Humans are the only animals to have developed the ability to represent numerical value with symbols. This capacity is thought to be built upon more basic quantification skills that allow us to accurately estimate the size (numerosity) of a set of objects. However, we know little about the coding of numerical value in the human brain. Kutter et al. have now used single cell recording methods to demonstrate the existence of distinct populations of ‘number neurons’ representing either symbolic numbers or numerosity in the human medial temporal lobe (MTL).

Credit: Image Source/Getty

The MTL is part of a wider ‘number network’ identified in human and non-human primate studies. To examine the contributions of individual MTL neurons to number encoding, the authors recorded the activity of 585 neurons from the MTL (including neurons in the amygdala, parahippocampal cortex, entorhinal cortex and hippocampus) of 9 subjects undergoing neurosurgery. During recording, the subjects participated in a task in which they were shown numerical values presented in a symbolic format (Arabic numerals) or a nonsymbolic format (random dot arrays).

The authors found that 16% of the recorded neurons were tuned to numerosity: they demonstrated peak activity when the subject viewed a dot array corresponding to one particular numerical value. Their responses to the presentation of other numerical values decreased progressively according to the size of the difference between the presented value and their preferred value, a property known as a numerical distance effect. 3% of the recorded neurons responded specifically to numerical values presented as symbols. These neurons exhibited a smaller numerical distance effect, suggesting a sharper tuning to numerical value. Only a few neurons were tuned to both nonsymbolic and symbolic numerical values.

To examine the encoding of numerical value at a population level, the authors trained a machine classifier to recognize numerical values on the basis of the firing of the population of selective neurons. The classifier was capable of decoding numerical information from a novel set of firing data at above chance levels for both symbolic and nonsymbolic representations, indicating that MTL population activity can accurately encode numerical representations.

3% of the recorded neurons responded specifically to numerical values presented as symbols

These findings reveal the existence of two mostly separate populations of ‘number neurons’ representing symbolic and non-symbolic numerical value in the human MTL. Further analysis of the functional properties of these neurons and their place in the number network may provide clues to the mechanisms underlying our unique symbolic numerical abilities.