Predicting Stimulus-Lockedsingle Unit Spiking From

Predicting Stimulus-LockedSingle Unit Spiking from

Cortical Local Field Potentials

Supplemental Figures

Edgar E. Galindo-Leon1 and Robert C. Liu1,2

1 Department of Biology, EmoryUniversity, 1510 Clifton Road NE, Atlanta, GA30322

2 Center for Behavioral Neuroscience, GeorgiaStateUniversity, PO Box 3966, Atlanta, GA30302

Figure S1.

(see caption next page)

Figure S1. Spline-interpolated “despiking” does not introduce artifacts.

(A) Despiking procedure. Example of “raw LFP” signal (LFP1597)with spike-contamination (blue continuous). The extracellular signal was sampled at a rate of24414.0625 samples/s and online band pass filtered for recording of LFP ([2 to 1000 Hz] with a notch filter at 60 Hz). The co-recorded single unit SU1412 fired a spike at 418.7 ms (black vertical hash), whose waveform can be noticed in the rawLFP. ”Despiking” consisted ofremoving the [-0.5 4] ms window around the spike onset (cyan box) and replacing itwitha spline-interpolation (bluedashed). The same procedure was applied around a random spike time as well (magenta vertical hash and gray box). The offline pre-processing of the LFP signal to create the “original LFP” (red continuous)consisted of decimation (order 24) followed by low- pass (firpm, Parks-McClellan optimal equiripple FIR filter, transition band between 90 and 100 Hz) forward and backward filtering (filtfilt). By eye, this procedure, whether performed on a real spike or a fake spike, appeared to eliminate the high frequency leakage of the spike into the 100 Hz band.

(B)No artifacts from despiking in the spike-triggered spectrogram. Spike-triggered spectrogramsaround real spike times inthe rawLFP signal (B1) andthe originalLFP(B2) for another example site (SU1365; LFP1542). In this case, the spike leakage was large enough to create visible contamination. This isobserved as a “nose” at t=0 in (B1). Despiking to create the original LFP seems to completely remove this artifact without introducing additional contamination. To confirm this, the “despiking” algorithm was performed at random times as well (B3), as explained in (A), and produced a “spike”-triggered spectrogram similar to the real despiked version (B2).

(C) No artifacts from despiking in the STPDT. The spike-triggered probability distribution of trajectories (STPDT) for the site in (A)is depicted for the original LFP representation(C1)and for the phase representation(C3). Hot colors represent high probability. The same for random “spike” times as in (A) are shown in (C2) and (C4). Thesepanels shows that “despiking”per se does not introduce any artifact in the probability distributions, which would show up as some spike-locked structure in (C2) or (C4).

(D) Backdrifts (points of negative phase increment from one time point to another) are not systematically associated with spikes or despiking. The panels show thereal (D1) and random (D2)spike-triggered distribution of the first derivative of the phase trajectoriesfor the site in (A). Both are for despiked signals, and show no time-dependent signatures locked to the “spike” time. Both also have minimal proportion of negative phase derivatives (below dashed black line). This suggests that spikes are not systematically associated with backdrifts, whether created by our despiking or not. Furthermore, compared to a non-despiked signal at random times (D3), there is again no difference, indicating that the “despiking” method itself does not introduce additional backdrifts systematically.

Figure S2.

LFP-SU interaction in , , and - bands (top, middle and bottom rows, respectively). (A) Hilbert phase representations of a single trial (same trial as in Figures 1 and 2). Panels in columns (B) and (D)show the STPDTs of two different cells (SU1381 and SU1356, respectively). The respective phase-locking distributions are shown in columns (C) and (E).