On the correspondence of electrical and optical physiology in in vivo population-scale two-photon calcium imaging

Abstract

Multiphoton calcium imaging is commonly used to monitor the spiking of large populations of neurons. Recovering action potentials from fluorescence necessitates calibration experiments, often with simultaneous imaging and cell-attached recording.Here we performed calibration for imaging conditions matching those of the Allen Brain Observatory. We developed a novel crowd-sourced, algorithmic approach to quality control. Our final data set was 50 recordings from 35 neurons in 3 mouse lines. Our calibration indicated that 3 or more spikes were required to produce consistent changes in fluorescence. Moreover, neither a simple linear model nor a more complex biophysical model accurately predicted fluorescence for small numbers of spikes (1-3). We observed increases in fluorescence corresponding to prolonged depolarizations, particularly in Emx1-IRES-Cre mouse line crosses. Our results indicate that deriving spike times from fluorescence measurements may be an intractable problem in some mouse lines.Introduction 1 Systems neuroscience demands a steady increase in the number of simultaneously 2 recorded neurons. Over the last five decades, the number of recorded neurons has 3 doubled approximately every 7 years, mimicking Moore's Law [1]. Electrophysiology has 4 been the de facto gold standard for measuring neural activity for more than 80 years, 5 starting with pioneering work on the squid giant axon by Young, Curtis, Cole, Hodgkin, 6 and Huxley [2][3][4]. However, when the express goal is to push the envelope on the 7 number of simultaneously recorded neurons in vivo, intracellular electrophysiology is 8 impractical, and extracellular electrophysiology has other limitations, e.g. that the 9 precise origin of the signal (positional, genetic, morphological, and physiological details 10 of the generating neuron) are typically unknown [5]. Simultaneous recordings from large 11 numbers of genetically-identified neurons or experiments that necessitate recording from 12 the same neurons repeatedly over many days are difficult to achieve with 13 electrophysiology and are more tractable with imaging techniques. 14 One commonly used imaging technique is 2-photon calcium imaging. Anticipated15 from first principles by Maria Göppert-Mayer [6, 7], and experimentally pioneered by 16 Denk, Strickler, and Webb [8], 2-photon imaging emerged as a complementary technique 17 for the acquisition of neural activity with single-cell resolution from hundreds of neurons 18 in the living brain. Modern technology has enabled the development of genetically 19 1/32 encoded calcium indicators (GECIs), which can be introduced via viral transfection or 20transgenic strategy, and are expressed in a promoter-specific manner. The development 21 of GCaMP6 in particular, with its promise of elusive single spike sensitivity [9], heralded 22 a new era of massive optophysiological surveys [10], recording large numbers of neurons 23 from specific, genetically distinct populations. 24Unfortunately, the fluorescence of a calcium indicator is only indirectly li...

Reconciling functional differences in populations of neurons recorded with two-photon imaging and electrophysiology

“…Unexpectedly, the forward model did not change responsiveness metrics. We initially hypothesized that the lower responsiveness in ophys was due to the fact that single-spike events are often not translated into detectable calcium transients (Ledochowitsch et al, 2019). Instead, our observation of unchanged responsiveness following the forward model suggests that differences between modalities are more likely due to ephys sampling bias-i.e., extracellular recordings missing small or low-firing rate cells, or merging spike trains from nearby cells.…”

“…This algorithm identified the onset time and the corresponding magnitude of fluorescence change (magnitudes are reported on the same scale as ∆F/F and are therefore unitless). For the most part, such "events" do not represent individual spikes (Ledochowitsch et al, 2019;Huang et al, 2019), but rather are heavily biased towards indicating short bouts of high firing rate, e.g. bursting.…”

“…However, this comparison is also confounded by the fact that one event is not equivalent to one spike. There is rich information contained in the event amplitudes, which have a non-linear-albeit on average monotonic-relationship with the underlying number of spikes within a window (Ledochowitsch et al, 2019). Integrating over all 250 ms windows in the ophys dataset (N = 40 million intervals with at least one event, out of 1.1 billion possible intervals), event amplitudes vary over 4 orders of magnitude, with most amplitudes falling in the middle of this range ( Figure 2B; the truncated shoulder at the lowest end of this range is due to the noisedependent event detection regularization, see Methods for details).…”

“…two-photon-imaging recordings from layer 2/3 of mouse visual cortex (Huang et al, 2019;Ledochowitsch et al, 2019). Additionally, three parameters were calibrated on the fluorescence traces from the ophys dataset to capture the neuron-to-neuron variance of these parameters: the average amplitude of a fluorescence transient in response to a spike burst (A), the decay time of the fluorescence transients (τ), and the level of Gaussian noise in the signal (σ) ( Figure 4B).…”

Scaling of information in large neural populations reveals signatures of information-limiting correlations

“…To compare our required population size estimates to the total number of neurons in mouse V1, we conservatively estimated the need for about 48,000 neurons (see Methods) to achieve drift direction discrimination performance that most likely exceeds that of the animals (Abdolrahmani, Lyamzin, Aoki, & Benucci, 2019;Glickfeld, Histed, & Maunsell, 2013). Our use of time-deconvolved calcium activity as a noisy proxy for spike counts (T.-W. Chen et al, 2013;Ledochowitsch et al, 2019) makes these estimates upper bounds on required population sizes (see SI). Nonetheless, they compare favorably to the number of neurons in mouse V1, whose estimates range from 283,000 to 655,500 (Herculano-Houzel, Watson, & Paxinos, 2013;Keller, Erö, & Markram, 2018).…”