Memories all the way down.
Could memories be an emergent cellular property?
The 60s were a weird time, and no one was weirder than a group of scientists studying the transfer of memories via cannibalism in flatworms. Flatworms, specifically planaria or arrowhead worms, are famous for their ability to regenerate and be conditioned or to learn.
Integrating these two features led James V. McConnell to design an... interesting experiment. He conditioned his worms to fear light by shocking them, then crushed the trained worms and fed them to new, naive worms. These naive worms supposedly gained the same fear of light or were more easily trained, having consumed the information. He fractionated his educated, mangled worms and found that RNA fraction induces this memory transfer. Weird. I know.
Memory transfer via cannibalism was admittedly a highly debated, somewhat fringe result, but it sparked something: a new reductionist approach to memories. Dealing with memories and experimenting with creating, curating, and reflecting upon them was now a tangible, measured thing that could be boiled down to the fundamental building blocks of life.
Could memories be a biological trait? Ubiquitous across cellular life and subjected to the same processes as protein synthesis and DNA replication? Why are we avoiding the same robust, straightforward approaches we employ to study other phenomenological traits to this?
Let's start simple, or with what reassembles simplicity for most scientists—the humble E. coli. E coli cells have a very straightforward life; they eat food. If the food is nowhere to be found, they go searching for said food, and when they find it, they can return to, well, eating it. The two main activities in this search are Run- a swim maneuver that utilizes all of their flagella intertwined into a single, directional propellant, and Tumble - a floating state in which all the flagella cancel one another and the cell is immobilized. The cells use tumble when conditions are good, and they want to stay and Run when they move toward food or up a chemoattractant gradient. I said something pretty wild here because, at the very least, to perform this activity, a cell needs to have:
A sense of the state it is in
Temporal memory of the state it was in
A way to decide if things are improving (run) or worsening (tumble).
So cells, both in the context of a network and as a singular biological entity fending for itself, have memories; they utilize them, maintain them, and sometimes tell others about them. At the very least, they consist of a temporal association and reference comparison.
Ok. this is a very rudimentary definition of memory. It only spans a temporal, situational state shift. It definitely doesn't change the phenotype of the behavior of the population. A simple algorithm that requires very little of what we would constitute as a memory.
Slime mold, a form of social amoeba, takes it much further than E.coli. This communal organism, in which the border between self and others is not quite clear, can form webs of mycelia that span areas of the forest floor. These webs are quite literally pulsing, with cells moving through this convoluted network in search of nutrients or moisture. Some bad neighborhoods where food has been depleted are abandoned while new prosperous areas experience cellular migration. This movement is fascinating to watch, but it becomes completely surreal when new food is introduced to once-abandoned feeding grounds. Cells flood back through the network, illustrating a memory that is not only confined to experienced changes but also spatial ones.
Another intriguing aspect of incorporating spatial memory into the reality perception of an individual is the slime mold's ability to integrate two types of memory. There is strong evidence that slime mold can shuttle nuclei in a given transcriptomic state to an area where those transcripts are needed. This genetic gymnastics act requires a lot. You have to know the state of a given nucleus and a place where that state is required and facilitate the best solution where these two will meet. One that requires a much more complex internal representation of the environment and multiple transcriptional states of many nuclei.
"but you don't MEAN the bacteria FEEL anything, right? It doesn't remember anything? It just is!" This is a common statement when discussing these forms of behavior with molecular, evolutionary, or system biologists. Is this relevant, given what we know about natural selection? Population dynamics? Surly single-cell organisms are this zen-like creature in an unpermutable state—only changing slightly and in an open circuit with no feedback. If it happens to find itself in an advantageous state, it will propagate; otherwise, it will die. That's it. Solved.
You will know what I mean if you have ever attended a single-cell conference. One of the more popular slides to open a talk is one of a car with the parts constituting it spread out in front of it to illustrate complexity. It does exactly the opposite. A car is predictable, organized, and fragile. The car won't fix itself; throwing a wrench into its engine will not make it drive faster; changes made to it need to be designed and programmed. A cell is viewed the same: an elaborate machine with many moving parts working together to avoid death.
Only it's not. Organisms are malleable, adaptable, and great at solving problems not only as a population but as individuals hacking away at a problem changing to accommodate the infinite insults hurled at them by environmental changes or, much more frequently, internal errors and inadequacies. These problems are accommodated by alterations to all levels of biology, from RNA to changing behavior, all without mutations or the appearance of new genes.
But do cells remember stuff anyway? Why would they? Perhaps let's get a bit formal about it. Knowing something about reality increases your ability to predict, based on the information flowing in, the upcoming events. Predicting well is the same as lowering uncertainty, helping you prepare and make decisions. Important stuff, as the cost of a wrong decision for a cell, might be detrimental. Going out of lag, switching from glucose to maltose, swarming, or staying put is costly and dangerous for a biological entity when done based on wrong information. Memory is useful here because it integrates these wrong and correct decisions into an ongoing model of reality.
Let's think of the epigenetic landscape, or the ability to produce changes within the life state of an individual sans mutations, as the place of not random changes but directed stress-correlated behavioral changes. What will direct such changes? What will make a bacteria behave differently than the swarm? Standing variation, of course, or natural selection? Both are powerful, but let's imagine a situation where both fail for an individual. That individual is dying; it is now quite literally to save his life, is changing, and by some freak accident, he manages to alleviate his stress and survive. This solution now found is part of the population heterogeneity. Natural selection's role now, in this case, is not to generate new solutions but rather to hold on to these found solutions and commit them to, you guessed it, memory via genetic assimilation.
Ok, so for single cells, fending for themselves, memories of solutions and temporal or spatial sequences have a great fitness value. Still, when integrated into a cohort of cells in a higher organism, these individual "memories" are presumed meaningless as the network of neurons is the one capable of cognition and memory.
Let's circle back to our flatworms and their regenerative properties. Later, much more robust experiments tried to poke at exactly this notion, in which a neuron is not an agent of memory but a part of a network. These works tackle the top-down, brain-centric approach, pardon the pun, head first by taking trained worms and decapitating them. As we know, planaria are remarkable regenerators, and it comes as no surprise they can regrow their head; what is surprising is that that regrown head, complete with a new brain, manages to regenerate not only tissue but the memory of the conditioning the body has gone through. That's right. Memories stored in individual somatic neurons can induce full-on behavior in the semi-new, functioning worm. This illustrates precisely what I mean: an elaborate memory confined to some out-of-network rouge neurons.
Biology is special. It's wonderfully weird and is plastered with many things that could not be. Why are we limiting ourselves from thinking or experimenting with robust biological ideas easily measured and quantified from a different discipline? Perhaps the time has come for magic-free, wonder-filled science, where experiments lead the way instead of convention; hey, in the worst-case scenario, we’ll be proven wrong. Single-cell life has surprised us with so many things; so many capabilities were boiled down to this fundamental component of life. We should find out how deep the memory hole goes.
This work was written as a part of the ideas matter fellowship! I would like to thank Niko McCarty and Ethan Freedman for their time, attention and truly super helpful comments.
