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

Joshua H. Siegle, Peter Ledochowitsch, Xiaoxuan Jia, Daniel Millman, Gabriel Koch Ocker, Shiella Caldejon, Linzy Casal, Andy Cho, Daniel J. Denman, Séverine Durand, Peter A. Groblewski, Gregg Heller, India Kato, Sara Kivikas, Jérôme Lecoq, Chelsea Nayan, Kiet Ngo, Philip R. Nicovich, Kat North, Tamina K. Ramirez, Jackie Swapp, Xana Waughman, Ali Williford, Shawn R. Olsen, Christof Koch, Michael A. Buice, Saskia de Vries

TL;DR

This paper tackles the challenge of comparing neural activity data from two common techniques: electrophysiology (recording electrical spikes) and two-photon calcium imaging (measuring calcium influx as a proxy for activity). They found that while basic preferences are similar, electrophysiology tends to overestimate neuron responsiveness, and calcium imaging overestimates selectivity. By applying a "spikes-to-calcium" model, they could reconcile some of these differences, highlighting the inherent biases of each method.

What Problem Does This Paper Address?

Neuroscientists often use two primary methods to record neural activity: electrophysiology, which directly measures the electrical spikes (action potentials) neurons fire, and two-photon calcium imaging, which tracks changes in intracellular calcium levels as an indirect indicator of neuronal activity. While both offer single-cell resolution and can record from large populations, they have inherent differences in what they measure and how they sample neural populations. This leads to systematic discrepancies in the functional properties (like responsiveness and selectivity) derived from each method, making it difficult to directly compare findings across studies or integrate data from different labs. This paper aims to understand and reconcile these differences by comparing data from both modalities collected under highly standardized conditions.

Methods

The authors employed a multi-pronged approach:

  • Direct Comparison of Datasets: They utilized two large, publicly available datasets from the Allen Institute: one from two-photon calcium imaging and another from dense extracellular electrophysiology, both recorded from mouse visual cortex under standardized conditions.
  • Metric Calculation: They calculated key functional metrics for neurons:
  • Responsiveness: Whether a neuron reliably responds to a stimulus.
  • Preference: Which stimulus condition elicits the strongest response.
  • Selectivity: How sharply a neuron's response is tuned to a specific stimulus feature (e.g., orientation).
  • Spikes-to-Calcium Forward Model: They used a biophysically inspired model (MLSpike) to simulate calcium imaging signals from electrophysiology spike trains. This allowed them to transform ephys data into a format comparable to imaging data.
  • Event Detection Analysis: They investigated how different algorithms for detecting calcium events from fluorescence traces (e.g., exact lo, non-negative deconvolution) impact functional metrics.
  • Sub-sampling and Filtering: They controlled for potential confounds by sub-sampling datasets to match laminar distributions, running behavior, and filtering ephys data based on contamination levels (ISI violations).

Datasets

  • Allen Brain Observatory Visual Coding dataset: Two-photon calcium imaging data from 63,521 neurons in six cortical areas across 14 transgenic lines. The analysis focused on neurons in five areas (V1, LM, AL, PM, AM) from 10 lines expressing GCaMP6f in excitatory neurons.
  • Allen Brain Observatory Neuropixels dataset: Extracellular electrophysiology recordings from 99,180 units in cortical and subcortical structures from 58 mice. The analysis focused on units from five visual cortical areas (V1, LM, AL, PM, and AM) exhibiting regular spiking waveforms.
  • Novel electrophysiology dataset: Collected from transgenic mice expressing GCaMP6f and wild-type mice, used for specific forward model parameter fitting and comparisons.

Key Insights and Ideas

  • Preference is Robust, Responsiveness and Selectivity are Biased: Basic preferences (like preferred temporal frequency) are remarkably consistent between the two modalities. However, electrophysiology tends to report higher responsiveness, while calcium imaging reports higher selectivity.
  • Calcium Dynamics Sparsify Responses: The slow kinetics of calcium indicators and the event detection process inherently "sparsify" neural signals, meaning they tend to boost the amplitude of bursty activity while downplaying single spikes. This is a fundamental difference from electrophysiology, which directly measures spikes.
  • Event Detection Matters: The choice of algorithm and parameters for extracting calcium events from fluorescence traces significantly impacts derived functional metrics, particularly selectivity.
  • Electrophysiology Selectivity Bias: Electrophysiology data can be biased by the presence of contaminating spikes from nearby neurons, which can artificially inflate responsiveness and reduce selectivity.
  • Forward Model as a Reconciliation Tool: The spikes-to-calcium forward model is crucial for bridging the gap, allowing for more direct comparisons of selectivity, but it doesn't fully resolve differences in responsiveness without further constraints.

Key Results

  • Responsiveness Discrepancy: Electrophysiology consistently identified a larger fraction of responsive neurons across all stimulus types compared to calcium imaging.
  • Selectivity Discrepancy: Calcium imaging consistently showed higher stimulus selectivity (sharper tuning) than electrophysiology.
  • Forward Model Impact: Applying the spikes-to-calcium forward model to electrophysiology data significantly improved the match in selectivity distributions between the two modalities, reducing Jensen-Shannon distances from 0.41 to 0.14. However, it had a negligible impact on responsiveness.
  • Filtering Ephys Data: Filtering electrophysiology data to include only the "cleanest" units (lowest ISI violation scores) was the most effective way to reconcile selectivity and responsiveness metrics between the two modalities.
  • Sub-sampling Imaging Data: Including only the most active neurons in the imaging dataset improved the match in responsiveness but worsened the match in selectivity.
  • Functional Classification: When analyzing functional classes based on response reliability across multiple stimuli, electrophysiology appeared to have a much larger proportion of neurons responding to all stimuli. However, by increasing the responsiveness threshold for electrophysiology data, the class distributions became remarkably similar to those from imaging data.

Limitations and Future Work

  • Residual Differences: Even after applying the forward model and filtering, some discrepancies in selectivity remained, suggesting potential undetected contamination in ephys data or non-stationarities in the spikes-to-calcium transfer function.
  • Transgene Expression: While the study found that GCaMP6f expression did not substantially bias functional metrics, the authors acknowledge that other indicators or viral expression methods might behave differently.
  • Future Directions: The authors suggest future work could involve:
  • Using higher-density silicon probes (Neuropixels Ultra) to better disambiguate spike waveforms and reduce contamination.
  • Combining two-photon imaging with electrophysiology in more sophisticated ways (e.g., spike-triggered imaging) for direct ground-truth comparisons.
  • Investigating the impact of stimulus-dependent changes in calcium concentration that might not be captured by current forward models.
  • Exploring alternative calcium indicators (like GCaMP6s) and viral expression methods.
  • Further understanding the impact of elevated activity around implanted probes on recorded signals.

Summary updated

Abstract

Extracellular electrophysiology and two-photon calcium imaging are widely used methods for measuring physiological activity with single-cell resolution across large populations of cortical neurons. While each of these two modalities has distinct advantages and disadvantages, neither provides complete, unbiased information about the underlying neural population. Here, we compare evoked responses in visual cortex recorded in awake mice under highly standardized conditions using either imaging of genetically expressed GCaMP6f or electrophysiology with silicon probes. Across all stimulus conditions tested, we observe a larger fraction of responsive neurons in electrophysiology and higher stimulus selectivity in calcium imaging, which was partially reconciled by applying a spikes-to-calcium forward model to the electrophysiology data. However, the forward model could only reconcile differences in responsiveness when restricted to neurons with low contamination and an event rate above a minimum threshold. This work established how the biases of these two modalities impact functional metrics that are fundamental for characterizing sensory-evoked responses.
BibTeX
@inproceedings{siegle2021reconciling,
  author = {Joshua H. Siegle and Peter Ledochowitsch and Xiaoxuan Jia and Daniel Millman and Gabriel Koch Ocker and Shiella Caldejon and Linzy Casal and Andy Cho and Daniel J. Denman and Séverine Durand and Peter A. Groblewski and Gregg Heller and India Kato and Sara Kivikas and Jérôme Lecoq and Chelsea Nayan and Kiet Ngo and Philip R. Nicovich and Kat North and Tamina K. Ramirez and Jackie Swapp and Xana Waughman and Ali Williford and Shawn R. Olsen and Christof Koch and Michael A. Buice and Saskia de Vries},
  title = {{Reconciling functional differences in populations of neurons recorded with two-photon imaging and electrophysiology}},
  booktitle = {eLife},
  year = {2021},
  doi = {10.7554/elife.69068},
}
Figure