Concept Cells Help Your Brain Abstract Information and Build Memories
Abstract representations of individuals, objects and ideas are stored in individual brain cells known as concept neurons. Research suggests that they are central to memory.
Carlos Arrojo for Quanta Magazine
Introduction
Imagine you’re on a first date, sipping a martini at a bar. You eat an olive and patiently listen to your date tell you about his job at a bank. Your brain is processing this scene, in part, by breaking it down into concepts. Bar. Date. Martini. Olive. Bank. Deep in your brain, neurons known as concept cells are firing.
You might have concept cells that fire for martinis but not for olives. Or ones that fire for bars — perhaps even that specific bar, if you’ve been there before. The idea of a “bank” also has its own set of concept cells, maybe millions of them. And there, in that dimly lit bar, you’re starting to form concept cells for your date, whether you like him or not. Those cells will fire when something reminds you of him.
Concept neurons fire for their concept no matter how it is presented: in real life or a photo, in text or speech, on television or in a podcast. “It’s more abstract, really different from what you’re seeing,” said Elizabeth Buffalo, a neuroscientist at the University of Washington.
For decades, neuroscientists mocked the idea that the brain could have such intense selectivity, down to the level of an individual neuron: How could there be one or more neurons for each of the seemingly countless concepts we engage with over a lifetime? “It’s inefficient. It’s not economic,” people broadly agreed, according to the neurobiologist Florian Mormann at the University of Bonn.
But when researchers identified concept cells in the early 2000s, the laughter started to fade. Over the past 20 years, they have established that concept cells not only exist but are critical to the way the brain abstracts and stores information. New studies, including one recently published in Nature Communications, have suggested that they may be central to how we form and retrieve memory.
We know that the brain processes information about the outside world through the complex dynamics of circuits of neurons, said the mathematician Valeriy Makarov Slizneva of the Complutense University of Madrid, who has done theoretical calculations to prove that concept cells exist. However, it’s also possible that individual cells have essential roles in the brain’s reconstruction of reality.
“Nature has used, over time, simple but efficient concepts instead of dealing with complex distributed calculations,” he said. “We are simpler than we thought.”
Parody Brought to Life
The concept of a concept cell was a joke to neuroscientists — until it wasn’t.
In 1969, the neuroscientist Jerome Lettvin gave what became a famous lecture at the Massachusetts Institute of Technology. With a mocking tone, he told his students a story about a fictional neurosurgeon seeing a fictional patient who had a difficult relationship with his mother. To help, the neurosurgeon deleted a cell in his patient’s brain that coded for his mother, thereby erasing all memory of her. Satisfied with his accomplishment, he continued his research by looking for “grandmother cells.”
“Since then, people keep talking about grandmother cells,” said Rodrigo Quian Quiroga, a neuroscientist at the University of Leicester. Theoretically, a grandmother cell is a single neuron, hidden somewhere among the 86 billion in your brain, that codes for one of your grandmothers. You delete it and — poof — everything you know about that person disappears from your brain.
It wasn’t a notion that anyone took seriously. A cell for every single person you’ve ever encountered? “Isn’t that ridiculous?” said the neuroscientist Christof Koch of the Allen Institute for Brain Science in Seattle. “People had pooh-poohed the entire idea.”
But not everyone. In the 1990s, a research group at the University of California, Los Angeles, led by the neurosurgeon Itzhak Fried, developed a new kind of electrode that could look at the activity of individual neurons — an unprecedented level of resolution at the time. A scientist as much as a surgeon, Fried had always been curious about memory and our mental lives. “Somehow there is a transformation made of the entire external world into some representation” in the brain, he said. This representation could be reflected in vague and abstract concepts, absent of details from the real world. What could that look like?
Fried and Quiroga collaborated with Koch to investigate. During surgery on consenting epilepsy patients, they inserted electrodes into each patient’s medial temporal lobe, the part of the brain that includes the amygdala, entorhinal cortex and hippocampus, which is the hub for emotion and memory.
Mark Belan/Quanta Magazine
They then showed the patients images of objects. In 2000, the researchers reported that individual neurons seemed to represent broad categories, such as faces, scenes, houses or animals, by firing for multiple images within each category.
The results suggested that something like grandmother cells might exist — but only if these cells were responding to more than the images alone.
Cellular Conception
In the early 2000s, Quiroga was fiddling with an algorithm he had created to analyze electrode data that let him identify many more neurons than was previously possible — even cells that rarely fired and thus were harder to detect. “I can see neurons that people couldn’t see before … because I was using tricks that I learned from physics and math,” he said. “And then I said: ‘Well, I want to see what these neurons do.’”
At first he showed epilepsy patients images of scientists such as Richard Feynman and Albert Einstein to see whether neurons responded to individual people. When the patients couldn’t identify them, he tried showing them photos of more recognizable places and people, including Jennifer Aniston, a star of the hit sitcom Friends.
To his delight, he found a neuron that responded to the actor. That raised a new question: “Is it responding to this picture of Jennifer Aniston, or is it responding to the concept of ‘Jennifer Aniston’?” he recalled. In a follow-up experiment, he showed patients seven different pictures of Aniston, and found that the same neuron fired for all of the photos — but not for images of other actors or objects. He then started to identify neurons for other famous places and people. He found one that responded only to Halle Berry, and another that fired only for the Leaning Tower of Pisa.
Quiroga wrote out the name “Oprah Winfrey.” The same neurons that had fired for her picture also fired for her name. That meant that the neurons weren’t responding to features of the picture, such as brightness or color: They were context-independent. They were responding to Oprah as a concept.
He knew that his observation of one neuron firing didn’t mean there was only one neuron for every concept. If that were true, “the chance of finding it would be close to zero,” he said. “I used to joke that, if this were the case, I should quit science and start gambling because I would be the luckiest person ever.” He believed the brain must have many neurons for every concept, but he didn’t know how many.
In 2005, the team published their results in Nature, and the cells became known colloquially as “Jennifer Aniston cells.” At first, because of the long-running negative connotations around grandmother cells, “it was the hardest thing to get people to accept the possibility of such cells,” Koch said. In a related perspective piece, the neuroscientist Charles Connor wrote: “No one wants to be accused of believing in grandmother cells. But …”
Were these grandmother cells? “I’m very against this view,” Quiroga said. Sure, these cells were highly selective, firing only for Aniston or sometimes also for closely related people who might evoke her, such as other Friends cast members. However, the parodic concept of the grandmother cell assumed a one-to-one concept-to-cell ratio, and that was not the case with these cells.
A year after publishing their data, the team crunched some numbers. Based on an estimate from psychologists that the brain can distinguish about 20,000 semantic concepts, they calculated that millions of cells would code for every concept, and that each concept cell could code for dozens of different, though often related, concepts.
“When your brain encounters things all the time — even high-level things like dogs and cats and people and trees and ships — we will have neurons for that,” said Christof Koch, a neuroscientist at the Allen Institute for Brain Science. “I will certainly have lots of [concept] cells for Mr. Felix,” referring to his dog (pictured).
Courtesy of Christof Koch
For example, cells that fire for Harry Potter might also fire for his wizarding school pals Ron Weasley or Hermione Granger. Maybe they’d even fire for Gandalf, the wizard from The Lord of the Rings. “Same profession, different story,” Mormann said. “Sometimes you have narrow tuning to a single individual person and no one else, and sometimes you have wider tuning, maybe to a category like ‘wizards.’” The same concept cell might also fire for “wand” or “old men in gowns with beards,” he added.
Concept cells could code for anything and everything, but they are not used for object recognition. They’re too slow for that: These cells fire after a delay of about 300 milliseconds. “It’s unclear why it takes so long,” said Ueli Rutishauser, a neuroscientist at Cedars-Sinai Medical Center in Los Angeles. Rather, these cells seem to dip into a more internal process, forming an abstract representation informed by past experiences and memory.
Everyone has a different set of concepts and cells that encode them. Not everyone has seen Friends or follows celebrity culture. Instead, concept cells develop for people or objects that we care about or have some history with. “The representation depends on the past experience of that organism and things that have been associated before,” Buffalo said. For example, your brain could form an association between your date and the bar where you met him, so that your concept cells for the man might also fire for the bar. However, that’s only true if the bar is tightly coupled with the person, Mormann said: If it’s a place you go all the time, it’s unlikely that the same neuron will fire for both.
For years after publishing the work, Quiroga, who was not happy to be known as “the Jennifer Aniston neuron guy,” tried to get the term “concept cells” to stick. It failed to catch on until 2012, when he published a paper in Nature titled “Concept cells: the building blocks of declarative memory functions.”
The paper presented his hypothesis that the brain uses concept cells to convert information from the world into memory. The process requires abstraction: extracting relevant information from experience, stripping it of unnecessary detail and storing it. He proposed that concept cells, as abstract representations of ideas such as specific people or objects, might link together to form new associations (like words in a sentence) and serve as building blocks for memories (like a story composed of sentences).
“This is the skeleton of how we store memories,” Quiroga said.
Building a Memory
To many scientists, the idea that concept cells link up and interweave to form memories intuitively makes sense. Because memories are so important for our survival, it is “the best explanation for why our brain can afford the luxury of having such high specialization to independent semantic concepts,” said Sina Mackay, a graduate student at the University of Bonn who works with Mormann.
Indeed, in a recent study in Nature Communications, their team found the strongest experimental hints yet that concept cells may link specific objects to locations in our long-term memory. For decades, researchers have studied “place cells,” which store location information in our brains. The study found that the firing patterns of concept cells and place cells correlated with patients’ ability to remember an object’s location. Concept cells are the “what” to our memories, while place cells are the “where,” the authors wrote.
Concept cells are also linked to working memory, which is activated temporarily when you’re grocery shopping or remembering a phone number. This type of memory is “low capacity, and it has high demands,” Rutishauser said. “If you get slightly distracted, it’s gone.” In 2017, his team found that concept cells remain active for several seconds as you try to hold items in working memory. And in a study published in Neuron at the end of 2024, his team found that working memories are more likely to migrate into long-term memory when patients’ concept cells are active.
Working memory also turns on when you imagine a scenario or tell a story. “Shrek and Jennifer Aniston walk into a bar. … Maybe Shrek orders a beer,” suggested Pieter Roelfsema, who studies vision, perception and memory at the Netherlands Institute for Neuroscience. As you read this sentence, the concepts of Aniston, Shrek and a bar stitch together, one by one. It’s likely that concept cells play a role in this imagining. “You are building something in your working memory that is incrementally becoming richer and maybe more realistic,” he said, “and then the story unfolds.”
Roelfsema’s group recently found that concept cells respond to pronouns as well. In the study, the pronoun “he,” standing in for “Shrek,” lit up the same concept cells as “Shrek” did. “The [pronoun] then directs attention to the concept ‘Shrek,’ who is going to be the subject of the next sentence,” Roelfsema said. “I think that’s just beautiful you can measure that.”
Swiss Army Cells
Researchers debate how concept neurons fit in with other models of how the brain represents the external world. They are “a fantastic discovery,” said György Buzsáki, a neuroscientist at New York University who has researched the hippocampus for decades. However, the representation of concepts occurs at different scales in the brain — at the level of a single neuron and also at the level of cellular populations, he said. “What is more important?” he asked.
One obstacle to finding an answer is that concept cells are hard to locate. Currently they can only be studied in a clinical setting, where patients are undergoing surgery to have electrodes implanted for medical reasons. That limits who can study the cells and how.
Plus, it’s not easy to define them, said Cory Miller, a neuroscientist at the University of California, San Diego. Part of the problem is the vague definition of “concept” itself — no one can say whether we have concept cells for experiences such as emotions, for example.
One intriguing possibility is that the varied hippocampal cells can be remapped to do different jobs in different contexts. “When you start looking at the history and the overall picture, then you start scratching your head,” Buzsáki said. “There are time cells, there are place cells, there are border cells, boundary vector cells, there are concept cells. … Then at one point you say, ‘Oh, that cannot be right, there are a limited number of neurons in the hippocampus.’”
It’s possible that these neurons can play different roles and take on different identities based on the task at hand, Buffalo said. When it needs to be a concept cell for Jennifer Aniston, that’s what it is. When it needs to be a place cell to help you navigate toward the martini at the bar, it is a place cell. “That cell is like a Swiss Army knife,” Miller suggested.
The few groups with access to patients and the technology to record the activity of single neurons are excitedly continuing their experiments. Mormann wants to understand just how abstract concept cells can get: In preliminary data, he found some concept cells that respond to broad, amorphous concepts, such as government and taxes, but more that respond to concrete concepts, such as Jennifer Aniston. Quiroga, meanwhile, is hoping to prove that concept cells are specific to humans — a hotly debated idea with potentially profound implications. If no other animal can represent concepts in the brain, he said, “I would say this is the basis of our intelligence.”
Now that you’ve read this article, it’s possible that you have formed concept cells that code for concept cells — a concept that we are somehow able to wrap our brains around.
