Spike inference from calcium imaging data acquired with GCaMP8 indicators

Neuroscience can only be reproducible when its key methods are quantitative and interpretable. Calcium imaging is such a key method which, however, records neuronal activity only indirectly and is therefore difficult to interpret. These difficulties arise primarily from the kinetics, nonlinearity, and sensitivity of the calcium indicator, but also depend on the methods for calcium signal analysis. Here, we evaluate the ability of the recently developed calcium indicator GCaMP8 to reveal neuronal spiking, and we investigate how existing spike inference methods (CASCADE, OASIS, MLSpike) should be adapted for optimal performance. We demonstrate, both for principal cells and interneurons, that algorithms require fine-tuning to obtain optimal results with GCaMP8 data. Specifically, supervised algorithms adapted for GCaMP8 result in more linear and therefore more accurate recovery of complex spiking events. In addition, our analysis of cortical ground truth recordings shows that GCaMP8s and GCaMP8m – but not GCaMP6, GCaMP7f or GCaMP8f – are able to reliably detect isolated action potentials for realistic noise levels. Finally, we demonstrate that, due to their fast rise times, GCaMP8 indicators support shorter closed-loop latencies for real-time detection of neuronal activity. Together, our study provides demonstrations, tools, and guidelines to optimally process and quantitatively interpret calcium signals obtained with GCaMP8.