How synaptic input location shapes frequency modulation in a single neuron

[MakotoMiyakoshi]

The bottom-right plots show time-domain responses, so you have to Fourier-transform them in your mind; narrower the peak, flatter the PSD.

Rall W, Burke RE, Smith TG, Nelson PG, Frank K. 1967. Dendritic location of synapses and possible mechanisms for the monosynaptic EPSP in motoneurons. J Neurophysiol. 30:1169-1193.
DOI: 10.1152/jn.1967.30.5.1169, PMID: 4293410

Lindén H, Pettersen KH, Einevoll GT. 2010. Intrinsic dendritic filtering gives low-pass power spectra of local field potentials. J Comput Neurosci. 29:423-444.
DOI: 10.1007/s10827-010-0245-4, PMID: 20502952

[mgy42]

This is beautiful stuff, thank you for sharing, Makoto. I think we could certainly use this information to convincingly argue that a synaptic distance mechanism also contributes to the 1/f shape (potentially in addition to E:I balance, i.e., that we can see 1/f changes without changes in the relative contribution of EPSPs/IPSPs, just by changing the distance of E inputs from the soma). Now I will just have to double check how this differs from the dendritic filtering idea which has already been floated by others.

[MakotoMiyakoshi]

@mgy42 Yes, that figure from Buzsaki's classical paper (which can further date back actually) shows it too. So if you have theory-laden eyes, you see it!

@amisepa The cellular differences is another layer of the problem. We need to rely on quantitative approach to determine what is the dominant factor. This is a typical spatial scale problem too--we do not know, for example, how the mesoscopic-level anatomy and physiology scale to our scalp EEG. That's why Voytek's FOOOF was an exceptional success.

If you see, for example, Stephanie Jones' Human Neocortical Neurosolver, they model only basket cells as inhibitory. Interestingly, their model has 'proximal' and 'distal' for both excitatory and inhibitory. I know the developers, so if you want I can ask them for their opinion about the difference between distal and proximal basket cells.

But aren't they overkill for scalp EEG anyway? Do you have any particular reason why these mechanisms matters to your model or hypothesis?

[amisepa]

@MakotoMiyakoshi

Since the goal of our project is scalp EEG level, I see your cable theory result as identifying synaptic distance as an important variable, but not yet explaining how it would be controlled or expressed at the population level.

In the figures you and Mate discussed here, the spatial location of input is effectively imposed from the outside by stimulation. In ongoing brain activity, my intuition is that this role is played internally by inhibitory circuits. Interneurons do not inject current of course, but they regulate where along the dendritic tree excitation can effectively contribute, thereby shifting the dominant synaptic distance that shapes dendritic filtering.

Because distal EPSPs and IPSPs are more strongly lowpass filtered than proximal ones, changing where excitation is permitted along the dendrites should bias the 1/f slope of the summed extracellular field. At the scalp, this would not reflect single cells, but a population level shift in which dendritic compartments dominate the signal.

Whether feedforward versus feedback processing truly maps onto proximal versus distal inhibition would need to be validated by experts in inhibitory circuitry, and the hypothesis should be adjusted if that mapping is wrong.

But this is why I suggested a simple within-subject experiment: if tasks engage different inhibitory control regimes, dendritic filtering predicts systematic changes in the 1/f slope measurable at the scalp, even though all cellular details are heavily averaged.

[mgy42]

Just wanted to add that I like Cedric's summary here. In general, my question still remains whether we can think of any within-subject manipulation where the dendritic compartment dominating the signal changes systematically (so that it might be visible in scalp level data too). It would be an added bonus if the E:I balance view of aperiodic activity predicted a different change for that particular manipulation that the dendritic filtring idea (so we had competing hypotheses), but based on Cedric's summary, this might not be a realistic expectation...!

[MakotoMiyakoshi]

Hi @amisepa and @mgy42,

If we want to test the distance-dependent low-pass filter effect of the dendritic filter, we first need to find a good model. I would not start with a general healthy adult resting EEG.

For example, we know developing brains do not have myelinations in axons. Then we can immediately predict that those small kids should show steeper 1/f curve because their axons are leaky and low-pass filtered. In fact, this prediction was supported here

I know lemniscal and extralemniscal thalamo-cortical projection target different cortical layers: the former goes to Layere III, while the latter goes to Layer I, III, IV, and VI. Maybe we can take advantage of this cortical laminar structure to 'bias', if not exactly pin-pointing, the synaptic input locations relative to soma. I'm not super knowledgable about the tasks, but there must be a good one if you are serious to do it.

My point is not necessarily in the distance dependency, but the fact that dendrites and unmyelinated axons show general low-pass filter effect. Experimental designs may be optimized so that we can maximize experimental effect. At very least, I can show that E/I balance is not everything when we talk about 1/f-ness.

[MakotoMiyakoshi]

One important addition: the cable theory-driven low-pass filter effect can be correlated with amplitude modulation: distal input means not only more low-pass filter effect but also larger pole distance i.e., larger dipole moment. (EFB p. 368) Thus, it is predicted that more low-pass filter effect is associated with larger amplitude generation.

[MarjanZamani1990]

Periodic and Aperiodic Components as Emergent Signatures of Synaptic and Spiking Dynamics

In my view, neural field potentials measured by EEG arise from the interaction of two fundamentally distinct but dynamically coupled electrophysiological processes: subthreshold synaptic activity and suprathreshold action potentials. While classical EEG theory has traditionally emphasized the dominance of synaptic potentials due to their spatial summation and slow temporal dynamics, this perspective can be understood as a methodological approximation rather than an absolute biophysical constraint.

Subthreshold post-synaptic potentials (EPSPs/IPSPs) are distributed along dendritic trees and are strongly shaped by dendritic cable properties. Dendrites appear to act as distance-dependent low-pass filters: synaptic inputs proximal to the soma tend to retain relatively higher-frequency components and narrower temporal profiles, whereas distal inputs are more likely to undergo temporal broadening, attenuation of fast components, and a reduction in effective bandwidth. The aggregation of synaptic activity across spatial scales and heterogeneous time constants may naturally give rise to a scale-free, aperiodic (1/f-like) spectral background.

Action potentials, in contrast, are actively regenerated, nonlinear events with intrinsically sharp temporal profiles and therefore extremely broad spectral bandwidth. From a biophysical standpoint, they can be considered among the highest-bandwidth signals in neural tissue. Their limited visibility in scalp EEG is generally attributed not to an absence of signal generation, but to macroscopic constraints such as spatial cancellation of oppositely oriented dipoles, temporal jitter across neuronal populations, and the effects of volume conduction.

Importantly, the common assertion that action potentials “cancel out” at the scalp implicitly assumes averaging over temporal windows that are long relative to spike duration (on the order of ~1 ms). When neural activity is examined within sufficiently short temporal windows, complete cancellation may not be guaranteed. Under conditions of partial synchrony, spatial clustering, or reduced broadband contamination, action-potential-related activity could plausibly leave a measurable residual field—not necessarily as a time-locked ERP waveform, but rather as a statistical or spectral signature.

Accordingly, action potentials may contribute indirectly to EEG through increases in high-frequency broadband power, modulation of the aperiodic spectral slope, and transient enhancements of signal bandwidth. Within this framework, empirical observations of increased gamma-band activity following ICA-based artifact removal can be interpreted mechanistically: ICA does not generate neural gamma activity, but may unmask localized, high-bandwidth neural signals that were previously obscured by artifacts or spatial mixing.

From this perspective, periodic oscillations and aperiodic components should not be viewed as mutually exclusive entities, but rather as complementary projections of underlying neural dynamics. Rhythmic activity likely reflects locally synchronized circuit behavior with preserved temporal structure, whereas the aperiodic component may index more global network states, shaped jointly by dendritic filtering, synaptic integration, excitation–inhibition balance, and the collective, imperfectly cancelling activity of action potentials.

Although individual action potentials are unlikely to be directly observable at the scalp, their broadband nature suggests that their influence is not entirely eliminated, but rather redistributed—potentially leaving an indirect yet meaningful imprint on the EEG signal.

[EugenMasherov]

It seems to me that action potentials can be observed in the EEG under certain conditions. Synchronized in time, they form spikes or sharp waves. Another situation in which their contribution to the EEG can be seen is the breach effect, high-frequency activity at the site of a cranial defect.
And I believe that the third component, ionic fluctuations, cannot be neglected. Unlike dipole EPSPs and IPSPs and quadrupole action potentials, they create a monopole source. Its amplitude is small, but the potential of a monopole decays more slowly than that of a dipole or quadrupole. And all three of these sources not only contribute to the overall EEG but also interact with each other nonlinearly.

[MakotoMiyakoshi]

Hi @Marjanz69 and @EugenMasherov,

Murakami and Okada (2006) as well as Reimann et al. (2013) indicate that action potentials make nontrivial contribution to EEG and MEG. However, traditionally this view has not been a main stream; Electric Fields of the Brain (2006) dismisses contribution of action potentials and counts only stationary, subthreshold membrane potential changes due to synaptic inputs as source of EEG signals.

Eugen shared a potential observation of quadrupole in scalp EEG of an epilepsy patient. Standard textbooks do not necessarily cover all observations we make, I guess.