Oscillatory integration windows in neurons - PubMed (original) (raw)
Oscillatory integration windows in neurons
Nitin Gupta et al. Nat Commun. 2016.
Abstract
Oscillatory synchrony among neurons occurs in many species and brain areas, and has been proposed to help neural circuits process information. One hypothesis states that oscillatory input creates cyclic integration windows: specific times in each oscillatory cycle when postsynaptic neurons become especially responsive to inputs. With paired local field potential (LFP) and intracellular recordings and controlled stimulus manipulations we directly test this idea in the locust olfactory system. We find that inputs arriving in Kenyon cells (KCs) sum most effectively in a preferred window of the oscillation cycle. With a computational model, we show that the non-uniform structure of noise in the membrane potential helps mediate this process. Further experiments performed in vivo demonstrate that integration windows can form in the absence of inhibition and at a broad range of oscillation frequencies. Our results reveal how a fundamental coincidence-detection mechanism in a neural circuit functions to decode temporally organized spiking.
Figures
Figure 1. Experiment design.
(a) Recording configuration: In locusts, oscillating waves of odour-evoked spikes are carried by projection neurons from the antennal lobe to Kenyon cells (KCs). A blunt electrode in the mushroom body monitored the odour-evoked oscillatory local field potential (LFP Rec) while an intracellular electrode in a KC (KC Rec) recorded membrane potential and injected pulses of current (_I_inj). (b) Experiment strategy: By injecting pairs of current pulses into KCs (upper and lower panels show two such pairs) we tested the effect of phase position on input summation. (c) Example experiment from _n_=16 KCs: LFP oscillations (purple, top) were induced by a 2-s odour presentation (black, bottom). The second trace (blue) shows a simultaneous intracellular recording from a KC. In each trial, the KC received 19 pairs of intracellular current pulses (_I_inj, red) with the following parameters: pulse-width=5 ms; inter-pulse interval=25 ms (onset-to-onset); inter-pair interval=125 ms (onset-to-onset). Each neuron received at least 20 trials. Inset: A 210-ms period spanning the 10th and the 11th pulse pairs. Although current injections caused large artefacts in the recording, spikes (marked with asterisks) could be identified between or superimposed with the stereotyped artefacts.
Figure 2. KCs have cyclic integration windows.
(a) Average membrane potential plotted as a function of the LFP phase (ϕ) shows robust subthreshold oscillations (_V_osc) with upward deflections in the first half of the cycle; _n_=16 KCs, error bars: s.e.m. (b) Residual depolarization in different time intervals after the onset of the first pulse (the pulse ends at 5 ms). (c) Residual depolarization in the 20–25 ms interval after the first pulse (that is, just before the onset of the second pulse) did not depend on the phase of the oscillation cycle. (d) Spiking responses (mean±s.e.m.) evoked by the first and the second current pulses, _R_1(ϕ) and _R_2(ϕ), respectively, are most likely to occur in the first half of the cycle. (e) Phase-dependent summation is measured by comparing responses to the first pulse (_R_1(ϕ)) and the second pulse (_R_2(ϕ)) occurring at the same phase (not to the two pulses in the same pair). The effect of summation, _R_2(ϕ)−_R_1(ϕ), is most effective in the first half of the cycle. (f) To test possible contributions of after-hyperpolarizations to phase-dependent summation, _R_2(ϕ)−_R_1(ϕ) was computed after excluding responses to pairs of pulses in which the first pulse evoked a spike. This data set lacking after-hyperpolarizations also showed robust phase-dependent summation. All error bars represent s.e.m.
Figure 3. Computational model shows noise contributes to phase-dependent summation.
(a) Top: representative membrane potential trace from a model KC. The membrane potential (V) has three independent components: subthreshold oscillations (_V_osc) generated by the integration of oscillatory input from PNs; noise and a pair of current pulses (_I_inj) injected into the cell. In this example, the second current pulse elicited a spike. Oscillations in both LFP and _V_osc are caused by the same oscillatory input from PNs, but are delayed in _V_osc (see Methods). (b) The model's average membrane potential as a function of the LFP phase (ϕ). (c) _R_1(ϕ) and _R_2(ϕ) show a phase-dependency similar to _V_osc. The effect of summation, _R_2(ϕ)−_R_1(ϕ), also shows a similar phase-dependency. (d) Blue lines: correlation between _R_2–_R_1 and subthreshold oscillations was robust to variations in model parameters, such as the amplitude of the injected current pulses (_I_inj), the amplitude of the subthreshold oscillations, and the amplitude of noise. Dots: parameter values used in simulations shown in a-c. Dashed line: correlations larger than this value (0.576) are statistically significant (P<0.05). Red line: fraction of phase bins (of 12) in which _R_1(_ϕ_)>0. (e) A graphical description of phase-dependent summation in the presence of noise. The membrane potential oscillates (V_osc) when given cyclic input. Noise causes the actual voltage V at a given phase to vary from the mean. The probability distribution of V at any phase must be approximately Gaussian because of the accumulation of noise. The area of this distribution above the spiking threshold equals the probability of spiking at that phase. If the cell receives two current pulses, the distribution of V during the second pulse shifts upward by Δ_V relative to the distribution of V during the first pulse. At the phase where V_osc is lowest (left), shifting the distribution by Δ_V produces a small change in area above the curve (_R_2−_R_1, shown in checkerboard pattern). At the phase where V_osc is highest (right), the same shift Δ_V produces a larger change in the area (compare the two checkerboard areas). Thus, owing to the nonlinear shape of the noise distribution, the first pulse makes a larger contribution to the second pulse when the second pulse occurs at phases close to the peak of _V_osc. (f) If the noise distribution is made uniform (see Methods), _R_2−_R_1 loses its phase-dependence and correlation with subthreshold oscillations.
Figure 4. Integration windows can form in the absence of inhibition.
(a) A representative experiment from _n_=10 KCs, and (b) the experiment design are shown. Purely excitatory oscillatory waveforms (_I_osc), and 25 pairs of 5-ms pulses (_I_inj), were injected directly into the KC through the intracellular electrode, without eliciting network inhibition. Membrane potential oscillations were induced by _I_osc. Inset: the mean membrane potential (20 trials) of the indicated segment. (c) Current injections evoked robust oscillations in the subthreshold membrane potential of the KC. Lacking inhibition, KCs still formed integration windows: _R_2−_R_1 showed the same phase-dependence as seen previously with odour stimulation (see Fig. 2). Error bars represent s.e.m.
Figure 5. KCs form integration windows robustly even at frequencies other than 20 Hz.
(a) The computational model predicted phase-dependent summation for a wide range of oscillation cycle periods (this panel uses the same format as Fig. 3d). (b) In vivo test: experiment design is similar to the one shown in Fig. 4: Oscillatory input with an arbitrary frequency of 11.9 Hz (84-ms period) was injected into the KC along with pairs of current pulses. Bottom: membrane potential oscillations generated by the current in a representative neuron (first 500 ms of current injection is shown), from _n_=10 KCs. (c) Current injections at 11.9 Hz generated robust oscillations in the subthreshold membrane potential of the KC. KCs exhibited integration windows: _R_2−_R_1 showed the same phase-dependence as seen previously with 20-Hz oscillations (Fig. 2). Error bars represent s.e.m.
Figure 6. Integration windows are robust to variations in the inter-stimulus interval.
(a) The computational model predicted phase-dependent summation for a range of intervals between the two pulse inputs. (b) In vivo test: the experiment shown in Fig. 1, including odour-elicited oscillations, was repeated with two different inter-pulse intervals (IPIs); LFP electrode is not shown. (c) For 15-ms IPI, subthreshold oscillations and the effect of summation are phase-locked and correlated (_n_=18 KCs). (d) The same result was obtained for 35-ms IPI (_n_=13 KCs). Error bars represent s.e.m.
References
- Buzsáki G. Rhythms of the Brain Oxford University Press (2006).
- Kay L. M. Circuit oscillations in odor perception and memory. Prog. Brain Res. 208, 223–251 (2014). -PubMed
- Stopfer M., Bhagavan S., Smith B. H. & Laurent G. Impaired odour discrimination on desynchronization of odour-encoding neural assemblies. Nature 390, 70–74 (1997). -PubMed
Publication types
MeSH terms
LinkOut - more resources
Full Text Sources
Other Literature Sources