Detection of interneuronal ripple oscillations without filtering artefacts using the

Band-pass filtering is generally used for demonstrating oscillations in different frequency bands (theta, gamma, and ripple, etc.). For example, network ripple oscillations are typically presented by the use of band-pass filtering in the 100-300 Hz frequency range. However, according to the Fourier transformation theory, band-pass filtering can artificially generate ripple oscillations from short non-oscillating depolarizing humps, which last for a time period of roughly one SPW event (Figure 15A and 15B).

SPW-EPSP events recorded in FS-PV INs (n=62) were separated into three groups based on the amplitude of the interneuronal ripple oscillations, which were then averaged (no-oscillation, n=10; medium-oscillation, n=22 and high-oscillation groups, n=30) (Figure 15A). Instead of aligning the EPSPs to their peaks, transients were

“randomly” and temporally shifted to eliminate phase synchrony of interneuronal fast-ripple oscillations before averaging. This random temporal shift was achieved by aligning all the transients to arbitrary threshold amplitude (7 mV, thr.) (Figure 15A).

After baseline subtraction, the group of highly-oscillating SPW-EPSPs showed no oscillations and was similar to the non-oscillating group (Figure 15B left, red and black, respectively). These data, therefore, show that the random temporal shift in 15D effectively eliminated phase synchrony and significantly reduced oscillations in the average trace. In contrast, when a band-pass filter (Chebyshev, 100-300 Hz) was applied instead of baseline subtraction, it generated large oscillations in the ripple

frequency range in each of the three groups of SPW-EPSPs, irrespective of whether the SPW-EPSP belonged to the no-oscillation, medium-oscillation, or high-oscillation group (Figure 15B, right). Moreover, band-pass filtering over-estimates oscillation amplitudes in the subgroup of SPW-EPSPs which have no, or very low, oscillations (Figures 15C-E). When Chebysev band-pass filtering was used on the same EPSP groups, the separation between the three groups was less evident (Figure 15D). In addition, band-pass filtering can generate extra oscillation cycles and artificial phase shifts when the biological oscillations are irregular, e.g. SPW-R-associated interneuronal ripple oscillations (Figures 16B-C).

Off-line Gaussian filtering (100 Hz) was used on the raw whole-cell traces to generate a “baseline”, which was then subtracted from the raw traces to preserve ripple oscillations (referred to as baseline-subtraction method, Figure 16A). In contrast to band-pass filtering, the baseline-subtraction method did not generate any extra oscillation cycles, phase shifts (Figures 16A-C), or false oscillations from the relatively slow depolarizing humps which characterize SPW-associated EPSPs (Figures 15A-E).

Therefore, the baseline-subtraction allowed us to detect individual ripple oscillation cycles in the somatic membrane voltage with the required phase and amplitude precision (Figure 16A).

Figure 15. Band-pass filtering versus baseline subtraction methods in the detection of interneuronal ripple oscillations. A: Three groups of randomly and temporally shifted SPW-EPSPs eliminate phase synchrony of interneuronal fast-ripple oscillations before averaging (no-oscillation, n=10; medium-(no-oscillation, n=22 and high-oscillation groups, n=30). B, Left: The same three groups of SPW-EPSPs as in A, but after baseline subtraction. Right: SPW-EPSPs after band-pass filter (Chebyshev, 100-300 Hz) was applied C: The same three groups of EPSPs as in A following baseline subtraction but without the random shift. D: Same as in C, but Chebysev band-pass filtering was used on the same EPSP groups instead of baseline subtraction E: Overlay of the non-oscillating groups in C and D.

A

Figure 16. Comparison of the baseline subtraction method and the generally used band-pass filtering for the detection of individual oscillatory cycles during interneuronal ripple oscillations in somatic membrane potential. A: A somatic EPSP with interneuronal ripple oscillations was recorded from an FS-PV IN during an SPW event (red). The high frequency component was removed by Gaussian filtering (cut-off frequency 100 Hz), which yielded a smooth baseline trace with a hump (gray). Oscillations could be seen (black) after subtracting this Gaussian-filtered baseline trace (gray) from the original trace (red). The peak of the SPW-EPSP is indicated with a triangle. Dashed blue lines indicate individual oscillatory events which could be reliably detected by this method. B: Comparison of the baseline subtraction method (black) and band-pass filtering (red). Black trace is the same as in A. Note that the use of a Chebyshev band-pass filter (100-300 Hz) resulted in extra oscillation peaks (red arrows), and also shifted the phase of responses (blue asterisks). C: Six further examples of baseline-subtracted and band-pass filtered SPW-EPSPs. Similarly to A, all oscillatory events in black traces were also visible on the raw traces (data not shown), but the Chebyshev band-pass filter generated undesired extra oscillatory events. In the enlarged image, the red arrows indicate these events.

derivative is too high. These artefacts could be observed in two cases in our measurements, but in both cases the impact of the artefact on the detection of ripple oscillations could easily be excluded during analysis. In the first case, when EPSPs start with a rapid increase following two-photon glutamate uncaging, the baseline-subtraction method generates a two-phase artefact limited to the initiation phase of the EPSPs. However, interneuronal ripple oscillations occur a few milliseconds after this period, and they could therefore be isolated temporally (see for example Figure 35A). I have marked this two-phase artefact with blue dotted boxes in figures. In the second case, when an AP was rising on uncaging or SPW-associated subthreshold depolarizations, the baseline had to be locally corrected before baseline subtraction to avoid generation of an artefact before and after the AP. Therefore, APs were replaced with spline interpolated curves before the baseline was smoothed and substracted.

Traces of the interpolated periods were marked with dashed lines in the figures (see Figures 29B and 36A).

The ripple frequency of individual traces and averages was determined either by Fourier transformation (fFourier) or by measuring the time between oscillation peaks after baseline subtraction in an interval centered on the peak of raw events (fmax).