Number-Selective Neurons in the Medial Temporal Lobe

Post by Deborah Joye

What’s the science?

How does our brain represent numbers? Pre-verbal infants and non-human animals can’t recognize numeric symbols, but they can estimate numerosity: how many objects are in a group. Numerosity is intuitive, and young children leverage this to associate number values with specific symbols (i.e. Arabic numerals). There are ‘number networks’ in the brain that play a role in learning to use number symbols to solve math problems. The medial temporal lobe is one brain area connected to these number networks, including the hippocampus, parahippocampal cortex, and amygdala. Evidence from monkeys suggest that single neurons can preferentially respond to a specific numerosity and there is some evidence that humans may have neurons like this too. This week in Neuron, Kutter and colleagues record brain activity from participants performing a simple number task and demonstrate that single neurons in the human medial temporal lobe preferentially respond to specific numerosity and numeric symbols.

How did they do it?

Nine epilepsy patients had electrodes implanted in varying regions of their medial temporal lobes as part of their clinical treatment, and signals were recorded while they completed a simple calculation task. The authors presented numerosities (e.g.: a group of three dots) or number symbols (e.g.: “3”) and calculation rules (“+” or “add”) on a computer screen, each followed by blank delay frame to assess working memory activity. After presentation of a number, a rule cue, and another number, participants were asked to solve the calculation. After the task, the authors isolated recordings from individual neurons, then grouped them together according to brain region and time-matched them to the task sequence. They then analyzed how individual cells responded to numerosity and number symbols in the early part of the task (the presentation of the first number and delay frame) and later in the task (the presentation of the calculation rule and subsequent number). They also assessed whether single cells responded to both types of number representations or responded to the calculation component of the task (i.e. addition versus subtraction and whether the rule was a symbol or a word). Finally, to examine how the number information was encoded at the level of the neuronal population, the authors trained a machine learning algorithm (multi-class support vector machine ) to predict numerical value based on a subset of the recordings.

What did they find?

When examining neural activity from the first part of the task, the authors found that a significant proportion of individual cells in the medial temporal lobe responded to specific numerosities or number symbols, but not both simultaneously. More cells responded preferentially to numerosity than to symbolic numbers (29% vs. 6%), and only about 1% of all recorded neurons exhibited responses to both. Individual cells also had longer lasting responses to numerosity: they responded while the number value was shown, as well as the following delay period. In comparison, symbolic number-preferring cells responded only while the numeric symbol was present. However, neurons that preferred number symbols exhibited a greater selectivity for preferred values, showing a large response to the preferred number symbol and the same significantly decreased response to any other number symbol. This number discernment holds true on the population level as well. Their machine learning algorithm reliably predicted numerical value based on recordings from neuron populations, though it was more accurate for numerosity recordings. Interestingly, the closer the two values were on the number line, the more similar the population response pattern. This demonstrates a cellular version of the behavioral “numerical distance effect,” the idea that it’s easier to tell two numbers apart the further they are from each other.


When analyzing recordings from the later portion of the task (presentation of the calculation rule and second number), the authors found that a significant proportion of cells maintained preferential response to specific numerosities, though a much smaller proportion than responded to the first number presentation. Cells that responded to both the first and second numerosity also tended to prefer the same number value. Only a chance proportion of neurons responded preferentially to number symbols during the latter part of the task. Finally, the authors found that a very small proportion of cells selectively responded to the calculation rule (subtraction vs. addition) and of those, a subset responded preferentially to the format of the rule (symbol vs. word).

What’s the impact?

This is the first study to provide direct evidence that single neurons encode specific numerosity or number symbols. Past research has proposed two hypotheses: a “summation code” where cell firing systematically increases or decreases as numeric value increases or decreases or a “labelled-line code” where numerosity-selective neurons respond preferentially to specific number values. This study supports the latter hypothesis. Neurons demonstrate a unique capacity to prefer specific numeric symbols with a high degree of selectivity. These findings suggest that we may learn number symbols by linking them to evolutionarily conserved numerosity representations.


Kutter et al. Single neurons in the human brain encode numbers. Neuron (2018). Access the original scientific publication here.