Construction of Training Sets for Valid Calibration of in Vivo Cyclic Voltammetric Data by Principal Component Analysis.

Principal component regression, a multivariate calibration technique, is an invaluable tool for the analysis of voltammetric data collected in vivo with acutely implanted microelectrodes. This method utilizes training sets to separate cyclic voltammograms into contributions from multiple electroactive species. The introduction of chronically implanted microelectrodes permits longitudinal measurements at the same electrode and brain location over multiple recordings. The reliability of these measurements depends on a consistent calibration methodology. One published approach has been the use of training sets built with data from separate electrodes and animals to evaluate neurochemical signals in multiple subjects. Alternatively, responses to unpredicted rewards have been used to generate calibration data. This study addresses these approaches using voltammetric data from three different experiments in freely moving rats obtained with acutely implanted microelectrodes. The findings demonstrate critical issues arising from the misuse of principal component regression that result in significant underestimates of concentrations and improper statistical model validation that, in turn, can lead to inaccurate data interpretation. Therefore, the calibration methodology for chronically implanted microelectrodes needs to be revisited and improved before measurements can be considered reliable.