Memory Transfer: What Worm Experiments Really Show
A 1960s worm experiment claimed memories could pass between animals. Explore why modern researchers failed to replicate it and what newer studies suggest.
Can Memories Move Between Animals? The Worm Experiment That Still Has No Easy Answer
In the 1960s, planarian flatworms looked like a way to answer a big question in a surprisingly direct way: could an animal pass a learned response to another animal just by being eaten? Researchers reported that worms trained to react to light could pass that response to other worms - and that even body fragments regenerated after cutting could show signs of the original learning. It was the kind of idea that feels both simple and hard to shake.
Then the results got messy. Years later, when researchers tried to repeat the work, their planarians often wouldn't learn the light response in the first place. That matters beyond one strange corner of neuroscience. It forces a harder look at how experiments are measured, how much trust older results deserve, and what a "successful replication" is actually telling us.
This article follows the original claim, the modern effort to reproduce it, and the newer lines of research that have kept memory transfer in the conversation. The goal isn't to declare that memories can - or can't - move between animals. It's to separate what the evidence supports from what still remains an open, technically difficult problem.
Why did the planarian experiment matter
Memory is often explained through brain activity: neurons change connections, and those changes help encode experience. That model has produced a lot of useful research, but it still leaves a question hanging in the air. Are neural connections the whole story - or do other biological processes play a role in how organisms store and use information?
Planarians made the question feel unusually testable. These flatworms have a simple nervous system, and they can also regenerate. Cut one in half and - under the right conditions - each piece can grow back into a whole animal. With that in hand, researchers could ask whether a learned response depends on the brain, on the rest of the body, or on something that can survive regeneration.
In the 1960s, psychologist James McConnell and colleagues reported that planarians could be conditioned to respond to a light associated with an electric shock. The response was a visible contraction, sometimes described as a "scrunch." McConnell reported that worms regenerated from trained animals retained the response, including pieces that hadn't contained the original head. He also claimed that untrained worms could pick up the response after consuming trained worms.
If that had been right, the implications would stretch well beyond flatworms. The idea was that at least some information tied to learning might be stored outside the brain and potentially carried by biological material. McConnell suggested RNA could be involved. At the time, scientists were deeply interested in how molecules like DNA, RNA, and proteins store and transmit biological information, so the idea landed as provocative rather than pulled from nowhere.
Why a simple animal was useful
A model organism helps when its biology makes a question easier to test. Planarians seemed to offer a kind of shortcut: train an animal, divide it, see what regenerates, and compare the results with animals that never got trained. The approach is dramatic, but the underlying question is straightforward: is the "memory" tied to the brain, or something else that the animal rebuilds?
Modern research uses the same logic. A simpler nervous system can make it easier to connect stimulus, mechanism, and behavior. But simplicity doesn't automatically make the result easy to interpret. To understand what you're seeing, the behavior has to be defined clearly, measured consistently, and separated from other possibilities - stress, movement, differences in handling, and so on.
Why the story outgrew the evidence
The claim also had a marketing-friendly shape. Memory "pills" and meals that could transmit knowledge are easy images to repeat, and McConnell was good at getting attention. He appeared on television and discussed ideas such as "piano-lesson pills" and "professor burgers."
That spotlight brought students and public interest, but it also blurred the distance between a reported laboratory observation and some future technology. The real experiments were about a specific conditioned response in worms, and the proposed mechanism was still speculative. Once the story became shorthand for "memory stored in food," the details were easy to lose.
What did McConnell's experiments actually test
The first step was classical conditioning. Researchers repeatedly paired a light stimulus with an electric shock. After training, they looked for planarians to contract when exposed to the light alone. In theory, that outcome would mean the animals learned an association between the light and an unpleasant stimulus.
McConnell then used regeneration to test whether the response would survive cutting. The head fragments weren't the only ones reported to retain the behavior; researchers also described responses in worms regenerated from tail pieces. If that was accurate, it would challenge the idea that memory lives only in the brain.
After that came the most sensational step. McConnell and others prepared material from trained worms and fed it to untrained ones. They reported that the recipients showed the conditioned response sooner than expected, as if some part of the training had been inherited. Some researchers also tried other transfer methods, including injecting material from trained animals into untrained recipients.
These steps shouldn't be treated as one lump. A worm that responds after conditioning, a regenerated fragment that responds, and an untrained worm that responds after consuming another worm are different claims with different controls. Even if one piece holds up, it doesn't automatically prove the others - or prove that something like a transferable "memory" exists in the everyday sense of the word.
The measurement problem hidden in the experiment
The core evidence was visible behavior. Researchers had to decide whether a worm's movement matched the contraction they counted as the learned response. That sounds simple until you remember how subtle the movement can be, especially in small animals that react to being handled and can shift through a bunch of different turns and contractions.
If observers disagree about what counts as a scrunch, then the same video can lead to different conclusions. If the people scoring the behavior know which worms were trained, their expectations can nudge how ambiguous movements get labeled. This isn't about calling anyone dishonest. It's about the ordinary risks of behavioral research, especially when scoring depends on human judgment.
A solid replication needs more than "same general idea." It needs a clear protocol for training, timing, scoring, and excluding cases. Blinding observers to experimental groups, recording footage for independent review, and defining the response before data collection can shrink the space for interpretation. Those safeguards don't guarantee the "right" answer, but they make disagreement easier to trace.
Why the RNA explanation was a hypothesis
McConnell thought RNA might be part of the transfer mechanism. RNA does have important roles in gene expression and cellular activity, and the idea was plausible enough to test - particularly in an era of intense interest in molecular biology. But a proposed mechanism isn't proven just because an observed effect could fit it.
To argue that RNA carried a learned response, researchers would have to isolate the relevant material, show it was necessary and/or sufficient for the effect, and rule out competing explanations. A recipient responding after consuming worm material doesn't, by itself, identify the active molecule - or show that the recipient acquired a memory in the same way the donor learned it.
Why couldn't modern researchers reproduce the results
In 2025, Sam Gershman's Harvard research group set out to reproduce a planarian-conditioning protocol associated with Alan Jacobson, a researcher from the original period. The team followed the published method, but the worms didn't reliably learn the light-and-shock association. The result was reported as a preprint in 2026, and it wasn't a failure of "memory transfer" so much as a failure of the basic training stage. In other words, they couldn't establish the training effect that would make a transfer test meaningful.
That distinction matters. If donor worms haven't demonstrably learned the response, then a recipient's behavior can't tell you whether a learned response was transferred. A replication that fails early doesn't automatically settle every later claim. Still, it challenges the assumption that the original conditioning result can be reproduced reliably using the described methods.
The team tried different stimuli, consulted other planarian labs, and used machine-learning analysis to review footage. They also went back to historical details - checking equipment and contacting Daniel and Reeva Kimble, former members of McConnell's lab. The Kimbles remembered the experiments as successful and had kept an archive from that period, The Worm Runner's Digest. Their recollections were useful context, but memories of past experiments aren't a substitute for new measurements carried out independently.
Replication is a diagnostic, not a verdict
A failed replication can point to several different issues. The original result might have been a false positive. The published method might miss a detail that matters in practice. The worms, their environment, or the equipment could differ. Or the outcome might depend on a condition the modern team hasn't identified.
In this case, the group collected planarians from multiple locations, including regions associated with the earlier work, and tested different sources. They still didn't find animals that learned under the attempted protocols. Failure across sources makes a simple explanation based on one bad batch less likely, but it still doesn't prove that no planarian can learn under any conditions.
This is the practical value of replication: it shows which parts of a claim survive contact with new teams, new equipment, and documented procedures. It isn't a courtroom ruling delivered by one experiment. Researchers have to interpret both positive and negative results alongside methods, statistical uncertainty, and independent attempts.
The likely weak point may be behavioral scoring
One explanation raised by the modern researchers is that early investigators might have interpreted ordinary turns or contractions as the specific learned response. That's hard to resolve from historical papers alone, especially if the original behavior wasn't recorded in a way that can be rescored today.
Behavioral measurement has improved since the 1960s, but "automation" doesn't remove judgment entirely. A machine-learning system can process video at scale, but its output depends on the training data, labels, and definitions it's given. If the model learns from an unclear version of the behavior, it may reproduce the ambiguity quickly rather than resolve it.
The deeper point is methodological. When a result depends on a small movement, teams need to specify what they're measuring before they look at the outcome. Video archives, blinded scoring, and inter-rater comparisons can help. So can publishing enough procedural detail for another lab to recreate the setup. Without those elements, a famous result can remain hard to separate from a measurement artifact.
What do newer memory-transfer studies suggest
The planarian story isn't the only reason scientists have asked whether biological material can influence learned behavior. More recent studies in other organisms have reported effects that keep the broad question alive - while still putting limits on what you can responsibly claim.
In 2018, neuroscientist David Glanzman's group at UCLA reported transferring sensitization in the sea slug Aplysia californica using genetic material from trained animals. The work suggested RNA might help trigger a change associated with learning. But it didn't show that a complete, detailed memory - like a specific autobiographical event - could be copied from one animal to another.
In 2021, geneticist Coleen Murphy's group at Princeton reported that Caenorhabditis elegans, a microscopic roundworm with 302 neurons, could learn to avoid a harmful bacterium. They found that exposure to material from trained worms could affect the behavior of untrained ones. Follow-up work has pointed to possible roles for biological carriers, including a retrotransposon called Cer1 and extracellular vesicles that contain genetic material.
These findings are scientifically interesting because they offer more specific mechanisms to test. They still don't establish a general-purpose memory-transfer process, and the organisms and behaviors differ from McConnell's planarians. A mechanism that changes a simple avoidance response in one species can't automatically be assumed to transfer a complex learned behavior in another.
What counts as a transferred memory
"Memory transfer" can mean a few different things. A molecule from a trained animal might change a recipient's sensitivity, stress response, or tendency to avoid a stimulus. That could be a real biological effect you can measure - without reproducing the donor's full memory.
To say the claim clearly, researchers need to show three things: that the donor learned something, that a defined material from the donor changes the recipient's behavior, and that the change is specific to the training rather than a general effect of exposure. Controls matter here: material from untrained donors, different stimuli, and independent behavioral scoring can help separate "information transfer" from nonspecific changes.
The strongest line of evidence would connect experience to mechanism to behavior. Researchers would identify the biological carrier, show how it changes the recipient, and demonstrate that interfering with the carrier eliminates the effect. It's a higher standard than just showing two groups behave differently after exposure, but it's what turns an intriguing observation into a credible mechanism.
What this means for neuroscience and technology
These experiments are not a demonstration of near-term memory-upload technology. There's no evidence here that people could transfer skills, knowledge, or personal experiences by consuming biological material. Treating that as an implied next step would blur a narrow experimental result into a completely different engineering problem.
The more useful connection to technology is methodological. Modern work combines biological experiments with video measurement, computational analysis, and molecular tools. Each layer can strengthen the evidence, but each also adds failure modes - unclear labels, protocols that are hard to reproduce, or molecular signals that correlate with behavior without causing it.
For researchers, product teams, and anyone evaluating a technical claim, the same rule applies: inspect the whole chain, not the headline. What was measured? What comparison was used? Which steps were replicated? What alternative explanations remain? Those questions are more useful than treating an unusual result as either proof of a breakthrough or proof of fraud.
Conclusion: keep the claim narrower than the headline
The planarian story remains compelling because it sits right on the edge between a bold idea and a stubborn experimental problem. McConnell's work suggested that learned information might survive outside the brain and perhaps pass between animals. Modern researchers couldn't reliably reproduce the conditioning that the claim depended on, leaving the most memorable conclusion unresolved.
Newer work in sea slugs and C. elegans gives scientists clearer molecular questions to pursue. It doesn't automatically validate every old planarian result, and it doesn't establish that memories can be transferred in the everyday sense. What the evidence supports is narrower: biological material may influence certain learned behaviors in some organisms, and the mechanisms deserve careful, targeted testing.
For now, the most useful conclusion is also the least sensational. Treat memory transfer as a research question, not an established capability. Follow the experimental chain from training to measurement to mechanism, and keep asking which parts independent teams can reproduce. That standard is what turns an enduring scientific mystery into something sturdier than a headline.