comics/archive/phd051809s.gif
(sorry for plaintext URl, poating from phone and can't recall what the HN markup is for a link.)
Since there's no known way to prevent TGA, no treatment, and apparently no lasting damage, it has never (AFAIK) been researched, certainly not researched much, but it strikes me as the kind of thing that should be studied. If scientists could figure out how memory formation is completely blocked in episodes of TGA, it seems like that would provide useful data about how memory formation normally works.
I suspect the issue may be the lack of a good biological model. I'd guess it would be a very attractive and obvious research target for anyone aiming to investigate memory formation and recall. However, systematic studies generally require a model system that can be reliably reproduced.
Possibly, but I strongly suspect humans wouldn't be very good models in this case given how much we don't know about memory in general. Rather you'd want a robust non-human animal model of some sort, with a straightforward way of inducing TGA (or something that appeared to be substantially similar) on demand. Think rodent with optogenetic modifications or similar.
Unfortunately it only surfaces in one way. If I listen to a podcast while I’m doing something, then listen to that same podcast again within a week or so, I will have regular, intrusive visual recall of exactly what I was doing the last time I heard whatever passage the podcast I was listening to. The recall is synchronized down to the word with the podcast, and is vividly detailed and colorful and includes some sensation of feeling as well.
It’s weird and cool at the same time. One of these days I’d like to do an experiment where I videotape myself listening to the podcast, wait a week, then listen to it again and record myself describing what I see in my mind to test how accurate it is.
For example, I occasionally play Pubg Mobile and while I do it I ikte listening to some podcasts on a topic I do work. When it comes to apply something that I learned from the podcast I will can have a recall on where exactly I was in the game when this particular topic was discussed.
Something similar was as a teenager: I could remember when/where I learned or first heard a new word. Sadly that ability has been waning.
The power of the market compels your good idea to turn evil.
Instead, the material+process should be adapted to support long term memory. We know how to do it, already have spaced repetition and mnemonics, sadly teachers mostly ignore them.
It would be really cool to know how these neurons are different, and how and when they differentiate.
If you're interested in the biological basis of memory, I suggest reading 'In search of memory' by Eric Kandel. It's a biographic tale of Eric Kandel winning his nobel and along the way you learn a ton about how the brain works.
This is very interesting, and for me a new take on memory formation. It seems to have the advantage over LTP that it can work as a conincidene detector over longer time-spans.
"The epigenome and three-dimensional (3D) genomic architecture are emerging as key factors in the dynamic regulation of different transcriptional programs required for neuronal functions. In this study, we used an activity-dependent tagging system in mice to determine the epigenetic state, 3D genome architecture and transcriptional landscape of engram cells over the lifespan of memory formation and recall. Our findings reveal that memory encoding leads to an epigenetic priming event, marked by increased accessibility of enhancers without the corresponding transcriptional changes. Memory consolidation subsequently results in spatial reorganization of large chromatin segments and promoter–enhancer interactions. Finally, with reactivation, engram neurons use a subset of de novo long-range interactions, where primed enhancers are brought in contact with their respective promoters to upregulate genes involved in local protein translation in synaptic compartments. Collectively, our work elucidates the comprehensive transcriptional and epigenomic landscape across the lifespan of memory formation and recall in the hippocampal engram ensemble."
I think back to a historical abstract:
"We wish to suggest a structure for the salt of deoxyribose nucleic acid (D.N.A.). This structure has novel features which are of considerable biological interest."
This paper describes a process significantly more complex than the static molecular structure of DNA, and its abstract is written for the audience of researchers sharing the authors' own specialism. That's actually more useful than the converse, because it enables members of that audience quickly to evaluate the contents for relevance to their own work, in a way that would be much more difficult, perhaps impossible, were it written more concisely.
I really don't understand why a non-expert would expect experts who are writing for other experts to dumb things down for them. You might as well complain that a graduate level physics textbook isn't readily understandable to someone with no mathematical background!
And unfortunately, because this was Nature Neuroscience, the intended audience is more specific, which is why you see more jargon there.
Here is an extracted example from the latest issue of Science, for which the intended audience is more general:
Psychotic disorders such as schizophrenia impose enormous human, social, and economic burdens. The prognosis of psychotic disorders has not substantially improved over the past decades because our understanding of the underlying neurobiology has remained stagnant. Indeed, the subjective nature of hallucinations, a defining symptom of psychosis, presents an enduring challenge for their rigorous study in humans and translation to preclinical animal models. Here, we developed a cross-species computational psychiatry approach to directly relate human and rodent behavior and used this approach to study the neural basis of hallucination-like perception in mice.
https://science.sciencemag.org/content/372/6537/eabf4740
It’s not a perfect comparison because Science abstracts are apparently much larger (this is just the Intro!) but it does convey my point about language.
One concrete example: "Our work shows" would work just as well as "our work elucidates".
It's ultimately up to the author, but that kind of word choice really doesn't add anything and makes the paper less accessible to non-native speakers.
I hate institution's PR release hyperbole everything to the max.
And the senior author of the paper has a few listed on PubPeer https://pubpeer.com/search?q=li-huei+tsai
i’m not surprised that a mechanism like this is involved in memory storage/retrieval.
im left with questions about the heritability of these epigenetic parts of memory.
It started decades ago with seminal work on place & grid cells in spatial navigation (https://www.nobelprize.org/prizes/medicine/2014/may-britt-mo...) which was later shown to be generalizable to multidimensional "concept spaces".
It's probably the most exciting field in neuroscience right now, it also maps very nicely into mathematics of high-dimensional spaces:
"Navigating cognition: Spatial codes for human thinking" https://science.sciencemag.org/content/362/6415/eaat6766.abs...
"Organizing conceptual knowledge in humans with a gridlike code" https://science.sciencemag.org/content/352/6292/1464
"The Hippocampus Encodes Distances in Multidimensional Feature Space" https://www.sciencedirect.com/science/article/pii/S096098221...
"A non-spatial account of place and grid cells based on clustering models of concept learning" https://www.nature.com/articles/s41467-019-13760-8
"A learned map for places and concepts in the human MTL" https://www.biorxiv.org/content/10.1101/2020.06.15.152504v1....
"What Is a Cognitive Map? Organizing Knowledge for Flexible Behavior" https://www.sciencedirect.com/science/article/pii/S089662731...
"A map of abstract relational knowledge in the human hippocampal–entorhinal cortex" https://elifesciences.org/articles/17086
"Map-Like Representations of an Abstract Conceptual Space in the Human Brain" https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7884611/
"Knowledge Across Reference Frames: Cognitive Maps and Image Spaces" https://www.sciencedirect.com/science/article/pii/S136466132...
"Concept formation as a computational cognitive process" https://www.sciencedirect.com/science/article/pii/S235215462...
"Efficient and flexible representation of higher-dimensional cognitive variables with grid cells" https://journals.plos.org/ploscompbiol/article?id=10.1371/jo...
"The cognitive map in humans: spatial navigation and beyond" https://www.nature.com/articles/nn.4656
"A general model of hippocampal and dorsal striatal learning and decision making" https://www.pnas.org/content/117/49/31427.short
"On the Integration of Space, Time, and Memory" https://www.sciencedirect.com/science/article/pii/S089662731...
Slime mold has memory, and it doesn't have a neural network.
if one reduces "memory" to anything that exhibits sustained interplay between hysteresis and goal optimization then yes, neurons may lie below the level of explanation you wish to use, and could thus be substituted with some other signaling medium, but this is not the same as claiming they are not involved at all.