Trial by trial, machine learning approach identifies temporally discrete Aδ- and C-fibre mediated laser evoked potentials that predict pain behaviour in rats

Abstract

AbstractLaser evoked potentials (LEPs) – the EEG response to temporally-discrete thermal stimuli – are commonly used in experimental pain studies in humans. Such stimuli selectively activate nociceptors and produce EEG features which correlate with pain intensity. The rodent LEP has been proposed to be a translational biomarker of nociception and pain, however its validity has been questioned because of reported differences in the classes of nociceptive fibres mediating the response. Here we use a machine learning, trial by trial analysis approach on wavelet-denoised LEPs generated by stimulation of the plantar hindpaw of rats. The LEP amplitude was more strongly related to behavioural response than to laser stimulus energy. A simple decision tree classifier using LEP features was able to predict behavioural responses with 73% accuracy. An examination of the features used by the classifier showed that mutually exclusive short and long latency LEP peaks were clearly seen in single-trial data, yet were not evident in grand average data pooled from multiple trials. This bimodal distribution of LEP latencies was mirrored in the paw withdrawal latencies which were preceded and predicted by the LEP responses. The proportion of short latency events was increased after intradermal application of high dose capsaicin (to defunctionalise TRPV1 expressing nociceptors), suggesting they were mediated by Aδ-fibres (specifically AMH-I). These findings demonstrate that both C- and Aδ-fibres contribute to rodent LEPs and concomitant behavioural responses, providing a real-time assay of specific fibre function in conscious animals. Single-trial analysis approaches can improve the utility of LEPs as a translatable biomarker of pain.