(J-391) Application of principal component analysis to multiparametric electrophysiology datasets obtained from hiPSC sensory co-cultures to identify novel analgesics in a moderate-throughput system

Introduction:: Chronic pain affects 1 in 5 adults and is the number one cause of long-term disability in the world. Opioids are prescribed to manage chronic pain but are associated with deleterious side effects, tolerance, and abuse. The Pain Research Lab at UMass Lowell has developed an innovative DRG tissue model of acute and chronic nociception using human induced pluripotent stem cell (hiPSC)-derived sensory neurons and glia co-cultured on microelectrode arrays (MEAs). This novel approach enables non-invasive, moderate-throughput, high-content phenotypic screening of prospective therapeutic compounds. In this work, we used Principal Component Analysis (PCA) and K-Means Clustering to analyze multiparametric electrophysiological datasets and optimize the definition of an analgesic screening ‘hit’. We then compared the number and quality (based on Z’ assay quality metrics) of multiparametric hits to those identified based on changes in extracellular action potential (EAP) firing rates alone.

Materials and Methods:: EAP recordings were performed using a 48-well Maestro Pro (Axion Biosystems, GA, USA) system, with EAPs defined by filtered continuous recording threshold crossings at ±5.5σ. EAP time stamps were then used to calculate a number of parameters related to rates and patterns of EAP firing (i.e., active electrode yield, mean firing rate, weighted mean firing rate, interspike interval coefficient of variation, burst rate, network burst rate, and area under the normalized cross-correlogram (AUNCC)). PCA, K-means Clustering (to determine the number and confidence of ‘hits’), and plotting were carried out using a boutique Python script.

Results, Conclusions, and Discussions::

Results

Unsurprisingly, firing rate parameters such as mean firing rate and weighted mean firing rate exhibited high covariance (0.998) and contributed near-equally (Eigen values of 46.3 and 24.5%) to data variance in 2D PC space. Likewise, EAP pattern parameters (AUNCC, interspike interval coefficient of variation, and network burst rate) exhibited reasonably high covariance (0.46 and 0.27, respectively) and PC variance contributions (14.2 and 4.64%, respectively). PCA and K-means was run on 9 multi-well plates. K-means clustering successfully identified statistically separable clusters in all cases, as summarized in Table 1. PCA was likely to identify fewer hits than based on mean firing rate alone, but reproducibly identified positive control compounds (i.e., lidocaine).

Conclusions

We are able to record and pre-process EAP firing data for multi-parametric analysis in boutique Python script. PCA plus K-means clustering provides a promising framework for defining statistical models to identify prospective analgesic hits. Future work will focus on identification of compound class signatures based on PCA shifts in pooled moderate-throughput data sets.

Discussions

Multi-well MEA systems yield high-content (i.e., multiparametric) datasets that are traditionally reduced to a single value per well – mean firing rate. This largely ignores the effect compounds may have on network-level activity, versus the effects on intrinsic activity from single neurons. This also ignores the modulation of EAP patterns, which is known to affect transmission efficiency and downstream firing in the spinal cord and brain. Our approach, which considers multiple parameters related to firing rates and patterns, may add needed richness to data sets intended to identify novel analgesic compounds and reduce false-negative hit rates.

Acknowledgements (Optional): : Special thanks to Neal Lojek, Rasha El Ghazal, and Jai Singh Rajput for their training, assistance, and mentorship

References (Optional): :