Viable fast distinction of human temporal lobe cortical layer 2/3 neurons based on electrical properties

Direct targeting of different neuron types in patch-clamp experiments in human brain slices is challenging because morphologically more intact neurons are deep in the slice, where visibility gets poorer, and we currently mostly lack genetic or fluorescent labels. Therefore, we aimed for an approximate but fast alignment of patch-clamped neurons to physiologically relevant neuronal classes.

We combined electrophysiological properties derived from somatic whole-cell recordings and morphological reconstructions of layer 2/3 neurons of the human temporal lobe. Using predefined electrophysiological and morphological features, we performed an unbiased principal component analysis (PCA) with subsequent hierarchical clustering of neurons for dimensionality reduction.

PCA revealed 6 stable groups separated by a high Euclidean distance: 2 pyramidal neuron (PYN) and 4 interneuron (IN) subgroups. Post-hoc statistical comparison between INs and PYNs allowed selection of 9 useful electrophysiological features to train a support vector machine (SVM) for distinction between the two neuronal types. The SVM reliably predicted PYN and IN in our own incomplete dataset and in an openly available dataset (Allen institute) gathered on a different experimental basis.

Our approach appears superior to subjective, experimenter-based classification, as it enables the simultaneous integration of multiple parameters. It does not rely on protracted post-hoc analysis such as molecular or anatomical classifications and therewith enables immediate assignment during the experiment also for less experienced experimenters.

Data from earlier experiments-comprising large numbers of neurons but acquired under suboptimal recording conditions-could be (re)analyzed using this framework.