Can Single Cells Learn?

To Gelber, the experiments demonstrated that Paramecium was learning to associate the wire with food, a conclusion that challenged scientists’ belief that only highly evolved, multicellular animals with central nervous systems were capable of such behavior. More fundamentally, her results suggested that at least some of the biological machinery needed for learning and other cognitive processes might exist not in the connections among neurons in an animal brain, but within individual cells themselves. “Possibly the biochemical and cellular physiological processes which encode new responses are continuous through the phyla,” Gelber speculated in one 1962 paper, “and therefore would be reasonably similar for a protozoan and a mammal.”

Seven decades after Gelber started her experiments, a group of researchers at Harvard University are arguing that her ideas deserve a revival.

Her conclusions touched a nerve in the wider scientific community. While some colleagues viewed her ideas with interest, her many critics argued that her experiments omitted vital controls needed to rule out other, simpler explanations for her results, such as tropism—an essentially automatic response or movement of an organism relative to a stimulus, such as a wire or food. More egregiously, some claimed, she seemed to be ignoring a well-accepted division between us and them: “Gelber freely applies to Protozoa concepts (reinforcement and approach response) and situations (food presentation) developed with higher metazoan animals,” wrote Paramecium researcher Donald Jensen in 1957. “I feel that such application overestimates the sensory and motor capabilities of this organism.”

Gelber’s work had all but disappeared from view by the time she died in 1991; a search for her name in academic papers between 1980 and 2020 suggests that she was barely mentioned during that time. But now, seven decades after she started her experiments, a group of researchers at Harvard University are arguing that her ideas deserve a revival. “I think that Gelber was really onto something,” says Sam Gershman, a Harvard cognitive neuroscientist who coauthored a recent eLife review of Gelber’s work, noting that he sees parallels between her ideas and how some researchers are currently thinking about information storage in single neurons. 

In addition to a handful of recent experiments on learning in single-celled organisms, there are now several converging lines of evidence from multicellular organisms, Gershman says, to suggest that at least some types of memory can be encoded in intracellular changes, such as epigenetic modifications to DNA or changes in genetic regulatory networks. He adds that, while the idea that unicellular creatures might learn “still doesn’t go down well” among some Paramecium biologists, he’s hoping that a more open-minded attitude could help researchers identify general rules about how complex behaviors arise across the animal kingdom, and, ultimately, better understand what learning and memory really are. 

Associative Learning in Ciliates © IKUMI KAYAMA, STUDIO KAYAMA Psychologist Beatrice Gelber conducted experiments in the 1950s and 1960s to test for associative learning in Paramecium aurelia. In her main trials (top row), she reported that, while the single-celled ciliates (green) ignored a bare wire dipped into their microscope slide, they swam over when she coated the wire in bacteria (red), which Paramecium eats. After several such sessions, she put bare wire back into the liquid, and reported that the protozoans still swam over, suggesting to her that they’d learned to associate the wire with food. Gelber ran controls to try to exclude other explanations: in one, she didn’t present the wire or any food during the training period (bottom row); in another, she presented bare wire instead of bacteria-covered wire (middle row). Neither of the control conditions resulted in learning, she concluded. Her critics said that she failed to control for other confounding factors, such as bacteria-induced changes to the liquid. Many other researchers rejected her findings on the grounds that it was physiologically implausible that such a biologically simple organism could learn at all. See full infographic: WEB

The evidence for unicellular learning

Central to the controversy about tests of learning in single-celled organisms is the age-old difficulty of designing an experiment that cleanly distinguishes one explanation for a result from another. In Gelber’s case, this meant demonstrating that her protozoa were adapting their behavior to some new stimulus because they’d associated it with a particular reward, and not because they were responding instinctively to chemical or other signals from the bacteria or the wire, for example. Experiments carried out by her critics, who concluded that Gelber’s findings were irreproducible, had their own problems on this front, Gershman says, and Gelber did carry out several careful controls that he says strengthened her conclusions. (See illustration.) Nevertheless, “the criticism stuck,” says Gershman, “because it fit with people’s predisposition” to assume unicellular organisms simply weren’t equipped to learn.

These experimental challenges are still present for researchers interested in finding examples of learning outside of multicellular animals—something that Jeremy Gunawardena, a systems biologist at Harvard Medical School and a coauthor on the eLife paper with Gershman, became familiar with when his lab started work on the single-celled, trumpet-shape ciliate Stentor roeseli a couple years ago. 

You wouldn’t have seen papers published in eLife and Current Biology ten years ago about learning in a single-cell organism. I really think people are getting interested.

—Audrey Dussutour, University of Toulouse and CNRS  

Like Gershman, Gunawardena says that he’s fascinated by how unicellular organisms challenge scientists’ understanding of the requirements for learning and other complex behaviors. To explore these ideas, his team set out to replicate experiments by Herbert Spencer Jennings, a zoologist who had studied S. roeseli decades before Gelber started her work on Paramecium. Using carmine dye as an irritant, Jennings had found S. roeseli responded differently upon repeated exposures, suggesting to him they were in some way learning from past experience. His findings, like Gelber’s, were criticized and deemed irreproducible in the mid-20th century—something that “really stuck in my craw,” Gunawardena says, not least because Jennings’s principal detractors hadn’t even used the same species of Stentor in their failed efforts to replicate his work. 

In their own study, Gunawardena and his colleagues used a needle to deliver polystyrene beads, an irritant the team had found to be more effective at eliciting a response than carmine dye, to S. roeseli cells that had settled on a microscope slide. In response, the protozoa showed various avoidance behaviors, such as bending away from the beads, curling over, or swimming away altogether. The researchers found, as Jennings had reported, that the cells seemed to show a behavioral hierarchy—at first responding in less dramatic ways, such as gently bending, but responding to subsequent provocation by swimming away or contracting. (See illustration.) While the behavior was not as complex as the associative learning observed in animals such as Pavlov’s dogs, the findings did imply that S. roeseli was adapting its responses based on its previous experience, the researchers concluded in a 2019 paper in Current Biology; Jennings, it seemed, had been vindicated.

See “Single-Celled Organism Appears to Make Decisions”

Another unicellular organism that appears to show some elementary form of learning is the slime mold Physarum polycephalum. It’s an unusual single-celled organism in that it can contain multiple nuclei, explains Audrey Dussutour, a biologist at the University of Toulouse and the Centre National de la Recherche Scientifique (CNRS) in France whose 2017 book, Le Blob, positions P. polycephalum as a model to understand complex behaviors in “nonneural organisms.” A few years ago, her team showed that this mold displays a type of nonassociative learning called habituation, in which an organism gets used to a particular stimulus and stops responding to it. A multicellular example would be a mouse that, after being startled by a sudden loud noise, responds less and less to the same noise the more times it hears it.

Dussutour’s group showed that P. polycephalum became habituated to quinine and caffeine—two compounds that the slime mold normally stays clear of—if the compounds were placed on a bridge that gave the mold access to food. It took a long time for the molds to explore over the coated bridges to begin with, Dussutour says, but once they started to do so, they seemed to stop minding the stimuli they’d previously avoided. (See illustration.) Consistent with observations of habituation in multicellular animals, she adds, the molds “recovered” their aversion to the compounds if they went a couple of days without encountering them. The team performed careful controls to show that the habituation was specific to those compounds, and not just a fatigue response to sensory overload. 

The existence of these and a handful of other studies on such a controversial subject is evidence of how the discussion around nonneural cognition is changing, says Dussutour. “You wouldn’t have seen papers published in eLife and Current Biology ten years ago about learning in a single-cell organism,” she says. “I really think people are getting interested.” Gershman says he hopes there will be more to come, adding that his lab has its own experiments with Paramecium planned. 

Nonassociative Learning in Ciliates and Slime Molds © IKUMI KAYAMA, STUDIO KAYAMA Following experiments by Herbert Spencer Jennings at the beginning of the 20th century, researchers at Harvard Medical School recently tested for a form of behavioral adaptation called behavioral hierarchy in the protozoan Stentor roeseli. They irritated cells that had settled on a microscope slide (1) by administering pulses of polystyrene beads every few minutes. They found, as Jennings reported, that S. roeseli behaved differently depending on what had happened previously. At first, the animals responded by bending away from the beads (2), or beating their cilia (3). But after a while, they took more dramatic steps—contracting (4), or eventually swimming away (5). The researchers report in their paper that the behavior, while not a complex form of learning, suggests that S. roeseli is making decisions about what to do based on previous experience.

© IKUMI KAYAMA, STUDIO KAYAMA Experiments by researchers at CNRS in France suggest that the unicellular slime mold Physarum polycephalum, which moves by protruding the edges of its body, shows a simple form of nonassociative learning called habituation. Mold grown in a dish connected by a bridge to a plate of food will normally grow over the bridge (1). Replacing the normal bridge with one covered in a substance the slime mold usually avoids, such as quinine (purple), drastically reduces the mold’s movement over the bridge. After several days of encountering quinine-covered bridges, however, the mold habituates and grows almost as normal over them (3). This behavioral change isn’t permanent—if the mold encounters normal bridges again (4), it appears to forget its habituation, and later shows the same aversion behavior on encountering another quinine-covered bridge (5). See full infographic: WEB

Still, not everyone’s convinced there’s something to see here. Some biologists continue to regard protozoa as “stimulus-response devices that maybe seem capable of sophisticated behavior, but really there are simplistic mechanistic explanations for those behaviors,” Gershman says. “That has been the research program for over a century.” 

Judith Van Houten, a biologist at the University of Vermont and an expert on Paramecium, recently wrote to Gershman and his colleagues explaining that she found Gelber’s experiments to be flawed and that claims of associative learning are incompatible with science’s understanding of this protozoan. She declined to be interviewed for this article, but writes in an email to The Scientist that “all behavioral studies of Paramecium must be grounded in the established understanding of their physiology, which is based on elegant and long-standing studies” from around the world.

The search for intracellular mechanisms of memory

Learning typically requires some kind of storage of information about the environment, so one way in which researchers such as Gershman hope to move the conversation forward is through exploring possible mechanisms by which memories could form in individual cells—whether that’s a unicellular organism swimming in a puddle, or an individual cell within a multicellular animal. Lumping the two into a single category isn’t as outlandish as it might sound, says Gunawardena. “A lot of the machinery is universal.” 

He notes that Paramecium, for example, produces calcium-based action potentials in response to certain stimuli and displays receptors for GABA, a well-studied neurotransmitter in multicellular animals, and has therefore often been referred to as a “swimming neuron” in the scientific literature. “I think if we were to uncover mechanisms in single-celled organisms, those same mechanisms might be operative in multicellular organisms.” Gunawardena adds that his lab is also planning experiments on behavioral adaptation and habituation in isolated mammalian cells.

One candidate for the kind of universal intracellular information storage that Gunawardena and others are after is RNA, which is produced and modified throughout an organism’s lifetime. The idea, which has so far been explored primarily in simple multicellular rather than unicellular organisms, originated in the 1960s when biologist James McConnell claimed that he could transfer memories among flatworms by taking RNA molecules from one individual and injecting them into another. Once again, the research was deemed irreproducible by most of the scientific community, and it soon faded from mainstream scientific discourse.

C. elegans

© ISTOCK.COM, HEITIPAVES

Among the labs now revisiting the concept is Coleen Murphy’s group at Princeton University. Murphy and her colleagues study the roundworm C. elegans, which can learn to avoid dangerous bacteria in its environment after being exposed to those bacteria. In a preprint posted on bioRxiv at the end of last year, the team reported that a worm that hasn’t encountered a particular bacterium can nevertheless learn to avoid it after being exposed to mashed-up bits of worms that have. The researchers identified tiny particles that appear to contain RNAs—although there wasn’t enough of the genetic material  for sequencing—as being critical to this transfer. 

Another team, led by David Glanzman at the University of California, Los Angeles (UCLA), reported evidence a couple of years ago that RNA seems to carry at least some types of memories in California sea hares (Aplysia californica), a type of marine snail. In one set of experiments, the team extracted RNA from nerve cells in snails that had experienced an electric shock, and injected that RNA into snails that hadn’t. After receiving the injection, the recipient snails, like their donors, behaved more cautiously, showing longer cowering behaviors after being tapped by a researcher compared with animals that received an injection of RNA from unshocked snails. Moreover, cultured Aplysia nerve cells treated with RNA from shocked snails were more easily excited by an electric current than were cells treated with RNA from unshocked snails.   

Glanzman and his colleagues speculated that the extracted RNA might be transferring memory between the organisms by inducing epigenetic changes to the DNA of recipient snails’ neurons and subsequently changing the animals’ behavior—an idea that he acknowledged to The Scientist at the time was “probably going to strike most of my colleagues as extremely improbable.” There have now been a handful of studies showing changes in patterns of DNA methylation or histone modifications in vertebrates during various learning tasks, Gershman notes, although he adds that these epigenetic changes have typically been assumed by neuroscientists to play a supporting role in the formation of memories, rather than storing those memories themselves. 

The great insight is that these are all instances of one fundamental capacity that exists in a wide range of systems: the ability to alter your future behavior based on your past experience.

—Michael Levin, Tufts University

Dussutour says her team is hoping to adapt these ideas to the study of unicellular organisms, and is now collaborating with molecular biologists to see whether RNA-based mechanisms might underlie habituation in slime molds. Other researchers are continuing to develop hypotheses about how physical modifications to different pieces of cellular machinery could provide opportunities for single-cell memory formation. Possible mechanisms include changes to the cytoskeleton and cycles of enzymatic phosphorylation and dephosphorylation of intracellular proteins, Gershman says. And just this year, researchers in Germany reported that P. physarum may use its own cell morphology to store information about where it previously found food.

Yet another proposed mechanism of single-cell learning bypasses the need for physical modifications to cellular components altogether. While those modifications are like “hardware changes,” there may also be “software changes,” says regenerative and developmental biologist Michael Levin. His group at Tufts University has been studying gene regulatory networks, which control gene expression, in individual cells. In a computational study published earlier this year, Levin and colleagues explored how these networks could shift their responses to certain stimuli or inputs without requiring underlying physical changes—much like how a computer doesn’t need to physically change its hardware when it records a piece of information typed into word processor. 

In the simplest version of such a network, genes are assumed to be activated or inactivated by interactions with other genes or by stimuli from the external environment. Memory arises because the current state of genes in the network is dependent on all the interactions and inputs that occurred until now. In some situations the team has studied, this means that the network can be trained to learn certain associations and adapt its future behavior “not because we’ve changed the connections between genes A and B. . . . It’s simply that certain experiences change the overall stable state of the system in a way that changes how it reacts to those stimuli in the future.” Levin says. “This is a thing that people get very tweaked by.”  

Paramecium caudatum

© ISTOCK.COM, MICRO_PHOTO

For some neuroscientists, these or other mechanisms of intracellular information storage could even offer an alternative, or at least a complement, to the more traditional, multicellular theories of how memory and learning work in humans. There have been “grumblings for years about the inadequacy of the current way we understand memory in the brain,” Gershman says. The leading idea, known as the synaptic plasticity theory, holds that memories are stored in connections between neurons, and that learning arises from changes in the relative strength of these connections.

See “Flexible Synaptic Strength May Underpin Mammal Brains’ Complexity”

But it’s been criticized by various scientists, including Randy Gallistel, an emeritus professor at Rutgers University and coauthor on the eLife paper, and UCLA’s Glanzman, for failing to fully explain real-life data. While it’s far from clear that newly proposed intracellular mechanisms will pick up the slack, Gershman says, they’re challenging researchers to rethink conventional theories of cognition.

Moving beyond ideologies

Although researchers who spoke to The Scientist say that they see value in the overlap between studies on unicellular organisms and those on multicellular ones, they acknowledge that the debate about the boundaries of learning and other cognitive processes is far from settled. Dussutour, for one, suggests that it might be less controversial for researchers to discuss their findings in single cells if they didn’t couch them in terms borrowed from traditional animal behavior research, perhaps implying an equivalency that’s yet to be demonstrated. “I think . . . people get upset because we call it learning,” she says of her and others’ studies on unicellular organisms. “It was the same thing in plants when people were talking about neurobiology,” she adds, referring to a heated debate that bubbles up now and then over whether plants can be said to show animal-like cognition. She would be as happy calling what she observes in slime molds something such as “adaptation” instead of “learning,” she says, adding that rather than trying to categorize particular behaviors as one thing or another, for her, “the more interesting venture is how they do that.” 

Levin, however, argues that it’s precisely the use of general concepts that helps researchers identify and talk about parallels that might otherwise be missed. One could say that “there’s memory and there’s shmemory,” he suggests, and thus satisfy people who’d prefer to keep traditional cognitive neuroscience out of it. But in doing so, “you miss an opportunity for the most powerful tool in science, which is unification.” With researchers now building artificially intelligent systems in living and nonliving media, single cells can hardly be viewed as the strangest examples of learning, he adds. “The great insight is that these are all instances of one fundamental capacity that exists in a wide range of systems: the ability to alter your future behavior based on your past experience.” 

Gelber herself would likely have approved of this holistic view, having discussed in several papers how studies of Paramecium might provide general insights into information storage and behavior across living things. Gershman, who says that a newly added Wikipedia entry on Gelber’s life and work came into being after he and his colleagues started asking about her on Twitter, tells The Scientist that he thinks it’s a pity her ideas were never taken more seriously, whatever qualms people had with her experiments. Revisiting her forgotten research has “made me keenly aware of the sociology of science, and the way in which things can get prematurely rejected,” he says. “Research paradigms can create a kind of tunnel vision, where we end up with this sort of accumulation of information about one particular way of thinking about some phenomenon, and then that ends up basically stymying any efforts to study alternatives.”