Fig 1.
Photocurrent contamination confounds mapping of nearby synaptic connections.
a, Synaptic connectivity mapping proceeds by patching a single downstream cell and targeting surrounding cells with two-photon excitation. When the downstream cell expresses opsin, the observed currents contain a mix of synaptic current and direct photocurrent. Left panel: cartoon of experimental setup; red circle represents radius in which direct photocurrent will be evoked. Middle panel: photocurrent artifacts are added to any observed synaptic current signals. Right panel: such artifacts subsequently corrupt connectivity inference. b, Schematic of grid experimental design. Stimulation laser was fired randomly in a grid pattern to excite cells expressing ChroME2f. c, Representative traces from a proximal (top) and distal (bottom) stimulus location in a grid experiment. Without accounting for the photocurrent artifact, both locations are naively categorized as being connected to the patched cell. The top traces represent putative photocurrent-only responses: excitatory currents which begin concurrently with laser onset, with amplitude modulated by laser power. The bottom traces represent prototypical PSC responses. Note the steeper onset profile, which occurs a few milliseconds after stimulation. PSC response latency is modulated by laser power, but amplitude is not. Purple and red traces denote 30 and 50 mW stimulation, respectively. d, Inferred connectivity map for an example grid experiment. The cluster of strong inferred weights in the center is likely artifactual. Caption continued on next page. e, In the targeted design, excitatory cells expressed ChroME2s fused to mRuby3, allowing us to detect and segment presynaptic target cells. f, Representative traces from a proximal (top) and distal (bottom) target cell in a targeted experiment. The top traces contain a putative mixture of photocurrent and synaptic current. Bottom traces represent putative PSC-only response. g, Inferred connectivity from a representative targeted experiment. Green cross denotes location of the postsynaptic cell.
Fig 2.
A constrained low-rank model separates synaptic currents from photocurrents.
a, Scaled photocurrent traces from the grid experiments with ChroME2f. Colors correspond to different datasets; stimulation power used is shown on the right. Even with the same opsin, there is substantial cell-to-cell variability in the photocurrent waveform. b, Same as a, but for targeted experiments using ChroME2s. c, Schematic of the low-rank approach. The matrix of PSC traces (left) can be approximated as a single waveform scaled by a vector of amplitudes. d, The overapproximation constraint stops the estimated waveform from picking up PSCs. e, PhoRC estimates a baseline component which accounts for photocurrents from the prior trial. f, Simulated example showing the effect of the overapproximation constraint. Left: simulated traces containing a mixture of photocurrents and PSCs. Right: Photocurrent estimates for the traces shown on the left. The overapproximation constraint allows PhoRC to extract the photocurrent waveform while ignoring PSCs.
Fig 3.
PhoRC captures photocurrent artifacts across protocols and datasets.
a, Left: Inferred weights for the grid design, before and after subtraction. Weights are shown for a single Z plane. Right: Inferred weights before and after subtraction for a targeted experiment. Weights are shown for the entire Z stack. Green cross marks location of the postsynaptic cell. Applying PhoRC reduced the inferred synaptic strength of the connection nearest the postsynaptic cell, which was contaminated by photocurrent. b, Traces selected by taking 10 trials with the largest estimated synaptic current components, and ten trials with the largest estimated photocurrent components. Left panels: Observed currents, photocurrent estimates, and subtracted traces for a ChroME2f experiment. Right panels: same as in left panels but for a ChroME2s experiment. PhoRC successfully infers the photocurrent waveform in both cases, despite the use of different opsins. Estimates ignore PSCs which have higher latency than photocurrents. c, Left panels: average evoked waveforms for putative connections before and after subtraction for a ChroME2f experiment. Right panels: same as left, but for the ChroME2s experiment. In both cases, connection waveforms before subtraction display signs of photocurrent contamination, whereas waveforms obtained after subtraction have PSC-like profiles. d, PhoRC performance across grid datasets. Each column shows a different dataset. Responses are averaged across planes. From top to bottom: raw maps, photocurrent estimates, and inferred weights from the CAVIaR pipeline.
Fig 4.
Validating photocurrent subtraction with synaptic transmission block.
a, We used the grid design to map synaptic connections in a two part experiment. During the control block, we recorded photocurrents along with synaptic currents as in prior experiments. b, Raw and subtracted traces from two ROIs during the control block. ROI 1 is far from the putative location of the postsynaptic cell, while ROI 2 is nearby. c, Raw and subtracted grid maps from the control block. The colorbar for the “control raw” map has been truncated to improve legibility. ROIs correspond to the traces shown in b. d, After bath application of NBQX, we repeated the mapping experiment on the same patched cell, recording isolated photocurrents. e, Raw and subtracted traces from the NBQX block for the same two ROIs as in b,c. As expected, nearly all currents are removed from ROI 2. f, Raw and subtracted grid maps from the NBQX block. As expected, the subtracted maps during the NBQX block appear blank. ROIs correspond to the traces shown in e.
Fig 5.
Validating photocurrent subtraction with simulated mapping experiments.
a, We simulated complete circuit mapping experiments in which stimulating some cells (green) evoked direct photocurrent. We then used a connectivity inference pipeline to assess how photocurrent subtraction affected connectivity inference. b, Scatterplots of inferred vs. true weights before (left panel) and after (center panel) photocurrent subtraction. Vertical axis shows estimated weight, horizontal axis shows true weight. After subtraction, the photocurrent-inducing cells (green) are pushed closer to the identity line, while the non-photocurrent-inducing connections (purple) are largely unaffected. c, False positive rate as a function of the number of cells which induce direct photocurrents before (orange) and after (blue) subtraction. d, Traces from an example false positive in which residual current after subtraction caused us to falsely infer a connection. This typically occured due to residual photocurrent from prior trials. e, Schematic of low-latency PSCs, which can appear as scaled copies of one another when they are driven reliably. f, Example of over-subtraction that can occur in rare instances of reliable low-latency PSCs. g, We swept the minimum PSC latency used in a simulated experiment, and found that false negatives caused by PhoRC were most prevalent at very low latencies. However, even in this case they occurred rarely.
Fig 6.
Validating photocurrent subtraction via simulated-plus-real “hybrid” datasets.
a, Using a sparsely expressing preparation, we mapped synaptic connectivity using the grid design. We then added simulated photocurrents to these recorded PSCs in order to validate the low-rank model. b Ground truth PSC map, observed map containing real PSCs and simulated photocurrents, and map after subtraction. Bottom right shows the true vs. estimated responses, before and after applying PhoRC. c, Example traces corresponding to the maps shown in c. Subtracted PSCs closely mirror ground-truth. d,e Same as b,c for a second hybrid dataset.
Fig 7.
PhoRC allows ensemble mapping in the presence of photocurrent artifacts.
a, Cartoon depicting the photocurrent contamination problem with ensemble grid stimulation. Red circle denotes region where stimulation will evoke direct photocurrent. If one grid point in an ensemble evokes photocurrent, the entire measurement will be corrupted. b, Inferred grid connections before and after photocurrent subtraction. ROI denotes location of a putative connection. c, Observed and subtracted traces from the ROI selected in b. Traces from this ROI contain direct photocurrents whenever a different target in the ensemble evokes direct photocurrent. d, Inferred waveforms for each putative connection before and after subtraction. e-h, Same as a-d but for the targeted mapping design. e, Cartoon depicting the photocurrent contamination problem with ensemble targeted stimulation. As in a, red circle denotes region where stimulation evokes direct photocurrent. f, Schematic of inferred connectivity before and after subtracting photocurrents, Z projection. ROI corresponds to a region far from the postsynaptic cell (green cross), but photocurrents are still present due to the use of ensemble mapping. g, Observed and subtracted currents from the ROI shown in f. h, Inferred waveforms for each putative connection before and after subtraction.