Abstract
Octopus cells in the posteroventral cochlear nucleus exhibit characteristic onset responses to broad band transients but are little investigated in response to more complex sound stimuli. In this paper, we propose a phenomenological, but biophysically motivated, modeling approach that allows to simulate responses of large populations of octopus cells to arbitrary sound pressure waves. The model depends on only few parameters and reproduces basic physiological characteristics like onset firing and phase locking to amplitude modulations. Simulated responses to speech stimuli suggest that octopus cells are particularly sensitive to high-frequency transients in natural sounds and their sustained firing to phonemes provides a population code for sound level.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
The auditory brainstem consists of multiple afferent pathways that process different features of sound. Besides the spectral pattern and the temporal fine structure of a sound, it is particularly transients and amplitude modulations (AM), i.e., fluctuations in sound intensity on an intermediate time scale on the order of 10 ms, that provide most information about the identity of a natural sound stimulus [9, 12]. Octopus cells of the posterior-ventral cochlear nucleus (PVCN) are generally thought to encode such amplitude modulations of high-frequency sound stimuli by means of their temporal spike patterns [17, 19]. They are thus likely to play a central role in the processing of natural sounds, including conspecific vocalizations [14, 15, 18].
Octopus cell spikes only occur at the onset of broad band transients [3,4,5,6, 15, 21] but phase-lock persistently to amplitude modulations in a specific AM frequency band. Mechanistically, this firing behavior is thought to arise from integrating across auditory nerve fibers (ANFs) [10, 13, 16] with a broad range of characteristic frequencies [4, 20, 24]. This suggests that the main computation underlying AM extraction is most likely based on the tonotopic pattern of afferent arborization. In addition, octopus cells have remarkably low input resistances of only few Mega Ohms [5, 15] leading to fast enough membrane time constants for processing of fast transient as well as slow amplitude fluctuations. The short membrane time constants are generated by a high density of low-threshold potassium channels [23], which in addition to reducing integration time also endow the neurons with differentiation properties [1, 22] that further facilitate AM locking.
Computational theories of octopus cell function thus require to analyze the interplay between cellular biophysical properties and the circuit parameters describing ANF population inputs. Here, we propose an efficient phenomenological model for octopus cell spiking with only few parameters that are either constrained by direct physiological measurements or functional properties. We find that octopus cell spiking over a wide range of best frequencies can be robustly explained by only small changes in these parameters. Our model thus provides a computationally efficient and robust tool to simulate octopus cell spike responses to any kind of sound stimulus. The model can therefore be used to emulate population inputs to downstream structures in the auditory pathway, the ventral nucleus of the lateral lemniscus and the inferior colliculus.
2 Model
The general structure of the proposed effective model is outlined in Fig. 1. In short, the sound stimulus is translated to simulated ANF firing rates \(r_i(t)\), where i labels the respective frequency channel. The ANF rates are then translated into the octopus cell input by a weighted sum over frequency channels with weight factors \(g_i\). The cellular membrane potential is derived from these inputs by a combination of differentiation and low-pass filtering. Finally, the output rate R(t) of the octopus cell is obtained by a sigmoidal transformation of the pseudo potential P(t). Spike trains can subsequently be obtained by using R(t) as the density of an inhomogeneous Poisson process. All individual transformations will be explained in detail in the following paragraphs.
2.1 Arborization
By integrating over ANFs with multiple characteristic frequencies the information about the stimulus’ fine structure is removed, whereas envelope information is preserved. The model generates the cochlear output of an array of ANFs using the (Zilany–Bruce–Carney) model described in [8, 26, 27] with parameters tuned to cat physiology. The model runs with the numerical sampling frequency of 100 kHz. For a model octopus cell with characteristic frequency \(f_c\), we simulate 9 frequency channels (corresponding to characteristic cochlear locations) that are logarithmically spaced in the two octave intervals \(f_c/2\) to \(2\, f_c\). For each ANF frequency, we simulate high, medium and low spontaneous rate fibers and linearly combine them with the fractions, 0.16, 0.24, and 0.6, respectively [11]. The resulting ANF firing probability densities \(r_{i}(t)\) are linearly combined to the octopus cell’s input current
The weights \(g_i\) are obtained from a log-normal function around the octopus cell’s characteristic frequency (CF, here denoted as \(f_{c}\)):
For high frequency cells it is necessary to introduce an additional frequency shift \(f_{0}\) to properly fit the observed characteristic frequency, compensating for the overlap of peripheral filters. The parameter \(\varDelta \) describes the width (in octaves) of the arborization and will be the essential fit parameter to model the afferent arborization.
Receptive fields of model octopus cells using the plain periphery model either showed unphysiologically strong low-frequency tails [compared to data from [24]], even for cells with high characteristic frequencies, or amplitude modulation locking of the model was distorted by low frequency components from the tails of the ANF receptive fields. For simplicity, we removed these low-frequency components by applying an additional high pass filter
to the sound pressure wave before the periphery model, with \(f_\mathrm{hp}= 450\) Hz, the Heaviside step function \(\varTheta (t)\), and \(\varDelta t=1/(100\,\mathrm{kHz})\). This filter suppresses most of the low-frequency-tail of the receptive fields, while still preserving the general response patterns of the model octopus cells (see Discussion for biological feasibility). For numerical convolution, we restricted the kernel \(k_\mathrm{hp}\) to a duration of \(3/f_\mathrm{hp}\).
2.2 Pseudo potential
The octopus cells respond to rising envelopes of the sound stimuli, which, following [1], we model via a differentiation. The kinetics of the membrane potential response is accounted for by an additional second-order low pass filter
that is supposed to reflect the combination of synaptic and potassium channel kinetics as well as membrane filtering. The frequency \(f_\mathrm{lp}\) is the second fit parameter of the model. This leads to the pseudo potential
which we consider to reflect the behavior of the membrane voltage. The parameter \(d_{a}\) allows to include an additional non-derivative component. If \(d_a\) is large, the cell has a more primary-like response, whereas for small \(d_a\) the response is more onset type. We found that the general dependence on \(d_a\) is weak in that for all values \(2\pi \, d_{a}\lessapprox 700/\text {s}\) the overall response pattern is onset. Cells with higher CF thereby required larger values to ensure sufficient AM entrainment. For all simulations shown in this paper we satisfy these demands by choosing \(2\pi \, d_{a}= \lfloor 50/\mathrm{s} + \frac{f_c-1.3\,\mathrm{kHz}}{10\,\mathrm{Hz}}~\mathrm{1/s}\rfloor _+\).
The magnitude of the impedance profile of the pseudo potential P,
is illustrated in Fig. 2a for the best fitting frequency parameter \(f_\mathrm{lp}=300\) Hz (see Fig. 4 below) and illustrates band bass characteristics with peak frequency at 300 Hz (for \(d_\mathrm{a}=0\)). The time constant \(1/(2\pi \, f_\mathrm{lp}) \approx 0.5\) ms fits into the range of membrane time constants in octopus cells [6] and in the auditory pathway in general [2] further supporting the model design.
2.3 Spike generation
To derive the spike rate R(t) for an inhomogeneous Poisson process from P(t), we use a sigmoidal function
fixing the rate parameter for the maximal firing rate \(R_{\max }\) of the neuron and allowing \(Q=(\frac{R_{\max }}{R_{T}})^{\gamma }-1\) to set the firing rate \(R_{T}\) at the threshold potential T. \(R_{\max }=12\) spikes/ms is set to fit the experimentally reported firing probability density when adding a refractory period of 2 ms [24]. Further on, we set \(R_{T} = 9\) spikes/ms which yields roughly a 10% probability of an onset spike for pure tone presentation which generates a response P equal to T.
There is only little data published on the pure tone threshold levels \(L_0\) of octopus cells [24], however, these appear to be in a range of 30–60 dB while cells with high characteristic frequencies tend to have higher thresholds. We therefore assumed \(L_0\) to grow linearly as
The threshold parameter T is then taken as the peak value of P(t) for a 12 ms pure tone at CF and threshold level \(L_{0}\). The resulting threshold parameters are shown in Fig. 2b.
The slope parameters \(\beta \) and \(\gamma \) of the sigmoid take values of \(2.5/(6\pi )\cdot 10^{-6}\) and 1/2, respectively. This parameter choice generates relatively steep activation functions, which ensure suppression of spontaneous and sustained firing owing to the non-derivative component from Eq. (5). Furthermore, since in model cells with low CF the pseudo potential P(t) fluctuates in only a small interval, the steep slope also ensures that the small dynamic range of P is translated to the full output range of R.
The parameters of the sigmoid not only ensure that the model responds reliably and temporally precise to pure-tone stimuli above threshold. The specific choice of the parameters \(\beta \) and T was also checked to not introduce unphysiological islands in the receptive fields.
2.4 Basic response properties
During early electrophysiological experiments as well as while designing our model, pure tone stimuli are a necessary simplification to probe octopus cell responses in a controlled way (also as compared to AM stimuli) although these stimuli only provide sketches of these cells’ function under realistic acoustic conditions. Nevertheless, pure tones provide a valuable good first benchmark for modelling. We thus confirmed that the model’s responses can fit octopus cells’ localized pure tone receptive fields (Fig. 2c), onset responses to pure tones (Fig. 3 left), and locking to sinusoidally amplitude modulated sounds stimulus (Fig. 3 right). As by design, the model only fires at the onset of the pure tone, whereas it phase-locks to each cycle of the amplitude modulated sound.
3 Mathematical analysis of a simplified model
To better understand the model dynamics, we analyze its linear part P(t) for an amplitude modulated pure tone stimulus
Here, \(\varTheta \) represents the Heaviside function that implements the onset at time \(t=t_0\). The carrier frequency is denoted by \(f_s\). The amplitude A(t) is assumed to vary much more slowly than the carrier and thus can be taken as constant \(A(t) \approx A\) during the period \(1/f_s\).
Due to its combination of low pass and band pass filtering the response of a single ANF to a pure tone stimulus (with frequency \(f_s\)) can be approximated as
with a constant component \(a_{i}\) and the oscillatory component proportional to \(b_{i}\). With this, Eq. (1) for an octopus cell with CF \(f_c\) can be rewritten as
with
and
Due to the filtering properties of the periphery model \(a_{i}\) grows with \(f_{i}\) while \(b_{i}\) decays. Since octopus cells seem to be most consistently modeled with input mostly from high frequency ANFs [25], one can neglect the \(b_{g}\)-component of Eq. (10) and simplify it to:
The pseudo potential P is mostly governed by the derivative component of Eq. (5) [1]. Thus for simplicity, in the next paragraph we will assume that \(d_{a} \approx 0\).
Using the input approximation (13) for a non-modulated pure tone (\(A(t)=A=\mathrm{const.}\)), Eq. (5) simplifies to a delta-like pure tone (PT) response
Conversely, for an amplitude modulated stimulus, the differentiation of the input approximation (13) by Eq. (5) yields an additional additive component extracting the derivative of A(t),
The term \(\propto \frac{\text {d}}{\text {d}t} A(t)\) lets the octopus neurons fire on the positive slope of their inputs. This behavior enhances fast amplitude modulations, particularly at high modulation frequencies (as concluded from a previous octopus cell model in [7]). It thus accounts for phase locking to sinusoidally amplitude modulated tones up to a certain modulation frequency (governed by the second-order low-pass kernel \(k_\mathrm{lp}\)). Therefore, by design, the model can replicate the essential firing characteristics for pure tone and AM stimuli.
What remains open is, how to find the parameters that reliably produce such physiological responses. Since \(R_\mathrm{max}, Q, \beta , T, d_a, \gamma \) are constrained by basic cellular measurements (see above) we next will explore the effect of the two thus far unconstrained parameters \(f_\mathrm{lp}\) and \(\varDelta \) by simulations.
4 Simulations
The above considerations leave only two parameters that are not directly constrained by the desired firing pattern.
-
1.
The width of the arborization \(\varDelta \) (Eq. 1) and
-
2.
the cut-off frequency \(f_\mathrm{lp}\) of the low pass filter Eq. (4).
Both were scanned by simulations shown in Fig. 4 monitoring the quality of three functional criteria for a given parameter configuration.
The first criterion (upper panels) is the mean number of spikes for an 80 dB SPL pure tone stimulus at the octopus cell’s best frequency. A mean spike count of 1 indicates that the cell produces the experimentally reported onset response. A mean spike count of about one is realized across CF by all \(\varDelta \)-values between 0.4 and 1 and \(f_\mathrm{lp}>250\) Hz. The exception are cells with very low CF, which exhibit low levels of sustained firing in general.
A second criterion is the entrainment E, which is defined as the fraction of spikes per modulation cycle of a sinusoidal amplitude-modulated noise stimulus (in this case with a modulation frequency of \(f_{m}=300\) Hz). High values of E are thus desirable. We find generally low values of E at \(\varDelta <0.8\) and at \(f_\mathrm{lp}>350\) Hz, which excludes these parameter regimes from further considerations.
As a last criterion we use the vector strength \(V=\frac{1}{n}|\sum _{j=1}^{n}{\exp ({2\, \pi i f_{m} t_{j}}}|)\), where \(t_{j}\) is the time of the j-th spike and n is the total number of spikes. It indicates the cell’s ability to phase-lock to amplitude modulations. High values of V indicate good phase locking and are preferable for choosing parameters. Low values of V are generally only observable for \(\varDelta <1.0\) in the mid-CF range, and for \(f_\mathrm{lp}<100\) Hz, which would also exclude these parameter regimes.
As a result of these considerations we suggest as a good choice for the model parameters \(f_\mathrm{lp}=300\) Hz and \(\varDelta = 0.9\), where none of the criteria exhibit extensive regions of low unfavorable values along the CF axis.
4.1 Modulation transfer functions
Phase-locking of the model responses to varying AM frequencies are experimentally quantified by modulation transfer functions [17]. We thus also applied AM stimuli with different carrier frequencies to our model (Fig. 5) reproducing experimentally reported results.
Cells show a typical best modulation frequency with maximal firing rate [17] and the best modulation frequency correlates with the characteristic frequency in the tested range as expected from peripheral filtering. All cells phase lock very well to the AM stimuli over a broad frequency range as indicated by vector strength. Above some threshold frequency cells cease to phase lock, and this threshold frequency again correlates with characteristic frequency for the range tested.
4.2 Transients
Octopus cells have been proposed to particularly encode sound transients [5, 15] that occur on a much faster time scale than the amplitude modulations in the stimuli tested so far. The transition from amplitude modulations to transients can be explored by manipulating spectral width of the carrier noise and modulation depth. We first simulated model responses to amplitude modulated noise with increasing bandwidth and full modulation depth and compared them to the responses to a sharp-onset noise-burst stimulation (Fig. 6). For fixed sound level and increasing spectral width, octopus cells gain in response rate (entrainment) and vector strength, surpassing ANF vector strengths. This finding corroborates that octopus cells are best driven by sharp broad band transients.
To also quantify how well octopus cells are able to detect amplitude modulations, we next varied modulation depths (Fig. 7). Phase locking (if present over all CFs) only started to clearly deviate from control levels (zero modulation depth) at at least 60% modulation depth and almost achieved vector strengths obtained with noise-burst stimulations. Conversely, entrainment required at least about 80% to show marked differences to controls but, particularly for high frequencies, remained below entrainment obtained from noise burst responses.
We thus conclude that amplitude modulations generally yield good phase-locking of octopus cells, whereas they do not reach the efficiency of sharp transients in evoking secure responses, particularly for narrow band stimuli and at characteristic frequencies above 5 kHz.
4.3 Complex stimuli
To compare our results with bandpass noise to natural stimuli, we also applied it to two exemplary speech sounds. Figure 8 shows the results for the speech stimuli “‘time”’ and “‘space”’ from a male speaker, presented at 50 dB and 60 dB SPL.
In line with our observations from bandpass noise, the qualitative observation from these plots is that the octopus cells fire selectively on sharp transients, the strong modulations as well as parts of the stimulus with a broad spectrum, as typical for consonants. Cells with different characteristic frequencies thereby select different parts of the stimuli: The pattern is bipartite, in that cells with characteristic frequency above some threshold only show responses to sharp transients. This frequency threshold is raised by increasing stimulus level, and also reflects the recruitment of high frequency ANFs. As a result, the threshold of sustained octopus cell firing to phonemes approximately coincides with the CF at which the stimulus intensity matches the cells’ pure tone threshold according to Eq. (8).
5 Discussion
Owing to the high technical challenges in both in-vivo and in-vitro physiology preparations, octopus cells, despite their putative crucial role in the processing of natural sounds [15, 18], have not been investigated to a similarly large extent as other cells in the ascending auditory brainstem, like, e.g., bushy cells, or principal cells of the MNTB and SOC. We therefore still have a very incomplete picture on how the octopus cell pathway processes sound. Here, we present a modeling approach complementing detailed physiological studies comprising a phenomenological computational model for a population of octopus cells. The model is constrained by the well-studied responses of octopus cells to pure tones and amplitude-modulated noise and implements their basic known physiological operation of a differentiation of its input. Such a phenomenological approach allows us to test functional hypotheses on large sets of natural stimuli and thereby generate new hypotheses for follow-up experimental studies.
The disadvantage of a phenomenological approach is, however, that making connections to the underlying biological substrate is not always straightforward. For example, we do not have a good justification for requiring an additional high-pass filter applied before the established periphery model [8, 26, 27] other than improving of how the model fits the frequency tuning curve and phase locking data. A possible explanation may be additional mechanisms that suppress low frequency inputs to octopus cells. Despite this drawback, our model makes a clear experimentally-testable prediction, viz., the neurons should cease sustained firing to ongoing amplitude modulations of a complex stimulus (such as those evoked by phonemes) if the stimulus intensity (integrated over the whole word) falls roughly below the cell’s threshold to pure tone stimulation at CF, whereas they should only fire to sharp transients if the stimulus intensity is above the cell’s CF pure tone threshold (Fig. 8).
So far, the pathway originating from octopus cells was hard to probe in a functional manner, since the link between acoustic stimulus and synaptic input was unclear. A major benefit of our model is that it allows us to generate physiologically realistic inputs for physiological studies of such downstream neuronal structures, e.g., the ventral nucleus of the lateral lemniscus or the inferior colliculus.
Our simulations so far have already shown that the octopus model responds strongly and selectively to rapid signal onsets in speech stimuli (Figs. 6, 8). This suggests that speech stimuli, with abundant sharp transients and strong modulations, seem to be a suitable test set for eliciting rich activity for octopus cell recordings. Moreover, ethologically relevant sounds, such as conspecific vocalization, rustling, and predator noise all contain transients and broad band contributions and thus octopus cell populations are likely to play a crucial role for audition in natural environments in general.
References
Ferragamo MJ, Oertel D (2002) Octopus cells of the mammalian ventral cochlear nucleus sense the rate of depolarization. J Neurophysiol 87(5):2262–2270
Fischer L, Leibold C, Felmy F (2018) Resonance properties in auditory brainstem neurons. Front Cell Neurosci 12:8
Friauf E, Ostwald J (1988) Divergent projections of physiologically characterized rat ventral cochlear nucleus neurons as shown by intra-axonal injection of horseradish peroxidase. Exp Brain Res 73(2):263–284
Godfrey DA, Kiang NY, Norris BE (1975) Single unit activity in the posteroventral cochlear nucleus of the cat. J Comp Neurol 162(2):247–268
Golding NL, Ferragamo MJ, Oertel D (1999) Role of intrinsic conductances underlying responses to transients in octopus cells of the cochlear nucleus. J Neurosci 19(8):2897–2905
Golding NL, Robertson D, Oertel D (1995) Recordings from slices indicate that octopus cells of the cochlear nucleus detect coincident firing of auditory nerve fibers with temporal precision. J Neurosci 15(4):3138–3153
Hemmert W, Holmberg M, Ramacher U (2005) Temporal sound processing by cochlear nucleus octopus neurons. In: Duch W, Kacprzyk J, Oja E, Zadrozny S (eds) Artificial neural networks: biological inspirations—ICANN 2005. Springer, Berlin, pp 583–588
Ibrahim RA, Bruce IC (2010) Effects of peripheral tuning on the auditory nerve’s representation of speech envelope and temporal fine structure cues. In: Lopez-Poveda EA, Palmer AR, Meddis R (eds) The neurophysiological bases of auditory perception. Springer, New York, pp 429–438
Joris PX, Schreiner CE, Rees A (2004) Neural processing of amplitude-modulated sounds. Physiol Rev 84(2):541–577
Kane EC (1973) Octopus cells in the cochlear nucleus of the cat: heterotypic synapses upon homeotypic neurons. Int J Neurosci 5(6):251–279
Liberman MC (1978) Auditory-nerve response from cats raised in a low-noise chamber. J Acoust Soc Am 63(2):442–455
Machens CK, Wehr MS, Zador AM (2004) Linearity of cortical receptive fields measured with natural sounds. J Neurosci 24(5):1089–1100
McGinley MJ, Liberman MC, Bal R, Oertel D (2012) Generating synchrony from the asynchronous: compensation for cochlear traveling wave delays by the dendrites of individual brainstem neurons. J Neurosci 32(27):9301–9311
Oertel D (2005) Importance of timing for understanding speech. Focus on “perceptual consequences of disrupted auditory nerve activity”. J Neurophysiol 93(6):3044–3045
Oertel D, Bal R, Gardner SM, Smith PH, Joris PX (2000) Detection of synchrony in the activity of auditory nerve fibers by octopus cells of the mammalian cochlear nucleus. Proc Natl Acad Sci U S A 97(22):11773–11779
Oertel D, Cao XJ, Ison JR, Allen PD (2017) Cellular computations underlying detection of gaps in sounds and lateralizing sound sources. Trends Neurosci 40(10):613–624
Rhode WS (1994) Temporal coding of 200% amplitude modulated signals in the ventral cochlear nucleus of cat. Hear Res 77(1–2):43–68
Rhode WS (1998) Neural encoding of single-formant stimuli in the ventral cochlear nucleus of the chinchilla. Hear Res 117(1–2):39–56
Rhode WS, Greenberg S (1994) Encoding of amplitude modulation in the cochlear nucleus of the cat. J Neurophysiol 71(5):1797–1825
Rhode WS, Smith PH (1986) Encoding timing and intensity in the ventral cochlear nucleus of the cat. J Neurophysiol 56(2):261–286
Rhode WS, Smith PH, Oertel D (1983) Physiological response properties of cells labeled intracellularly with horseradish peroxidase in cat dorsal cochlear nucleus. J Comp Neurol 213(4):426–447
Rothman JS, Manis PB (2003) Differential expression of three distinct potassium currents in the ventral cochlear nucleus. J Neurophysiol 89(6):3070–3082
Rothman JS, Manis PB (2003) The roles potassium currents play in regulating the electrical activity of ventral cochlear nucleus neurons. J Neurophysiol 89(6):3097–3113
Smith PH, Massie A, Joris PX (2005) Acoustic stria: anatomy of physiologically characterized cells and their axonal projection patterns. J Comp Neurol 482(4):349–371
Spencer MJ, Grayden DB, Bruce IC, Meffin H, Burkitt AN (2012) An investigation of dendritic delay in octopus cells of the mammalian cochlear nucleus. Front Comput Neurosci 6:83
Zilany MS, Bruce IC, Carney LH (2014) Updated parameters and expanded simulation options for a model of the auditory periphery. J Acoust Soc Am 135(1):283–286
Zilany MS, Bruce IC, Nelson PC, Carney LH (2009) A phenomenological model of the synapse between the inner hair cell and auditory nerve: long-term adaptation with power-law dynamics. J Acoust Soc Am 126(5):2390–2412
Funding
Open Access funding enabled and organized by Projekt DEAL. This work was supported by the German Research Association (DFG) under Grant Number LE2250/6-1.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare no conflicts of interest.
Code availability
The paper is based on custom code that is made available upon reasonable request to the corresponding author.
Additional information
Communicated by Benjamin Lindner.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Rebhan, M., Leibold, C. A phenomenological spiking model for octopus cells in the posterior–ventral cochlear nucleus. Biol Cybern 115, 331–341 (2021). https://doi.org/10.1007/s00422-021-00881-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00422-021-00881-x