Inferring norepinephrine dynamics from partial observations reveals the temporal structure of elevations during arousal

Abstract

Hemodynamic artifacts present a significant challenge for two-photon fluorescence imaging of genetically encoded reporters, particularly when the timescale of relevant measurements matches those of vascular dynamics. This is an acute challenge for sensors in which the hemodynamic artifact is of comparable magnitude to the biological signal of interest. However, standard correction methods, such as isobestic recording or repeated experiments, are often impractical. Here we introduce a tiered framework for inferring norepinephrine (NE) dynamics across varying levels of recording information. First, we verify that dual-channel recording using an inert fluorescent reporter alongside the neuromodulator indicator enables direct hemodynamic correction within the same recording session. For contexts in which a dedicated reference channel is unavailable, we trained an LSTM-based model that predicts and removes hemodynamic contributions post-hoc from the recorded NE signal and behavioral variables. Finally, we show that key features of NE dynamics can be recovered from behavioral variables alone, providing an estimate of neuromodulatory state even when fluorescence recordings are unavailable. These methods enabled simultaneous multi-spectral measurements of axonal activity and neuromodulator release via simultaneous two-photon imaging of LC noradrenergic axons and extracellular NE in the same field of view. Cortical NE signals are graded with respect to behavioral intensity, scaling with both locomotion duration and pupil dilation amplitude. As expected, axonal activity precedes increases in ambient NE levels, but NE peaks later within a run and remains elevated after axonal activity has subsided, suggesting that extracellular NE integrates LC output over time rather than tracking instantaneous LC firing. Together, these findings demonstrate that accurate hemodynamic correction is essential for interpreting NE dynamics, and reveal a clearer view of the temporal structure of cortical norepinephrine signaling.