SPARC: Noise Analysis and Signal Drift Correction in Long‐Period Multi‐Unit Recordings of Action Potentials in Autonomic Nerves

Abstract

Recordings of spike activity in small peripheral autonomic nerves close to organs can provide real‐time information about the functional states of those organs (afferent traffic) and the neural systems that control them (efferent traffic). However, maintaining long‐term multifiber recordings from autonomic nerves in vivo poses significant and unique technical challenges. Many visceral nerves consist almost entirely of C fiber axons in which action potentials produce very small currents with little spatial segregation, making individual units difficult to distinguish based on spike shape in multifiber filaments recorded conventionally under mineral oil, especially when signal‐to‐noise ratios are low. Ongoing activity can be due to hundreds of active axons, each firing at relatively low frequency. Slow changes in recording conditions can have large effects on the amplitudes of spikes and the signal‐to‐noise ratio over time, potentially causing spurious/artefactual changes in recorded spiking frequency when activity is quantified via thresholding. Previous approaches to drift compensation in recordings from neuronal somata or myelinated axons have utilized spatial information from multi‐electrode arrays or tracked changes in the amplitudes of single units differentiated by spike shape. Such approaches are not useful when recording the activity of C fibers due to their lack of spatial segregation and similar spike shapes. We have developed a method to track drift in signal and noise separately and to compensate for drift without the need for spike sorting. The method is robust to changes in the signal‐to‐noise ratio and can be applied online during data acquisition. It employs a recursive piecewise linear model that predicts resistance as a function of recording time. The implicit assumption is that the current source is constant and, conversely, the extracellular path resistance of the system is dynamic, being impacted by events such as mechanical disruption, or subtle movement of aqueous conducting fluid. We evaluated our method using recordings of gastrointestinal afferent axons during responses to inflammation. We also developed a simulation suite to generate artificial data that mimics the conditions in our recordings and a conventional spike sorting algorithm optimized to distinguish activity in single units and spike families in low noise C fiber recordings lasting hours. We have used these tools to validate the performance of our drift compensation algorithm and found it could compensate for drift effectively without spike sorting and could also improve the efficiency of spike sorting routines in long recordings where drift was present. The simulation suite, the analysis algorithms, and a graphical user interface are freely available via online repositories. We anticipate that, in future, as recording technologies improve, our algorithm will allow online assessment of activity in multi‐unit recordings in vivo that may provide a basis for closed‐loop therapy systems.Support or Funding InformationFunded by the Defense Advanced Research Projects Agency (DARPA) BTO, Contract No. N66001‐15‐2‐4060.